Datasets:

---

ftopal

/

huggingface-datasets

like 0

[ "task_categories:text-generation", "task_ids:language-modeling", "annotations_creators:machine-generated", "language_creators:expert-generated", "multilinguality:monolingual", "size_categories:n>1M", "language:en", "license:cc-by-4.0", "region:us" ]

[]

[ "en" ]

[ "# Dataset Card for BioLang", "## Table of Contents\n- [Dataset Card for [Dataset Name]](#dataset-card-for-dataset-name)\n - Table of Contents\n - Dataset Description\n - Dataset Summary\n - Supported Tasks and Leaderboards\n - Languages\n - Dataset Structure\n - Data Instances\n - Data Fields\n - Data Splits\n - Dataset Creation\n - Curation Rationale\n - Source Data\n - Initial Data Collection and Normalization\n - Who are the source language producers?\n - Annotations\n - Annotation process\n - Who are the annotators?\n - Personal and Sensitive Information\n - Considerations for Using the Data\n - Social Impact of Dataset\n - Discussion of Biases\n - Other Known Limitations\n - Additional Information\n - Dataset Curators\n - Licensing Information\n - Citation Information\n - Contributions", "## Dataset Description\n\n- Homepage: URL\n- Repository: URL\n- Paper:\n- Leaderboard:\n- Point of Contact: thomas.lemberger@URL\n- Download Size: 5_299_878_661", "### Dataset Summary\n\nBioLang is a dataset is based on abstracts from the open access section of EuropePubMed Central to train language models in the domain of biology. The dataset can be used for random masked language modeling or for language modeling using only specific part-of-speech maksing. More details on generation and use of the dataset at URL .", "### Supported Tasks and Leaderboards\n\n- 'MLM': masked language modeling\n- 'DET': part-of-speach masked language model, with determinants ('DET') tagged\n- 'SMALL': part-of-speech masked language model, with \"small\" words ('DET', 'CCONJ', 'SCONJ', 'ADP', 'PRON') tagged\n- 'VERB': part-of-speach masked language model, with verbs ('VERB') tagged", "### Languages\n\nEnglish", "## Dataset Structure", "### Data Instances", "### Data Fields\n\n'MLM':\n- 'input_ids': a 'list' of 'int32' features.\n- 'special_tokens_mask': a 'list' of 'int8' features.\n\n'DET', 'VERB', 'SMALL':\n- 'input_ids': a 'list' of 'int32' features.\n- 'tag_mask': a 'list' of 'int8' features.", "### Data Splits\n\n- 'train':\n - features: ['input_ids', 'special_tokens_mask'],\n - num_rows: 12_005_390\n- 'test':\n - features: ['input_ids', 'special_tokens_mask'],\n - num_rows: 37_112\n- 'validation':\n - features: ['input_ids', 'special_tokens_mask'],\n - num_rows: 36_713", "## Dataset Creation", "### Curation Rationale\n\nThe dataset was assembled to train language models in the field of cell and molecular biology. To expand the size of the dataset and to include many examples with highly technical language, abstracts were complemented with figure legends (or figure 'captions').", "### Source Data", "#### Initial Data Collection and Normalization\n\nThe xml content of papers were downloaded in January 2021 from the open access section of EuropePMC. Figure legends and abstracts were extracted from the JATS XML, tokenized with the 'roberta-base' tokenizer and part-of-speech tagged with Spacy's 'en_core_web_sm' model (URL).\n\nMore details at URL", "#### Who are the source language producers?\n\nExperts scientists.", "### Annotations", "#### Annotation process\n\nPart-of-speech was tagged automatically.", "#### Who are the annotators?\n\nSpacy's 'en_core_web_sm' model (URL) was used for part-of-speech tagging.", "### Personal and Sensitive Information", "## Considerations for Using the Data", "### Social Impact of Dataset", "### Discussion of Biases", "### Other Known Limitations", "## Additional Information", "### Dataset Curators\n\nThomas Lemberger", "### Licensing Information\n\nCC-BY 4.0", "### Contributions\n\nThanks to @tlemberger for adding this dataset." ]

[ "TAGS\n#task_categories-text-generation #task_ids-language-modeling #annotations_creators-machine-generated #language_creators-expert-generated #multilinguality-monolingual #size_categories-n>1M #language-English #license-cc-by-4.0 #region-us \n", "# Dataset Card for BioLang", "## Table of Contents\n- [Dataset Card for [Dataset Name]](#dataset-card-for-dataset-name)\n - Table of Contents\n - Dataset Description\n - Dataset Summary\n - Supported Tasks and Leaderboards\n - Languages\n - Dataset Structure\n - Data Instances\n - Data Fields\n - Data Splits\n - Dataset Creation\n - Curation Rationale\n - Source Data\n - Initial Data Collection and Normalization\n - Who are the source language producers?\n - Annotations\n - Annotation process\n - Who are the annotators?\n - Personal and Sensitive Information\n - Considerations for Using the Data\n - Social Impact of Dataset\n - Discussion of Biases\n - Other Known Limitations\n - Additional Information\n - Dataset Curators\n - Licensing Information\n - Citation Information\n - Contributions", "## Dataset Description\n\n- Homepage: URL\n- Repository: URL\n- Paper:\n- Leaderboard:\n- Point of Contact: thomas.lemberger@URL\n- Download Size: 5_299_878_661", "### Dataset Summary\n\nBioLang is a dataset is based on abstracts from the open access section of EuropePubMed Central to train language models in the domain of biology. The dataset can be used for random masked language modeling or for language modeling using only specific part-of-speech maksing. More details on generation and use of the dataset at URL .", "### Supported Tasks and Leaderboards\n\n- 'MLM': masked language modeling\n- 'DET': part-of-speach masked language model, with determinants ('DET') tagged\n- 'SMALL': part-of-speech masked language model, with \"small\" words ('DET', 'CCONJ', 'SCONJ', 'ADP', 'PRON') tagged\n- 'VERB': part-of-speach masked language model, with verbs ('VERB') tagged", "### Languages\n\nEnglish", "## Dataset Structure", "### Data Instances", "### Data Fields\n\n'MLM':\n- 'input_ids': a 'list' of 'int32' features.\n- 'special_tokens_mask': a 'list' of 'int8' features.\n\n'DET', 'VERB', 'SMALL':\n- 'input_ids': a 'list' of 'int32' features.\n- 'tag_mask': a 'list' of 'int8' features.", "### Data Splits\n\n- 'train':\n - features: ['input_ids', 'special_tokens_mask'],\n - num_rows: 12_005_390\n- 'test':\n - features: ['input_ids', 'special_tokens_mask'],\n - num_rows: 37_112\n- 'validation':\n - features: ['input_ids', 'special_tokens_mask'],\n - num_rows: 36_713", "## Dataset Creation", "### Curation Rationale\n\nThe dataset was assembled to train language models in the field of cell and molecular biology. To expand the size of the dataset and to include many examples with highly technical language, abstracts were complemented with figure legends (or figure 'captions').", "### Source Data", "#### Initial Data Collection and Normalization\n\nThe xml content of papers were downloaded in January 2021 from the open access section of EuropePMC. Figure legends and abstracts were extracted from the JATS XML, tokenized with the 'roberta-base' tokenizer and part-of-speech tagged with Spacy's 'en_core_web_sm' model (URL).\n\nMore details at URL", "#### Who are the source language producers?\n\nExperts scientists.", "### Annotations", "#### Annotation process\n\nPart-of-speech was tagged automatically.", "#### Who are the annotators?\n\nSpacy's 'en_core_web_sm' model (URL) was used for part-of-speech tagging.", "### Personal and Sensitive Information", "## Considerations for Using the Data", "### Social Impact of Dataset", "### Discussion of Biases", "### Other Known Limitations", "## Additional Information", "### Dataset Curators\n\nThomas Lemberger", "### Licensing Information\n\nCC-BY 4.0", "### Contributions\n\nThanks to @tlemberger for adding this dataset." ]

[ "task_categories:text-classification", "task_ids:multi-class-classification", "task_ids:named-entity-recognition", "task_ids:parsing", "annotations_creators:expert-generated", "language_creators:expert-generated", "multilinguality:monolingual", "size_categories:10K<n<100K", "language:en", "license:cc-by-4.0", "region:us" ]

[]

[ "en" ]

[ "# Dataset Card for sd-nlp", "## Table of Contents\n- [Dataset Card for [Dataset Name]](#dataset-card-for-dataset-name)\n - Table of Contents\n - Dataset Description\n - Dataset Summary\n - Supported Tasks and Leaderboards\n - Languages\n - Dataset Structure\n - Data Instances\n - Data Fields\n - Data Splits\n - Dataset Creation\n - Curation Rationale\n - Source Data\n - Initial Data Collection and Normalization\n - Who are the source language producers?\n - Annotations\n - Annotation process\n - Who are the annotators?\n - Personal and Sensitive Information\n - Considerations for Using the Data\n - Social Impact of Dataset\n - Discussion of Biases\n - Other Known Limitations\n - Additional Information\n - Dataset Curators\n - Licensing Information\n - Citation Information\n - Contributions", "## Dataset Description\n\n- Homepage: URL\n- Repository: URL\n- Paper:\n- Leaderboard:\n- Point of Contact: thomas.lemberger@URL", "### Dataset Summary\n\nThis dataset is based on the content of the SourceData (URL) database, which contains manually annotated figure legends written in English and extracted from scientific papers in the domain of cell and molecular biology (Liechti et al, Nature Methods, 2017, URL \n\nThe dataset is pre-tokenized with the 'roberta-base' tokenizer. \n\nAdditional details at URL", "### Supported Tasks and Leaderboards\n\nTags are provided as IOB2-style tags).\n\n'PANELIZATION': figure captions (or figure legends) are usually composed of segments that each refer to one of several 'panels' of the full figure. Panels tend to represent results obtained with a coherent method and depicts data points that can be meaningfully compared to each other. 'PANELIZATION' provide the start (B-PANEL_START) of these segments and allow to train for recogntion of the boundary between consecutive panel lengends.\n\n'NER': biological and chemical entities are labeled. Specifically the following entities are tagged:\n- 'SMALL_MOLECULE': small molecules\n- 'GENEPROD': gene products (genes and proteins)\n- 'SUBCELLULAR': subcellular components\n- 'CELL': cell types and cell lines.\n- 'TISSUE': tissues and organs\n- 'ORGANISM': species\n- 'EXP_ASSAY': experimental assays\n\n'ROLES': the role of entities with regard to the causal hypotheses tested in the reported results. The tags are:\n- 'CONTROLLED_VAR': entities that are associated with experimental variables and that subjected to controlled and targeted perturbations.\n- 'MEASURED_VAR': entities that are associated with the variables measured and the object of the measurements.\n\n'BORING': entities are marked with the tag 'BORING' when they are more of descriptive value and not directly associated with causal hypotheses ('boring' is not an ideal choice of word, but it is short...). Typically, these entities are so-called 'reporter' geneproducts, entities used as common baseline across samples, or specify the context of the experiment (cellular system, species, etc...).", "### Languages\n\nThe text in the dataset is English.", "## Dataset Structure", "### Data Instances", "### Data Fields\n\n- 'input_ids': token id in 'roberta-base' tokenizers' vocabulary provided as a'list' of 'int' \n- 'label_ids':\n - 'entity_types': 'list' of 'strings' for the IOB2 tags for entity type; possible value in '[\"O\", \"I-SMALL_MOLECULE\", \"B-SMALL_MOLECULE\", \"I-GENEPROD\", \"B-GENEPROD\", \"I-SUBCELLULAR\", \"B-SUBCELLULAR\", \"I-CELL\", \"B-CELL\", \"I-TISSUE\", \"B-TISSUE\", \"I-ORGANISM\", \"B-ORGANISM\", \"I-EXP_ASSAY\", \"B-EXP_ASSAY\"]'\n - 'geneprod_roles': 'list' of 'strings' for the IOB2 tags for experimental roles; values in '[\"O\", \"I-CONTROLLED_VAR\", \"B-CONTROLLED_VAR\", \"I-MEASURED_VAR\", \"B-MEASURED_VAR\"]'\n - 'boring': 'list' of 'strings' for IOB2 tags for entities unrelated to causal design; values in '[\"O\", \"I-BORING\", \"B-BORING\"]'\n - 'panel_start': 'list' of 'strings' for IOB2 tags '[\"O\", \"B-PANEL_START\"]'", "### Data Splits\n\n- train:\n - features: ['input_ids', 'labels', 'tag_mask'],\n - num_rows: 48_771\n- test:\n - features: ['input_ids', 'labels', 'tag_mask'],\n - num_rows: 13_801\n- validation:\n - features: ['input_ids', 'labels', 'tag_mask'],\n - num_rows: 7_178", "## Dataset Creation", "### Curation Rationale\n\nThe dataset was built to train models for the automatic extraction of a knowledge graph based from the scientific literature. The dataset can be used to train models for text segmentation, named entity recognition and semantic role labeling.", "### Source Data", "#### Initial Data Collection and Normalization\n\nFigure legends were annotated according to the SourceData framework described in Liechti et al 2017 (Nature Methods, 2017, URL The curation tool at URL was used to segment figure legends into panel legends, tag enities, assign experiemental roles and normalize with standard identifiers (not available in this dataset). The source data was downloaded from the SourceData API (URL) on 21 Jan 2021.", "#### Who are the source language producers?\n\nThe examples are extracted from the figure legends from scientific papers in cell and molecular biology.", "### Annotations", "#### Annotation process\n\nThe annotations were produced manually with expert curators from the SourceData project (URL)", "#### Who are the annotators?\n\nCurators of the SourceData project.", "### Personal and Sensitive Information\n\nNone known.", "## Considerations for Using the Data", "### Social Impact of Dataset\n\nNot applicable.", "### Discussion of Biases\n\nThe examples are heavily biased towards cell and molecular biology and are enriched in examples from papers published in EMBO Press journals (URL)", "### Other Known Limitations", "## Additional Information", "### Dataset Curators\n\nThomas Lemberger, EMBO.", "### Licensing Information\n\nCC BY 4.0", "### Contributions\n\nThanks to @tlemberger for adding this dataset." ]

[ "TAGS\n#task_categories-text-classification #task_ids-multi-class-classification #task_ids-named-entity-recognition #task_ids-parsing #annotations_creators-expert-generated #language_creators-expert-generated #multilinguality-monolingual #size_categories-10K<n<100K #language-English #license-cc-by-4.0 #region-us \n", "# Dataset Card for sd-nlp", "## Table of Contents\n- [Dataset Card for [Dataset Name]](#dataset-card-for-dataset-name)\n - Table of Contents\n - Dataset Description\n - Dataset Summary\n - Supported Tasks and Leaderboards\n - Languages\n - Dataset Structure\n - Data Instances\n - Data Fields\n - Data Splits\n - Dataset Creation\n - Curation Rationale\n - Source Data\n - Initial Data Collection and Normalization\n - Who are the source language producers?\n - Annotations\n - Annotation process\n - Who are the annotators?\n - Personal and Sensitive Information\n - Considerations for Using the Data\n - Social Impact of Dataset\n - Discussion of Biases\n - Other Known Limitations\n - Additional Information\n - Dataset Curators\n - Licensing Information\n - Citation Information\n - Contributions", "## Dataset Description\n\n- Homepage: URL\n- Repository: URL\n- Paper:\n- Leaderboard:\n- Point of Contact: thomas.lemberger@URL", "### Dataset Summary\n\nThis dataset is based on the content of the SourceData (URL) database, which contains manually annotated figure legends written in English and extracted from scientific papers in the domain of cell and molecular biology (Liechti et al, Nature Methods, 2017, URL \n\nThe dataset is pre-tokenized with the 'roberta-base' tokenizer. \n\nAdditional details at URL", "### Supported Tasks and Leaderboards\n\nTags are provided as IOB2-style tags).\n\n'PANELIZATION': figure captions (or figure legends) are usually composed of segments that each refer to one of several 'panels' of the full figure. Panels tend to represent results obtained with a coherent method and depicts data points that can be meaningfully compared to each other. 'PANELIZATION' provide the start (B-PANEL_START) of these segments and allow to train for recogntion of the boundary between consecutive panel lengends.\n\n'NER': biological and chemical entities are labeled. Specifically the following entities are tagged:\n- 'SMALL_MOLECULE': small molecules\n- 'GENEPROD': gene products (genes and proteins)\n- 'SUBCELLULAR': subcellular components\n- 'CELL': cell types and cell lines.\n- 'TISSUE': tissues and organs\n- 'ORGANISM': species\n- 'EXP_ASSAY': experimental assays\n\n'ROLES': the role of entities with regard to the causal hypotheses tested in the reported results. The tags are:\n- 'CONTROLLED_VAR': entities that are associated with experimental variables and that subjected to controlled and targeted perturbations.\n- 'MEASURED_VAR': entities that are associated with the variables measured and the object of the measurements.\n\n'BORING': entities are marked with the tag 'BORING' when they are more of descriptive value and not directly associated with causal hypotheses ('boring' is not an ideal choice of word, but it is short...). Typically, these entities are so-called 'reporter' geneproducts, entities used as common baseline across samples, or specify the context of the experiment (cellular system, species, etc...).", "### Languages\n\nThe text in the dataset is English.", "## Dataset Structure", "### Data Instances", "### Data Fields\n\n- 'input_ids': token id in 'roberta-base' tokenizers' vocabulary provided as a'list' of 'int' \n- 'label_ids':\n - 'entity_types': 'list' of 'strings' for the IOB2 tags for entity type; possible value in '[\"O\", \"I-SMALL_MOLECULE\", \"B-SMALL_MOLECULE\", \"I-GENEPROD\", \"B-GENEPROD\", \"I-SUBCELLULAR\", \"B-SUBCELLULAR\", \"I-CELL\", \"B-CELL\", \"I-TISSUE\", \"B-TISSUE\", \"I-ORGANISM\", \"B-ORGANISM\", \"I-EXP_ASSAY\", \"B-EXP_ASSAY\"]'\n - 'geneprod_roles': 'list' of 'strings' for the IOB2 tags for experimental roles; values in '[\"O\", \"I-CONTROLLED_VAR\", \"B-CONTROLLED_VAR\", \"I-MEASURED_VAR\", \"B-MEASURED_VAR\"]'\n - 'boring': 'list' of 'strings' for IOB2 tags for entities unrelated to causal design; values in '[\"O\", \"I-BORING\", \"B-BORING\"]'\n - 'panel_start': 'list' of 'strings' for IOB2 tags '[\"O\", \"B-PANEL_START\"]'", "### Data Splits\n\n- train:\n - features: ['input_ids', 'labels', 'tag_mask'],\n - num_rows: 48_771\n- test:\n - features: ['input_ids', 'labels', 'tag_mask'],\n - num_rows: 13_801\n- validation:\n - features: ['input_ids', 'labels', 'tag_mask'],\n - num_rows: 7_178", "## Dataset Creation", "### Curation Rationale\n\nThe dataset was built to train models for the automatic extraction of a knowledge graph based from the scientific literature. The dataset can be used to train models for text segmentation, named entity recognition and semantic role labeling.", "### Source Data", "#### Initial Data Collection and Normalization\n\nFigure legends were annotated according to the SourceData framework described in Liechti et al 2017 (Nature Methods, 2017, URL The curation tool at URL was used to segment figure legends into panel legends, tag enities, assign experiemental roles and normalize with standard identifiers (not available in this dataset). The source data was downloaded from the SourceData API (URL) on 21 Jan 2021.", "#### Who are the source language producers?\n\nThe examples are extracted from the figure legends from scientific papers in cell and molecular biology.", "### Annotations", "#### Annotation process\n\nThe annotations were produced manually with expert curators from the SourceData project (URL)", "#### Who are the annotators?\n\nCurators of the SourceData project.", "### Personal and Sensitive Information\n\nNone known.", "## Considerations for Using the Data", "### Social Impact of Dataset\n\nNot applicable.", "### Discussion of Biases\n\nThe examples are heavily biased towards cell and molecular biology and are enriched in examples from papers published in EMBO Press journals (URL)", "### Other Known Limitations", "## Additional Information", "### Dataset Curators\n\nThomas Lemberger, EMBO.", "### Licensing Information\n\nCC BY 4.0", "### Contributions\n\nThanks to @tlemberger for adding this dataset." ]

[ "language:pt", "region:us" ]

[]

[ "pt" ]

[ "### Languages\n\n\nThe BCP-47 code for the dataset's language is pt.\n\n\nDataset Structure\n-----------------", "### Data Instances\n\n\nA sample from this dataset looks as follows:", "### Dataset Fields\n\n\nThe dataset has the following fields (also called \"features\"):", "### Dataset Splits\n\n\nThis dataset is split into a train and validation split. The split sizes are as follow:" ]

[ "TAGS\n#language-Portuguese #region-us \n", "### Languages\n\n\nThe BCP-47 code for the dataset's language is pt.\n\n\nDataset Structure\n-----------------", "### Data Instances\n\n\nA sample from this dataset looks as follows:", "### Dataset Fields\n\n\nThe dataset has the following fields (also called \"features\"):", "### Dataset Splits\n\n\nThis dataset is split into a train and validation split. The split sizes are as follow:" ]

[ "annotations_creators:found", "language_creators:found", "multilinguality:monolingual", "size_categories:10M<n<100M", "source_datasets:original", "language:en", "license:mit", "named entities", "similarity", "paraphrasing", "synonyms", "wikipedia", "arxiv:2004.04906", "arxiv:2202.13581", "region:us" ]

[ "2004.04906", "2202.13581" ]

[ "en" ]

[ "### Training Data Generation\n\n\nThe training pairs in this dataset are generated by taking each example from the corpus as a positive example, and creating a new negative example from the article title of the positive example and a random link text from a different article.\n\n\nThe articles featured in each split are disjoint from the other splits, and each split has the same number of positive (semantically the same) and negative (semantically different) examples.\n\n\nFor more details on the dataset motivation, see the paper. If you use this dataset in your work, please cite it using the ArXiv reference.\n\n\nGeneration scripts can be found in the GitHub repo." ]

[ "TAGS\n#annotations_creators-found #language_creators-found #multilinguality-monolingual #size_categories-10M<n<100M #source_datasets-original #language-English #license-mit #named entities #similarity #paraphrasing #synonyms #wikipedia #arxiv-2004.04906 #arxiv-2202.13581 #region-us \n", "### Training Data Generation\n\n\nThe training pairs in this dataset are generated by taking each example from the corpus as a positive example, and creating a new negative example from the article title of the positive example and a random link text from a different article.\n\n\nThe articles featured in each split are disjoint from the other splits, and each split has the same number of positive (semantically the same) and negative (semantically different) examples.\n\n\nFor more details on the dataset motivation, see the paper. If you use this dataset in your work, please cite it using the ArXiv reference.\n\n\nGeneration scripts can be found in the GitHub repo." ]

[ "region:us" ]

[]

[]

[ "### Dataset Summary\n\n\nFIG-Loneliness is a dataset for fine-grained loneliness characterization and model training. This dataset consists of 2633 lonely and 3000 non-lonely Reddit posts annotated by trained human annotators. For the lonely posts, we provide fine-grained category labels for the forms of loneliness including duration, context and interpersonal relationships, and for the coping strategies of the authors including reaching out, seeking advice, seeking validation and non-directed interaction.", "### Supported Tasks and Leaderboards", "### Languages\n\n\nThe language supported is English.\n\n\nDataset Structure\n-----------------", "### Loading\n\n\nTo load the dataset, first clone this dataset repo:\n\n\nThen we can load datasets using Huggingface Datasets API:", "### Data Instances\n\n\nThe 'train\\_set' split contains 3,943 instances. The 'dev\\_set' split contains 1,126. The 'test\\_set' split contains 564 instances.", "### Data Fields\n\n\nEach instance contains 8 fields: 'idx', 'unique\\_id', 'text', 'lonely', 'temporal', 'interaction', 'context\\_pri', and 'interpersonal\\_pri'.", "### Data Splits\n\n\nThe entire dataset is split into 3,943 training instances, 1,126 dev instances, and 564 test instances.\n\n\nDataset Creation\n----------------", "### Curation Rationale\n\n\nThe data curation rationale is to capture loneliness expressions not only from a wider user base but also from users who specifically belong to the young adult age group (a vulnerable group for loneliness). We sampled data from Reddit and subsequently annotated the data with loneliness labels.", "### Source Data", "#### Initial Data Collection and Normalization\n\n\nBy using Reddit’s Pushshift API, we collected all posts from two loneliness specific subreddits (*r/loneliness*, *r/lonely*) and two subreddits for young adults (*r/youngadults*, *r/college*) from 2018 to 2020.", "#### Who are the source language producers?", "### Annotations", "#### Who are the annotators?\n\n\nAnnotation labels were provided by trained undergraduate research assistants and Amazon’s Mechanical Turk workers (MTurkers) with a Master certification.", "#### Annotation process\n\n\nFor the potential lonely samples:\nWe had research assistants labeled the sampled potential lonely posts. Each post was labeled by three of research assistants. A posts was first labeled on whether it contains an expression on self-disclosure of loneliness. If the majority of the annotators labeled a post as not containing such expression, the post would be discarded, otherwise it is further labeled according to a codebook that contains the following categories: (1) *duration*: the duration of the loneliness experience (transient, enduring, and ambiguous), (2) *context*: the contexts of the experience (social, physical, somatic, and romantic), (3) *interpresonal*: the interpersonal relationships involved in the experience (romantic, family, friendship, and peers), and (4) *interaction*: user interaction styles (seeking advice, providing support, seeking validation/affirmation, reaching out and non-directed interaction). The codebook is intended for dissecting different forms of loneliness and users’ coping strategies in the loneliness discourse. We also included a ‘not applicable’ (NA) label to accommodate situations that are not suitable for classification. For each category, the annotators gave *one value* which they think would best capture the source of loneliness in the post or the poster’s interaction intent.\n\n\nFor the potential non-lonely samples:\nMTurkers were instructed to classify whether the Reddit posters express loneliness. Each post was annotated by three MTurkers, and only posts labeled as non-lonely by the majority would remain in the final annotated dataset.\n\n\nAll the labeled posts and annotations were included in FIG-Loneliness, which consists of roughly 3000 lonely and 3000 non-lonely posts.", "#### Dataset Codebook\n\n\nSee code rule and example posts for the category labels here.", "### Personal and Sensitive Information\n\n\nConsiderations for Using the Data\n---------------------------------", "### Social Impact of Dataset", "### Discussion of Biases", "### Other Known Limitations\n\n\nSee Limitation and Data Disclaimer here\n\n\nAdditional Information\n----------------------", "### Dataset Curator", "### Licensing Information\n\n\nCC BY-NC 4.0\n\n\nJiang, Y., Jiang, Y., Leqi, L., & Winkielman, P. (2022). Many Ways to Be Lonely: Fine-Grained Characterization of Loneliness and Its Potential Changes in COVID-19. Proceedings of the International AAAI Conference on Web and Social Media, 16(1), 405-416. Retrieved from URL" ]

[ "TAGS\n#region-us \n", "### Dataset Summary\n\n\nFIG-Loneliness is a dataset for fine-grained loneliness characterization and model training. This dataset consists of 2633 lonely and 3000 non-lonely Reddit posts annotated by trained human annotators. For the lonely posts, we provide fine-grained category labels for the forms of loneliness including duration, context and interpersonal relationships, and for the coping strategies of the authors including reaching out, seeking advice, seeking validation and non-directed interaction.", "### Supported Tasks and Leaderboards", "### Languages\n\n\nThe language supported is English.\n\n\nDataset Structure\n-----------------", "### Loading\n\n\nTo load the dataset, first clone this dataset repo:\n\n\nThen we can load datasets using Huggingface Datasets API:", "### Data Instances\n\n\nThe 'train\\_set' split contains 3,943 instances. The 'dev\\_set' split contains 1,126. The 'test\\_set' split contains 564 instances.", "### Data Fields\n\n\nEach instance contains 8 fields: 'idx', 'unique\\_id', 'text', 'lonely', 'temporal', 'interaction', 'context\\_pri', and 'interpersonal\\_pri'.", "### Data Splits\n\n\nThe entire dataset is split into 3,943 training instances, 1,126 dev instances, and 564 test instances.\n\n\nDataset Creation\n----------------", "### Curation Rationale\n\n\nThe data curation rationale is to capture loneliness expressions not only from a wider user base but also from users who specifically belong to the young adult age group (a vulnerable group for loneliness). We sampled data from Reddit and subsequently annotated the data with loneliness labels.", "### Source Data", "#### Initial Data Collection and Normalization\n\n\nBy using Reddit’s Pushshift API, we collected all posts from two loneliness specific subreddits (*r/loneliness*, *r/lonely*) and two subreddits for young adults (*r/youngadults*, *r/college*) from 2018 to 2020.", "#### Who are the source language producers?", "### Annotations", "#### Who are the annotators?\n\n\nAnnotation labels were provided by trained undergraduate research assistants and Amazon’s Mechanical Turk workers (MTurkers) with a Master certification.", "#### Annotation process\n\n\nFor the potential lonely samples:\nWe had research assistants labeled the sampled potential lonely posts. Each post was labeled by three of research assistants. A posts was first labeled on whether it contains an expression on self-disclosure of loneliness. If the majority of the annotators labeled a post as not containing such expression, the post would be discarded, otherwise it is further labeled according to a codebook that contains the following categories: (1) *duration*: the duration of the loneliness experience (transient, enduring, and ambiguous), (2) *context*: the contexts of the experience (social, physical, somatic, and romantic), (3) *interpresonal*: the interpersonal relationships involved in the experience (romantic, family, friendship, and peers), and (4) *interaction*: user interaction styles (seeking advice, providing support, seeking validation/affirmation, reaching out and non-directed interaction). The codebook is intended for dissecting different forms of loneliness and users’ coping strategies in the loneliness discourse. We also included a ‘not applicable’ (NA) label to accommodate situations that are not suitable for classification. For each category, the annotators gave *one value* which they think would best capture the source of loneliness in the post or the poster’s interaction intent.\n\n\nFor the potential non-lonely samples:\nMTurkers were instructed to classify whether the Reddit posters express loneliness. Each post was annotated by three MTurkers, and only posts labeled as non-lonely by the majority would remain in the final annotated dataset.\n\n\nAll the labeled posts and annotations were included in FIG-Loneliness, which consists of roughly 3000 lonely and 3000 non-lonely posts.", "#### Dataset Codebook\n\n\nSee code rule and example posts for the category labels here.", "### Personal and Sensitive Information\n\n\nConsiderations for Using the Data\n---------------------------------", "### Social Impact of Dataset", "### Discussion of Biases", "### Other Known Limitations\n\n\nSee Limitation and Data Disclaimer here\n\n\nAdditional Information\n----------------------", "### Dataset Curator", "### Licensing Information\n\n\nCC BY-NC 4.0\n\n\nJiang, Y., Jiang, Y., Leqi, L., & Winkielman, P. (2022). Many Ways to Be Lonely: Fine-Grained Characterization of Loneliness and Its Potential Changes in COVID-19. Proceedings of the International AAAI Conference on Web and Social Media, 16(1), 405-416. Retrieved from URL" ]

[ "multilinguality:monolingual", "size_categories:1K<n<10K", "language:en", "license:cc-by-4.0", "region:us" ]

[]

[ "en" ]

[ "# Dataset Card for quoteli3", "## Dataset Description\n\n- Homepage: URL\n- Repository: URL\n- Paper: Muzny, Grace, et al. \"A two-stage sieve approach for quote attribution.\" Proceedings of the 15th Conference of the European Chapter of the Association for Computational Linguistics: Volume 1, Long Papers. 2017.", "### Dataset Summary \n\nThis dataset is based on the quoteli3 dataset by Muzny et al. (2017). It contains annotated quotes for three pieces of literature: Chekhov\\\\\\\\\\\\\\\\\\\\\\\\'s The Steppe, Austen\\\\\\\\\\\\\\\\\\\\\\\\'s Emma and Pride and Prejudice.", "### Languages\n\nThe text in the dataset is English.", "## Dataset Structure\n\nTraining data: \n-Quotes (1575, 11)\n-Characters (32, 6)\n\nTest data:\n-Quotes (1513, 11)\n-Characters (145, 6)", "### Data Splits\n-Quotes:\n - train:\n - features: ['mention', 'oid', 'speaker', 'connection', 'id', 'answer', 'answer_mention {'answer', 'answer_start', 'answer_end', 'answer_in_context'}, 'question', 'context', 'large_context', 'book_title'],\n - num_rows: 1575\n - test:\n - features: ['mention', 'oid', 'speaker', 'connection', 'id', 'answer', 'answer_mention {'answer', 'answer_start', 'answer_end', 'answer_in_context'}, 'question', 'context', 'large_context', 'book_title'],\n - num_rows: 1513\n-Characters:\n - train:\n - features: ['aliases', 'description', 'gender', 'name', 'id', 'book_title'],\n - num_rows: 32\n - test:\n - features: ['aliases', 'description', 'gender', 'name', 'id', 'book_title'],\n - num_rows: 146" ]

[ "TAGS\n#multilinguality-monolingual #size_categories-1K<n<10K #language-English #license-cc-by-4.0 #region-us \n", "# Dataset Card for quoteli3", "## Dataset Description\n\n- Homepage: URL\n- Repository: URL\n- Paper: Muzny, Grace, et al. \"A two-stage sieve approach for quote attribution.\" Proceedings of the 15th Conference of the European Chapter of the Association for Computational Linguistics: Volume 1, Long Papers. 2017.", "### Dataset Summary \n\nThis dataset is based on the quoteli3 dataset by Muzny et al. (2017). It contains annotated quotes for three pieces of literature: Chekhov\\\\\\\\\\\\\\\\\\\\\\\\'s The Steppe, Austen\\\\\\\\\\\\\\\\\\\\\\\\'s Emma and Pride and Prejudice.", "### Languages\n\nThe text in the dataset is English.", "## Dataset Structure\n\nTraining data: \n-Quotes (1575, 11)\n-Characters (32, 6)\n\nTest data:\n-Quotes (1513, 11)\n-Characters (145, 6)", "### Data Splits\n-Quotes:\n - train:\n - features: ['mention', 'oid', 'speaker', 'connection', 'id', 'answer', 'answer_mention {'answer', 'answer_start', 'answer_end', 'answer_in_context'}, 'question', 'context', 'large_context', 'book_title'],\n - num_rows: 1575\n - test:\n - features: ['mention', 'oid', 'speaker', 'connection', 'id', 'answer', 'answer_mention {'answer', 'answer_start', 'answer_end', 'answer_in_context'}, 'question', 'context', 'large_context', 'book_title'],\n - num_rows: 1513\n-Characters:\n - train:\n - features: ['aliases', 'description', 'gender', 'name', 'id', 'book_title'],\n - num_rows: 32\n - test:\n - features: ['aliases', 'description', 'gender', 'name', 'id', 'book_title'],\n - num_rows: 146" ]

[ "task_categories:text-generation", "task_categories:fill-mask", "task_ids:language-modeling", "task_ids:masked-language-modeling", "multilinguality:monolingual", "size_categories:unknown", "source_datasets:extended|mc4", "language:fi", "region:us" ]

[]

[ "fi" ]

[ "# Dataset Card for mC4 Finnish Cleaned", "## Table of Contents\n- Dataset Description\n - Dataset Summary\n - Supported Tasks\n - Languages\n- Dataset Structure\n - Data Instances\n - Data Fields\n - Data Splits\n- Dataset Creation\n - Curation Rationale\n - Source Data\n - Annotations\n - Personal and Sensitive Information\n- Considerations for Using the Data\n - Social Impact of Dataset\n - Discussion of Biases\n - Other Known Limitations\n- Additional Information\n - Dataset Curators\n - Licensing Information\n - Citation Information", "## Dataset Description\n\n- Homepage: \n- Repository: \n- Paper: \n- Leaderboard: \n- Point of Contact:", "### Dataset Summary\n\nmC4 Finnish cleaned is cleaned version of the original mC4 Finnish split.", "### Supported Tasks and Leaderboards\n\nmC4 Finnish is mainly intended to pretrain Finnish language models and word representations.", "### Languages\n\nFinnish", "## Dataset Structure", "### Data Instances", "### Data Fields\n\nThe data have several fields:\n\n- url: url of the source as a string\n- text: text content as a string\n- timestamp: timestamp as a string\n- perplexity_kenlm_full: perplexity of the text calculated by KenLM model", "### Data Splits\n\nTrain Validation", "## Dataset Creation", "### Curation Rationale", "### Source Data", "#### Initial Data Collection and Normalization", "#### Who are the source language producers?", "### Annotations", "#### Annotation process", "#### Who are the annotators?", "### Personal and Sensitive Information", "## Considerations for Using the Data", "### Social Impact of Dataset", "### Discussion of Biases", "### Other Known Limitations", "## Additional Information", "### Dataset Curators", "### Licensing Information" ]

[ "TAGS\n#task_categories-text-generation #task_categories-fill-mask #task_ids-language-modeling #task_ids-masked-language-modeling #multilinguality-monolingual #size_categories-unknown #source_datasets-extended|mc4 #language-Finnish #region-us \n", "# Dataset Card for mC4 Finnish Cleaned", "## Table of Contents\n- Dataset Description\n - Dataset Summary\n - Supported Tasks\n - Languages\n- Dataset Structure\n - Data Instances\n - Data Fields\n - Data Splits\n- Dataset Creation\n - Curation Rationale\n - Source Data\n - Annotations\n - Personal and Sensitive Information\n- Considerations for Using the Data\n - Social Impact of Dataset\n - Discussion of Biases\n - Other Known Limitations\n- Additional Information\n - Dataset Curators\n - Licensing Information\n - Citation Information", "## Dataset Description\n\n- Homepage: \n- Repository: \n- Paper: \n- Leaderboard: \n- Point of Contact:", "### Dataset Summary\n\nmC4 Finnish cleaned is cleaned version of the original mC4 Finnish split.", "### Supported Tasks and Leaderboards\n\nmC4 Finnish is mainly intended to pretrain Finnish language models and word representations.", "### Languages\n\nFinnish", "## Dataset Structure", "### Data Instances", "### Data Fields\n\nThe data have several fields:\n\n- url: url of the source as a string\n- text: text content as a string\n- timestamp: timestamp as a string\n- perplexity_kenlm_full: perplexity of the text calculated by KenLM model", "### Data Splits\n\nTrain Validation", "## Dataset Creation", "### Curation Rationale", "### Source Data", "#### Initial Data Collection and Normalization", "#### Who are the source language producers?", "### Annotations", "#### Annotation process", "#### Who are the annotators?", "### Personal and Sensitive Information", "## Considerations for Using the Data", "### Social Impact of Dataset", "### Discussion of Biases", "### Other Known Limitations", "## Additional Information", "### Dataset Curators", "### Licensing Information" ]

[ "region:us" ]

[]

[]

[ "# Dataset Card for HumAID", "## Table of Contents\n- Dataset Description\n - Dataset Summary\n - Supported Tasks\n - Languages\n- Dataset Structure\n - Data Instances\n - Data Fields\n - Data Splits\n- Dataset Creation\n - Curation Rationale\n - Source Data\n - Annotations\n - Personal and Sensitive Information\n- Considerations for Using the Data\n - Social Impact of Dataset\n - Discussion of Biases\n - Other Known Limitations\n- Additional Information\n - Dataset Curators\n - Licensing Information\n - Citation Information", "## Dataset Description\n\n- Homepage: URL\n- Repository: URL\n- Paper: URL", "### Dataset Summary\n\nThe HumAID Twitter dataset consists of several thousands of manually annotated tweets that has been collected during 19 major natural disaster events including earthquakes, hurricanes, wildfires, and floods, which happened from 2016 to 2019 across different parts of the World. The annotations in the provided datasets consists of following humanitarian categories. The dataset consists only english tweets and it is the largest dataset for crisis informatics so far.\n Humanitarian categories \n- Caution and advice\n- Displaced people and evacuations\n- Dont know cant judge\n- Infrastructure and utility damage\n- Injured or dead people\n- Missing or found people\n- Not humanitarian\n- Other relevant information\n- Requests or urgent needs\n- Rescue volunteering or donation effort\n- Sympathy and support\n\nThe resulting annotated dataset consists of 11 labels.", "### Supported Tasks and Benchmark\nThe dataset can be used to train a model for multiclass tweet classification for disaster response. The benchmark results can be found in URL\n\nDataset is also released with event-wise and JSON objects for further research.\nFull set of the dataset can be found in URL", "### Languages\n\nEnglish", "## Dataset Structure", "### Data Instances", "### Data Fields\n\n* tweet_text: corresponds to the tweet text.\n* class_label: corresponds to a label assigned to a given tweet text", "### Data Splits\n\n* Train\n* Development\n* Test", "## Dataset Creation", "### Source Data", "#### Initial Data Collection and Normalization\n\nTweets has been collected during several disaster events.", "### Annotations", "#### Annotation process\n\nAMT has been used to annotate the dataset. Please check the paper for a more detail.", "#### Who are the annotators?\n\n- crowdsourced", "## Additional Information", "### Dataset Curators\n\nAuthors of the paper.", "### Licensing Information\n\n- cc-by-nc-4.0" ]

[ "TAGS\n#region-us \n", "# Dataset Card for HumAID", "## Table of Contents\n- Dataset Description\n - Dataset Summary\n - Supported Tasks\n - Languages\n- Dataset Structure\n - Data Instances\n - Data Fields\n - Data Splits\n- Dataset Creation\n - Curation Rationale\n - Source Data\n - Annotations\n - Personal and Sensitive Information\n- Considerations for Using the Data\n - Social Impact of Dataset\n - Discussion of Biases\n - Other Known Limitations\n- Additional Information\n - Dataset Curators\n - Licensing Information\n - Citation Information", "## Dataset Description\n\n- Homepage: URL\n- Repository: URL\n- Paper: URL", "### Dataset Summary\n\nThe HumAID Twitter dataset consists of several thousands of manually annotated tweets that has been collected during 19 major natural disaster events including earthquakes, hurricanes, wildfires, and floods, which happened from 2016 to 2019 across different parts of the World. The annotations in the provided datasets consists of following humanitarian categories. The dataset consists only english tweets and it is the largest dataset for crisis informatics so far.\n Humanitarian categories \n- Caution and advice\n- Displaced people and evacuations\n- Dont know cant judge\n- Infrastructure and utility damage\n- Injured or dead people\n- Missing or found people\n- Not humanitarian\n- Other relevant information\n- Requests or urgent needs\n- Rescue volunteering or donation effort\n- Sympathy and support\n\nThe resulting annotated dataset consists of 11 labels.", "### Supported Tasks and Benchmark\nThe dataset can be used to train a model for multiclass tweet classification for disaster response. The benchmark results can be found in URL\n\nDataset is also released with event-wise and JSON objects for further research.\nFull set of the dataset can be found in URL", "### Languages\n\nEnglish", "## Dataset Structure", "### Data Instances", "### Data Fields\n\n* tweet_text: corresponds to the tweet text.\n* class_label: corresponds to a label assigned to a given tweet text", "### Data Splits\n\n* Train\n* Development\n* Test", "## Dataset Creation", "### Source Data", "#### Initial Data Collection and Normalization\n\nTweets has been collected during several disaster events.", "### Annotations", "#### Annotation process\n\nAMT has been used to annotate the dataset. Please check the paper for a more detail.", "#### Who are the annotators?\n\n- crowdsourced", "## Additional Information", "### Dataset Curators\n\nAuthors of the paper.", "### Licensing Information\n\n- cc-by-nc-4.0" ]

[ "region:us" ]

[]

[]

[]

[ "TAGS\n#region-us \n" ]

[ "region:us" ]

[]

[]

[]

[ "TAGS\n#region-us \n" ]

[ "language:en", "license:mit", "program-synthesis", "region:us" ]

[]

[ "en" ]

[ "# Program Synthesis Data\n\nGenerated program synthesis datasets used to train dreamcoder.\n\nCurrently just supports text & list data. \n\n![](URL" ]

[ "TAGS\n#language-English #license-mit #program-synthesis #region-us \n", "# Program Synthesis Data\n\nGenerated program synthesis datasets used to train dreamcoder.\n\nCurrently just supports text & list data. \n\n![](URL" ]

[ "region:us" ]

[]

[]

[]

[ "TAGS\n#region-us \n" ]

[ "language:code", "region:us" ]

[]

[ "code" ]

[ "# Python State Changes\n\nState changes from the execution of single lines of Python code.\nAll code was taken from Python HackerRank solutions.\n\nScraped from my dataset of traced HackerRank solutions. URL", "## Get an overview of the dataset from seeing the frequency of different ASTs.\n URL" ]

[ "TAGS\n#language-code #region-us \n", "# Python State Changes\n\nState changes from the execution of single lines of Python code.\nAll code was taken from Python HackerRank solutions.\n\nScraped from my dataset of traced HackerRank solutions. URL", "## Get an overview of the dataset from seeing the frequency of different ASTs.\n URL" ]

[ "region:us" ]

[]

[]

[]

[ "TAGS\n#region-us \n" ]

[ "region:us" ]

[]

[]

[ "# Wiki Sentences\n\nA dataset of all english sentences in Wikipedia.\n\nTaken from the OPTIMUS project. URL\n\nThe dataset is 11.8GB so best to load it using streaming:" ]

[ "TAGS\n#region-us \n", "# Wiki Sentences\n\nA dataset of all english sentences in Wikipedia.\n\nTaken from the OPTIMUS project. URL\n\nThe dataset is 11.8GB so best to load it using streaming:" ]

[ "task_categories:other", "annotations_creators:automatically-created", "language_creators:unknown", "multilinguality:unknown", "size_categories:unknown", "source_datasets:original", "language:en", "license:apache-2.0", "reasoning", "arxiv:1908.05739", "arxiv:1906.05317", "region:us" ]

[ "1908.05739", "1906.05317" ]

[ "en" ]

[ "# Dataset Card for GEM/ART", "## Dataset Description\n\n- Homepage: URL\n- Repository: URL\n- Paper: URL\n- Leaderboard: N/A\n- Point of Contact: Chandra Bhagavatulla", "### Link to Main Data Card\n\nYou can find the main data card on the GEM Website.", "### Dataset Summary \n\nAbductive reasoning is inference to the most plausible explanation. For example, if Jenny finds her house in a mess when she returns from work, and remembers that she left a window open, she can hypothesize that a thief broke into her house and caused the mess, as the most plausible explanation.\nThis data loader focuses on abductive NLG: a conditional English generation task for explaining given observations in natural language. \n\nYou can load the dataset via:\n\nThe data loader can be found here.", "#### website\nWebsite", "#### paper\nOpenReview", "#### authors\nChandra Bhagavatula (AI2), Ronan Le Bras (AI2), Chaitanya Malaviya (AI2), Keisuke Sakaguchi (AI2), Ari Holtzman (AI2, UW), Hannah Rashkin (AI2, UW), Doug Downey (AI2), Wen-tau Yih (AI2), Yejin Choi (AI2, UW)", "## Dataset Overview", "### Where to find the Data and its Documentation", "#### Webpage\n\n\n\nWebsite", "#### Download\n\n\n\nGoogle Storage", "#### Paper\n\n\n\nOpenReview", "#### BibTex", "#### Contact Name\n\n\n\n\nChandra Bhagavatulla", "#### Contact Email\n\n\n\nchandrab@URL", "#### Has a Leaderboard?\n\n\n\nno", "### Languages and Intended Use", "#### Multilingual?\n\n\n\n\nno", "#### Covered Languages\n\n\n\n\n'English'", "#### Whose Language?\n\n\n\nCrowdworkers on the Amazon Mechanical Turk platform based in the U.S, Canada, U.K and Australia.", "#### License\n\n\n\n\napache-2.0: Apache License 2.0", "#### Intended Use\n\n\n\nTo study the viability of language-based abductive reasoning. Training and evaluating models to generate a plausible hypothesis to explain two given observations.", "#### Primary Task\n\n\n\nReasoning", "### Credit", "#### Curation Organization Type(s)\n\n\n\n'industry'", "#### Curation Organization(s)\n\n\n\nAllen Institute for AI", "#### Dataset Creators\n\n\n\nChandra Bhagavatula (AI2), Ronan Le Bras (AI2), Chaitanya Malaviya (AI2), Keisuke Sakaguchi (AI2), Ari Holtzman (AI2, UW), Hannah Rashkin (AI2, UW), Doug Downey (AI2), Wen-tau Yih (AI2), Yejin Choi (AI2, UW)", "#### Funding\n\n\n\nAllen Institute for AI", "#### Who added the Dataset to GEM?\n\n\n\nChandra Bhagavatula (AI2), Ronan LeBras (AI2), Aman Madaan (CMU), Nico Daheim (RWTH Aachen University)", "### Dataset Structure", "#### Data Fields\n\n\n\n- 'observation_1': A string describing an observation / event.\n- 'observation_2': A string describing an observation / event.\n- 'label': A string that plausibly explains why observation_1 and observation_2 might have happened.", "#### How were labels chosen?\n\n\n\nExplanations were authored by crowdworkers on the Amazon Mechanical Turk platform using a custom template designed by the creators of the dataset.", "#### Example Instance", "#### Data Splits\n\n\n\n- 'train': Consists of training instances. \n- 'dev': Consists of dev instances.\n- 'test': Consists of test instances.", "## Dataset in GEM", "### Rationale for Inclusion in GEM", "#### Why is the Dataset in GEM?\n\n\n\nAbductive reasoning is a crucial capability of humans and ART is the first dataset curated to study language-based abductive reasoning.", "#### Similar Datasets\n\n\n\nno", "#### Ability that the Dataset measures\n\n\n\nWhether models can reason abductively about a given pair of observations.", "### GEM-Specific Curation", "#### Modificatied for GEM?\n\n\n\nno", "#### Additional Splits?\n\n\n\nno", "### Getting Started with the Task", "#### Pointers to Resources\n\n\n\n- Paper\n- Code", "## Previous Results", "### Previous Results", "#### Measured Model Abilities\n\n\n\nWhether models can reason abductively about a given pair of observations.", "#### Metrics\n\n\n\n'BLEU', 'BERT-Score', 'ROUGE'", "#### Previous results available?\n\n\n\nno", "## Dataset Curation", "### Original Curation", "#### Sourced from Different Sources\n\n\n\nno", "### Language Data", "#### How was Language Data Obtained?\n\n\n\n'Crowdsourced'", "#### Where was it crowdsourced?\n\n\n\n'Amazon Mechanical Turk'", "#### Language Producers\n\n\n\nLanguage producers were English speakers in U.S., Canada, U.K and Australia.", "#### Topics Covered\n\n\n\nNo", "#### Data Validation\n\n\n\nvalidated by crowdworker", "#### Was Data Filtered?\n\n\n\nalgorithmically", "#### Filter Criteria\n\n\n\nAdversarial filtering algorithm as described in the paper", "### Structured Annotations", "#### Additional Annotations?\n\n\n\n\nautomatically created", "#### Annotation Service?\n\n\n\nno", "#### Annotation Values\n\n\n\nEach observation is associated with a list of COMET (URL inferences.", "#### Any Quality Control?\n\n\n\nnone", "### Consent", "#### Any Consent Policy?\n\n\n\nno", "### Private Identifying Information (PII)", "#### Contains PII?\n\n\n\n\nno PII", "#### Justification for no PII\n\n\n\nThe dataset contains day-to-day events. It does not contain names, emails, addresses etc.", "### Maintenance", "#### Any Maintenance Plan?\n\n\n\nno", "## Broader Social Context", "### Previous Work on the Social Impact of the Dataset", "#### Usage of Models based on the Data\n\n\n\nno", "### Impact on Under-Served Communities", "#### Addresses needs of underserved Communities?\n\n\n\nno", "### Discussion of Biases", "#### Any Documented Social Biases?\n\n\n\nno", "## Considerations for Using the Data", "### PII Risks and Liability", "#### Potential PII Risk\n\n\n\nNone", "### Licenses", "#### Copyright Restrictions on the Dataset\n\n\n\n'public domain'", "#### Copyright Restrictions on the Language Data\n\n\n\n'public domain'", "### Known Technical Limitations" ]

[ "TAGS\n#task_categories-other #annotations_creators-automatically-created #language_creators-unknown #multilinguality-unknown #size_categories-unknown #source_datasets-original #language-English #license-apache-2.0 #reasoning #arxiv-1908.05739 #arxiv-1906.05317 #region-us \n", "# Dataset Card for GEM/ART", "## Dataset Description\n\n- Homepage: URL\n- Repository: URL\n- Paper: URL\n- Leaderboard: N/A\n- Point of Contact: Chandra Bhagavatulla", "### Link to Main Data Card\n\nYou can find the main data card on the GEM Website.", "### Dataset Summary \n\nAbductive reasoning is inference to the most plausible explanation. For example, if Jenny finds her house in a mess when she returns from work, and remembers that she left a window open, she can hypothesize that a thief broke into her house and caused the mess, as the most plausible explanation.\nThis data loader focuses on abductive NLG: a conditional English generation task for explaining given observations in natural language. \n\nYou can load the dataset via:\n\nThe data loader can be found here.", "#### website\nWebsite", "#### paper\nOpenReview", "#### authors\nChandra Bhagavatula (AI2), Ronan Le Bras (AI2), Chaitanya Malaviya (AI2), Keisuke Sakaguchi (AI2), Ari Holtzman (AI2, UW), Hannah Rashkin (AI2, UW), Doug Downey (AI2), Wen-tau Yih (AI2), Yejin Choi (AI2, UW)", "## Dataset Overview", "### Where to find the Data and its Documentation", "#### Webpage\n\n\n\nWebsite", "#### Download\n\n\n\nGoogle Storage", "#### Paper\n\n\n\nOpenReview", "#### BibTex", "#### Contact Name\n\n\n\n\nChandra Bhagavatulla", "#### Contact Email\n\n\n\nchandrab@URL", "#### Has a Leaderboard?\n\n\n\nno", "### Languages and Intended Use", "#### Multilingual?\n\n\n\n\nno", "#### Covered Languages\n\n\n\n\n'English'", "#### Whose Language?\n\n\n\nCrowdworkers on the Amazon Mechanical Turk platform based in the U.S, Canada, U.K and Australia.", "#### License\n\n\n\n\napache-2.0: Apache License 2.0", "#### Intended Use\n\n\n\nTo study the viability of language-based abductive reasoning. Training and evaluating models to generate a plausible hypothesis to explain two given observations.", "#### Primary Task\n\n\n\nReasoning", "### Credit", "#### Curation Organization Type(s)\n\n\n\n'industry'", "#### Curation Organization(s)\n\n\n\nAllen Institute for AI", "#### Dataset Creators\n\n\n\nChandra Bhagavatula (AI2), Ronan Le Bras (AI2), Chaitanya Malaviya (AI2), Keisuke Sakaguchi (AI2), Ari Holtzman (AI2, UW), Hannah Rashkin (AI2, UW), Doug Downey (AI2), Wen-tau Yih (AI2), Yejin Choi (AI2, UW)", "#### Funding\n\n\n\nAllen Institute for AI", "#### Who added the Dataset to GEM?\n\n\n\nChandra Bhagavatula (AI2), Ronan LeBras (AI2), Aman Madaan (CMU), Nico Daheim (RWTH Aachen University)", "### Dataset Structure", "#### Data Fields\n\n\n\n- 'observation_1': A string describing an observation / event.\n- 'observation_2': A string describing an observation / event.\n- 'label': A string that plausibly explains why observation_1 and observation_2 might have happened.", "#### How were labels chosen?\n\n\n\nExplanations were authored by crowdworkers on the Amazon Mechanical Turk platform using a custom template designed by the creators of the dataset.", "#### Example Instance", "#### Data Splits\n\n\n\n- 'train': Consists of training instances. \n- 'dev': Consists of dev instances.\n- 'test': Consists of test instances.", "## Dataset in GEM", "### Rationale for Inclusion in GEM", "#### Why is the Dataset in GEM?\n\n\n\nAbductive reasoning is a crucial capability of humans and ART is the first dataset curated to study language-based abductive reasoning.", "#### Similar Datasets\n\n\n\nno", "#### Ability that the Dataset measures\n\n\n\nWhether models can reason abductively about a given pair of observations.", "### GEM-Specific Curation", "#### Modificatied for GEM?\n\n\n\nno", "#### Additional Splits?\n\n\n\nno", "### Getting Started with the Task", "#### Pointers to Resources\n\n\n\n- Paper\n- Code", "## Previous Results", "### Previous Results", "#### Measured Model Abilities\n\n\n\nWhether models can reason abductively about a given pair of observations.", "#### Metrics\n\n\n\n'BLEU', 'BERT-Score', 'ROUGE'", "#### Previous results available?\n\n\n\nno", "## Dataset Curation", "### Original Curation", "#### Sourced from Different Sources\n\n\n\nno", "### Language Data", "#### How was Language Data Obtained?\n\n\n\n'Crowdsourced'", "#### Where was it crowdsourced?\n\n\n\n'Amazon Mechanical Turk'", "#### Language Producers\n\n\n\nLanguage producers were English speakers in U.S., Canada, U.K and Australia.", "#### Topics Covered\n\n\n\nNo", "#### Data Validation\n\n\n\nvalidated by crowdworker", "#### Was Data Filtered?\n\n\n\nalgorithmically", "#### Filter Criteria\n\n\n\nAdversarial filtering algorithm as described in the paper", "### Structured Annotations", "#### Additional Annotations?\n\n\n\n\nautomatically created", "#### Annotation Service?\n\n\n\nno", "#### Annotation Values\n\n\n\nEach observation is associated with a list of COMET (URL inferences.", "#### Any Quality Control?\n\n\n\nnone", "### Consent", "#### Any Consent Policy?\n\n\n\nno", "### Private Identifying Information (PII)", "#### Contains PII?\n\n\n\n\nno PII", "#### Justification for no PII\n\n\n\nThe dataset contains day-to-day events. It does not contain names, emails, addresses etc.", "### Maintenance", "#### Any Maintenance Plan?\n\n\n\nno", "## Broader Social Context", "### Previous Work on the Social Impact of the Dataset", "#### Usage of Models based on the Data\n\n\n\nno", "### Impact on Under-Served Communities", "#### Addresses needs of underserved Communities?\n\n\n\nno", "### Discussion of Biases", "#### Any Documented Social Biases?\n\n\n\nno", "## Considerations for Using the Data", "### PII Risks and Liability", "#### Potential PII Risk\n\n\n\nNone", "### Licenses", "#### Copyright Restrictions on the Dataset\n\n\n\n'public domain'", "#### Copyright Restrictions on the Language Data\n\n\n\n'public domain'", "### Known Technical Limitations" ]

[ "annotations_creators:none", "language_creators:unknown", "multilinguality:unknown", "size_categories:unknown", "source_datasets:original", "language:de", "language:en", "language:fr", "language:es", "license:other", "region:us" ]

[]

[ "de", "en", "fr", "es" ]

[ "# Dataset Card for GEM/BiSECT", "## Dataset Description\n\n- Homepage: URL\n- Repository: URL\n- Paper: URL\n- Leaderboard: N/A\n- Point of Contact: Joongwon Kim, Mounica Maddela, Reno Kriz", "### Link to Main Data Card\n\nYou can find the main data card on the GEM Website.", "### Dataset Summary \n\nThis dataset is composed of 1 million complex sentences with the task to split and simplify them while retaining the full meaning. Compared to other simplification corpora, BiSECT requires more significant edits. BiSECT offers splits in English, German, French, and Spanish.\n\nYou can load the dataset via:\n\nThe data loader can be found here.", "#### website\nLink", "#### paper\nLink", "## Dataset Overview", "### Where to find the Data and its Documentation", "#### Webpage\n\n\n\nLink", "#### Download\n\n\n\nLink", "#### Paper\n\n\n\nLink", "#### BibTex", "#### Contact Name\n\n\n\n\nJoongwon Kim, Mounica Maddela, Reno Kriz", "#### Contact Email\n\n\n\njkim0118@URL, mmaddela3@URL, rkriz1@URL", "#### Has a Leaderboard?\n\n\n\nno", "### Languages and Intended Use", "#### Multilingual?\n\n\n\n\nyes", "#### Covered Languages\n\n\n\n\n'English', 'German', 'French', 'Spanish, Castilian'", "#### License\n\n\n\n\nother: Other license", "#### Intended Use\n\n\n\nSplit and Rephrase.", "#### Add. License Info\n\n\n\nThe dataset is not licensed by itself, and the source OPUS data consists solely of publicly available parallel corpora.", "#### Primary Task\n\n\n\nSimplification", "#### Communicative Goal\n\n\n\n\nTo rewrite a long, complex sentence into shorter, readable, meaning-equivalent sentences.", "### Credit", "### Dataset Structure", "#### Data Fields\n\n\n\n- 'gem_id' (string): a unique identifier for the instance\n- 'source_sentence' (string): sentence to be simplified\n- 'target_sentence' (string)\" simplified text that was split and rephrased", "#### Example Instance", "#### Data Splits\n\n\n\nFor the main English BiSECT dataset, the splits are as follows: 1. Train (n=928440) 2. Validation (n=9079) 3. Test (n=583) Additional challenge sets were derived from the data presented in the paper. Please refer to the challenge set sections. The train/validation/test splits for other languages are as follows: German (n=184638/n=864/n=735) Spanish (n=282944/n=3638/n=3081) French (n=491035/n=2400/n=1036)", "#### Splitting Criteria\n\n\n\nWhile all training data were derived from subsets of the OPUS corpora, different source subsets were used for training v.s., validation and testing. The training set comprised more web crawl data, whereas development and test sets comprised EMEA and EU texts. Details can be found in the BiSECT paper.", "## Dataset in GEM", "### Rationale for Inclusion in GEM", "#### Why is the Dataset in GEM?\n\n\n\nUnderstanding long and complex sentences is challenging for both humans and NLP models. The BiSECT dataset helps facilitate more research on Split and Rephrase as a task within itself, as well as how it can benefit downstream NLP applications.", "#### Similar Datasets\n\n\n\nyes", "#### Unique Language Coverage\n\n\n\nyes", "#### Difference from other GEM datasets\n\n\n\nBiSECT is the largest available corpora for the Split and Rephrase task. In addition, it has been shown that BiSECT is of higher quality than previous Split and Rephrase corpora and contains a wider variety of splitting operations.\n\nMost previous Split and Rephrase corpora (HSplit-Wiki, Cont-Benchmark, and Wiki-Benchmark) were manually written at a small scale and focused on evaluation, while the one corpus of comparable size, WikiSplit, contains around 25% of pairs contain significant errors. This is because Wikipedia editors are not only trying to split a sentence, but also often simultaneously modifying the sentence for other purposes, which results in changes of the initial meaning.", "### GEM-Specific Curation", "#### Modificatied for GEM?\n\n\n\nyes", "#### GEM Modifications\n\n\n\n'data points added'", "#### Modification Details\n\n\n\nThe original BiSECT training, validation, and test splits are maintained to ensure a fair comparison. Note that the original BiSECT test set was created by manually selecting 583 high-quality Split and Rephrase instances from 1000 random source-target pairs sampled from the EMEA and JRC-Acquis corpora from OPUS.\n\nAs the first challenge set, we include the HSPLIT-Wiki test set, containing 359 pairs. For each complex sentence, there are four reference splits; To ensure replicability, as reference splits, we again follow the BiSECT paper and present only the references from HSplit2-full.\n\nIn addition to the two evaluation sets used in the original BiSECT paper, we also introduce a second challenge set. For this, we initially consider all 7,293 pairs from the EMEA and JRC-Acquis corpora. From there, we classify each pair using the classification algorithm from Section 4.2 of the original BiSECT paper. The three classes are as follows:\n\n1. Direct Insertion: when a long sentence l contains two independent clauses and requires only minor changes in order to make a fluent and meaning-preserving split s.\n2. Changes near Split, when l contains one independent and one dependent clause, but modifications are restricted to the region where l is split.\n3. Changes across Sentences, where major changes are required throughout l in order to create a fluent split s.\nWe keep only pairs labeled as Type 3, and after filtering out pairs with significant length differences (signaling potential content addition/deletion), we present a second challenge set of 1,798 pairs.", "#### Additional Splits?\n\n\n\nno", "### Getting Started with the Task", "#### Pointers to Resources\n\n\n\nThe dataset can be downloaded from the original repository by the authors.\n\nThe original BiSECT paper proposes several transformer-based models that can be used as baselines, which also compares against Copy512, an LSTM-based model and the previous state-of-the-art.\n\nThe common metric used for automatic evaluation of Split and Rephrase, and sentence simplification more generally is SARI. The BiSECT paper also evaluates using BERTScore. Note that automatic evaluations tend to not correlate well with human judgments, so a human evaluation for quality is generally expected for publication. The original BiSECT paper provides templates for collecting quality annotations from Amazon Mechanical Turk.", "## Previous Results", "### Previous Results", "#### Measured Model Abilities\n\n\n\nText comprehension (needed to generate meaning-equivalent output) and notions of complexity (what is more 'readable' in terms of syntactic structure, lexical choice, punctuation).", "#### Metrics\n\n\n\n'Other: Other Metrics', 'BERT-Score'", "#### Other Metrics\n\n\n\nSARI is a metric used for evaluating automatic text simplification systems. The metric compares the predicted simplified sentences against the reference and the source sentences. It explicitly measures the goodness of words that are added, deleted and kept by the system.", "#### Proposed Evaluation\n\n\n\nExisting automatic metrics, such as BLEU (Papineni et al., 2002) and SAMSA (Sulem et al., 2018),\nare not optimal for the Split and Rephrase task as\nthey rely on lexical overlap between the output and\nthe target (or source) and underestimate the splitting capability of the models that rephrase often. \n\nAs such, the dataset creators focused on BERTScore (Zhang et al., 2020) and SARI (Xu et al., 2016). BERTScore captures meaning preservation and fluency\nwell (Scialom et al., 2021). SARI can provide three\nseparate F1/precision scores that explicitly measure the correctness of inserted, kept and deleted\nn-grams when compared to both the source and\nthe target. The authors used an extended version of SARI\nthat considers lexical paraphrases of the reference.", "#### Previous results available?\n\n\n\nyes", "## Dataset Curation", "### Original Curation", "#### Original Curation Rationale\n\n\n\nBiSECT was constructed to satisfy the need of a Split and Rephrase corpus that is both large-scale and high-quality. Most previous Split and Rephrase corpora (HSplit-Wiki, Cont-Benchmark, and Wiki-Benchmark) were manually written at a small scale and focused on evaluation, while the one corpus of comparable size, WikiSplit, contains around 25% of pairs contain significant errors. This is because Wikipedia editors are not only trying to split a sentence, but also often simultaneously modifying the sentence for other purposes, which results in changes of the initial meaning.", "#### Communicative Goal\n\n\n\nThe goal of Split and Rephrase is to break down longer sentences into multiple shorter sentences, which has downstream applications for many NLP tasks, including machine translation and dependency parsing.", "#### Sourced from Different Sources\n\n\n\nno", "### Language Data", "#### How was Language Data Obtained?\n\n\n\n'Found'", "#### Where was it found?\n\n\n\n'Other'", "#### Language Producers\n\n\n\nN/A.", "#### Topics Covered\n\n\n\nThere is a range of topics spanning domains such as web crawl and government documents (European Parliament, United Nations, EMEA).", "#### Data Validation\n\n\n\nvalidated by data curator", "#### Data Preprocessing\n\n\n\nThe construction of the BiSECT corpus relies on leveraging the sentence-level alignments from OPUS), a collection of bilingual parallel corpora over many language pairs. Given a target language A, this work extracts all 1-2 and 2-1 sentence alignments from parallel corpora between A and a set of foreign languages B.\n\nNext, the foreign sentences are translated into English using Google Translate’s Web API service to obtain sentence alignments between a single long sentence and two corresponding split sentences, both in the desired language.\n\nThe authors further filtered the data in a hybrid fashion.", "#### Was Data Filtered?\n\n\n\nhybrid", "#### Filter Criteria\n\n\n\nTo remove noise, the authors remove pairs where the single long sentence (l) contains a token with a punctuation after the first two and before the last two alphabetic characters. The authors also removed instances where l contains more than one unconnected component in its dependency tree, generated via SpaCy.", "### Structured Annotations", "#### Additional Annotations?\n\n\n\n\nnone", "#### Annotation Service?\n\n\n\nno", "### Consent", "#### Any Consent Policy?\n\n\n\nno", "#### Justification for Using the Data\n\n\n\nSince this data is collected from OPUS, all instances are already in the public domain.", "### Private Identifying Information (PII)", "#### Contains PII?\n\n\n\n\nunlikely", "#### Categories of PII\n\n\n\n'generic PII'", "#### Any PII Identification?\n\n\n\nno identification", "### Maintenance", "#### Any Maintenance Plan?\n\n\n\nno", "## Broader Social Context", "### Previous Work on the Social Impact of the Dataset", "#### Usage of Models based on the Data\n\n\n\nno", "### Impact on Under-Served Communities", "#### Addresses needs of underserved Communities?\n\n\n\nyes", "#### Details on how Dataset Addresses the Needs\n\n\n\nThe data as provided in GEMv2 is in English, which is a language with abundant existing resources. However, the original paper also provides Split and Rephrase pairs for French, Spanish, and German, while providing a framework for leveraging bilingual corpora from any language pair found within OPUS.", "### Discussion of Biases", "#### Any Documented Social Biases?\n\n\n\nno", "#### Are the Language Producers Representative of the Language?\n\n\n\nThe language produced in the dataset is limited to what is captured in the used subset of the OPUS corpora, which might not represent the full distribution of speakers from all locations. For example, the corpora used are from a limited set of relatively formal domains, so it is possible that high performance on the BiSECT test set may not transfer to more informal text.", "## Considerations for Using the Data", "### PII Risks and Liability", "#### Potential PII Risk\n\n\n\nSince this data is collected from OPUS, all pairs are already in the public domain.", "### Licenses", "#### Copyright Restrictions on the Dataset\n\n\n\n'public domain'", "#### Copyright Restrictions on the Language Data\n\n\n\n'public domain'", "### Known Technical Limitations", "#### Technical Limitations\n\n\n\nThe creation of English BiSECT relies on translating non-English text back to English. While machine translation systems tend to perform well on high-resource languages, there is still a non-negligible chance that there these systems make errors; through a manual evaluation of a subset of BiSECT, it was found that 15% of pairs contained significant errors, while an additional 22% contained minor adequacy/fluency errors. This problem is exacerbated slightly when creating German BiSECT (22% significant errors, 24% minor errors), and these numbers would likely get larger if lower-resource languages were used." ]

[ "TAGS\n#annotations_creators-none #language_creators-unknown #multilinguality-unknown #size_categories-unknown #source_datasets-original #language-German #language-English #language-French #language-Spanish #license-other #region-us \n", "# Dataset Card for GEM/BiSECT", "## Dataset Description\n\n- Homepage: URL\n- Repository: URL\n- Paper: URL\n- Leaderboard: N/A\n- Point of Contact: Joongwon Kim, Mounica Maddela, Reno Kriz", "### Link to Main Data Card\n\nYou can find the main data card on the GEM Website.", "### Dataset Summary \n\nThis dataset is composed of 1 million complex sentences with the task to split and simplify them while retaining the full meaning. Compared to other simplification corpora, BiSECT requires more significant edits. BiSECT offers splits in English, German, French, and Spanish.\n\nYou can load the dataset via:\n\nThe data loader can be found here.", "#### website\nLink", "#### paper\nLink", "## Dataset Overview", "### Where to find the Data and its Documentation", "#### Webpage\n\n\n\nLink", "#### Download\n\n\n\nLink", "#### Paper\n\n\n\nLink", "#### BibTex", "#### Contact Name\n\n\n\n\nJoongwon Kim, Mounica Maddela, Reno Kriz", "#### Contact Email\n\n\n\njkim0118@URL, mmaddela3@URL, rkriz1@URL", "#### Has a Leaderboard?\n\n\n\nno", "### Languages and Intended Use", "#### Multilingual?\n\n\n\n\nyes", "#### Covered Languages\n\n\n\n\n'English', 'German', 'French', 'Spanish, Castilian'", "#### License\n\n\n\n\nother: Other license", "#### Intended Use\n\n\n\nSplit and Rephrase.", "#### Add. License Info\n\n\n\nThe dataset is not licensed by itself, and the source OPUS data consists solely of publicly available parallel corpora.", "#### Primary Task\n\n\n\nSimplification", "#### Communicative Goal\n\n\n\n\nTo rewrite a long, complex sentence into shorter, readable, meaning-equivalent sentences.", "### Credit", "### Dataset Structure", "#### Data Fields\n\n\n\n- 'gem_id' (string): a unique identifier for the instance\n- 'source_sentence' (string): sentence to be simplified\n- 'target_sentence' (string)\" simplified text that was split and rephrased", "#### Example Instance", "#### Data Splits\n\n\n\nFor the main English BiSECT dataset, the splits are as follows: 1. Train (n=928440) 2. Validation (n=9079) 3. Test (n=583) Additional challenge sets were derived from the data presented in the paper. Please refer to the challenge set sections. The train/validation/test splits for other languages are as follows: German (n=184638/n=864/n=735) Spanish (n=282944/n=3638/n=3081) French (n=491035/n=2400/n=1036)", "#### Splitting Criteria\n\n\n\nWhile all training data were derived from subsets of the OPUS corpora, different source subsets were used for training v.s., validation and testing. The training set comprised more web crawl data, whereas development and test sets comprised EMEA and EU texts. Details can be found in the BiSECT paper.", "## Dataset in GEM", "### Rationale for Inclusion in GEM", "#### Why is the Dataset in GEM?\n\n\n\nUnderstanding long and complex sentences is challenging for both humans and NLP models. The BiSECT dataset helps facilitate more research on Split and Rephrase as a task within itself, as well as how it can benefit downstream NLP applications.", "#### Similar Datasets\n\n\n\nyes", "#### Unique Language Coverage\n\n\n\nyes", "#### Difference from other GEM datasets\n\n\n\nBiSECT is the largest available corpora for the Split and Rephrase task. In addition, it has been shown that BiSECT is of higher quality than previous Split and Rephrase corpora and contains a wider variety of splitting operations.\n\nMost previous Split and Rephrase corpora (HSplit-Wiki, Cont-Benchmark, and Wiki-Benchmark) were manually written at a small scale and focused on evaluation, while the one corpus of comparable size, WikiSplit, contains around 25% of pairs contain significant errors. This is because Wikipedia editors are not only trying to split a sentence, but also often simultaneously modifying the sentence for other purposes, which results in changes of the initial meaning.", "### GEM-Specific Curation", "#### Modificatied for GEM?\n\n\n\nyes", "#### GEM Modifications\n\n\n\n'data points added'", "#### Modification Details\n\n\n\nThe original BiSECT training, validation, and test splits are maintained to ensure a fair comparison. Note that the original BiSECT test set was created by manually selecting 583 high-quality Split and Rephrase instances from 1000 random source-target pairs sampled from the EMEA and JRC-Acquis corpora from OPUS.\n\nAs the first challenge set, we include the HSPLIT-Wiki test set, containing 359 pairs. For each complex sentence, there are four reference splits; To ensure replicability, as reference splits, we again follow the BiSECT paper and present only the references from HSplit2-full.\n\nIn addition to the two evaluation sets used in the original BiSECT paper, we also introduce a second challenge set. For this, we initially consider all 7,293 pairs from the EMEA and JRC-Acquis corpora. From there, we classify each pair using the classification algorithm from Section 4.2 of the original BiSECT paper. The three classes are as follows:\n\n1. Direct Insertion: when a long sentence l contains two independent clauses and requires only minor changes in order to make a fluent and meaning-preserving split s.\n2. Changes near Split, when l contains one independent and one dependent clause, but modifications are restricted to the region where l is split.\n3. Changes across Sentences, where major changes are required throughout l in order to create a fluent split s.\nWe keep only pairs labeled as Type 3, and after filtering out pairs with significant length differences (signaling potential content addition/deletion), we present a second challenge set of 1,798 pairs.", "#### Additional Splits?\n\n\n\nno", "### Getting Started with the Task", "#### Pointers to Resources\n\n\n\nThe dataset can be downloaded from the original repository by the authors.\n\nThe original BiSECT paper proposes several transformer-based models that can be used as baselines, which also compares against Copy512, an LSTM-based model and the previous state-of-the-art.\n\nThe common metric used for automatic evaluation of Split and Rephrase, and sentence simplification more generally is SARI. The BiSECT paper also evaluates using BERTScore. Note that automatic evaluations tend to not correlate well with human judgments, so a human evaluation for quality is generally expected for publication. The original BiSECT paper provides templates for collecting quality annotations from Amazon Mechanical Turk.", "## Previous Results", "### Previous Results", "#### Measured Model Abilities\n\n\n\nText comprehension (needed to generate meaning-equivalent output) and notions of complexity (what is more 'readable' in terms of syntactic structure, lexical choice, punctuation).", "#### Metrics\n\n\n\n'Other: Other Metrics', 'BERT-Score'", "#### Other Metrics\n\n\n\nSARI is a metric used for evaluating automatic text simplification systems. The metric compares the predicted simplified sentences against the reference and the source sentences. It explicitly measures the goodness of words that are added, deleted and kept by the system.", "#### Proposed Evaluation\n\n\n\nExisting automatic metrics, such as BLEU (Papineni et al., 2002) and SAMSA (Sulem et al., 2018),\nare not optimal for the Split and Rephrase task as\nthey rely on lexical overlap between the output and\nthe target (or source) and underestimate the splitting capability of the models that rephrase often. \n\nAs such, the dataset creators focused on BERTScore (Zhang et al., 2020) and SARI (Xu et al., 2016). BERTScore captures meaning preservation and fluency\nwell (Scialom et al., 2021). SARI can provide three\nseparate F1/precision scores that explicitly measure the correctness of inserted, kept and deleted\nn-grams when compared to both the source and\nthe target. The authors used an extended version of SARI\nthat considers lexical paraphrases of the reference.", "#### Previous results available?\n\n\n\nyes", "## Dataset Curation", "### Original Curation", "#### Original Curation Rationale\n\n\n\nBiSECT was constructed to satisfy the need of a Split and Rephrase corpus that is both large-scale and high-quality. Most previous Split and Rephrase corpora (HSplit-Wiki, Cont-Benchmark, and Wiki-Benchmark) were manually written at a small scale and focused on evaluation, while the one corpus of comparable size, WikiSplit, contains around 25% of pairs contain significant errors. This is because Wikipedia editors are not only trying to split a sentence, but also often simultaneously modifying the sentence for other purposes, which results in changes of the initial meaning.", "#### Communicative Goal\n\n\n\nThe goal of Split and Rephrase is to break down longer sentences into multiple shorter sentences, which has downstream applications for many NLP tasks, including machine translation and dependency parsing.", "#### Sourced from Different Sources\n\n\n\nno", "### Language Data", "#### How was Language Data Obtained?\n\n\n\n'Found'", "#### Where was it found?\n\n\n\n'Other'", "#### Language Producers\n\n\n\nN/A.", "#### Topics Covered\n\n\n\nThere is a range of topics spanning domains such as web crawl and government documents (European Parliament, United Nations, EMEA).", "#### Data Validation\n\n\n\nvalidated by data curator", "#### Data Preprocessing\n\n\n\nThe construction of the BiSECT corpus relies on leveraging the sentence-level alignments from OPUS), a collection of bilingual parallel corpora over many language pairs. Given a target language A, this work extracts all 1-2 and 2-1 sentence alignments from parallel corpora between A and a set of foreign languages B.\n\nNext, the foreign sentences are translated into English using Google Translate’s Web API service to obtain sentence alignments between a single long sentence and two corresponding split sentences, both in the desired language.\n\nThe authors further filtered the data in a hybrid fashion.", "#### Was Data Filtered?\n\n\n\nhybrid", "#### Filter Criteria\n\n\n\nTo remove noise, the authors remove pairs where the single long sentence (l) contains a token with a punctuation after the first two and before the last two alphabetic characters. The authors also removed instances where l contains more than one unconnected component in its dependency tree, generated via SpaCy.", "### Structured Annotations", "#### Additional Annotations?\n\n\n\n\nnone", "#### Annotation Service?\n\n\n\nno", "### Consent", "#### Any Consent Policy?\n\n\n\nno", "#### Justification for Using the Data\n\n\n\nSince this data is collected from OPUS, all instances are already in the public domain.", "### Private Identifying Information (PII)", "#### Contains PII?\n\n\n\n\nunlikely", "#### Categories of PII\n\n\n\n'generic PII'", "#### Any PII Identification?\n\n\n\nno identification", "### Maintenance", "#### Any Maintenance Plan?\n\n\n\nno", "## Broader Social Context", "### Previous Work on the Social Impact of the Dataset", "#### Usage of Models based on the Data\n\n\n\nno", "### Impact on Under-Served Communities", "#### Addresses needs of underserved Communities?\n\n\n\nyes", "#### Details on how Dataset Addresses the Needs\n\n\n\nThe data as provided in GEMv2 is in English, which is a language with abundant existing resources. However, the original paper also provides Split and Rephrase pairs for French, Spanish, and German, while providing a framework for leveraging bilingual corpora from any language pair found within OPUS.", "### Discussion of Biases", "#### Any Documented Social Biases?\n\n\n\nno", "#### Are the Language Producers Representative of the Language?\n\n\n\nThe language produced in the dataset is limited to what is captured in the used subset of the OPUS corpora, which might not represent the full distribution of speakers from all locations. For example, the corpora used are from a limited set of relatively formal domains, so it is possible that high performance on the BiSECT test set may not transfer to more informal text.", "## Considerations for Using the Data", "### PII Risks and Liability", "#### Potential PII Risk\n\n\n\nSince this data is collected from OPUS, all pairs are already in the public domain.", "### Licenses", "#### Copyright Restrictions on the Dataset\n\n\n\n'public domain'", "#### Copyright Restrictions on the Language Data\n\n\n\n'public domain'", "### Known Technical Limitations", "#### Technical Limitations\n\n\n\nThe creation of English BiSECT relies on translating non-English text back to English. While machine translation systems tend to perform well on high-resource languages, there is still a non-negligible chance that there these systems make errors; through a manual evaluation of a subset of BiSECT, it was found that 15% of pairs contained significant errors, while an additional 22% contained minor adequacy/fluency errors. This problem is exacerbated slightly when creating German BiSECT (22% significant errors, 24% minor errors), and these numbers would likely get larger if lower-resource languages were used." ]

[ "task_categories:conversational", "annotations_creators:none", "language_creators:unknown", "multilinguality:unknown", "size_categories:unknown", "source_datasets:original", "language:zh", "license:apache-2.0", "dialog-response-generation", "region:us" ]

[]

[ "zh" ]

[ "### Link to Main Data Card\n\n\nYou can find the main data card on the GEM Website.", "### Dataset Summary\n\n\nCrossWOZ is a Chinese multi-domain task-oriented dialogue dataset . It contains 6K dialogue sessions and 102K utterances for 5 domains, including hotel, restaurant, attraction, metro, and taxi. About 60{%} of the dialogues have cross-domain user goals that favor inter-domain dependency and encourage natural transition across domains in conversation.\n\n\nYou can load the dataset via:\n\n\nThe data loader can be found here.", "#### website\n\n\nGithub", "#### paper\n\n\nACL Anthology", "#### authors\n\n\nQi Zhu, Kaili Huang, Zheng Zhang, Xiaoyan Zhu, and Minlie Huang from CoAI group, Tsinghua University\n\n\nDataset Overview\n----------------", "### Where to find the Data and its Documentation", "#### Webpage\n\n\nGithub", "#### Download\n\n\nGithub", "#### Paper\n\n\nACL Anthology", "#### BibTex", "#### Contact Name\n\n\nQi Zhu", "#### Contact Email\n\n\nzhuq96@URL", "#### Has a Leaderboard?\n\n\nno", "### Languages and Intended Use", "#### Multilingual?\n\n\nno", "#### Covered Languages\n\n\n'Chinese'", "#### License\n\n\napache-2.0: Apache License 2.0", "#### Intended Use\n\n\nCrossWOZ is the first large-scale Chinese Cross-Domain Wizard-of-Oz task-oriented dataset. It contains 6K dialogue sessions and 102K utterances for 5 domains, including hotel, restaurant, attraction, metro, and taxi. Moreover, the corpus contains rich annotation of dialogue states and dialogue acts at both user and system sides. We also provide a user simulator and several benchmark models for pipelined taskoriented dialogue systems, which will facilitate researchers to compare and evaluate their models on this corpus.", "#### Primary Task\n\n\nDialog Response Generation", "#### Communicative Goal\n\n\nGenerate a response according to the dialog context and database search results.", "### Credit", "#### Curation Organization Type(s)\n\n\n'academic'", "#### Curation Organization(s)\n\n\nTsinghua University", "#### Dataset Creators\n\n\nQi Zhu, Kaili Huang, Zheng Zhang, Xiaoyan Zhu, and Minlie Huang from CoAI group, Tsinghua University", "#### Funding\n\n\nNational Science Foundation of China, National Key R&D Program of China", "#### Who added the Dataset to GEM?\n\n\nQi Zhu (Tsinghua University)", "### Dataset Structure", "#### Data Fields\n\n\n* 'gem\\_id' (string): GEM-CrossWOZ-{split}-{id}\n* 'dialog\\_id' (string): dialog ID\n* 'sys\\_id' (string): system annotator ID\n* 'usr\\_id' (string): user annotation ID\n* 'type' (string): dialog type\n* 'task description' (list of strings): natural language descriptions of the user goal\n* 'goal' (list of tuples), includes:\n\t+ 'sub-goal id' (string)\n\t+ 'domain name' (string)\n\t+ 'slot name' (string)\n\t+ 'constraint' if filled, else 'requirement' (string)\n\t+ 'whether be mentioned in previous turns' (string)\n* 'messages' (list of dict): dialog turns. Each turn contains:\n\t+ 'content' (string): utterance\n\t+ 'role' (string): user or system\n\t+ 'dialog\\_act' (list of tuples), includes:\n\t\t- 'domain' (string)\n\t\t- 'intent' (string)\n\t\t- 'slot' (string)\n\t\t- 'value' (string)\n\t+ 'user\\_state' (list of tuples): same format as \"goal\", can be viewed as dynamic goal.\n\t+ 'sys\\_state\\_init' (dict): the first db query emitted, records user constraints faithfully. If the system find no result that matches, he/she may relax the constraints manually and search db multiple times.\n\t\t- 'domain' (dict): slot(string)-value(string) pairs\n\t\t- 'selectedResults' (list of string): db search result that would be used in this turn.\n\t+ 'sys\\_state' (dict): the final db query emitted, records the db used by the system in this turn. Same format as sys\\_state\\_init. Note that this may not satisfy all user constraints.\n* 'final\\_goal' (list of tuples): user state/goal at the end of dialog. same format as \"goal\".", "#### Example Instance", "#### Data Splits\n\n\n\nDataset in GEM\n--------------", "### Rationale for Inclusion in GEM", "#### Why is the Dataset in GEM?\n\n\nCrossWOZ is the first large-scale Chinese Cross-Domain Wizard-of-Oz task-oriented dataset.", "#### Similar Datasets\n\n\nyes", "#### Unique Language Coverage\n\n\nno", "#### Difference from other GEM datasets\n\n\nThe corpus contains rich annotation of dialogue states and dialogue acts at both user and system sides, which can be used in a wide range of tasks.", "#### Ability that the Dataset measures\n\n\nDialog understanding, dialog policy learning", "### GEM-Specific Curation", "#### Modificatied for GEM?\n\n\nyes", "#### GEM Modifications\n\n\n'other'", "#### Modification Details\n\n\nTo adapt to hugging face Datasets, we 1) separate user annotators' ID and system annotations' ID; 2) we convert the data type in goal/user state to string.", "#### Additional Splits?\n\n\nno", "### Getting Started with the Task", "#### Pointers to Resources\n\n\nCode", "#### Technical Terms\n\n\nAccording to the type of user goal, we group the dialogues in the training set into five categories:\n\n\n* S: 417 dialogues have only one sub-goal in HAR domains.\n* M: 1573 dialogues have multiple sub-goals (2-3) in HAR domains. However, these sub-goals do not have cross-domain informable slots.\n* M+T: 691 dialogues have multiple sub-goals in HAR domains and at least one sub-goal in the metro or taxi domain (3-5 sub-goals). The sub-goals in HAR domains do not have cross-domain informable slots.\n* CM: 1,759 dialogues have multiple sub-goals (2-5) in HAR domains with cross-domain informable slots.\n* CM+T: 572 dialogues have multiple sub-goals in HAR domains with cross-domain informable slots and at least one sub-goal in the metro or taxi domain (3-5 sub-goals).\n\n\nPrevious Results\n----------------", "### Previous Results", "#### Measured Model Abilities\n\n\nDialog understanding, dialog policy learning", "#### Metrics\n\n\n'BLEU'", "#### Proposed Evaluation\n\n\nBLEU evaluates the generation quality.", "#### Previous results available?\n\n\nyes", "#### Other Evaluation Approaches\n\n\nInform rate: how many entities in the gold response appear in the generated response.", "#### Relevant Previous Results\n\n\nBLEU on MultiWOZ dataset.\n\n\nDataset Curation\n----------------", "### Original Curation", "#### Original Curation Rationale\n\n\nGather human-to-human dialog in Chinese.", "#### Communicative Goal\n\n\nGenerate a response according to the dialog context and database search results.", "#### Sourced from Different Sources\n\n\nno", "### Language Data", "#### How was Language Data Obtained?\n\n\n'Crowdsourced'", "#### Where was it crowdsourced?\n\n\n'Participatory experiment'", "#### Language Producers\n\n\nAn usr/sys ID indicates the creator of different data points.", "#### Topics Covered\n\n\ndomains: attraction, hotel, restaurant, metro, taxi", "#### Data Validation\n\n\nvalidated by data curator", "#### Was Data Filtered?\n\n\nnot filtered", "### Structured Annotations", "#### Additional Annotations?\n\n\nnone", "#### Annotation Service?\n\n\nno", "### Consent", "#### Any Consent Policy?\n\n\nyes", "#### Consent Policy Details\n\n\nAnnotators agree using the dataset for research purpose.", "#### Other Consented Downstream Use\n\n\nAny", "### Private Identifying Information (PII)", "#### Contains PII?\n\n\nunlikely", "#### Categories of PII\n\n\n'generic PII'", "#### Any PII Identification?\n\n\nno identification", "### Maintenance", "#### Any Maintenance Plan?\n\n\nno\n\n\nBroader Social Context\n----------------------", "### Previous Work on the Social Impact of the Dataset", "#### Usage of Models based on the Data\n\n\nno", "### Impact on Under-Served Communities", "#### Addresses needs of underserved Communities?\n\n\nyes", "#### Details on how Dataset Addresses the Needs\n\n\nCrossWOZ is the first large-scale Chinese Cross-Domain Wizard-of-Oz task-oriented dataset. The corpus contains rich annotation of dialogue states and dialogue acts at both user and system sides, which can be used in a wide range of tasks.", "### Discussion of Biases", "#### Any Documented Social Biases?\n\n\nno", "#### Are the Language Producers Representative of the Language?\n\n\nYes\n\n\nConsiderations for Using the Data\n---------------------------------", "### PII Risks and Liability", "#### Potential PII Risk\n\n\nNo", "### Licenses", "#### Copyright Restrictions on the Dataset\n\n\n'open license - commercial use allowed'", "#### Copyright Restrictions on the Language Data\n\n\n'open license - commercial use allowed'", "### Known Technical Limitations", "#### Technical Limitations\n\n\nNo", "#### Unsuited Applications\n\n\nModel may not handle unknown values in the dialog", "#### Discouraged Use Cases\n\n\nResponses can be diverse, which is not captured by BLEU" ]

[ "TAGS\n#task_categories-conversational #annotations_creators-none #language_creators-unknown #multilinguality-unknown #size_categories-unknown #source_datasets-original #language-Chinese #license-apache-2.0 #dialog-response-generation #region-us \n", "### Link to Main Data Card\n\n\nYou can find the main data card on the GEM Website.", "### Dataset Summary\n\n\nCrossWOZ is a Chinese multi-domain task-oriented dialogue dataset . It contains 6K dialogue sessions and 102K utterances for 5 domains, including hotel, restaurant, attraction, metro, and taxi. About 60{%} of the dialogues have cross-domain user goals that favor inter-domain dependency and encourage natural transition across domains in conversation.\n\n\nYou can load the dataset via:\n\n\nThe data loader can be found here.", "#### website\n\n\nGithub", "#### paper\n\n\nACL Anthology", "#### authors\n\n\nQi Zhu, Kaili Huang, Zheng Zhang, Xiaoyan Zhu, and Minlie Huang from CoAI group, Tsinghua University\n\n\nDataset Overview\n----------------", "### Where to find the Data and its Documentation", "#### Webpage\n\n\nGithub", "#### Download\n\n\nGithub", "#### Paper\n\n\nACL Anthology", "#### BibTex", "#### Contact Name\n\n\nQi Zhu", "#### Contact Email\n\n\nzhuq96@URL", "#### Has a Leaderboard?\n\n\nno", "### Languages and Intended Use", "#### Multilingual?\n\n\nno", "#### Covered Languages\n\n\n'Chinese'", "#### License\n\n\napache-2.0: Apache License 2.0", "#### Intended Use\n\n\nCrossWOZ is the first large-scale Chinese Cross-Domain Wizard-of-Oz task-oriented dataset. It contains 6K dialogue sessions and 102K utterances for 5 domains, including hotel, restaurant, attraction, metro, and taxi. Moreover, the corpus contains rich annotation of dialogue states and dialogue acts at both user and system sides. We also provide a user simulator and several benchmark models for pipelined taskoriented dialogue systems, which will facilitate researchers to compare and evaluate their models on this corpus.", "#### Primary Task\n\n\nDialog Response Generation", "#### Communicative Goal\n\n\nGenerate a response according to the dialog context and database search results.", "### Credit", "#### Curation Organization Type(s)\n\n\n'academic'", "#### Curation Organization(s)\n\n\nTsinghua University", "#### Dataset Creators\n\n\nQi Zhu, Kaili Huang, Zheng Zhang, Xiaoyan Zhu, and Minlie Huang from CoAI group, Tsinghua University", "#### Funding\n\n\nNational Science Foundation of China, National Key R&D Program of China", "#### Who added the Dataset to GEM?\n\n\nQi Zhu (Tsinghua University)", "### Dataset Structure", "#### Data Fields\n\n\n* 'gem\\_id' (string): GEM-CrossWOZ-{split}-{id}\n* 'dialog\\_id' (string): dialog ID\n* 'sys\\_id' (string): system annotator ID\n* 'usr\\_id' (string): user annotation ID\n* 'type' (string): dialog type\n* 'task description' (list of strings): natural language descriptions of the user goal\n* 'goal' (list of tuples), includes:\n\t+ 'sub-goal id' (string)\n\t+ 'domain name' (string)\n\t+ 'slot name' (string)\n\t+ 'constraint' if filled, else 'requirement' (string)\n\t+ 'whether be mentioned in previous turns' (string)\n* 'messages' (list of dict): dialog turns. Each turn contains:\n\t+ 'content' (string): utterance\n\t+ 'role' (string): user or system\n\t+ 'dialog\\_act' (list of tuples), includes:\n\t\t- 'domain' (string)\n\t\t- 'intent' (string)\n\t\t- 'slot' (string)\n\t\t- 'value' (string)\n\t+ 'user\\_state' (list of tuples): same format as \"goal\", can be viewed as dynamic goal.\n\t+ 'sys\\_state\\_init' (dict): the first db query emitted, records user constraints faithfully. If the system find no result that matches, he/she may relax the constraints manually and search db multiple times.\n\t\t- 'domain' (dict): slot(string)-value(string) pairs\n\t\t- 'selectedResults' (list of string): db search result that would be used in this turn.\n\t+ 'sys\\_state' (dict): the final db query emitted, records the db used by the system in this turn. Same format as sys\\_state\\_init. Note that this may not satisfy all user constraints.\n* 'final\\_goal' (list of tuples): user state/goal at the end of dialog. same format as \"goal\".", "#### Example Instance", "#### Data Splits\n\n\n\nDataset in GEM\n--------------", "### Rationale for Inclusion in GEM", "#### Why is the Dataset in GEM?\n\n\nCrossWOZ is the first large-scale Chinese Cross-Domain Wizard-of-Oz task-oriented dataset.", "#### Similar Datasets\n\n\nyes", "#### Unique Language Coverage\n\n\nno", "#### Difference from other GEM datasets\n\n\nThe corpus contains rich annotation of dialogue states and dialogue acts at both user and system sides, which can be used in a wide range of tasks.", "#### Ability that the Dataset measures\n\n\nDialog understanding, dialog policy learning", "### GEM-Specific Curation", "#### Modificatied for GEM?\n\n\nyes", "#### GEM Modifications\n\n\n'other'", "#### Modification Details\n\n\nTo adapt to hugging face Datasets, we 1) separate user annotators' ID and system annotations' ID; 2) we convert the data type in goal/user state to string.", "#### Additional Splits?\n\n\nno", "### Getting Started with the Task", "#### Pointers to Resources\n\n\nCode", "#### Technical Terms\n\n\nAccording to the type of user goal, we group the dialogues in the training set into five categories:\n\n\n* S: 417 dialogues have only one sub-goal in HAR domains.\n* M: 1573 dialogues have multiple sub-goals (2-3) in HAR domains. However, these sub-goals do not have cross-domain informable slots.\n* M+T: 691 dialogues have multiple sub-goals in HAR domains and at least one sub-goal in the metro or taxi domain (3-5 sub-goals). The sub-goals in HAR domains do not have cross-domain informable slots.\n* CM: 1,759 dialogues have multiple sub-goals (2-5) in HAR domains with cross-domain informable slots.\n* CM+T: 572 dialogues have multiple sub-goals in HAR domains with cross-domain informable slots and at least one sub-goal in the metro or taxi domain (3-5 sub-goals).\n\n\nPrevious Results\n----------------", "### Previous Results", "#### Measured Model Abilities\n\n\nDialog understanding, dialog policy learning", "#### Metrics\n\n\n'BLEU'", "#### Proposed Evaluation\n\n\nBLEU evaluates the generation quality.", "#### Previous results available?\n\n\nyes", "#### Other Evaluation Approaches\n\n\nInform rate: how many entities in the gold response appear in the generated response.", "#### Relevant Previous Results\n\n\nBLEU on MultiWOZ dataset.\n\n\nDataset Curation\n----------------", "### Original Curation", "#### Original Curation Rationale\n\n\nGather human-to-human dialog in Chinese.", "#### Communicative Goal\n\n\nGenerate a response according to the dialog context and database search results.", "#### Sourced from Different Sources\n\n\nno", "### Language Data", "#### How was Language Data Obtained?\n\n\n'Crowdsourced'", "#### Where was it crowdsourced?\n\n\n'Participatory experiment'", "#### Language Producers\n\n\nAn usr/sys ID indicates the creator of different data points.", "#### Topics Covered\n\n\ndomains: attraction, hotel, restaurant, metro, taxi", "#### Data Validation\n\n\nvalidated by data curator", "#### Was Data Filtered?\n\n\nnot filtered", "### Structured Annotations", "#### Additional Annotations?\n\n\nnone", "#### Annotation Service?\n\n\nno", "### Consent", "#### Any Consent Policy?\n\n\nyes", "#### Consent Policy Details\n\n\nAnnotators agree using the dataset for research purpose.", "#### Other Consented Downstream Use\n\n\nAny", "### Private Identifying Information (PII)", "#### Contains PII?\n\n\nunlikely", "#### Categories of PII\n\n\n'generic PII'", "#### Any PII Identification?\n\n\nno identification", "### Maintenance", "#### Any Maintenance Plan?\n\n\nno\n\n\nBroader Social Context\n----------------------", "### Previous Work on the Social Impact of the Dataset", "#### Usage of Models based on the Data\n\n\nno", "### Impact on Under-Served Communities", "#### Addresses needs of underserved Communities?\n\n\nyes", "#### Details on how Dataset Addresses the Needs\n\n\nCrossWOZ is the first large-scale Chinese Cross-Domain Wizard-of-Oz task-oriented dataset. The corpus contains rich annotation of dialogue states and dialogue acts at both user and system sides, which can be used in a wide range of tasks.", "### Discussion of Biases", "#### Any Documented Social Biases?\n\n\nno", "#### Are the Language Producers Representative of the Language?\n\n\nYes\n\n\nConsiderations for Using the Data\n---------------------------------", "### PII Risks and Liability", "#### Potential PII Risk\n\n\nNo", "### Licenses", "#### Copyright Restrictions on the Dataset\n\n\n'open license - commercial use allowed'", "#### Copyright Restrictions on the Language Data\n\n\n'open license - commercial use allowed'", "### Known Technical Limitations", "#### Technical Limitations\n\n\nNo", "#### Unsuited Applications\n\n\nModel may not handle unknown values in the dialog", "#### Discouraged Use Cases\n\n\nResponses can be diverse, which is not captured by BLEU" ]

[ "task_categories:summarization", "annotations_creators:unknown", "language_creators:unknown", "multilinguality:unknown", "size_categories:unknown", "source_datasets:original", "language:fr", "license:other", "region:us" ]

[]

[ "fr" ]

[ "# Dataset Card for GEM/OrangeSum", "## Dataset Description\n\n- Homepage: \n- Repository: URL\n- Paper: URL\n- Leaderboard: N/A\n- Point of Contact:", "### Link to Main Data Card\n\nYou can find the main data card on the GEM Website.", "### Dataset Summary \n\nOrangeSum is a French summarization dataset inspired by XSum. It features two subtasks: abstract generation and title generation. The data was sourced from \"Orange Actu\" articles between 2011 and 2020. \n\nYou can load the dataset via:\n\nThe data loader can be found here.", "#### paper\nACL Anthology", "## Dataset Overview", "### Where to find the Data and its Documentation", "#### Download\n\n\n\nGithub", "#### Paper\n\n\n\nACL Anthology", "#### BibTex", "#### Has a Leaderboard?\n\n\n\nno", "### Languages and Intended Use", "#### Multilingual?\n\n\n\n\nno", "#### Covered Languages\n\n\n\n\n'French'", "#### License\n\n\n\n\nother: Other license", "#### Primary Task\n\n\n\nSummarization", "### Credit", "### Dataset Structure", "## Dataset in GEM", "### Rationale for Inclusion in GEM", "#### Similar Datasets\n\n\n\nno", "### GEM-Specific Curation", "#### Modificatied for GEM?\n\n\n\nno", "#### Additional Splits?\n\n\n\nno", "### Getting Started with the Task", "#### Pointers to Resources\n\n\n\nPapers about abstractive summarization using seq2seq models:\n\n- Abstractive Text Summarization using Sequence-to-sequence RNNs and Beyond\n- Get To The Point: Summarization with Pointer-Generator Networks\n- BART: Denoising Sequence-to-Sequence Pre-training for Natural Language Generation, Translation, and Comprehension\n- BARThez: a Skilled Pretrained French Sequence-to-Sequence Model\n\nPapers about (pretrained) Transformers:\n\n- Attention is All you Need\n- BERT: Pre-training of Deep Bidirectional Transformers for Language Understanding", "#### Technical Terms\n\n\n\nNo unique technical words in this data card.", "## Previous Results", "### Previous Results", "#### Measured Model Abilities\n\n\n\nThe ability of the model to generate human like titles and abstracts for given news articles.", "#### Metrics\n\n\n\n'ROUGE', 'BERT-Score'", "#### Proposed Evaluation\n\n\n\nAutomatic Evaluation: Rouge-1, Rouge-2, RougeL and BERTScore were used.\n\nHuman evalutaion: a human evaluation study was conducted with 11 French native speakers. The evaluators were PhD students from the computer science department of the university of the authors, working in NLP and other fields of AI. They volunteered after receiving an email announcement. the best-Worst Scaling (Louviere et al.,2015) was used. Two summaries from two different systems, along with their input document, were presented to a human annotator who had to decide which one was better. The evaluators were asked to base their judgments on accuracy (does the summary contain accurate facts?), informativeness (is important in-formation captured?) and fluency (is the summary written in well-formed French?).", "#### Previous results available?\n\n\n\nno", "## Broader Social Context", "### Previous Work on the Social Impact of the Dataset", "#### Usage of Models based on the Data\n\n\n\nno", "### Impact on Under-Served Communities", "#### Addresses needs of underserved Communities?\n\n\n\nno", "### Discussion of Biases", "#### Any Documented Social Biases?\n\n\n\nno", "#### Are the Language Producers Representative of the Language?\n\n\n\nThe dataset contains news articles written by professional authors.", "## Considerations for Using the Data", "### PII Risks and Liability", "### Licenses", "#### Copyright Restrictions on the Dataset\n\n\n\n'open license - commercial use allowed'", "#### Copyright Restrictions on the Language Data\n\n\n\n'open license - commercial use allowed'", "### Known Technical Limitations" ]

[ "TAGS\n#task_categories-summarization #annotations_creators-unknown #language_creators-unknown #multilinguality-unknown #size_categories-unknown #source_datasets-original #language-French #license-other #region-us \n", "# Dataset Card for GEM/OrangeSum", "## Dataset Description\n\n- Homepage: \n- Repository: URL\n- Paper: URL\n- Leaderboard: N/A\n- Point of Contact:", "### Link to Main Data Card\n\nYou can find the main data card on the GEM Website.", "### Dataset Summary \n\nOrangeSum is a French summarization dataset inspired by XSum. It features two subtasks: abstract generation and title generation. The data was sourced from \"Orange Actu\" articles between 2011 and 2020. \n\nYou can load the dataset via:\n\nThe data loader can be found here.", "#### paper\nACL Anthology", "## Dataset Overview", "### Where to find the Data and its Documentation", "#### Download\n\n\n\nGithub", "#### Paper\n\n\n\nACL Anthology", "#### BibTex", "#### Has a Leaderboard?\n\n\n\nno", "### Languages and Intended Use", "#### Multilingual?\n\n\n\n\nno", "#### Covered Languages\n\n\n\n\n'French'", "#### License\n\n\n\n\nother: Other license", "#### Primary Task\n\n\n\nSummarization", "### Credit", "### Dataset Structure", "## Dataset in GEM", "### Rationale for Inclusion in GEM", "#### Similar Datasets\n\n\n\nno", "### GEM-Specific Curation", "#### Modificatied for GEM?\n\n\n\nno", "#### Additional Splits?\n\n\n\nno", "### Getting Started with the Task", "#### Pointers to Resources\n\n\n\nPapers about abstractive summarization using seq2seq models:\n\n- Abstractive Text Summarization using Sequence-to-sequence RNNs and Beyond\n- Get To The Point: Summarization with Pointer-Generator Networks\n- BART: Denoising Sequence-to-Sequence Pre-training for Natural Language Generation, Translation, and Comprehension\n- BARThez: a Skilled Pretrained French Sequence-to-Sequence Model\n\nPapers about (pretrained) Transformers:\n\n- Attention is All you Need\n- BERT: Pre-training of Deep Bidirectional Transformers for Language Understanding", "#### Technical Terms\n\n\n\nNo unique technical words in this data card.", "## Previous Results", "### Previous Results", "#### Measured Model Abilities\n\n\n\nThe ability of the model to generate human like titles and abstracts for given news articles.", "#### Metrics\n\n\n\n'ROUGE', 'BERT-Score'", "#### Proposed Evaluation\n\n\n\nAutomatic Evaluation: Rouge-1, Rouge-2, RougeL and BERTScore were used.\n\nHuman evalutaion: a human evaluation study was conducted with 11 French native speakers. The evaluators were PhD students from the computer science department of the university of the authors, working in NLP and other fields of AI. They volunteered after receiving an email announcement. the best-Worst Scaling (Louviere et al.,2015) was used. Two summaries from two different systems, along with their input document, were presented to a human annotator who had to decide which one was better. The evaluators were asked to base their judgments on accuracy (does the summary contain accurate facts?), informativeness (is important in-formation captured?) and fluency (is the summary written in well-formed French?).", "#### Previous results available?\n\n\n\nno", "## Broader Social Context", "### Previous Work on the Social Impact of the Dataset", "#### Usage of Models based on the Data\n\n\n\nno", "### Impact on Under-Served Communities", "#### Addresses needs of underserved Communities?\n\n\n\nno", "### Discussion of Biases", "#### Any Documented Social Biases?\n\n\n\nno", "#### Are the Language Producers Representative of the Language?\n\n\n\nThe dataset contains news articles written by professional authors.", "## Considerations for Using the Data", "### PII Risks and Liability", "### Licenses", "#### Copyright Restrictions on the Dataset\n\n\n\n'open license - commercial use allowed'", "#### Copyright Restrictions on the Language Data\n\n\n\n'open license - commercial use allowed'", "### Known Technical Limitations" ]

[ "task_categories:conversational", "annotations_creators:crowd-sourced", "language_creators:unknown", "multilinguality:unknown", "size_categories:unknown", "source_datasets:original", "language:zh", "license:cc-by-4.0", "dialog-response-generation", "region:us" ]

[]

[ "zh" ]

[ "# Dataset Card for GEM/RiSAWOZ", "## Dataset Description\n\n- Homepage: URL\n- Repository: URL\n- Paper: URL\n- Leaderboard: N/A\n- Point of Contact: Deyi Xiong", "### Link to Main Data Card\n\nYou can find the main data card on the GEM Website.", "### Dataset Summary \n\nRiSAWOZ is a Chinese dialog dataset. It can be used to study various dialogue tasks, such as Dialogue State Tracking, Dialogue Context-to-Text Generation, Coreference Resolution and Unified Generative Ellipsis and Coreference Resolution.\n\nYou can load the dataset via:\n\nThe data loader can be found here.", "#### website\nWebsite", "#### paper\nACL Anthology", "#### authors\nJun Quan (Soochow University, Suzhou, China), Shian Zhang (Soochow University, Suzhou, China), Qian Cao(Soochow University, Suzhou, China), Zizhong Li (Tianjin University, Tianjin, China), Deyi Xiong (Tianjin University, Tianjin, China)", "## Dataset Overview", "### Where to find the Data and its Documentation", "#### Webpage\n\n\n\nWebsite", "#### Download\n\n\n\nGithub", "#### Paper\n\n\n\nACL Anthology", "#### BibTex", "#### Contact Name\n\n\n\n\nDeyi Xiong", "#### Contact Email\n\n\n\ndyxiong@URL", "#### Has a Leaderboard?\n\n\n\nno", "### Languages and Intended Use", "#### Multilingual?\n\n\n\n\nno", "#### Covered Dialects\n\n\n\nOnly Mandarin Chinese is covered in this dataset.", "#### Covered Languages\n\n\n\n\n'Mandarin Chinese'", "#### License\n\n\n\n\ncc-by-4.0: Creative Commons Attribution 4.0 International", "#### Intended Use\n\n\n\nRiSAWOZ can be used to support the study under various dialogue tasks, such as Natural Language Understanding, Dialogue State Tracking, Dialogue Context-to-Text Generation, Coreference Resolution and Unified Generative Ellipsis and Coreference Resolution.", "#### Primary Task\n\n\n\nDialog Response Generation", "#### Communicative Goal\n\n\n\n\nGenerate system response given dialogue context across multiple domains.", "### Credit", "#### Curation Organization Type(s)\n\n\n\n'academic'", "#### Curation Organization(s)\n\n\n\nSoochow University and Tianjin University", "#### Dataset Creators\n\n\n\nJun Quan (Soochow University, Suzhou, China), Shian Zhang (Soochow University, Suzhou, China), Qian Cao(Soochow University, Suzhou, China), Zizhong Li (Tianjin University, Tianjin, China), Deyi Xiong (Tianjin University, Tianjin, China)", "#### Funding\n\n\n\nthe National Key Research and Development Project", "#### Who added the Dataset to GEM?\n\n\n\nTianhao Shen (Tianjin University, Tianjin, China), Chaobin You (Tianjin University, Tianjin, China), Deyi Xiong (Tianjin University, Tianjin, China)", "### Dataset Structure", "#### Data Fields\n\n\n\n- gem_id (string): GEM-RiSAWOZ-{split}-{id}\n- dialogue_id (string): dialogue ID\n- goal (string): natural language descriptions of the user goal\n- domains (list of strings): domains mentioned in current dialogue session\n- dialogue (list of dicts): dialog turns and corresponding annotations. Each turn includes:\n - turn_id (int): turn ID\n - turn_domain (list of strings): domain mentioned in current turn\n - user_utterance (string): user utterance\n - system_utterance (string): system utterance\n - belief_state (dict): dialogue state, including:\n - inform slot-values (dict): the slots and corresponding values informed until current turn\n - turn_inform (dict): the slots and corresponding values informed in current turn\n - turn request (dict): the slots requested in current turn\n - user_actions (list of lists): user dialogue acts in current turn\n - user_actions (list of lists): system dialogue acts in current turn\n - db_results (list of strings): database search results\n - segmented_user_utterance (string): word segmentation result of user utterance\n - segmented_system_utterance (string): word segmentation result of system utterance", "#### Example Instance", "#### Data Splits\n\n\n\n- Train:\n - dialogues: 10000\n - turns: 134580\n - Avg. turns: 13.5\n- Valid:\n - dialogues: 600\n - turns: 8116\n - Avg. turns: 13.5\n- Test:\n - dialogues: 600\n - turn: 9286\n - Avg. turns: 15.5", "#### \n\n\n\nthe most complex example (i.e. with the most domains):", "## Dataset in GEM", "### Rationale for Inclusion in GEM", "#### Why is the Dataset in GEM?\n\n\n\nRiSAWOZ is the first large-scale multi-domain Chinese Wizard-of-Oz dataset with rich semantic annotations.", "#### Similar Datasets\n\n\n\nyes", "#### Unique Language Coverage\n\n\n\nno", "#### Difference from other GEM datasets\n\n\n\nThe corpus contains rich semantic annotations, such as ellipsis and coreference, in addition to traditional dialogue annotations (dialogue states, dialogue acts, etc.), which can be used in various tasks in dialogue system.", "#### Ability that the Dataset measures\n\n\n\nNatural Language Understanding, Dialogue State Tracking, Dialogue Context-to-Text Generation, Coreference Resolution, Unified Generative Ellipsis and Coreference Resolution", "### GEM-Specific Curation", "#### Modificatied for GEM?\n\n\n\nno", "#### Additional Splits?\n\n\n\nno", "### Getting Started with the Task", "#### Pointers to Resources\n\n\n\nWebsite", "#### Technical Terms\n\n\n\n- In task-oriented dialogue system, the Natural Language Understanding (NLU) module aims to convert the user utterance into the representation that computer can understand, which includes intent and dialogue act (slot & value) detection.\n- Dialogue State Tracking (DST) is a core component in task-oriented dialogue systems, which extracts dialogue states (user goals) embedded in dialogue context. It has progressed toward open-vocabulary or generation-based DST where state-of-the-art models can generate dialogue states from dialogue context directly.\n- Context-to-Text Generation: encoding dialogue context to decode system response.\n- Coreference Resolution: predict coreference clusters where all mentions are referring to the same entity for each dialogue.\n- Unified Generative Ellipsis and Coreference Resolution: generating omitted or referred expressions from the dialogue context.", "## Previous Results", "### Previous Results", "#### Measured Model Abilities\n\n\n\nNatural Language Understanding, Dialogue State Tracking, Dialogue Context-to-Text Generation, Coreference Resolution, Unified Generative Ellipsis and Coreference Resolution", "#### Metrics\n\n\n\n'Other: Other Metrics'", "#### Other Metrics\n\n\n\n- Natural Language Understanding:\n - F1 score: F1 score of user intent.\n- Dialogue State Tracking:\n - Joint Accuracy: accuracy of turn-level dialogue states.\n- Dialogue Context-to-Text Generation:\n - inform rate: measures the percentage that the output contains the appropriate entity the user asks for.\n - success rate: estimates the proportion that all the requested attributes have been answered.\n - BLEU: the BLEU score of generated system response.\n - Combined Score: (inform + success) ∗ 0.5 + BLEU as an overall\n quality.\n- Coreference Resolution:\n - MUC F1 Score: a link-based metric. Mentions in the same entity/cluster are considered “linked”. MUC penalizes the missing links and incorrect links, each with the same weight.\n - B3 F1 Score: a mention-based metric.The evaluation score depends on the fraction of the correct mentions included in the response entities (i.e. entities created by the system).\n - CEAFφ4 F1 Score: a metric which assumes each key entity should only be mapped to one response entity, and vice versa. It aligns the key entities (clusters) with the response entities in the best way, and compute scores from that alignment.\n - Average F1 Score: an average F1 score of the above three metrics.\n- Unified Generative Ellipsis and Coreference Resolution:\n - Exact Match Rate: measures whether the generated utterances exactly match the ground-truth utterances.\n - BLEU: the BLEU score of generated utterances\n - Resolution F1: comparing machine-generated words with ground-truth words only from the ellipsis/coreference part of user utterances.", "#### Proposed Evaluation\n\n\n\nsee \"Definitions of other metrics\"", "#### Previous results available?\n\n\n\nyes", "#### Other Evaluation Approaches\n\n\n\nsame as our dataset", "#### Relevant Previous Results\n\n\n\nJoint Accuracy, Inform Rate, Success Rate, BLEU Score and Combined Score on MultiWOZ and CrossWOZ dataset.", "## Dataset Curation", "### Original Curation", "#### Original Curation Rationale\n\n\n\nGather human-to-human dialog in Chinese.", "#### Communicative Goal\n\n\n\nGenerate system response given dialogue context across multiple domains.", "#### Sourced from Different Sources\n\n\n\nno", "### Language Data", "#### How was Language Data Obtained?\n\n\n\n'Crowdsourced'", "#### Where was it crowdsourced?\n\n\n\n'Other crowdworker platform'", "#### Topics Covered\n\n\n\ndomains: Attraction, Restaurant, Hotel, Flight, Train, Weather, Movie, TV, Computer, Car, Hospital, Courses", "#### Data Validation\n\n\n\nvalidated by data curator", "#### Was Data Filtered?\n\n\n\nhybrid", "#### Filter Criteria\n\n\n\nRule-based and manual selection criteria", "### Structured Annotations", "#### Additional Annotations?\n\n\n\n\ncrowd-sourced", "#### Number of Raters\n\n\n\n51<n<100", "#### Rater Qualifications\n\n\n\nChinese native speaker", "#### Raters per Training Example\n\n\n\n3", "#### Raters per Test Example\n\n\n\n3", "#### Annotation Service?\n\n\n\nno", "#### Annotation Values\n\n\n\n- dialogue_id (string): dialogue ID\n- goal (string): natural language descriptions of the user goal\n- domains (list of strings): domains mentioned in current dialogue session\n- turn_id (int): turn ID\n- turn_domain (list of strings): domain mentioned in current turn\n- belief_state (dict): dialogue state, including:\n - inform slot-values (dict): the slots and corresponding values informed until current turn\n - turn_inform (dict): the slots and corresponding values informed in current turn\n - turn request (dict): the slots requested in current turn\n- user_actions (list of lists): user dialogue acts in current turn\n- user_actions (list of lists): system dialogue acts in current turn\n- db_results (list of strings): database search results\n- segmented_user_utterance (string): word segmentation result of user utterance\n- segmented_system_utterance (string): word segmentation result of system utterance", "#### Any Quality Control?\n\n\n\nunknown", "### Consent", "#### Any Consent Policy?\n\n\n\nyes", "#### Consent Policy Details\n\n\n\nAnnotators agree using the dataset for research purpose.", "#### Other Consented Downstream Use\n\n\n\nAny", "### Private Identifying Information (PII)", "#### Contains PII?\n\n\n\n\nno PII", "#### Justification for no PII\n\n\n\nThe slots and values as well as utterances do not contain any personal information.", "### Maintenance", "#### Any Maintenance Plan?\n\n\n\nyes", "#### Maintenance Plan Details\n\n\n\nBuilding a leaderboard webpage to trace and display the latest results on the dataset", "#### Maintainer Contact Information\n\n\n\nDeyi Xiong (dyxiong@URL)", "#### Any Contestation Mechanism?\n\n\n\ncontact maintainer", "#### Contestation Form Link\n\n\n\nDeyi Xiong (dyxiong@URL)", "## Broader Social Context", "### Previous Work on the Social Impact of the Dataset", "#### Usage of Models based on the Data\n\n\n\nno", "### Impact on Under-Served Communities", "#### Addresses needs of underserved Communities?\n\n\n\nyes", "#### Details on how Dataset Addresses the Needs\n\n\n\nRiSAWOZ is the first large-scale multi-domain Chinese Wizard-of-Oz dataset with rich semantic annotations.", "### Discussion of Biases", "#### Any Documented Social Biases?\n\n\n\nno", "#### Are the Language Producers Representative of the Language?\n\n\n\nyes", "## Considerations for Using the Data", "### PII Risks and Liability", "#### Potential PII Risk\n\n\n\nNone", "### Licenses", "#### Copyright Restrictions on the Dataset\n\n\n\n'open license - commercial use allowed'", "#### Copyright Restrictions on the Language Data\n\n\n\n'open license - commercial use allowed'", "### Known Technical Limitations", "#### Technical Limitations\n\n\n\nNone", "#### Unsuited Applications\n\n\n\nUsing the trained model on domains that are not included in the 12 domains selected for this dataset.", "#### Discouraged Use Cases\n\n\n\nDesigning models that leverage unknown bias in the dataset to optimize specific metrics." ]

[ "TAGS\n#task_categories-conversational #annotations_creators-crowd-sourced #language_creators-unknown #multilinguality-unknown #size_categories-unknown #source_datasets-original #language-Chinese #license-cc-by-4.0 #dialog-response-generation #region-us \n", "# Dataset Card for GEM/RiSAWOZ", "## Dataset Description\n\n- Homepage: URL\n- Repository: URL\n- Paper: URL\n- Leaderboard: N/A\n- Point of Contact: Deyi Xiong", "### Link to Main Data Card\n\nYou can find the main data card on the GEM Website.", "### Dataset Summary \n\nRiSAWOZ is a Chinese dialog dataset. It can be used to study various dialogue tasks, such as Dialogue State Tracking, Dialogue Context-to-Text Generation, Coreference Resolution and Unified Generative Ellipsis and Coreference Resolution.\n\nYou can load the dataset via:\n\nThe data loader can be found here.", "#### website\nWebsite", "#### paper\nACL Anthology", "#### authors\nJun Quan (Soochow University, Suzhou, China), Shian Zhang (Soochow University, Suzhou, China), Qian Cao(Soochow University, Suzhou, China), Zizhong Li (Tianjin University, Tianjin, China), Deyi Xiong (Tianjin University, Tianjin, China)", "## Dataset Overview", "### Where to find the Data and its Documentation", "#### Webpage\n\n\n\nWebsite", "#### Download\n\n\n\nGithub", "#### Paper\n\n\n\nACL Anthology", "#### BibTex", "#### Contact Name\n\n\n\n\nDeyi Xiong", "#### Contact Email\n\n\n\ndyxiong@URL", "#### Has a Leaderboard?\n\n\n\nno", "### Languages and Intended Use", "#### Multilingual?\n\n\n\n\nno", "#### Covered Dialects\n\n\n\nOnly Mandarin Chinese is covered in this dataset.", "#### Covered Languages\n\n\n\n\n'Mandarin Chinese'", "#### License\n\n\n\n\ncc-by-4.0: Creative Commons Attribution 4.0 International", "#### Intended Use\n\n\n\nRiSAWOZ can be used to support the study under various dialogue tasks, such as Natural Language Understanding, Dialogue State Tracking, Dialogue Context-to-Text Generation, Coreference Resolution and Unified Generative Ellipsis and Coreference Resolution.", "#### Primary Task\n\n\n\nDialog Response Generation", "#### Communicative Goal\n\n\n\n\nGenerate system response given dialogue context across multiple domains.", "### Credit", "#### Curation Organization Type(s)\n\n\n\n'academic'", "#### Curation Organization(s)\n\n\n\nSoochow University and Tianjin University", "#### Dataset Creators\n\n\n\nJun Quan (Soochow University, Suzhou, China), Shian Zhang (Soochow University, Suzhou, China), Qian Cao(Soochow University, Suzhou, China), Zizhong Li (Tianjin University, Tianjin, China), Deyi Xiong (Tianjin University, Tianjin, China)", "#### Funding\n\n\n\nthe National Key Research and Development Project", "#### Who added the Dataset to GEM?\n\n\n\nTianhao Shen (Tianjin University, Tianjin, China), Chaobin You (Tianjin University, Tianjin, China), Deyi Xiong (Tianjin University, Tianjin, China)", "### Dataset Structure", "#### Data Fields\n\n\n\n- gem_id (string): GEM-RiSAWOZ-{split}-{id}\n- dialogue_id (string): dialogue ID\n- goal (string): natural language descriptions of the user goal\n- domains (list of strings): domains mentioned in current dialogue session\n- dialogue (list of dicts): dialog turns and corresponding annotations. Each turn includes:\n - turn_id (int): turn ID\n - turn_domain (list of strings): domain mentioned in current turn\n - user_utterance (string): user utterance\n - system_utterance (string): system utterance\n - belief_state (dict): dialogue state, including:\n - inform slot-values (dict): the slots and corresponding values informed until current turn\n - turn_inform (dict): the slots and corresponding values informed in current turn\n - turn request (dict): the slots requested in current turn\n - user_actions (list of lists): user dialogue acts in current turn\n - user_actions (list of lists): system dialogue acts in current turn\n - db_results (list of strings): database search results\n - segmented_user_utterance (string): word segmentation result of user utterance\n - segmented_system_utterance (string): word segmentation result of system utterance", "#### Example Instance", "#### Data Splits\n\n\n\n- Train:\n - dialogues: 10000\n - turns: 134580\n - Avg. turns: 13.5\n- Valid:\n - dialogues: 600\n - turns: 8116\n - Avg. turns: 13.5\n- Test:\n - dialogues: 600\n - turn: 9286\n - Avg. turns: 15.5", "#### \n\n\n\nthe most complex example (i.e. with the most domains):", "## Dataset in GEM", "### Rationale for Inclusion in GEM", "#### Why is the Dataset in GEM?\n\n\n\nRiSAWOZ is the first large-scale multi-domain Chinese Wizard-of-Oz dataset with rich semantic annotations.", "#### Similar Datasets\n\n\n\nyes", "#### Unique Language Coverage\n\n\n\nno", "#### Difference from other GEM datasets\n\n\n\nThe corpus contains rich semantic annotations, such as ellipsis and coreference, in addition to traditional dialogue annotations (dialogue states, dialogue acts, etc.), which can be used in various tasks in dialogue system.", "#### Ability that the Dataset measures\n\n\n\nNatural Language Understanding, Dialogue State Tracking, Dialogue Context-to-Text Generation, Coreference Resolution, Unified Generative Ellipsis and Coreference Resolution", "### GEM-Specific Curation", "#### Modificatied for GEM?\n\n\n\nno", "#### Additional Splits?\n\n\n\nno", "### Getting Started with the Task", "#### Pointers to Resources\n\n\n\nWebsite", "#### Technical Terms\n\n\n\n- In task-oriented dialogue system, the Natural Language Understanding (NLU) module aims to convert the user utterance into the representation that computer can understand, which includes intent and dialogue act (slot & value) detection.\n- Dialogue State Tracking (DST) is a core component in task-oriented dialogue systems, which extracts dialogue states (user goals) embedded in dialogue context. It has progressed toward open-vocabulary or generation-based DST where state-of-the-art models can generate dialogue states from dialogue context directly.\n- Context-to-Text Generation: encoding dialogue context to decode system response.\n- Coreference Resolution: predict coreference clusters where all mentions are referring to the same entity for each dialogue.\n- Unified Generative Ellipsis and Coreference Resolution: generating omitted or referred expressions from the dialogue context.", "## Previous Results", "### Previous Results", "#### Measured Model Abilities\n\n\n\nNatural Language Understanding, Dialogue State Tracking, Dialogue Context-to-Text Generation, Coreference Resolution, Unified Generative Ellipsis and Coreference Resolution", "#### Metrics\n\n\n\n'Other: Other Metrics'", "#### Other Metrics\n\n\n\n- Natural Language Understanding:\n - F1 score: F1 score of user intent.\n- Dialogue State Tracking:\n - Joint Accuracy: accuracy of turn-level dialogue states.\n- Dialogue Context-to-Text Generation:\n - inform rate: measures the percentage that the output contains the appropriate entity the user asks for.\n - success rate: estimates the proportion that all the requested attributes have been answered.\n - BLEU: the BLEU score of generated system response.\n - Combined Score: (inform + success) ∗ 0.5 + BLEU as an overall\n quality.\n- Coreference Resolution:\n - MUC F1 Score: a link-based metric. Mentions in the same entity/cluster are considered “linked”. MUC penalizes the missing links and incorrect links, each with the same weight.\n - B3 F1 Score: a mention-based metric.The evaluation score depends on the fraction of the correct mentions included in the response entities (i.e. entities created by the system).\n - CEAFφ4 F1 Score: a metric which assumes each key entity should only be mapped to one response entity, and vice versa. It aligns the key entities (clusters) with the response entities in the best way, and compute scores from that alignment.\n - Average F1 Score: an average F1 score of the above three metrics.\n- Unified Generative Ellipsis and Coreference Resolution:\n - Exact Match Rate: measures whether the generated utterances exactly match the ground-truth utterances.\n - BLEU: the BLEU score of generated utterances\n - Resolution F1: comparing machine-generated words with ground-truth words only from the ellipsis/coreference part of user utterances.", "#### Proposed Evaluation\n\n\n\nsee \"Definitions of other metrics\"", "#### Previous results available?\n\n\n\nyes", "#### Other Evaluation Approaches\n\n\n\nsame as our dataset", "#### Relevant Previous Results\n\n\n\nJoint Accuracy, Inform Rate, Success Rate, BLEU Score and Combined Score on MultiWOZ and CrossWOZ dataset.", "## Dataset Curation", "### Original Curation", "#### Original Curation Rationale\n\n\n\nGather human-to-human dialog in Chinese.", "#### Communicative Goal\n\n\n\nGenerate system response given dialogue context across multiple domains.", "#### Sourced from Different Sources\n\n\n\nno", "### Language Data", "#### How was Language Data Obtained?\n\n\n\n'Crowdsourced'", "#### Where was it crowdsourced?\n\n\n\n'Other crowdworker platform'", "#### Topics Covered\n\n\n\ndomains: Attraction, Restaurant, Hotel, Flight, Train, Weather, Movie, TV, Computer, Car, Hospital, Courses", "#### Data Validation\n\n\n\nvalidated by data curator", "#### Was Data Filtered?\n\n\n\nhybrid", "#### Filter Criteria\n\n\n\nRule-based and manual selection criteria", "### Structured Annotations", "#### Additional Annotations?\n\n\n\n\ncrowd-sourced", "#### Number of Raters\n\n\n\n51<n<100", "#### Rater Qualifications\n\n\n\nChinese native speaker", "#### Raters per Training Example\n\n\n\n3", "#### Raters per Test Example\n\n\n\n3", "#### Annotation Service?\n\n\n\nno", "#### Annotation Values\n\n\n\n- dialogue_id (string): dialogue ID\n- goal (string): natural language descriptions of the user goal\n- domains (list of strings): domains mentioned in current dialogue session\n- turn_id (int): turn ID\n- turn_domain (list of strings): domain mentioned in current turn\n- belief_state (dict): dialogue state, including:\n - inform slot-values (dict): the slots and corresponding values informed until current turn\n - turn_inform (dict): the slots and corresponding values informed in current turn\n - turn request (dict): the slots requested in current turn\n- user_actions (list of lists): user dialogue acts in current turn\n- user_actions (list of lists): system dialogue acts in current turn\n- db_results (list of strings): database search results\n- segmented_user_utterance (string): word segmentation result of user utterance\n- segmented_system_utterance (string): word segmentation result of system utterance", "#### Any Quality Control?\n\n\n\nunknown", "### Consent", "#### Any Consent Policy?\n\n\n\nyes", "#### Consent Policy Details\n\n\n\nAnnotators agree using the dataset for research purpose.", "#### Other Consented Downstream Use\n\n\n\nAny", "### Private Identifying Information (PII)", "#### Contains PII?\n\n\n\n\nno PII", "#### Justification for no PII\n\n\n\nThe slots and values as well as utterances do not contain any personal information.", "### Maintenance", "#### Any Maintenance Plan?\n\n\n\nyes", "#### Maintenance Plan Details\n\n\n\nBuilding a leaderboard webpage to trace and display the latest results on the dataset", "#### Maintainer Contact Information\n\n\n\nDeyi Xiong (dyxiong@URL)", "#### Any Contestation Mechanism?\n\n\n\ncontact maintainer", "#### Contestation Form Link\n\n\n\nDeyi Xiong (dyxiong@URL)", "## Broader Social Context", "### Previous Work on the Social Impact of the Dataset", "#### Usage of Models based on the Data\n\n\n\nno", "### Impact on Under-Served Communities", "#### Addresses needs of underserved Communities?\n\n\n\nyes", "#### Details on how Dataset Addresses the Needs\n\n\n\nRiSAWOZ is the first large-scale multi-domain Chinese Wizard-of-Oz dataset with rich semantic annotations.", "### Discussion of Biases", "#### Any Documented Social Biases?\n\n\n\nno", "#### Are the Language Producers Representative of the Language?\n\n\n\nyes", "## Considerations for Using the Data", "### PII Risks and Liability", "#### Potential PII Risk\n\n\n\nNone", "### Licenses", "#### Copyright Restrictions on the Dataset\n\n\n\n'open license - commercial use allowed'", "#### Copyright Restrictions on the Language Data\n\n\n\n'open license - commercial use allowed'", "### Known Technical Limitations", "#### Technical Limitations\n\n\n\nNone", "#### Unsuited Applications\n\n\n\nUsing the trained model on domains that are not included in the 12 domains selected for this dataset.", "#### Discouraged Use Cases\n\n\n\nDesigning models that leverage unknown bias in the dataset to optimize specific metrics." ]

[ "task_categories:table-to-text", "annotations_creators:automatically-created", "language_creators:unknown", "multilinguality:unknown", "size_categories:unknown", "source_datasets:original", "language:en", "language:de", "license:cc-by-4.0", "data-to-text", "region:us" ]

[]

[ "en", "de" ]

[ "# Dataset Card for GEM/RotoWire_English-German", "## Dataset Description\n\n- Homepage: URL\n- Repository: URL\n- Paper: URL\n- Leaderboard: N/A\n- Point of Contact: Hiroaki Hayashi", "### Link to Main Data Card\n\nYou can find the main data card on the GEM Website.", "### Dataset Summary \n\nThis dataset is a data-to-text dataset in the basketball domain. The input are tables in a fixed format with statistics about a game (in English) and the target is a German translation of the originally English description. The translations were done by professional translators with basketball experience. The dataset can be used to evaluate the cross-lingual data-to-text capabilities of a model with complex inputs. \n\nYou can load the dataset via:\n\nThe data loader can be found here.", "#### website\nWebsite", "#### paper\nACL Anthology", "#### authors\nGraham Neubig (Carnegie Mellon University), Hiroaki Hayashi (Carnegie Mellon University)", "## Dataset Overview", "### Where to find the Data and its Documentation", "#### Webpage\n\n\n\nWebsite", "#### Download\n\n\n\nGithub", "#### Paper\n\n\n\nACL Anthology", "#### BibTex", "#### Contact Name\n\n\n\n\nHiroaki Hayashi", "#### Contact Email\n\n\n\nhiroakih@URL", "#### Has a Leaderboard?\n\n\n\nno", "### Languages and Intended Use", "#### Multilingual?\n\n\n\n\nyes", "#### Covered Languages\n\n\n\n\n'English', 'German'", "#### License\n\n\n\n\ncc-by-4.0: Creative Commons Attribution 4.0 International", "#### Intended Use\n\n\n\nFoster the research on document-level generation technology and contrast the methods for different types of inputs.", "#### Primary Task\n\n\n\nData-to-Text", "#### Communicative Goal\n\n\n\n\nDescribe a basketball game given its box score table (and possibly a summary in a foreign language).", "### Credit", "#### Curation Organization Type(s)\n\n\n\n'academic'", "#### Curation Organization(s)\n\n\n\nCarnegie Mellon University", "#### Dataset Creators\n\n\n\nGraham Neubig (Carnegie Mellon University), Hiroaki Hayashi (Carnegie Mellon University)", "#### Funding\n\n\n\nGraham Neubig", "#### Who added the Dataset to GEM?\n\n\n\nHiroaki Hayashi (Carnegie Mellon University)", "### Dataset Structure", "#### Data Fields\n\n\n\n- 'id' ('string'): The identifier from the original dataset.\n- 'gem_id' ('string'): The identifier from GEMv2.\n- 'day' ('string'): Date of the game (Format: 'MM_DD_YY')\n- 'home_name' ('string'): Home team name.\n- 'home_city' ('string'): Home team city name.\n- 'vis_name' ('string'): Visiting (Away) team name.\n- 'vis_city' ('string'): Visiting team (Away) city name.\n- 'home_line' ('Dict[str, str]'): Home team statistics (e.g., team free throw percentage).\n- 'vis_line' ('Dict[str, str]'): Visiting team statistics (e.g., team free throw percentage).\n- 'box_score' ('Dict[str, Dict[str, str]]'): Box score table. (Stat_name to [player ID to stat_value].)\n- 'summary_en' ('List[string]'): Tokenized target summary in English.\n- 'sentence_end_index_en' ('List[int]'): Sentence end indices for 'summary_en'.\n- 'summary_de' ('List[string]'): Tokenized target summary in German.\n- 'sentence_end_index_de' ('List[int]'): ): Sentence end indices for 'summary_de'.\n- (Unused) 'detok_summary_org' ('string'): Original summary provided by RotoWire dataset.\n- (Unused) 'summary' ('List[string]'): Tokenized summary of 'detok_summary_org'.\n- (Unused) 'detok_summary' ('string'): Detokenized (with organizer's detokenizer) summary of 'summary'.", "#### Reason for Structure\n\n\n\n- Structured data are directly imported from the original RotoWire dataset.\n- Textual data (English, German) are associated to each sample.", "#### Example Instance", "#### Data Splits\n\n\n\n- Train\n- Validation\n- Test", "#### Splitting Criteria\n\n\n\n- English summaries are provided sentence-by-sentence to professional German translators with basketball knowledge to obtain sentence-level German translations.\n- Split criteria follows the original RotoWire dataset.", "#### \n\n\n\n- The (English) summary length in the training set varies from 145 to 650 words, with an average of 323 words.", "## Dataset in GEM", "### Rationale for Inclusion in GEM", "#### Why is the Dataset in GEM?\n\n\n\nThe use of two modalities (data, foreign text) to generate a document-level text summary.", "#### Similar Datasets\n\n\n\nyes", "#### Unique Language Coverage\n\n\n\nyes", "#### Difference from other GEM datasets\n\n\n\nThe potential use of two modalities (data, foreign text) as input.", "#### Ability that the Dataset measures\n\n\n\n- Translation\n- Data-to-text verbalization\n- Aggregation of the two above.", "### GEM-Specific Curation", "#### Modificatied for GEM?\n\n\n\nyes", "#### GEM Modifications\n\n\n\n'other'", "#### Modification Details\n\n\n\n- Added GEM ID in each sample.\n- Normalize the number of players in each sample with \"N/A\" for consistent data loading.", "#### Additional Splits?\n\n\n\nno", "### Getting Started with the Task", "#### Pointers to Resources\n\n\n\n- Challenges in Data-to-Document Generation\n- Data-to-Text Generation with Content Selection and Planning\n- Findings of the Third Workshop on Neural Generation and Translation", "#### Technical Terms\n\n\n\n- Data-to-text\n- Neural machine translation (NMT)\n- Document-level generation and translation (DGT)", "## Previous Results", "### Previous Results", "#### Measured Model Abilities\n\n\n\n- Textual accuracy towards the gold-standard summary.\n- Content faithfulness to the input structured data.", "#### Metrics\n\n\n\n'BLEU', 'ROUGE', 'Other: Other Metrics'", "#### Other Metrics\n\n\n\nModel-based measures proposed by (Wiseman et al., 2017):\n- Relation Generation\n- Content Selection\n- Content Ordering", "#### Proposed Evaluation\n\n\n\nTo evaluate the fidelity of the generated content to the input data.", "#### Previous results available?\n\n\n\nyes", "#### Other Evaluation Approaches\n\n\n\nN/A.", "#### Relevant Previous Results\n\n\n\nSee Table 2 to 7 of (URL for previous results for this dataset.", "## Dataset Curation", "### Original Curation", "#### Original Curation Rationale\n\n\n\nA random subset of RotoWire dataset was chosen for German translation annotation.", "#### Communicative Goal\n\n\n\nFoster the research on document-level generation technology and contrast the methods for different types of inputs.", "#### Sourced from Different Sources\n\n\n\nyes", "#### Source Details\n\n\n\nRotoWire", "### Language Data", "#### How was Language Data Obtained?\n\n\n\n'Created for the dataset'", "#### Creation Process\n\n\n\nProfessional German language translators were hired to translate basketball summaries from a subset of RotoWire dataset.", "#### Language Producers\n\n\n\nTranslators are familiar with basketball terminology.", "#### Topics Covered\n\n\n\nBasketball (NBA) game summaries.", "#### Data Validation\n\n\n\nvalidated by data curator", "#### Data Preprocessing\n\n\n\nSentence-level translations were aligned back to the original English summary sentences.", "#### Was Data Filtered?\n\n\n\nnot filtered", "### Structured Annotations", "#### Additional Annotations?\n\n\n\n\nautomatically created", "#### Annotation Service?\n\n\n\nno", "#### Annotation Values\n\n\n\nSentence-end indices for the tokenized summaries. Sentence boundaries can help users accurately identify aligned sentences in both languages, as well as allowing an accurate evaluation that involves sentence boundaries (ROUGE-L).", "#### Any Quality Control?\n\n\n\nvalidated through automated script", "#### Quality Control Details\n\n\n\nToken and number overlaps between pairs of aligned sentences are measured.", "### Consent", "#### Any Consent Policy?\n\n\n\nno", "#### Justification for Using the Data\n\n\n\nReusing by citing the original papers:\n- Sam Wiseman, Stuart M. Shieber, Alexander M. Rush:\nChallenges in Data-to-Document Generation. EMNLP 2017.\n- Hiroaki Hayashi, Yusuke Oda, Alexandra Birch, Ioannis Konstas, Andrew Finch, Minh-Thang Luong, Graham Neubig, Katsuhito Sudoh. Findings of the Third Workshop on Neural Generation and Translation. WNGT 2019.", "### Private Identifying Information (PII)", "#### Contains PII?\n\n\n\n\nunlikely", "#### Categories of PII\n\n\n\n'generic PII'", "#### Any PII Identification?\n\n\n\nno identification", "### Maintenance", "#### Any Maintenance Plan?\n\n\n\nno", "## Broader Social Context", "### Previous Work on the Social Impact of the Dataset", "#### Usage of Models based on the Data\n\n\n\nno", "### Impact on Under-Served Communities", "#### Addresses needs of underserved Communities?\n\n\n\nno", "### Discussion of Biases", "#### Any Documented Social Biases?\n\n\n\nno", "#### Are the Language Producers Representative of the Language?\n\n\n\n- English text in this dataset is from Rotowire, originally written by writers at URL that are likely US-based.\n- German text is produced by professional translators proficient in both English and German.", "## Considerations for Using the Data", "### PII Risks and Liability", "#### Potential PII Risk\n\n\n\n- Structured data contain real National Basketball Association player and organization names.", "### Licenses", "#### Copyright Restrictions on the Dataset\n\n\n\n'open license - commercial use allowed'", "#### Copyright Restrictions on the Language Data\n\n\n\n'open license - commercial use allowed'", "### Known Technical Limitations", "#### Technical Limitations\n\n\n\nPotential overlap of box score tables between splits. This was extensively studied and pointed out by [1].\n\n[1]: Thomson, Craig, Ehud Reiter, and Somayajulu Sripada. \"SportSett: Basketball-A robust and maintainable data-set for Natural Language Generation.\" Proceedings of the Workshop on Intelligent Information Processing and Natural Language Generation. 2020.", "#### Unsuited Applications\n\n\n\nUsers may interact with a trained model to learn about a NBA game in a textual manner. On generated texts, they may observe factual errors that contradicts the actual data that the model conditions on.\nFactual errors include wrong statistics of a player (e.g., 3PT), non-existent injury information.", "#### Discouraged Use Cases\n\n\n\nPublishing the generated text as is. Even if the model achieves high scores on the evaluation metrics, there is a risk of factual errors mentioned above." ]

[ "TAGS\n#task_categories-table-to-text #annotations_creators-automatically-created #language_creators-unknown #multilinguality-unknown #size_categories-unknown #source_datasets-original #language-English #language-German #license-cc-by-4.0 #data-to-text #region-us \n", "# Dataset Card for GEM/RotoWire_English-German", "## Dataset Description\n\n- Homepage: URL\n- Repository: URL\n- Paper: URL\n- Leaderboard: N/A\n- Point of Contact: Hiroaki Hayashi", "### Link to Main Data Card\n\nYou can find the main data card on the GEM Website.", "### Dataset Summary \n\nThis dataset is a data-to-text dataset in the basketball domain. The input are tables in a fixed format with statistics about a game (in English) and the target is a German translation of the originally English description. The translations were done by professional translators with basketball experience. The dataset can be used to evaluate the cross-lingual data-to-text capabilities of a model with complex inputs. \n\nYou can load the dataset via:\n\nThe data loader can be found here.", "#### website\nWebsite", "#### paper\nACL Anthology", "#### authors\nGraham Neubig (Carnegie Mellon University), Hiroaki Hayashi (Carnegie Mellon University)", "## Dataset Overview", "### Where to find the Data and its Documentation", "#### Webpage\n\n\n\nWebsite", "#### Download\n\n\n\nGithub", "#### Paper\n\n\n\nACL Anthology", "#### BibTex", "#### Contact Name\n\n\n\n\nHiroaki Hayashi", "#### Contact Email\n\n\n\nhiroakih@URL", "#### Has a Leaderboard?\n\n\n\nno", "### Languages and Intended Use", "#### Multilingual?\n\n\n\n\nyes", "#### Covered Languages\n\n\n\n\n'English', 'German'", "#### License\n\n\n\n\ncc-by-4.0: Creative Commons Attribution 4.0 International", "#### Intended Use\n\n\n\nFoster the research on document-level generation technology and contrast the methods for different types of inputs.", "#### Primary Task\n\n\n\nData-to-Text", "#### Communicative Goal\n\n\n\n\nDescribe a basketball game given its box score table (and possibly a summary in a foreign language).", "### Credit", "#### Curation Organization Type(s)\n\n\n\n'academic'", "#### Curation Organization(s)\n\n\n\nCarnegie Mellon University", "#### Dataset Creators\n\n\n\nGraham Neubig (Carnegie Mellon University), Hiroaki Hayashi (Carnegie Mellon University)", "#### Funding\n\n\n\nGraham Neubig", "#### Who added the Dataset to GEM?\n\n\n\nHiroaki Hayashi (Carnegie Mellon University)", "### Dataset Structure", "#### Data Fields\n\n\n\n- 'id' ('string'): The identifier from the original dataset.\n- 'gem_id' ('string'): The identifier from GEMv2.\n- 'day' ('string'): Date of the game (Format: 'MM_DD_YY')\n- 'home_name' ('string'): Home team name.\n- 'home_city' ('string'): Home team city name.\n- 'vis_name' ('string'): Visiting (Away) team name.\n- 'vis_city' ('string'): Visiting team (Away) city name.\n- 'home_line' ('Dict[str, str]'): Home team statistics (e.g., team free throw percentage).\n- 'vis_line' ('Dict[str, str]'): Visiting team statistics (e.g., team free throw percentage).\n- 'box_score' ('Dict[str, Dict[str, str]]'): Box score table. (Stat_name to [player ID to stat_value].)\n- 'summary_en' ('List[string]'): Tokenized target summary in English.\n- 'sentence_end_index_en' ('List[int]'): Sentence end indices for 'summary_en'.\n- 'summary_de' ('List[string]'): Tokenized target summary in German.\n- 'sentence_end_index_de' ('List[int]'): ): Sentence end indices for 'summary_de'.\n- (Unused) 'detok_summary_org' ('string'): Original summary provided by RotoWire dataset.\n- (Unused) 'summary' ('List[string]'): Tokenized summary of 'detok_summary_org'.\n- (Unused) 'detok_summary' ('string'): Detokenized (with organizer's detokenizer) summary of 'summary'.", "#### Reason for Structure\n\n\n\n- Structured data are directly imported from the original RotoWire dataset.\n- Textual data (English, German) are associated to each sample.", "#### Example Instance", "#### Data Splits\n\n\n\n- Train\n- Validation\n- Test", "#### Splitting Criteria\n\n\n\n- English summaries are provided sentence-by-sentence to professional German translators with basketball knowledge to obtain sentence-level German translations.\n- Split criteria follows the original RotoWire dataset.", "#### \n\n\n\n- The (English) summary length in the training set varies from 145 to 650 words, with an average of 323 words.", "## Dataset in GEM", "### Rationale for Inclusion in GEM", "#### Why is the Dataset in GEM?\n\n\n\nThe use of two modalities (data, foreign text) to generate a document-level text summary.", "#### Similar Datasets\n\n\n\nyes", "#### Unique Language Coverage\n\n\n\nyes", "#### Difference from other GEM datasets\n\n\n\nThe potential use of two modalities (data, foreign text) as input.", "#### Ability that the Dataset measures\n\n\n\n- Translation\n- Data-to-text verbalization\n- Aggregation of the two above.", "### GEM-Specific Curation", "#### Modificatied for GEM?\n\n\n\nyes", "#### GEM Modifications\n\n\n\n'other'", "#### Modification Details\n\n\n\n- Added GEM ID in each sample.\n- Normalize the number of players in each sample with \"N/A\" for consistent data loading.", "#### Additional Splits?\n\n\n\nno", "### Getting Started with the Task", "#### Pointers to Resources\n\n\n\n- Challenges in Data-to-Document Generation\n- Data-to-Text Generation with Content Selection and Planning\n- Findings of the Third Workshop on Neural Generation and Translation", "#### Technical Terms\n\n\n\n- Data-to-text\n- Neural machine translation (NMT)\n- Document-level generation and translation (DGT)", "## Previous Results", "### Previous Results", "#### Measured Model Abilities\n\n\n\n- Textual accuracy towards the gold-standard summary.\n- Content faithfulness to the input structured data.", "#### Metrics\n\n\n\n'BLEU', 'ROUGE', 'Other: Other Metrics'", "#### Other Metrics\n\n\n\nModel-based measures proposed by (Wiseman et al., 2017):\n- Relation Generation\n- Content Selection\n- Content Ordering", "#### Proposed Evaluation\n\n\n\nTo evaluate the fidelity of the generated content to the input data.", "#### Previous results available?\n\n\n\nyes", "#### Other Evaluation Approaches\n\n\n\nN/A.", "#### Relevant Previous Results\n\n\n\nSee Table 2 to 7 of (URL for previous results for this dataset.", "## Dataset Curation", "### Original Curation", "#### Original Curation Rationale\n\n\n\nA random subset of RotoWire dataset was chosen for German translation annotation.", "#### Communicative Goal\n\n\n\nFoster the research on document-level generation technology and contrast the methods for different types of inputs.", "#### Sourced from Different Sources\n\n\n\nyes", "#### Source Details\n\n\n\nRotoWire", "### Language Data", "#### How was Language Data Obtained?\n\n\n\n'Created for the dataset'", "#### Creation Process\n\n\n\nProfessional German language translators were hired to translate basketball summaries from a subset of RotoWire dataset.", "#### Language Producers\n\n\n\nTranslators are familiar with basketball terminology.", "#### Topics Covered\n\n\n\nBasketball (NBA) game summaries.", "#### Data Validation\n\n\n\nvalidated by data curator", "#### Data Preprocessing\n\n\n\nSentence-level translations were aligned back to the original English summary sentences.", "#### Was Data Filtered?\n\n\n\nnot filtered", "### Structured Annotations", "#### Additional Annotations?\n\n\n\n\nautomatically created", "#### Annotation Service?\n\n\n\nno", "#### Annotation Values\n\n\n\nSentence-end indices for the tokenized summaries. Sentence boundaries can help users accurately identify aligned sentences in both languages, as well as allowing an accurate evaluation that involves sentence boundaries (ROUGE-L).", "#### Any Quality Control?\n\n\n\nvalidated through automated script", "#### Quality Control Details\n\n\n\nToken and number overlaps between pairs of aligned sentences are measured.", "### Consent", "#### Any Consent Policy?\n\n\n\nno", "#### Justification for Using the Data\n\n\n\nReusing by citing the original papers:\n- Sam Wiseman, Stuart M. Shieber, Alexander M. Rush:\nChallenges in Data-to-Document Generation. EMNLP 2017.\n- Hiroaki Hayashi, Yusuke Oda, Alexandra Birch, Ioannis Konstas, Andrew Finch, Minh-Thang Luong, Graham Neubig, Katsuhito Sudoh. Findings of the Third Workshop on Neural Generation and Translation. WNGT 2019.", "### Private Identifying Information (PII)", "#### Contains PII?\n\n\n\n\nunlikely", "#### Categories of PII\n\n\n\n'generic PII'", "#### Any PII Identification?\n\n\n\nno identification", "### Maintenance", "#### Any Maintenance Plan?\n\n\n\nno", "## Broader Social Context", "### Previous Work on the Social Impact of the Dataset", "#### Usage of Models based on the Data\n\n\n\nno", "### Impact on Under-Served Communities", "#### Addresses needs of underserved Communities?\n\n\n\nno", "### Discussion of Biases", "#### Any Documented Social Biases?\n\n\n\nno", "#### Are the Language Producers Representative of the Language?\n\n\n\n- English text in this dataset is from Rotowire, originally written by writers at URL that are likely US-based.\n- German text is produced by professional translators proficient in both English and German.", "## Considerations for Using the Data", "### PII Risks and Liability", "#### Potential PII Risk\n\n\n\n- Structured data contain real National Basketball Association player and organization names.", "### Licenses", "#### Copyright Restrictions on the Dataset\n\n\n\n'open license - commercial use allowed'", "#### Copyright Restrictions on the Language Data\n\n\n\n'open license - commercial use allowed'", "### Known Technical Limitations", "#### Technical Limitations\n\n\n\nPotential overlap of box score tables between splits. This was extensively studied and pointed out by [1].\n\n[1]: Thomson, Craig, Ehud Reiter, and Somayajulu Sripada. \"SportSett: Basketball-A robust and maintainable data-set for Natural Language Generation.\" Proceedings of the Workshop on Intelligent Information Processing and Natural Language Generation. 2020.", "#### Unsuited Applications\n\n\n\nUsers may interact with a trained model to learn about a NBA game in a textual manner. On generated texts, they may observe factual errors that contradicts the actual data that the model conditions on.\nFactual errors include wrong statistics of a player (e.g., 3PT), non-existent injury information.", "#### Discouraged Use Cases\n\n\n\nPublishing the generated text as is. Even if the model achieves high scores on the evaluation metrics, there is a risk of factual errors mentioned above." ]

[ "task_categories:text2text-generation", "task_ids:text-simplification", "annotations_creators:crowd-sourced", "language_creators:unknown", "multilinguality:unknown", "size_categories:unknown", "source_datasets:original", "language:it", "license:cc-by-4.0", "region:us" ]

[]

[ "it" ]

[ "# Dataset Card for GEM/SIMPITIKI", "## Dataset Description\n\n- Homepage: URL\n- Repository: URL\n- Paper: URL\n- Leaderboard: N/A\n- Point of Contact: Sara Tonelli", "### Link to Main Data Card\n\nYou can find the main data card on the GEM Website.", "### Dataset Summary \n\nSIMPITIKI is an Italian Simplification dataset. Its examples were selected from Italian Wikipedia such that their editing tracking descriptions contain any of the words \"Simplified\"/\"Simplify\"/\"Simplification\". \n\nYou can load the dataset via:\n\nThe data loader can be found here.", "#### website\nGithub", "#### paper\nWebsite", "#### authors\nSara Tonelli (Fondazione Bruno Kessler), Alessio Palmero Aprosio (Fondazione Bruno Kessler), Francesca Saltori (Fondazione Bruno Kessler)", "## Dataset Overview", "### Where to find the Data and its Documentation", "#### Webpage\n\n\n\nGithub", "#### Download\n\n\n\nGithub", "#### Paper\n\n\n\nWebsite", "#### BibTex", "#### Contact Name\n\n\n\n\nSara Tonelli", "#### Contact Email\n\n\n\nsatonelli@URL", "#### Has a Leaderboard?\n\n\n\nno", "### Languages and Intended Use", "#### Multilingual?\n\n\n\n\nno", "#### Covered Dialects\n\n\n\nNone", "#### Covered Languages\n\n\n\n\n'Italian'", "#### License\n\n\n\n\ncc-by-4.0: Creative Commons Attribution 4.0 International", "#### Intended Use\n\n\n\nThe purpose of the dataset is to train NLG models to simplify complex text by learning different types of transformations (verb to noun, noun to verbs, deletion, insertion, etc)", "#### Primary Task\n\n\n\nSimplification", "#### Communicative Goal\n\n\n\n\nThis dataset aims to enhance research in text simplification in Italian language with different text transformations.", "### Credit", "#### Curation Organization Type(s)\n\n\n\n'academic', 'independent'", "#### Curation Organization(s)\n\n\n\nFondazione Bruno Kessler (FBK)", "#### Dataset Creators\n\n\n\nSara Tonelli (Fondazione Bruno Kessler), Alessio Palmero Aprosio (Fondazione Bruno Kessler), Francesca Saltori (Fondazione Bruno Kessler)", "#### Funding\n\n\n\nEU Horizon 2020 Programme via the SIMPATICO Project (H2020-EURO-6-2015, n. 692819)", "#### Who added the Dataset to GEM?\n\n\n\nSebastien Montella (Orange Labs), Vipul Raheja (Grammarly Inc.)", "### Dataset Structure", "#### Data Fields\n\n\n\nEach sample comes with the following fields:\n\n- 'gem_id' (string): Unique sample ID\n-'text' (string): The raw text to be simplified\n-'simplified_text' (string): The simplified version of \"text\" field\n-'transformation_type' (string): Nature of transformation applied to raw text in order to simplify it.\n-'source_dataset' (string): Initial dataset source of sample. Values: 'itwiki' (for Italian Wikipedia) or 'tn' (manually annotated administrative documents from the Municipality of Trento, Italy)", "#### Reason for Structure\n\n\n\nThe dataset is organized as a pairs where the raw text (input) is associated with its simplified text (output). The editing transformation and the source dataset of each sample is also provided for advanced analysis.", "#### How were labels chosen?\n\n\n\nSIMPITIKI dataset selects documents from Italian Wikipedia such that their editing tracking descriptions contain any of the words \"Simplified\"/\"Simplify\"/\"Simplification\". For the Public Administration domain of the documents of the Municipality of Trento (Italy)", "#### Example Instance", "#### Data Splits\n\n\n\nSeveral splits are proposed to train models on different configurations:\n\n-\"train\": Training samples randomly selected from initial corpus. 816 training samples.\n-\"validation\": Validating samples randomly selected from initial corpus. 174 validating samples.\n-\"test\": Testing samples randomly selected from initial corpus. 176 validating samples.\n-\"challenge_seen_transformations_train\": This training challenge split includes specific transformations to simplify the raw text. Precisely, transformations are \"Split\", \"Merge\", \"Reordering\", \"Insert - Verb\", \"Insert - Other\", \"Delete - Verb\", \"Delete - Other\", \"Transformation - Lexical Substitution (word level)\", \"Transformation - Anaphoric replacement\", \"Transformation - Noun to Verb\", \"Transformation - Verbal Features\". 562 training samples.\n-\"challenge_seen_transformations_val\": This validating challenge split includes same transformations than the ones observed in training. Precisely, transformations are \"Split\", \"Merge\", \"Reordering\", \"Insert - Verb\", \"Insert - Other\", \"Delete - Verb\", \"Delete - Other\", \"Transformation - Lexical Substitution (word level)\", \"Transformation - Anaphoric replacement\", \"Transformation - Noun to Verb\", \"Transformation - Verbal Features\". 121 validating samples.\n-\"challenge_seen_transformations_test\": This testing challenge split includes same transformations than the ones observed in training. Precisely, transformations are \"Split\", \"Merge\", \"Reordering\", \"Insert - Verb\", \"Insert - Other\", \"Delete - Verb\", \"Delete - Other\", \"Transformation - Lexical Substitution (word level)\", \"Transformation - Anaphoric replacement\", \"Transformation - Noun to Verb\", \"Transformation - Verbal Features\". 127 testing samples.\n-\"challenge_unseen_transformations_test\" : \"Insert - Subject\", \"Delete - Subject\", \"Transformation - Lexical Substitution (phrase level)\", \"Transformation - Verb to Noun (nominalization)\", \"Transformation - Verbal Voice\". 356 testing samples.\n-\"challenge_itwiki_train\": This challenge split includes random samples from the Italian Wikipedia as source dataset. 402 training samples.\n-\"challenge_itwiki_val\": This validating challenge split includes random samples from the Italian Wikipedia as source dataset. 86 validating samples.\n-\"challenge_itwiki_test\": This testing challenge split includes random samples from the Italian Wikipedia as source dataset. 87 testing samples.\n-\"challenge_tn_test\": This testing challenge split includes all samples from the Municipality of Trento administrative documents ('tn') as source dataset. 591 testing samples.", "#### Splitting Criteria\n\n\n\nThe training ratio is set to 0.7. The validation and test somehow equally divide the remaining 30% of the dataset.", "## Dataset in GEM", "### Rationale for Inclusion in GEM", "#### Why is the Dataset in GEM?\n\n\n\nThis dataset promotes Simplification task for Italian language.", "#### Similar Datasets\n\n\n\nno", "#### Ability that the Dataset measures\n\n\n\nModels can be evaluated if they can simplify text regarding different simplification transformations.", "### GEM-Specific Curation", "#### Modificatied for GEM?\n\n\n\nyes", "#### Additional Splits?\n\n\n\nyes", "#### Split Information\n\n\n\nThe SIMPITIKI dataset provides a single file. Several splits are proposed to train models on different configurations:\n-\"train\": Training samples randomly selected from initial corpus. 816 training samples.\n-\"validation\": Validating samples randomly selected from initial corpus. 174 validating samples.\n-\"test\": Testing samples randomly selected from initial corpus. 176 validating samples.\n-\"challenge_seen_transformations_train\": This training challenge split includes specific transformations to simplify the raw text. Precisely, transformations are \"Split\", \"Merge\", \"Reordering\", \"Insert - Verb\", \"Insert - Other\", \"Delete - Verb\", \"Delete - Other\", \"Transformation - Lexical Substitution (word level)\", \"Transformation - Anaphoric replacement\", \"Transformation - Noun to Verb\", \"Transformation - Verbal Features\". 562 training samples.\n-\"challenge_seen_transformations_val\": This validating challenge split includes same transformations than the ones observed in training. Precisely, transformations are \"Split\", \"Merge\", \"Reordering\", \"Insert - Verb\", \"Insert - Other\", \"Delete - Verb\", \"Delete - Other\", \"Transformation - Lexical Substitution (word level)\", \"Transformation - Anaphoric replacement\", \"Transformation - Noun to Verb\", \"Transformation - Verbal Features\". 121 validating samples.\n-\"challenge_seen_transformations_test\": This testing challenge split includes same transformations than the ones observed in training. Precisely, transformations are \"Split\", \"Merge\", \"Reordering\", \"Insert - Verb\", \"Insert - Other\", \"Delete - Verb\", \"Delete - Other\", \"Transformation - Lexical Substitution (word level)\", \"Transformation - Anaphoric replacement\", \"Transformation - Noun to Verb\", \"Transformation - Verbal Features\". 127 testing samples.\n-\"challenge_unseen_transformations_test\" : \"Insert - Subject\", \"Delete - Subject\", \"Transformation - Lexical Substitution (phrase level)\", \"Transformation - Verb to Noun (nominalization)\", \"Transformation - Verbal Voice\". 356 testing samples.\n-\"challenge_itwiki_train\": This challenge split includes random samples from the Italian Wikipedia as source dataset. 402 training samples.\n-\"challenge_itwiki_val\": This validating challenge split includes random samples from the Italian Wikipedia as source dataset. 86 validating samples.\n-\"challenge_itwiki_test\": This testing challenge split includes random samples from the Italian Wikipedia as source dataset. 87 testing samples.\n-\"challenge_tn_test\": This testing challenge split includes all samples from the Municipality of Trento administrative documents ('tn') as source dataset. 591 testing samples.", "#### Split Motivation\n\n\n\nThe splits allows to investigate the generalization of models regarding editing/transformations (\"challenge_seen_transformations_test\" / \"challenge_unseen_transformations_test\") and for transfer learning to different domain (\"challenge_tn_test\")", "### Getting Started with the Task", "#### Pointers to Resources\n\n\n\n- Coster and Kauchak, Simple English Wikipedia: A New Text Simplification Task, Proceedings of the 49th Annual Meeting of the Association for Computational Linguistics, pages 665–669, Portland, Oregon, June 19-24, 2011\n- Xu et al, Optimizing Statistical Machine Translation for Text Simplification, Transactions of the Association for Computational Linguistics, vol. 4, pp. 401–415, 2016\n- Aprosio et al, Neural Text Simplification in Low-Resource Conditions Using Weak Supervision, Proceedings of the Workshop on Methods for Optimizing and Evaluating Neural Language Generation (NeuralGen), pages 37–44, Minneapolis, Minnesota, USA, June 6, 2019", "#### Technical Terms\n\n\n\nSimplification: Process that consists in transforming an input text to its simplified version.", "## Previous Results", "### Previous Results", "#### Measured Model Abilities\n\n\n\nThe splits allows to investigate the generalization of models regarding editing/transformations (\"challenge_seen_transformations_test\" / \"challenge_unseen_transformations_test\") and for transfer learning to different domain (\"challenge_tn_test\")", "#### Metrics\n\n\n\n'BLEU', 'Other: Other Metrics'", "#### Other Metrics\n\n\n\nFKBLEU (URL Combines Flesch-Kincaid Index and iBLEU metrics.\nSARI (URL Compares system output against references and against the input sentence. It explicitly measures the goodness of words that are added, deleted and kept by the systems \nWord-level F1", "#### Previous results available?\n\n\n\nno", "## Dataset Curation", "### Original Curation", "#### Original Curation Rationale\n\n\n\nMost of the resources for Text Simplification are in English. To stimulate research to different languages, SIMPITIKI proposes an Italian corpus with Complex-Simple sentence pairs.", "#### Communicative Goal\n\n\n\nText simplification allows a smooth reading of text to enhance understanding.", "#### Sourced from Different Sources\n\n\n\nyes", "#### Source Details\n\n\n\nItalian Wikipedia\n(Manually) Annotated administrative documents from the Municipality of Trento, Italy", "### Language Data", "#### How was Language Data Obtained?\n\n\n\n'Found'", "#### Where was it found?\n\n\n\n'Single website', 'Offline media collection'", "#### Language Producers\n\n\n\nSIMPITIKI is a combination of documents from Italian Wikipedia and from the Municipality of Trento, Italy.", "#### Topics Covered\n\n\n\nSamples from documents from the Municipality of Trento corpus are in the administrative domain.", "#### Data Validation\n\n\n\nvalidated by data curator", "#### Was Data Filtered?\n\n\n\nnot filtered", "### Structured Annotations", "#### Additional Annotations?\n\n\n\n\ncrowd-sourced", "#### Number of Raters\n\n\n\nunknown", "#### Rater Qualifications\n\n\n\nNative speaker", "#### Raters per Training Example\n\n\n\n0", "#### Raters per Test Example\n\n\n\n0", "#### Annotation Service?\n\n\n\nunknown", "#### Annotation Values\n\n\n\nAnnotators specified any of the tags as designed by Brunato et al. (URL\n-Split: Splitting a clause into two clauses.\n-Merge: Merge two or more clauses together.\n-Reordering: Word order changes.\n-Insert: Insertion of words or phrases that provide supportive information to the original sentence\n-Delete: dropping redundant information.\n-Transformation: Modification which can affect the sentence at the lexical, morpho-syntactic and syntactic level: Lexical substitution (word level) / Lexical substitution (phrase level) / Anaphoric replacement / Noun to Verb / Verb to Noun / Verbal voice / Verbal features/ morpho–syntactic and syntactic level, also giving rise to overlapping phenomena", "#### Any Quality Control?\n\n\n\nunknown", "### Consent", "#### Any Consent Policy?\n\n\n\nno", "#### Justification for Using the Data\n\n\n\nThe dataset is available online under the CC-BY 4.0 license.", "### Private Identifying Information (PII)", "#### Contains PII?\n\n\n\n\nlikely", "#### Categories of PII\n\n\n\n'generic PII'", "#### Any PII Identification?\n\n\n\nno identification", "### Maintenance", "#### Any Maintenance Plan?\n\n\n\nno", "## Broader Social Context", "### Previous Work on the Social Impact of the Dataset", "#### Usage of Models based on the Data\n\n\n\nno", "### Impact on Under-Served Communities", "#### Addresses needs of underserved Communities?\n\n\n\nyes", "#### Details on how Dataset Addresses the Needs\n\n\n\nThe creator of SIMPITIKI wants to promote text simplification for Italian because few resources are available in other languages than English.", "### Discussion of Biases", "#### Any Documented Social Biases?\n\n\n\nunsure", "## Considerations for Using the Data", "### PII Risks and Liability", "### Licenses", "#### Copyright Restrictions on the Dataset\n\n\n\n'research use only'", "#### Copyright Restrictions on the Language Data\n\n\n\n'research use only'", "### Known Technical Limitations", "#### Discouraged Use Cases\n\n\n\nThe risk of surface-based metrics (BLEU, chrf++, etc) for this task is that semantic adequacy is not respected when simplifying the input document." ]

[ "TAGS\n#task_categories-text2text-generation #task_ids-text-simplification #annotations_creators-crowd-sourced #language_creators-unknown #multilinguality-unknown #size_categories-unknown #source_datasets-original #language-Italian #license-cc-by-4.0 #region-us \n", "# Dataset Card for GEM/SIMPITIKI", "## Dataset Description\n\n- Homepage: URL\n- Repository: URL\n- Paper: URL\n- Leaderboard: N/A\n- Point of Contact: Sara Tonelli", "### Link to Main Data Card\n\nYou can find the main data card on the GEM Website.", "### Dataset Summary \n\nSIMPITIKI is an Italian Simplification dataset. Its examples were selected from Italian Wikipedia such that their editing tracking descriptions contain any of the words \"Simplified\"/\"Simplify\"/\"Simplification\". \n\nYou can load the dataset via:\n\nThe data loader can be found here.", "#### website\nGithub", "#### paper\nWebsite", "#### authors\nSara Tonelli (Fondazione Bruno Kessler), Alessio Palmero Aprosio (Fondazione Bruno Kessler), Francesca Saltori (Fondazione Bruno Kessler)", "## Dataset Overview", "### Where to find the Data and its Documentation", "#### Webpage\n\n\n\nGithub", "#### Download\n\n\n\nGithub", "#### Paper\n\n\n\nWebsite", "#### BibTex", "#### Contact Name\n\n\n\n\nSara Tonelli", "#### Contact Email\n\n\n\nsatonelli@URL", "#### Has a Leaderboard?\n\n\n\nno", "### Languages and Intended Use", "#### Multilingual?\n\n\n\n\nno", "#### Covered Dialects\n\n\n\nNone", "#### Covered Languages\n\n\n\n\n'Italian'", "#### License\n\n\n\n\ncc-by-4.0: Creative Commons Attribution 4.0 International", "#### Intended Use\n\n\n\nThe purpose of the dataset is to train NLG models to simplify complex text by learning different types of transformations (verb to noun, noun to verbs, deletion, insertion, etc)", "#### Primary Task\n\n\n\nSimplification", "#### Communicative Goal\n\n\n\n\nThis dataset aims to enhance research in text simplification in Italian language with different text transformations.", "### Credit", "#### Curation Organization Type(s)\n\n\n\n'academic', 'independent'", "#### Curation Organization(s)\n\n\n\nFondazione Bruno Kessler (FBK)", "#### Dataset Creators\n\n\n\nSara Tonelli (Fondazione Bruno Kessler), Alessio Palmero Aprosio (Fondazione Bruno Kessler), Francesca Saltori (Fondazione Bruno Kessler)", "#### Funding\n\n\n\nEU Horizon 2020 Programme via the SIMPATICO Project (H2020-EURO-6-2015, n. 692819)", "#### Who added the Dataset to GEM?\n\n\n\nSebastien Montella (Orange Labs), Vipul Raheja (Grammarly Inc.)", "### Dataset Structure", "#### Data Fields\n\n\n\nEach sample comes with the following fields:\n\n- 'gem_id' (string): Unique sample ID\n-'text' (string): The raw text to be simplified\n-'simplified_text' (string): The simplified version of \"text\" field\n-'transformation_type' (string): Nature of transformation applied to raw text in order to simplify it.\n-'source_dataset' (string): Initial dataset source of sample. Values: 'itwiki' (for Italian Wikipedia) or 'tn' (manually annotated administrative documents from the Municipality of Trento, Italy)", "#### Reason for Structure\n\n\n\nThe dataset is organized as a pairs where the raw text (input) is associated with its simplified text (output). The editing transformation and the source dataset of each sample is also provided for advanced analysis.", "#### How were labels chosen?\n\n\n\nSIMPITIKI dataset selects documents from Italian Wikipedia such that their editing tracking descriptions contain any of the words \"Simplified\"/\"Simplify\"/\"Simplification\". For the Public Administration domain of the documents of the Municipality of Trento (Italy)", "#### Example Instance", "#### Data Splits\n\n\n\nSeveral splits are proposed to train models on different configurations:\n\n-\"train\": Training samples randomly selected from initial corpus. 816 training samples.\n-\"validation\": Validating samples randomly selected from initial corpus. 174 validating samples.\n-\"test\": Testing samples randomly selected from initial corpus. 176 validating samples.\n-\"challenge_seen_transformations_train\": This training challenge split includes specific transformations to simplify the raw text. Precisely, transformations are \"Split\", \"Merge\", \"Reordering\", \"Insert - Verb\", \"Insert - Other\", \"Delete - Verb\", \"Delete - Other\", \"Transformation - Lexical Substitution (word level)\", \"Transformation - Anaphoric replacement\", \"Transformation - Noun to Verb\", \"Transformation - Verbal Features\". 562 training samples.\n-\"challenge_seen_transformations_val\": This validating challenge split includes same transformations than the ones observed in training. Precisely, transformations are \"Split\", \"Merge\", \"Reordering\", \"Insert - Verb\", \"Insert - Other\", \"Delete - Verb\", \"Delete - Other\", \"Transformation - Lexical Substitution (word level)\", \"Transformation - Anaphoric replacement\", \"Transformation - Noun to Verb\", \"Transformation - Verbal Features\". 121 validating samples.\n-\"challenge_seen_transformations_test\": This testing challenge split includes same transformations than the ones observed in training. Precisely, transformations are \"Split\", \"Merge\", \"Reordering\", \"Insert - Verb\", \"Insert - Other\", \"Delete - Verb\", \"Delete - Other\", \"Transformation - Lexical Substitution (word level)\", \"Transformation - Anaphoric replacement\", \"Transformation - Noun to Verb\", \"Transformation - Verbal Features\". 127 testing samples.\n-\"challenge_unseen_transformations_test\" : \"Insert - Subject\", \"Delete - Subject\", \"Transformation - Lexical Substitution (phrase level)\", \"Transformation - Verb to Noun (nominalization)\", \"Transformation - Verbal Voice\". 356 testing samples.\n-\"challenge_itwiki_train\": This challenge split includes random samples from the Italian Wikipedia as source dataset. 402 training samples.\n-\"challenge_itwiki_val\": This validating challenge split includes random samples from the Italian Wikipedia as source dataset. 86 validating samples.\n-\"challenge_itwiki_test\": This testing challenge split includes random samples from the Italian Wikipedia as source dataset. 87 testing samples.\n-\"challenge_tn_test\": This testing challenge split includes all samples from the Municipality of Trento administrative documents ('tn') as source dataset. 591 testing samples.", "#### Splitting Criteria\n\n\n\nThe training ratio is set to 0.7. The validation and test somehow equally divide the remaining 30% of the dataset.", "## Dataset in GEM", "### Rationale for Inclusion in GEM", "#### Why is the Dataset in GEM?\n\n\n\nThis dataset promotes Simplification task for Italian language.", "#### Similar Datasets\n\n\n\nno", "#### Ability that the Dataset measures\n\n\n\nModels can be evaluated if they can simplify text regarding different simplification transformations.", "### GEM-Specific Curation", "#### Modificatied for GEM?\n\n\n\nyes", "#### Additional Splits?\n\n\n\nyes", "#### Split Information\n\n\n\nThe SIMPITIKI dataset provides a single file. Several splits are proposed to train models on different configurations:\n-\"train\": Training samples randomly selected from initial corpus. 816 training samples.\n-\"validation\": Validating samples randomly selected from initial corpus. 174 validating samples.\n-\"test\": Testing samples randomly selected from initial corpus. 176 validating samples.\n-\"challenge_seen_transformations_train\": This training challenge split includes specific transformations to simplify the raw text. Precisely, transformations are \"Split\", \"Merge\", \"Reordering\", \"Insert - Verb\", \"Insert - Other\", \"Delete - Verb\", \"Delete - Other\", \"Transformation - Lexical Substitution (word level)\", \"Transformation - Anaphoric replacement\", \"Transformation - Noun to Verb\", \"Transformation - Verbal Features\". 562 training samples.\n-\"challenge_seen_transformations_val\": This validating challenge split includes same transformations than the ones observed in training. Precisely, transformations are \"Split\", \"Merge\", \"Reordering\", \"Insert - Verb\", \"Insert - Other\", \"Delete - Verb\", \"Delete - Other\", \"Transformation - Lexical Substitution (word level)\", \"Transformation - Anaphoric replacement\", \"Transformation - Noun to Verb\", \"Transformation - Verbal Features\". 121 validating samples.\n-\"challenge_seen_transformations_test\": This testing challenge split includes same transformations than the ones observed in training. Precisely, transformations are \"Split\", \"Merge\", \"Reordering\", \"Insert - Verb\", \"Insert - Other\", \"Delete - Verb\", \"Delete - Other\", \"Transformation - Lexical Substitution (word level)\", \"Transformation - Anaphoric replacement\", \"Transformation - Noun to Verb\", \"Transformation - Verbal Features\". 127 testing samples.\n-\"challenge_unseen_transformations_test\" : \"Insert - Subject\", \"Delete - Subject\", \"Transformation - Lexical Substitution (phrase level)\", \"Transformation - Verb to Noun (nominalization)\", \"Transformation - Verbal Voice\". 356 testing samples.\n-\"challenge_itwiki_train\": This challenge split includes random samples from the Italian Wikipedia as source dataset. 402 training samples.\n-\"challenge_itwiki_val\": This validating challenge split includes random samples from the Italian Wikipedia as source dataset. 86 validating samples.\n-\"challenge_itwiki_test\": This testing challenge split includes random samples from the Italian Wikipedia as source dataset. 87 testing samples.\n-\"challenge_tn_test\": This testing challenge split includes all samples from the Municipality of Trento administrative documents ('tn') as source dataset. 591 testing samples.", "#### Split Motivation\n\n\n\nThe splits allows to investigate the generalization of models regarding editing/transformations (\"challenge_seen_transformations_test\" / \"challenge_unseen_transformations_test\") and for transfer learning to different domain (\"challenge_tn_test\")", "### Getting Started with the Task", "#### Pointers to Resources\n\n\n\n- Coster and Kauchak, Simple English Wikipedia: A New Text Simplification Task, Proceedings of the 49th Annual Meeting of the Association for Computational Linguistics, pages 665–669, Portland, Oregon, June 19-24, 2011\n- Xu et al, Optimizing Statistical Machine Translation for Text Simplification, Transactions of the Association for Computational Linguistics, vol. 4, pp. 401–415, 2016\n- Aprosio et al, Neural Text Simplification in Low-Resource Conditions Using Weak Supervision, Proceedings of the Workshop on Methods for Optimizing and Evaluating Neural Language Generation (NeuralGen), pages 37–44, Minneapolis, Minnesota, USA, June 6, 2019", "#### Technical Terms\n\n\n\nSimplification: Process that consists in transforming an input text to its simplified version.", "## Previous Results", "### Previous Results", "#### Measured Model Abilities\n\n\n\nThe splits allows to investigate the generalization of models regarding editing/transformations (\"challenge_seen_transformations_test\" / \"challenge_unseen_transformations_test\") and for transfer learning to different domain (\"challenge_tn_test\")", "#### Metrics\n\n\n\n'BLEU', 'Other: Other Metrics'", "#### Other Metrics\n\n\n\nFKBLEU (URL Combines Flesch-Kincaid Index and iBLEU metrics.\nSARI (URL Compares system output against references and against the input sentence. It explicitly measures the goodness of words that are added, deleted and kept by the systems \nWord-level F1", "#### Previous results available?\n\n\n\nno", "## Dataset Curation", "### Original Curation", "#### Original Curation Rationale\n\n\n\nMost of the resources for Text Simplification are in English. To stimulate research to different languages, SIMPITIKI proposes an Italian corpus with Complex-Simple sentence pairs.", "#### Communicative Goal\n\n\n\nText simplification allows a smooth reading of text to enhance understanding.", "#### Sourced from Different Sources\n\n\n\nyes", "#### Source Details\n\n\n\nItalian Wikipedia\n(Manually) Annotated administrative documents from the Municipality of Trento, Italy", "### Language Data", "#### How was Language Data Obtained?\n\n\n\n'Found'", "#### Where was it found?\n\n\n\n'Single website', 'Offline media collection'", "#### Language Producers\n\n\n\nSIMPITIKI is a combination of documents from Italian Wikipedia and from the Municipality of Trento, Italy.", "#### Topics Covered\n\n\n\nSamples from documents from the Municipality of Trento corpus are in the administrative domain.", "#### Data Validation\n\n\n\nvalidated by data curator", "#### Was Data Filtered?\n\n\n\nnot filtered", "### Structured Annotations", "#### Additional Annotations?\n\n\n\n\ncrowd-sourced", "#### Number of Raters\n\n\n\nunknown", "#### Rater Qualifications\n\n\n\nNative speaker", "#### Raters per Training Example\n\n\n\n0", "#### Raters per Test Example\n\n\n\n0", "#### Annotation Service?\n\n\n\nunknown", "#### Annotation Values\n\n\n\nAnnotators specified any of the tags as designed by Brunato et al. (URL\n-Split: Splitting a clause into two clauses.\n-Merge: Merge two or more clauses together.\n-Reordering: Word order changes.\n-Insert: Insertion of words or phrases that provide supportive information to the original sentence\n-Delete: dropping redundant information.\n-Transformation: Modification which can affect the sentence at the lexical, morpho-syntactic and syntactic level: Lexical substitution (word level) / Lexical substitution (phrase level) / Anaphoric replacement / Noun to Verb / Verb to Noun / Verbal voice / Verbal features/ morpho–syntactic and syntactic level, also giving rise to overlapping phenomena", "#### Any Quality Control?\n\n\n\nunknown", "### Consent", "#### Any Consent Policy?\n\n\n\nno", "#### Justification for Using the Data\n\n\n\nThe dataset is available online under the CC-BY 4.0 license.", "### Private Identifying Information (PII)", "#### Contains PII?\n\n\n\n\nlikely", "#### Categories of PII\n\n\n\n'generic PII'", "#### Any PII Identification?\n\n\n\nno identification", "### Maintenance", "#### Any Maintenance Plan?\n\n\n\nno", "## Broader Social Context", "### Previous Work on the Social Impact of the Dataset", "#### Usage of Models based on the Data\n\n\n\nno", "### Impact on Under-Served Communities", "#### Addresses needs of underserved Communities?\n\n\n\nyes", "#### Details on how Dataset Addresses the Needs\n\n\n\nThe creator of SIMPITIKI wants to promote text simplification for Italian because few resources are available in other languages than English.", "### Discussion of Biases", "#### Any Documented Social Biases?\n\n\n\nunsure", "## Considerations for Using the Data", "### PII Risks and Liability", "### Licenses", "#### Copyright Restrictions on the Dataset\n\n\n\n'research use only'", "#### Copyright Restrictions on the Language Data\n\n\n\n'research use only'", "### Known Technical Limitations", "#### Discouraged Use Cases\n\n\n\nThe risk of surface-based metrics (BLEU, chrf++, etc) for this task is that semantic adequacy is not respected when simplifying the input document." ]

[ "task_categories:other", "annotations_creators:none", "language_creators:unknown", "multilinguality:unknown", "size_categories:unknown", "source_datasets:original", "language:en", "license:apache-2.0", "text-to-slide", "region:us" ]

[]

[ "en" ]

[ "# Dataset Card for GEM/SciDuet", "## Dataset Description\n\n- Homepage: URL\n- Repository: URL\n- Paper: URL\n- Leaderboard: N/A\n- Point of Contact: N/A", "### Link to Main Data Card\n\nYou can find the main data card on the GEM Website.", "### Dataset Summary \n\nThis dataset supports the document-to-slide generation task where a model has to generate presentation slide content from the text of a document. \n\nYou can load the dataset via:\n\nThe data loader can be found here.", "#### website\nHuggingface", "#### paper\nACL Anthology", "#### authors\nEdward Sun, Yufang Hou, Dakuo Wang, Yunfeng Zhang, Nancy Wang", "## Dataset Overview", "### Where to find the Data and its Documentation", "#### Webpage\n\n\n\nHuggingface", "#### Download\n\n\n\nGithub", "#### Paper\n\n\n\nACL Anthology", "#### BibTex", "#### Has a Leaderboard?\n\n\n\nno", "### Languages and Intended Use", "#### Multilingual?\n\n\n\n\nno", "#### Covered Languages\n\n\n\n\n'English'", "#### License\n\n\n\n\napache-2.0: Apache License 2.0", "#### Intended Use\n\n\n\nPromote research on the task of document-to-slides generation", "#### Primary Task\n\n\n\nText-to-Slide", "### Credit", "#### Curation Organization Type(s)\n\n\n\n'industry'", "#### Curation Organization(s)\n\n\n\nIBM Research", "#### Dataset Creators\n\n\n\nEdward Sun, Yufang Hou, Dakuo Wang, Yunfeng Zhang, Nancy Wang", "#### Funding\n\n\n\nIBM Research", "#### Who added the Dataset to GEM?\n\n\n\nYufang Hou (IBM Research), Dakuo Wang (IBM Research)", "### Dataset Structure", "#### How were labels chosen?\n\n\n\nThe original papers and slides (both are in PDF format) are carefully processed by a combination of PDF/Image processing tookits. The text contents from multiple slides that correspond to the same slide title are mreged.", "#### Data Splits\n\n\n\nTraining, validation and testing data contain 136, 55, and 81 papers from ACL Anthology and their corresponding slides, respectively.", "#### Splitting Criteria\n\n\n\nThe dataset integrated into GEM is the ACL portion of the whole dataset described in the paper, It contains the full Dev and Test sets, and a portion of the Train dataset. \nNote that although we cannot release the whole training dataset due to copyright issues, researchers can still use our released data procurement code to generate the training dataset from the online ICML/NeurIPS anthologies.", "## Dataset in GEM", "### Rationale for Inclusion in GEM", "#### Why is the Dataset in GEM?\n\n\n\nSciDuet is the first publicaly available dataset for the challenging task of document2slides generation, which requires a model has a good ability to \"understand\" long-form text, choose appropriate content and generate key points.", "#### Similar Datasets\n\n\n\nno", "#### Ability that the Dataset measures\n\n\n\ncontent selection, long-form text undersanding and generation", "### GEM-Specific Curation", "#### Modificatied for GEM?\n\n\n\nno", "#### Additional Splits?\n\n\n\nno", "### Getting Started with the Task", "## Previous Results", "### Previous Results", "#### Measured Model Abilities\n\n\n\ncontent selection, long-form text undersanding and key points generation", "#### Metrics\n\n\n\n'ROUGE'", "#### Proposed Evaluation\n\n\n\nAutomatical Evaluation Metric: ROUGE\nHuman Evaluation: (Readability, Informativeness, Consistency) \n1) Readability: The generated slide content is coherent, concise, and grammatically correct;\n2) Informativeness: The generated slide provides sufficient and necessary information that corresponds to the given slide title, regardless of its similarity to the original slide;\n3) Consistency: The generated slide content is similar to the original author’s reference slide.", "#### Previous results available?\n\n\n\nyes", "#### Other Evaluation Approaches\n\n\n\nROUGE + Human Evaluation", "#### Relevant Previous Results\n\n\n\nPaper \"D2S: Document-to-Slide Generation Via Query-Based\nText Summarization\" reports 20.47, 5.26 and 19.08 for ROUGE-1, ROUGE-2 and ROUGE-L (f-score).", "## Dataset Curation", "### Original Curation", "#### Original Curation Rationale\n\n\n\nProvide a benchmark dataset for the document-to-slides task.", "#### Sourced from Different Sources\n\n\n\nno", "### Language Data", "#### How was Language Data Obtained?\n\n\n\n'Other'", "#### Data Validation\n\n\n\nnot validated", "#### Data Preprocessing\n\n\n\nText on papers was extracted through Grobid. Figures andcaptions were extracted through pdffigures. Text on slides was extracted through IBM Watson Discovery package and OCR by pytesseract. Figures and tables that appear on slides and papers were linked through multiscale template matching by OpenCV. Further dataset\ncleaning was performed with standard string-based\nheuristics on sentence building, equation and floating caption removal, and duplicate line deletion.", "#### Was Data Filtered?\n\n\n\nalgorithmically", "#### Filter Criteria\n\n\n\nthe slide context text shouldn't contain additional format information such as \"* University\"", "### Structured Annotations", "#### Additional Annotations?\n\n\n\n\nnone", "#### Annotation Service?\n\n\n\nno", "### Consent", "#### Any Consent Policy?\n\n\n\nyes", "#### Consent Policy Details\n\n\n\nThe original dataset was open-sourced under Apache-2.0. \nSome of the original dataset creators are part of the GEM v2 dataset infrastructure team and take care of integrating this dataset into GEM.", "### Private Identifying Information (PII)", "#### Contains PII?\n\n\n\n\nyes/very likely", "#### Categories of PII\n\n\n\n'generic PII'", "#### Any PII Identification?\n\n\n\nno identification", "### Maintenance", "#### Any Maintenance Plan?\n\n\n\nno", "## Broader Social Context", "### Previous Work on the Social Impact of the Dataset", "#### Usage of Models based on the Data\n\n\n\nno", "### Impact on Under-Served Communities", "#### Addresses needs of underserved Communities?\n\n\n\nno", "### Discussion of Biases", "#### Any Documented Social Biases?\n\n\n\nunsure", "## Considerations for Using the Data", "### PII Risks and Liability", "### Licenses", "#### Copyright Restrictions on the Dataset\n\n\n\n'non-commercial use only'", "#### Copyright Restrictions on the Language Data\n\n\n\n'research use only'", "### Known Technical Limitations" ]

[ "TAGS\n#task_categories-other #annotations_creators-none #language_creators-unknown #multilinguality-unknown #size_categories-unknown #source_datasets-original #language-English #license-apache-2.0 #text-to-slide #region-us \n", "# Dataset Card for GEM/SciDuet", "## Dataset Description\n\n- Homepage: URL\n- Repository: URL\n- Paper: URL\n- Leaderboard: N/A\n- Point of Contact: N/A", "### Link to Main Data Card\n\nYou can find the main data card on the GEM Website.", "### Dataset Summary \n\nThis dataset supports the document-to-slide generation task where a model has to generate presentation slide content from the text of a document. \n\nYou can load the dataset via:\n\nThe data loader can be found here.", "#### website\nHuggingface", "#### paper\nACL Anthology", "#### authors\nEdward Sun, Yufang Hou, Dakuo Wang, Yunfeng Zhang, Nancy Wang", "## Dataset Overview", "### Where to find the Data and its Documentation", "#### Webpage\n\n\n\nHuggingface", "#### Download\n\n\n\nGithub", "#### Paper\n\n\n\nACL Anthology", "#### BibTex", "#### Has a Leaderboard?\n\n\n\nno", "### Languages and Intended Use", "#### Multilingual?\n\n\n\n\nno", "#### Covered Languages\n\n\n\n\n'English'", "#### License\n\n\n\n\napache-2.0: Apache License 2.0", "#### Intended Use\n\n\n\nPromote research on the task of document-to-slides generation", "#### Primary Task\n\n\n\nText-to-Slide", "### Credit", "#### Curation Organization Type(s)\n\n\n\n'industry'", "#### Curation Organization(s)\n\n\n\nIBM Research", "#### Dataset Creators\n\n\n\nEdward Sun, Yufang Hou, Dakuo Wang, Yunfeng Zhang, Nancy Wang", "#### Funding\n\n\n\nIBM Research", "#### Who added the Dataset to GEM?\n\n\n\nYufang Hou (IBM Research), Dakuo Wang (IBM Research)", "### Dataset Structure", "#### How were labels chosen?\n\n\n\nThe original papers and slides (both are in PDF format) are carefully processed by a combination of PDF/Image processing tookits. The text contents from multiple slides that correspond to the same slide title are mreged.", "#### Data Splits\n\n\n\nTraining, validation and testing data contain 136, 55, and 81 papers from ACL Anthology and their corresponding slides, respectively.", "#### Splitting Criteria\n\n\n\nThe dataset integrated into GEM is the ACL portion of the whole dataset described in the paper, It contains the full Dev and Test sets, and a portion of the Train dataset. \nNote that although we cannot release the whole training dataset due to copyright issues, researchers can still use our released data procurement code to generate the training dataset from the online ICML/NeurIPS anthologies.", "## Dataset in GEM", "### Rationale for Inclusion in GEM", "#### Why is the Dataset in GEM?\n\n\n\nSciDuet is the first publicaly available dataset for the challenging task of document2slides generation, which requires a model has a good ability to \"understand\" long-form text, choose appropriate content and generate key points.", "#### Similar Datasets\n\n\n\nno", "#### Ability that the Dataset measures\n\n\n\ncontent selection, long-form text undersanding and generation", "### GEM-Specific Curation", "#### Modificatied for GEM?\n\n\n\nno", "#### Additional Splits?\n\n\n\nno", "### Getting Started with the Task", "## Previous Results", "### Previous Results", "#### Measured Model Abilities\n\n\n\ncontent selection, long-form text undersanding and key points generation", "#### Metrics\n\n\n\n'ROUGE'", "#### Proposed Evaluation\n\n\n\nAutomatical Evaluation Metric: ROUGE\nHuman Evaluation: (Readability, Informativeness, Consistency) \n1) Readability: The generated slide content is coherent, concise, and grammatically correct;\n2) Informativeness: The generated slide provides sufficient and necessary information that corresponds to the given slide title, regardless of its similarity to the original slide;\n3) Consistency: The generated slide content is similar to the original author’s reference slide.", "#### Previous results available?\n\n\n\nyes", "#### Other Evaluation Approaches\n\n\n\nROUGE + Human Evaluation", "#### Relevant Previous Results\n\n\n\nPaper \"D2S: Document-to-Slide Generation Via Query-Based\nText Summarization\" reports 20.47, 5.26 and 19.08 for ROUGE-1, ROUGE-2 and ROUGE-L (f-score).", "## Dataset Curation", "### Original Curation", "#### Original Curation Rationale\n\n\n\nProvide a benchmark dataset for the document-to-slides task.", "#### Sourced from Different Sources\n\n\n\nno", "### Language Data", "#### How was Language Data Obtained?\n\n\n\n'Other'", "#### Data Validation\n\n\n\nnot validated", "#### Data Preprocessing\n\n\n\nText on papers was extracted through Grobid. Figures andcaptions were extracted through pdffigures. Text on slides was extracted through IBM Watson Discovery package and OCR by pytesseract. Figures and tables that appear on slides and papers were linked through multiscale template matching by OpenCV. Further dataset\ncleaning was performed with standard string-based\nheuristics on sentence building, equation and floating caption removal, and duplicate line deletion.", "#### Was Data Filtered?\n\n\n\nalgorithmically", "#### Filter Criteria\n\n\n\nthe slide context text shouldn't contain additional format information such as \"* University\"", "### Structured Annotations", "#### Additional Annotations?\n\n\n\n\nnone", "#### Annotation Service?\n\n\n\nno", "### Consent", "#### Any Consent Policy?\n\n\n\nyes", "#### Consent Policy Details\n\n\n\nThe original dataset was open-sourced under Apache-2.0. \nSome of the original dataset creators are part of the GEM v2 dataset infrastructure team and take care of integrating this dataset into GEM.", "### Private Identifying Information (PII)", "#### Contains PII?\n\n\n\n\nyes/very likely", "#### Categories of PII\n\n\n\n'generic PII'", "#### Any PII Identification?\n\n\n\nno identification", "### Maintenance", "#### Any Maintenance Plan?\n\n\n\nno", "## Broader Social Context", "### Previous Work on the Social Impact of the Dataset", "#### Usage of Models based on the Data\n\n\n\nno", "### Impact on Under-Served Communities", "#### Addresses needs of underserved Communities?\n\n\n\nno", "### Discussion of Biases", "#### Any Documented Social Biases?\n\n\n\nunsure", "## Considerations for Using the Data", "### PII Risks and Liability", "### Licenses", "#### Copyright Restrictions on the Dataset\n\n\n\n'non-commercial use only'", "#### Copyright Restrictions on the Language Data\n\n\n\n'research use only'", "### Known Technical Limitations" ]

[ "task_categories:conversational", "annotations_creators:none", "language_creators:unknown", "multilinguality:unknown", "size_categories:unknown", "source_datasets:original", "language:en", "license:cc-by-4.0", "dialog-response-generation", "arxiv:2012.12458", "region:us" ]

[ "2012.12458" ]

[ "en" ]

[ "# Dataset Card for GEM/Taskmaster", "## Dataset Description\n\n- Homepage: URL\n- Repository: URL\n- Paper: URL\n- Leaderboard: N/A\n- Point of Contact: Karthik Krishnamoorthi", "### Link to Main Data Card\n\nYou can find the main data card on the GEM Website.", "### Dataset Summary \n\nThis is a large task-oriented dialog dataset in which a model has to produce the response. The input contains the context and a structured representation of what the model is supposed to generate. The input is already pre-formatted as string, turning this into a pure text-to-text problem. \n\nYou can load the dataset via:\n\nThe data loader can be found here.", "#### website\nGithub", "#### paper\nArxiv", "#### authors\nGoogle researchers", "## Dataset Overview", "### Where to find the Data and its Documentation", "#### Webpage\n\n\n\nGithub", "#### Download\n\n\n\nGithub", "#### Paper\n\n\n\nArxiv", "#### BibTex", "#### Contact Name\n\n\n\n\nKarthik Krishnamoorthi", "#### Contact Email\n\n\n\nkrishnamoorthi@URL", "#### Has a Leaderboard?\n\n\n\nno", "### Languages and Intended Use", "#### Multilingual?\n\n\n\n\nno", "#### Covered Dialects\n\n\n\nNA", "#### Covered Languages\n\n\n\n\n'English'", "#### Whose Language?\n\n\n\nNA", "#### License\n\n\n\n\ncc-by-4.0: Creative Commons Attribution 4.0 International", "#### Intended Use\n\n\n\nDialogues", "#### Primary Task\n\n\n\nDialog Response Generation", "#### Communicative Goal\n\n\n\n\na movie ticketing dialog dataset with 23,789 annotated conversations.", "### Credit", "#### Curation Organization Type(s)\n\n\n\n'other'", "#### Curation Organization(s)\n\n\n\nNA", "#### Dataset Creators\n\n\n\nGoogle researchers", "#### Funding\n\n\n\nGoogle", "#### Who added the Dataset to GEM?\n\n\n\nTosin Adewumi (Luleå University of Technology)", "### Dataset Structure", "#### Data Fields\n\n\n\n- 'gem_id': The unique example id\n- 'context': The context of the conversation\n- 'target': A string representing the target\n-'references': A List representing the target(s)\n-'conversation_id': A unique ID of the conversation", "#### Reason for Structure\n\n\n\nNA", "#### How were labels chosen?\n\n\n\nNA", "#### Example Instance", "#### Data Splits\n\n\n\n-'train': 187182 examples\n-'dev': 23406 examples\n-'test': 23316 examples", "#### Splitting Criteria\n\n\n\nNA", "#### \n\n\n\nNA", "## Dataset in GEM", "### Rationale for Inclusion in GEM", "#### Why is the Dataset in GEM?\n\n\n\nDialogue generation that makes sense", "#### Similar Datasets\n\n\n\nyes", "#### Unique Language Coverage\n\n\n\nno", "#### Difference from other GEM datasets\n\n\n\nNA", "#### Ability that the Dataset measures\n\n\n\nNA", "### GEM-Specific Curation", "#### Modificatied for GEM?\n\n\n\nyes", "#### GEM Modifications\n\n\n\n'other'", "#### Modification Details\n\n\n\ngem_id field was added to the 3 data splits", "#### Additional Splits?\n\n\n\nno", "### Getting Started with the Task", "#### Pointers to Resources\n\n\n\nURL", "#### Technical Terms\n\n\n\nNA", "## Previous Results", "### Previous Results", "#### Measured Model Abilities\n\n\n\nBLEU: 60", "#### Metrics\n\n\n\n'BLEU'", "#### Proposed Evaluation\n\n\n\nautomatic evaluation", "#### Previous results available?\n\n\n\nyes", "#### Other Evaluation Approaches\n\n\n\nNA", "#### Relevant Previous Results\n\n\n\nNA", "## Dataset Curation", "### Original Curation", "#### Original Curation Rationale\n\n\n\nNA", "#### Communicative Goal\n\n\n\na movie ticketing dialog dataset with 23,789 annotated conversations.", "#### Sourced from Different Sources\n\n\n\nno", "### Language Data", "#### How was Language Data Obtained?\n\n\n\n'Crowdsourced'", "#### Where was it crowdsourced?\n\n\n\n'Participatory experiment'", "#### Language Producers\n\n\n\nNA", "#### Topics Covered\n\n\n\nTicketing", "#### Data Validation\n\n\n\nnot validated", "#### Was Data Filtered?\n\n\n\nnot filtered", "### Structured Annotations", "#### Additional Annotations?\n\n\n\n\nnone", "#### Annotation Service?\n\n\n\nno", "### Consent", "#### Any Consent Policy?\n\n\n\nno", "#### Justification for Using the Data\n\n\n\nNA", "### Private Identifying Information (PII)", "#### Contains PII?\n\n\n\n\nno PII", "#### Justification for no PII\n\n\n\nIt's based on ticketing without personal information", "### Maintenance", "#### Any Maintenance Plan?\n\n\n\nno", "## Broader Social Context", "### Previous Work on the Social Impact of the Dataset", "#### Usage of Models based on the Data\n\n\n\nno", "### Impact on Under-Served Communities", "#### Addresses needs of underserved Communities?\n\n\n\nno", "### Discussion of Biases", "#### Any Documented Social Biases?\n\n\n\nunsure", "#### Are the Language Producers Representative of the Language?\n\n\n\nNA", "## Considerations for Using the Data", "### PII Risks and Liability", "#### Potential PII Risk\n\n\n\nNA", "### Licenses", "#### Copyright Restrictions on the Dataset\n\n\n\n'open license - commercial use allowed'", "#### Copyright Restrictions on the Language Data\n\n\n\n'public domain'", "### Known Technical Limitations", "#### Technical Limitations\n\n\n\nNA", "#### Unsuited Applications\n\n\n\nNA", "#### Discouraged Use Cases\n\n\n\nNA" ]

[ "TAGS\n#task_categories-conversational #annotations_creators-none #language_creators-unknown #multilinguality-unknown #size_categories-unknown #source_datasets-original #language-English #license-cc-by-4.0 #dialog-response-generation #arxiv-2012.12458 #region-us \n", "# Dataset Card for GEM/Taskmaster", "## Dataset Description\n\n- Homepage: URL\n- Repository: URL\n- Paper: URL\n- Leaderboard: N/A\n- Point of Contact: Karthik Krishnamoorthi", "### Link to Main Data Card\n\nYou can find the main data card on the GEM Website.", "### Dataset Summary \n\nThis is a large task-oriented dialog dataset in which a model has to produce the response. The input contains the context and a structured representation of what the model is supposed to generate. The input is already pre-formatted as string, turning this into a pure text-to-text problem. \n\nYou can load the dataset via:\n\nThe data loader can be found here.", "#### website\nGithub", "#### paper\nArxiv", "#### authors\nGoogle researchers", "## Dataset Overview", "### Where to find the Data and its Documentation", "#### Webpage\n\n\n\nGithub", "#### Download\n\n\n\nGithub", "#### Paper\n\n\n\nArxiv", "#### BibTex", "#### Contact Name\n\n\n\n\nKarthik Krishnamoorthi", "#### Contact Email\n\n\n\nkrishnamoorthi@URL", "#### Has a Leaderboard?\n\n\n\nno", "### Languages and Intended Use", "#### Multilingual?\n\n\n\n\nno", "#### Covered Dialects\n\n\n\nNA", "#### Covered Languages\n\n\n\n\n'English'", "#### Whose Language?\n\n\n\nNA", "#### License\n\n\n\n\ncc-by-4.0: Creative Commons Attribution 4.0 International", "#### Intended Use\n\n\n\nDialogues", "#### Primary Task\n\n\n\nDialog Response Generation", "#### Communicative Goal\n\n\n\n\na movie ticketing dialog dataset with 23,789 annotated conversations.", "### Credit", "#### Curation Organization Type(s)\n\n\n\n'other'", "#### Curation Organization(s)\n\n\n\nNA", "#### Dataset Creators\n\n\n\nGoogle researchers", "#### Funding\n\n\n\nGoogle", "#### Who added the Dataset to GEM?\n\n\n\nTosin Adewumi (Luleå University of Technology)", "### Dataset Structure", "#### Data Fields\n\n\n\n- 'gem_id': The unique example id\n- 'context': The context of the conversation\n- 'target': A string representing the target\n-'references': A List representing the target(s)\n-'conversation_id': A unique ID of the conversation", "#### Reason for Structure\n\n\n\nNA", "#### How were labels chosen?\n\n\n\nNA", "#### Example Instance", "#### Data Splits\n\n\n\n-'train': 187182 examples\n-'dev': 23406 examples\n-'test': 23316 examples", "#### Splitting Criteria\n\n\n\nNA", "#### \n\n\n\nNA", "## Dataset in GEM", "### Rationale for Inclusion in GEM", "#### Why is the Dataset in GEM?\n\n\n\nDialogue generation that makes sense", "#### Similar Datasets\n\n\n\nyes", "#### Unique Language Coverage\n\n\n\nno", "#### Difference from other GEM datasets\n\n\n\nNA", "#### Ability that the Dataset measures\n\n\n\nNA", "### GEM-Specific Curation", "#### Modificatied for GEM?\n\n\n\nyes", "#### GEM Modifications\n\n\n\n'other'", "#### Modification Details\n\n\n\ngem_id field was added to the 3 data splits", "#### Additional Splits?\n\n\n\nno", "### Getting Started with the Task", "#### Pointers to Resources\n\n\n\nURL", "#### Technical Terms\n\n\n\nNA", "## Previous Results", "### Previous Results", "#### Measured Model Abilities\n\n\n\nBLEU: 60", "#### Metrics\n\n\n\n'BLEU'", "#### Proposed Evaluation\n\n\n\nautomatic evaluation", "#### Previous results available?\n\n\n\nyes", "#### Other Evaluation Approaches\n\n\n\nNA", "#### Relevant Previous Results\n\n\n\nNA", "## Dataset Curation", "### Original Curation", "#### Original Curation Rationale\n\n\n\nNA", "#### Communicative Goal\n\n\n\na movie ticketing dialog dataset with 23,789 annotated conversations.", "#### Sourced from Different Sources\n\n\n\nno", "### Language Data", "#### How was Language Data Obtained?\n\n\n\n'Crowdsourced'", "#### Where was it crowdsourced?\n\n\n\n'Participatory experiment'", "#### Language Producers\n\n\n\nNA", "#### Topics Covered\n\n\n\nTicketing", "#### Data Validation\n\n\n\nnot validated", "#### Was Data Filtered?\n\n\n\nnot filtered", "### Structured Annotations", "#### Additional Annotations?\n\n\n\n\nnone", "#### Annotation Service?\n\n\n\nno", "### Consent", "#### Any Consent Policy?\n\n\n\nno", "#### Justification for Using the Data\n\n\n\nNA", "### Private Identifying Information (PII)", "#### Contains PII?\n\n\n\n\nno PII", "#### Justification for no PII\n\n\n\nIt's based on ticketing without personal information", "### Maintenance", "#### Any Maintenance Plan?\n\n\n\nno", "## Broader Social Context", "### Previous Work on the Social Impact of the Dataset", "#### Usage of Models based on the Data\n\n\n\nno", "### Impact on Under-Served Communities", "#### Addresses needs of underserved Communities?\n\n\n\nno", "### Discussion of Biases", "#### Any Documented Social Biases?\n\n\n\nunsure", "#### Are the Language Producers Representative of the Language?\n\n\n\nNA", "## Considerations for Using the Data", "### PII Risks and Liability", "#### Potential PII Risk\n\n\n\nNA", "### Licenses", "#### Copyright Restrictions on the Dataset\n\n\n\n'open license - commercial use allowed'", "#### Copyright Restrictions on the Language Data\n\n\n\n'public domain'", "### Known Technical Limitations", "#### Technical Limitations\n\n\n\nNA", "#### Unsuited Applications\n\n\n\nNA", "#### Discouraged Use Cases\n\n\n\nNA" ]

[ "task_categories:text2text-generation", "task_ids:text-simplification", "annotations_creators:none", "language_creators:unknown", "multilinguality:unknown", "size_categories:unknown", "source_datasets:original", "language:en", "license:cc-by-4.0", "region:us" ]

[]

[ "en" ]

[ "# Dataset Card for GEM/cochrane-simplification", "## Dataset Description\n\n- Homepage: URL\n- Repository: URL\n- Paper: URL\n- Leaderboard: N/A\n- Point of Contact: Ashwin Devaraj", "### Link to Main Data Card\n\nYou can find the main data card on the GEM Website.", "### Dataset Summary \n\nCochrane is an English dataset for paragraph-level simplification of medical texts. Cochrane is a database of systematic reviews of clinical questions, many of which have summaries in plain English targeting readers without a university education. The dataset comprises about 4,500 of such pairs. \n\nYou can load the dataset via:\n\nThe data loader can be found here.", "#### website\nLink", "#### paper\nLink", "#### authors\nAshwin Devaraj (The University of Texas at Austin), Iain J. Marshall (King's College London), Byron C. Wallace (Northeastern University), Junyi Jessy Li (The University of Texas at Austin)", "## Dataset Overview", "### Where to find the Data and its Documentation", "#### Webpage\n\n\n\nLink", "#### Download\n\n\n\nLink", "#### Paper\n\n\n\nLink", "#### BibTex", "#### Contact Name\n\n\n\n\nAshwin Devaraj", "#### Contact Email\n\n\n\nashwin.devaraj@URL", "#### Has a Leaderboard?\n\n\n\nno", "### Languages and Intended Use", "#### Multilingual?\n\n\n\n\nno", "#### Covered Languages\n\n\n\n\n'English'", "#### License\n\n\n\n\ncc-by-4.0: Creative Commons Attribution 4.0 International", "#### Intended Use\n\n\n\nThe intended use of this dataset is to train models that simplify medical text at the paragraph level so that it may be more accessible to the lay reader.", "#### Primary Task\n\n\n\nSimplification", "#### Communicative Goal\n\n\n\n\nA model trained on this dataset can be used to simplify medical texts to make them more accessible to readers without medical expertise.", "### Credit", "#### Curation Organization Type(s)\n\n\n\n'academic'", "#### Curation Organization(s)\n\n\n\nThe University of Texas at Austin, King's College London, Northeastern University", "#### Dataset Creators\n\n\n\nAshwin Devaraj (The University of Texas at Austin), Iain J. Marshall (King's College London), Byron C. Wallace (Northeastern University), Junyi Jessy Li (The University of Texas at Austin)", "#### Funding\n\n\n\nNational Institutes of Health (NIH) grant R01-LM012086, National Science Foundation (NSF) grant IIS-1850153, Texas Advanced Computing Center (TACC) computational resources", "#### Who added the Dataset to GEM?\n\n\n\nAshwin Devaraj (The University of Texas at Austin)", "### Dataset Structure", "#### Data Fields\n\n\n\n- 'gem_id': string, a unique identifier for the example\n- 'doi': string, DOI identifier for the Cochrane review from which the example was generated\n- 'source': string, an excerpt from an abstract of a Cochrane review\n- 'target': string, an excerpt from the plain-language summary of a Cochrane review that roughly aligns with the source text", "#### Example Instance", "#### Data Splits\n\n\n\n- 'train': 3568 examples\n- 'validation': 411 examples\n- 'test': 480 examples", "## Dataset in GEM", "### Rationale for Inclusion in GEM", "#### Why is the Dataset in GEM?\n\n\n\nThis dataset is the first paragraph-level simplification dataset published (as prior work had primarily focused on simplifying individual sentences). Furthermore, this dataset is in the medical domain, which is an especially useful domain for text simplification.", "#### Similar Datasets\n\n\n\nno", "#### Ability that the Dataset measures\n\n\n\nThis dataset measures the ability for a model to simplify paragraphs of medical text through the omission non-salient information and simplification of medical jargon.", "### GEM-Specific Curation", "#### Modificatied for GEM?\n\n\n\nno", "#### Additional Splits?\n\n\n\nno", "### Getting Started with the Task", "## Previous Results", "### Previous Results", "#### Measured Model Abilities\n\n\n\nThis dataset measures the ability for a model to simplify paragraphs of medical text through the omission non-salient information and simplification of medical jargon.", "#### Metrics\n\n\n\n'Other: Other Metrics', 'BLEU'", "#### Other Metrics\n\n\n\nSARI measures the quality of text simplification", "#### Previous results available?\n\n\n\nyes", "#### Relevant Previous Results\n\n\n\nThe paper which introduced this dataset trained BART models (pretrained on XSum) with unlikelihood training to produce simplification models achieving maximum SARI and BLEU scores of 40 and 43 respectively.", "## Dataset Curation", "### Original Curation", "#### Sourced from Different Sources\n\n\n\nno", "### Language Data", "#### Data Validation\n\n\n\nnot validated", "#### Was Data Filtered?\n\n\n\nnot filtered", "### Structured Annotations", "#### Additional Annotations?\n\n\n\n\nnone", "#### Annotation Service?\n\n\n\nno", "### Consent", "#### Any Consent Policy?\n\n\n\nno", "### Private Identifying Information (PII)", "#### Contains PII?\n\n\n\n\nyes/very likely", "#### Any PII Identification?\n\n\n\nno identification", "### Maintenance", "#### Any Maintenance Plan?\n\n\n\nno", "## Broader Social Context", "### Previous Work on the Social Impact of the Dataset", "#### Usage of Models based on the Data\n\n\n\nno", "### Impact on Under-Served Communities", "#### Addresses needs of underserved Communities?\n\n\n\nyes", "#### Details on how Dataset Addresses the Needs\n\n\n\nThis dataset can be used to simplify medical texts that may otherwise be inaccessible to those without medical training.", "### Discussion of Biases", "#### Any Documented Social Biases?\n\n\n\nunsure", "#### Are the Language Producers Representative of the Language?\n\n\n\nThe dataset was generated from abstracts and plain-language summaries of medical literature reviews that were written by medical professionals and thus does was not generated by people representative of the entire English-speaking population.", "## Considerations for Using the Data", "### PII Risks and Liability", "### Licenses", "### Known Technical Limitations", "#### Technical Limitations\n\n\n\nThe main limitation of this dataset is that the information alignment between the abstract and plain-language summary is often rough, so the plain-language summary may contain information that isn't found in the abstract. Furthermore, the plain-language targets often contain formulaic statements like \"this evidence is current to [month][year]\" not found in the abstracts. Another limitation is that some plain-language summaries do not simplify the technical abstracts very much and still contain medical jargon.", "#### Unsuited Applications\n\n\n\nThe main pitfall to look out for is errors in factuality. Simplification work so far has not placed a strong emphasis on the logical fidelity of model generations with the input text, and the paper introducing this dataset does not explore modeling techniques to combat this. These kinds of errors are especially pernicious in the medical domain, and the models introduced in the paper do occasionally alter entities like disease and medication names." ]

[ "TAGS\n#task_categories-text2text-generation #task_ids-text-simplification #annotations_creators-none #language_creators-unknown #multilinguality-unknown #size_categories-unknown #source_datasets-original #language-English #license-cc-by-4.0 #region-us \n", "# Dataset Card for GEM/cochrane-simplification", "## Dataset Description\n\n- Homepage: URL\n- Repository: URL\n- Paper: URL\n- Leaderboard: N/A\n- Point of Contact: Ashwin Devaraj", "### Link to Main Data Card\n\nYou can find the main data card on the GEM Website.", "### Dataset Summary \n\nCochrane is an English dataset for paragraph-level simplification of medical texts. Cochrane is a database of systematic reviews of clinical questions, many of which have summaries in plain English targeting readers without a university education. The dataset comprises about 4,500 of such pairs. \n\nYou can load the dataset via:\n\nThe data loader can be found here.", "#### website\nLink", "#### paper\nLink", "#### authors\nAshwin Devaraj (The University of Texas at Austin), Iain J. Marshall (King's College London), Byron C. Wallace (Northeastern University), Junyi Jessy Li (The University of Texas at Austin)", "## Dataset Overview", "### Where to find the Data and its Documentation", "#### Webpage\n\n\n\nLink", "#### Download\n\n\n\nLink", "#### Paper\n\n\n\nLink", "#### BibTex", "#### Contact Name\n\n\n\n\nAshwin Devaraj", "#### Contact Email\n\n\n\nashwin.devaraj@URL", "#### Has a Leaderboard?\n\n\n\nno", "### Languages and Intended Use", "#### Multilingual?\n\n\n\n\nno", "#### Covered Languages\n\n\n\n\n'English'", "#### License\n\n\n\n\ncc-by-4.0: Creative Commons Attribution 4.0 International", "#### Intended Use\n\n\n\nThe intended use of this dataset is to train models that simplify medical text at the paragraph level so that it may be more accessible to the lay reader.", "#### Primary Task\n\n\n\nSimplification", "#### Communicative Goal\n\n\n\n\nA model trained on this dataset can be used to simplify medical texts to make them more accessible to readers without medical expertise.", "### Credit", "#### Curation Organization Type(s)\n\n\n\n'academic'", "#### Curation Organization(s)\n\n\n\nThe University of Texas at Austin, King's College London, Northeastern University", "#### Dataset Creators\n\n\n\nAshwin Devaraj (The University of Texas at Austin), Iain J. Marshall (King's College London), Byron C. Wallace (Northeastern University), Junyi Jessy Li (The University of Texas at Austin)", "#### Funding\n\n\n\nNational Institutes of Health (NIH) grant R01-LM012086, National Science Foundation (NSF) grant IIS-1850153, Texas Advanced Computing Center (TACC) computational resources", "#### Who added the Dataset to GEM?\n\n\n\nAshwin Devaraj (The University of Texas at Austin)", "### Dataset Structure", "#### Data Fields\n\n\n\n- 'gem_id': string, a unique identifier for the example\n- 'doi': string, DOI identifier for the Cochrane review from which the example was generated\n- 'source': string, an excerpt from an abstract of a Cochrane review\n- 'target': string, an excerpt from the plain-language summary of a Cochrane review that roughly aligns with the source text", "#### Example Instance", "#### Data Splits\n\n\n\n- 'train': 3568 examples\n- 'validation': 411 examples\n- 'test': 480 examples", "## Dataset in GEM", "### Rationale for Inclusion in GEM", "#### Why is the Dataset in GEM?\n\n\n\nThis dataset is the first paragraph-level simplification dataset published (as prior work had primarily focused on simplifying individual sentences). Furthermore, this dataset is in the medical domain, which is an especially useful domain for text simplification.", "#### Similar Datasets\n\n\n\nno", "#### Ability that the Dataset measures\n\n\n\nThis dataset measures the ability for a model to simplify paragraphs of medical text through the omission non-salient information and simplification of medical jargon.", "### GEM-Specific Curation", "#### Modificatied for GEM?\n\n\n\nno", "#### Additional Splits?\n\n\n\nno", "### Getting Started with the Task", "## Previous Results", "### Previous Results", "#### Measured Model Abilities\n\n\n\nThis dataset measures the ability for a model to simplify paragraphs of medical text through the omission non-salient information and simplification of medical jargon.", "#### Metrics\n\n\n\n'Other: Other Metrics', 'BLEU'", "#### Other Metrics\n\n\n\nSARI measures the quality of text simplification", "#### Previous results available?\n\n\n\nyes", "#### Relevant Previous Results\n\n\n\nThe paper which introduced this dataset trained BART models (pretrained on XSum) with unlikelihood training to produce simplification models achieving maximum SARI and BLEU scores of 40 and 43 respectively.", "## Dataset Curation", "### Original Curation", "#### Sourced from Different Sources\n\n\n\nno", "### Language Data", "#### Data Validation\n\n\n\nnot validated", "#### Was Data Filtered?\n\n\n\nnot filtered", "### Structured Annotations", "#### Additional Annotations?\n\n\n\n\nnone", "#### Annotation Service?\n\n\n\nno", "### Consent", "#### Any Consent Policy?\n\n\n\nno", "### Private Identifying Information (PII)", "#### Contains PII?\n\n\n\n\nyes/very likely", "#### Any PII Identification?\n\n\n\nno identification", "### Maintenance", "#### Any Maintenance Plan?\n\n\n\nno", "## Broader Social Context", "### Previous Work on the Social Impact of the Dataset", "#### Usage of Models based on the Data\n\n\n\nno", "### Impact on Under-Served Communities", "#### Addresses needs of underserved Communities?\n\n\n\nyes", "#### Details on how Dataset Addresses the Needs\n\n\n\nThis dataset can be used to simplify medical texts that may otherwise be inaccessible to those without medical training.", "### Discussion of Biases", "#### Any Documented Social Biases?\n\n\n\nunsure", "#### Are the Language Producers Representative of the Language?\n\n\n\nThe dataset was generated from abstracts and plain-language summaries of medical literature reviews that were written by medical professionals and thus does was not generated by people representative of the entire English-speaking population.", "## Considerations for Using the Data", "### PII Risks and Liability", "### Licenses", "### Known Technical Limitations", "#### Technical Limitations\n\n\n\nThe main limitation of this dataset is that the information alignment between the abstract and plain-language summary is often rough, so the plain-language summary may contain information that isn't found in the abstract. Furthermore, the plain-language targets often contain formulaic statements like \"this evidence is current to [month][year]\" not found in the abstracts. Another limitation is that some plain-language summaries do not simplify the technical abstracts very much and still contain medical jargon.", "#### Unsuited Applications\n\n\n\nThe main pitfall to look out for is errors in factuality. Simplification work so far has not placed a strong emphasis on the logical fidelity of model generations with the input text, and the paper introducing this dataset does not explore modeling techniques to combat this. These kinds of errors are especially pernicious in the medical domain, and the models introduced in the paper do occasionally alter entities like disease and medication names." ]

[ "task_categories:other", "annotations_creators:none", "language_creators:unknown", "multilinguality:unknown", "size_categories:unknown", "source_datasets:original", "language:en", "license:mit", "reasoning", "arxiv:1911.03705", "arxiv:1910.13461", "arxiv:2009.12677", "arxiv:2012.00366", "arxiv:1910.10683", "arxiv:2006.08315", "region:us" ]

[ "1911.03705", "1910.13461", "2009.12677", "2012.00366", "1910.10683", "2006.08315" ]

[ "en" ]

[ "### Link to Main Data Card\n\n\nYou can find the main data card on the GEM Website.", "### Dataset Summary\n\n\nCommonGen is an English text generation task to explicitly test machines for the ability of generative commonsense reasoning. Given a set of common concepts, the task is to generate a coherent sentence describing an everyday scenario using these concepts. CommonGen is challenging because it inherently requires 1) relational reasoning using background commonsense knowledge, and 2) compositional generalization ability to work on unseen concept combinations. The dataset, constructed through a combination of crowd-sourcing from AMT and existing caption corpora, consists of 30k concept-sets and 50k sentences in total. Note that the CommonGen test set is private and requires submission to the external leaderboard.\n\n\nYou can load the dataset via:\n\n\nThe data loader can be found here.", "#### website\n\n\nlink", "#### paper\n\n\nLink", "#### authors\n\n\nBill Yuchen Lin (USC), Wangchunshu Zhou (USC), Ming Shen (USC), Pei Zhou (USC), Chandra Bhagavatula (AllenAI), Yejin Choi (AllenAI + UW), Xiang Ren (USC)\n\n\nDataset Overview\n----------------", "### Where to find the Data and its Documentation", "#### Webpage\n\n\nlink", "#### Download\n\n\nLink", "#### Paper\n\n\nLink", "#### BibTex", "#### Contact Name\n\n\nBill Yuchen Lin", "#### Contact Email\n\n\nURL@URL", "#### Has a Leaderboard?\n\n\nyes", "#### Leaderboard Link\n\n\nLink", "#### Leaderboard Details\n\n\nThe model outputs are evaluated against the crowdsourced references, and ranked by SPICE score. The leaderboard also reports BLEU-4 and CIDEr scores.", "### Languages and Intended Use", "#### Multilingual?\n\n\nno", "#### Covered Dialects\n\n\nNo information is provided on regional restrictions and we thus assume that the covered dialects are those spoken by raters on Mechanical Turk.", "#### Covered Languages\n\n\n'English'", "#### Whose Language?\n\n\nThe concepts were extracted from multiple English image captioning datasets and the data was collected via Amazon Mechanical Turk. No information on regional restrictions is provided.", "#### License\n\n\nmit: MIT License", "#### Intended Use\n\n\nCommonGen is a constrained text generation task, associated with a benchmark dataset, to explicitly test machines for the ability of generative commonsense reasoning.", "#### Primary Task\n\n\nReasoning", "#### Communicative Goal\n\n\nThe speaker is required to produce a *coherent* sentence which mentions all of the source concepts, and which describes a *likely* situation that could be captured in a picture or video.", "### Credit", "#### Curation Organization Type(s)\n\n\n'academic', 'independent'", "#### Curation Organization(s)\n\n\nThe dataset was curated by a joint team of researchers from the University of Southern California and Allen Institute for Artificial Intelligence.", "#### Dataset Creators\n\n\nBill Yuchen Lin (USC), Wangchunshu Zhou (USC), Ming Shen (USC), Pei Zhou (USC), Chandra Bhagavatula (AllenAI), Yejin Choi (AllenAI + UW), Xiang Ren (USC)", "#### Funding\n\n\nThe research is based upon work supported in part by the Office of the Director of National Intelligence (ODNI), Intelligence Advanced Research Projects Activity (IARPA), the DARPA MCS program, and NSF SMA 18-29268.", "#### Who added the Dataset to GEM?\n\n\nYacine Jernite created the initial data card. It was later extended by Simon Mille. Sebastian Gehrmann migrated it to the GEMv2 format.", "### Dataset Structure", "#### Data Fields\n\n\nA data instance has the following fields:\n\n\n* 'concepts': a 'list' of 'string' values denoting the concept the system should write about. Has 3 to 5 items, constitutes the 'input' of the task.\n* 'target': a sentence 'string' mentioning all of the above mentioned 'concepts'. Constitutes the desired 'output' of the task.", "#### Example Instance", "#### Data Splits\n\n\nEach example in the dataset consists of a set of 3 to 5 concepts denoted by a single noun, verb, or adjective (the input), and a sentence using these concepts (the output). The dataset provides several such sentences for each such concept.", "#### Splitting Criteria\n\n\nThe dev and test set were created by sampling sets of concepts of size 4 or 5 (and as many of size 3 for the dev set) present in the source captioning datasets and having crowd-workers write reference sentences using these concepts.\n\n\nConversely, the training set has more concept sets of size 3 than of size 4 and 5, and uses the original captions from the source datasets as references.\n\n\nThe authors also ensured that the training, dev and test set have different combinations of unique concepts to ensure compositionality (details in Table 1).\n\n\nDataset in GEM\n--------------", "### Rationale for Inclusion in GEM", "#### Why is the Dataset in GEM?\n\n\nCommonGen is a medium sized corpus with a unique reasoning challenge and interesting evaluation possibilities.", "#### Similar Datasets\n\n\nno", "#### Ability that the Dataset measures\n\n\nCommonsense reasoning", "### GEM-Specific Curation", "#### Modificatied for GEM?\n\n\nyes", "#### GEM Modifications\n\n\n'other'", "#### Modification Details\n\n\n4 challenge sets for CommenGen were added to the GEM evaluation suite.", "#### Additional Splits?\n\n\nyes", "#### Split Information\n\n\n1. Data Shift\n\n\nWe created subsets of the training and development sets of ~500 randomly selected inputs each.\n\n\n2. Transformations\n\n\nWe applied input scrambling on a subset of 500 randomly selected test instances; the order of the concepts was randomly reassigned.\n\n\n3. Subpopulations\n\n\nWe created a subpopulation based on input length, taking into account the number of concepts the input test structures. By comparing inputs of different lengths, we can see to what extent systems are able to handle different input sizes", "#### Split Motivation\n\n\nGeneralization and Robustness", "### Getting Started with the Task", "#### Pointers to Resources\n\n\n* Two variants of BART, Knowledge Graph augemnted-BART and Enhanced Knowledge Injection Model for Commonsense Generation, hold the top two spots on the leaderboard, followed by a fine-tuned T5 model.\n* The following script shows how to download and load the data, fine-tune, and evaluate a model using the ROUGE, BLEU, and METEOR metrics: GEM sample script.\n\n\nPrevious Results\n----------------", "### Previous Results", "#### Measured Model Abilities\n\n\nCommonsense Reasoning", "#### Metrics\n\n\n'Other: Other Metrics', 'BLEU', 'ROUGE', 'METEOR'", "#### Other Metrics\n\n\n* SPICE: An evaluation metric for image captioning that is defined over scene graphs\n* CIDEr: An n-gram overlap metric based on cosine similarity between the TF-IDF weighted ngram counts", "#### Proposed Evaluation\n\n\nThe main metrics are captioning metrics since the original concept lists were extracted from captioning datasets. A human subject study with five graduate students was conducted and they were asked to rank the \"commonsense plausibility\" of two models at a time.", "#### Previous results available?\n\n\nyes", "#### Other Evaluation Approaches\n\n\nThe currently best performing model KFCNet (URL uses the same automatic evaluation but does not conduct any human evaluation.", "#### Relevant Previous Results\n\n\nThe most relevant results can be seen on the leaderboard\n\n\nDataset Curation\n----------------", "### Original Curation", "#### Original Curation Rationale\n\n\nThe dataset creators selected sets of concepts that appeared in image and video captions (as identified by a POS tagger) to ensure that a likely real-world scenario including the set could be imagined and constructed. Section 3.1 of the paper describes a sampling scheme which encourages diversity of sets while selecting common concepts.", "#### Communicative Goal\n\n\nThe speaker is required to produce a *coherent* sentence which mentions all of the source concepts, and which describes a *likely* situation that could be captured in a picture or video.", "#### Sourced from Different Sources\n\n\nyes", "#### Source Details\n\n\n* Flickr30k\n* MSCOCO\n* Conceptual Captions\n* Video captioning datasets:\n\t+ LSMDC\n\t+ ActivityNet\n\t+ VaTeX", "### Language Data", "#### How was Language Data Obtained?\n\n\n'Crowdsourced'", "#### Where was it crowdsourced?\n\n\n'Amazon Mechanical Turk'", "#### Language Producers\n\n\nThe training data consists of concept sets and captions for the source datasets. The concept sets are the sets of labels of the images or videos, selected with a heuristic to maximize diversity while ensuring that they represent likely scenarios.\n\n\nThe dev and test set sentences were created by Amazon Mechanical Turk crowd workers. The workers were shown an example generation and a set of 4 or 5 concept names along with their part-of-speech and asked to write:\n\n\n1. One sentence mentioning all of the concepts\n2. A rationale explaining how the sentence connects the concept\n\n\nA screenshot of the interface is provided in Figure 7 of the Appendix.", "#### Topics Covered\n\n\nInformation was not provided.", "#### Data Validation\n\n\nvalidated by data curator", "#### Was Data Filtered?\n\n\nalgorithmically", "#### Filter Criteria\n\n\nDuring the data collection, workers who provided rationales that were too short, failed to have good coverage of the input in their sentences, or workers whose output had a high perplexity under a GPT-2 model were disqualified from the pool and replaced with newcomers.", "### Structured Annotations", "#### Additional Annotations?\n\n\nnone", "#### Annotation Service?\n\n\nno", "### Consent", "#### Any Consent Policy?\n\n\nno", "#### Justification for Using the Data\n\n\nThe data was sourced from Mechanical Turk which means that raters were aware that their annotations may be publicly released for research purposes.", "### Private Identifying Information (PII)", "#### Contains PII?\n\n\nno PII", "#### Justification for no PII\n\n\nThe concepts are restricted to verbs, adjectives, and common nouns, and no personal information is given in the captions.", "### Maintenance", "#### Any Maintenance Plan?\n\n\nno\n\n\nBroader Social Context\n----------------------", "### Previous Work on the Social Impact of the Dataset", "#### Usage of Models based on the Data\n\n\nno", "### Impact on Under-Served Communities", "#### Addresses needs of underserved Communities?\n\n\nno", "### Discussion of Biases", "#### Any Documented Social Biases?\n\n\nno", "#### Are the Language Producers Representative of the Language?\n\n\nThe dataset is created using data from image captioning systems and might inherit some of the social biases represented therein (see e.g. Tang et al. 2020).\n\n\nAnother related concern is the exposure bias introduced by the initial selection of pictures and video, which are likely to over-represent situations that are common in the US at the expense of other parts of the world (Flickr, for example, is a US-based company founded in Canada). For more discussion of the potential impacts of exposure bias, see e.g. The Social Impact of Natural Language Processing.\n\n\nConsiderations for Using the Data\n---------------------------------", "### PII Risks and Liability", "#### Potential PII Risk\n\n\nThe concepts are restricted to verbs, adjectives, and common nouns, and no personal information is given in the captions.", "### Licenses", "#### Copyright Restrictions on the Dataset\n\n\n'open license - commercial use allowed'", "#### Copyright Restrictions on the Language Data\n\n\n'open license - commercial use allowed'", "### Known Technical Limitations", "#### Technical Limitations\n\n\nThe dataset is in English, a language with an abundance of existing resources.\n\n\nThe use of GPT-2 to validate development ant test sentences might be cause for similar concern, but we do note that the authors only use the model to discount very high perplexity sequences which is less likely to surface those biases.\n\n\nThe language in the development and test set is crowdsourced, which means that it was written by workers whose main goal was speed. This is likely to impact the quality and variety of the targets. The population of crowdsource workers is also not identically distributed as the the base population of the locations the workers come from, which may lead to different representation of situations or underlying expectations of what these situations are.", "#### Unsuited Applications\n\n\nDue to the overrepresentation of US-situations, the system may not work for users across the world. Moreover, only limited information on the dataset quality are provided and the system may fail as a result of unknown issues.", "#### Discouraged Use Cases\n\n\nAny system needs to be evaluated on a broader set of unseen concepts then provided in the dataset. Since the references for the test set are private, it is not known how well findings generalize beyond the collection methodology." ]

[ "TAGS\n#task_categories-other #annotations_creators-none #language_creators-unknown #multilinguality-unknown #size_categories-unknown #source_datasets-original #language-English #license-mit #reasoning #arxiv-1911.03705 #arxiv-1910.13461 #arxiv-2009.12677 #arxiv-2012.00366 #arxiv-1910.10683 #arxiv-2006.08315 #region-us \n", "### Link to Main Data Card\n\n\nYou can find the main data card on the GEM Website.", "### Dataset Summary\n\n\nCommonGen is an English text generation task to explicitly test machines for the ability of generative commonsense reasoning. Given a set of common concepts, the task is to generate a coherent sentence describing an everyday scenario using these concepts. CommonGen is challenging because it inherently requires 1) relational reasoning using background commonsense knowledge, and 2) compositional generalization ability to work on unseen concept combinations. The dataset, constructed through a combination of crowd-sourcing from AMT and existing caption corpora, consists of 30k concept-sets and 50k sentences in total. Note that the CommonGen test set is private and requires submission to the external leaderboard.\n\n\nYou can load the dataset via:\n\n\nThe data loader can be found here.", "#### website\n\n\nlink", "#### paper\n\n\nLink", "#### authors\n\n\nBill Yuchen Lin (USC), Wangchunshu Zhou (USC), Ming Shen (USC), Pei Zhou (USC), Chandra Bhagavatula (AllenAI), Yejin Choi (AllenAI + UW), Xiang Ren (USC)\n\n\nDataset Overview\n----------------", "### Where to find the Data and its Documentation", "#### Webpage\n\n\nlink", "#### Download\n\n\nLink", "#### Paper\n\n\nLink", "#### BibTex", "#### Contact Name\n\n\nBill Yuchen Lin", "#### Contact Email\n\n\nURL@URL", "#### Has a Leaderboard?\n\n\nyes", "#### Leaderboard Link\n\n\nLink", "#### Leaderboard Details\n\n\nThe model outputs are evaluated against the crowdsourced references, and ranked by SPICE score. The leaderboard also reports BLEU-4 and CIDEr scores.", "### Languages and Intended Use", "#### Multilingual?\n\n\nno", "#### Covered Dialects\n\n\nNo information is provided on regional restrictions and we thus assume that the covered dialects are those spoken by raters on Mechanical Turk.", "#### Covered Languages\n\n\n'English'", "#### Whose Language?\n\n\nThe concepts were extracted from multiple English image captioning datasets and the data was collected via Amazon Mechanical Turk. No information on regional restrictions is provided.", "#### License\n\n\nmit: MIT License", "#### Intended Use\n\n\nCommonGen is a constrained text generation task, associated with a benchmark dataset, to explicitly test machines for the ability of generative commonsense reasoning.", "#### Primary Task\n\n\nReasoning", "#### Communicative Goal\n\n\nThe speaker is required to produce a *coherent* sentence which mentions all of the source concepts, and which describes a *likely* situation that could be captured in a picture or video.", "### Credit", "#### Curation Organization Type(s)\n\n\n'academic', 'independent'", "#### Curation Organization(s)\n\n\nThe dataset was curated by a joint team of researchers from the University of Southern California and Allen Institute for Artificial Intelligence.", "#### Dataset Creators\n\n\nBill Yuchen Lin (USC), Wangchunshu Zhou (USC), Ming Shen (USC), Pei Zhou (USC), Chandra Bhagavatula (AllenAI), Yejin Choi (AllenAI + UW), Xiang Ren (USC)", "#### Funding\n\n\nThe research is based upon work supported in part by the Office of the Director of National Intelligence (ODNI), Intelligence Advanced Research Projects Activity (IARPA), the DARPA MCS program, and NSF SMA 18-29268.", "#### Who added the Dataset to GEM?\n\n\nYacine Jernite created the initial data card. It was later extended by Simon Mille. Sebastian Gehrmann migrated it to the GEMv2 format.", "### Dataset Structure", "#### Data Fields\n\n\nA data instance has the following fields:\n\n\n* 'concepts': a 'list' of 'string' values denoting the concept the system should write about. Has 3 to 5 items, constitutes the 'input' of the task.\n* 'target': a sentence 'string' mentioning all of the above mentioned 'concepts'. Constitutes the desired 'output' of the task.", "#### Example Instance", "#### Data Splits\n\n\nEach example in the dataset consists of a set of 3 to 5 concepts denoted by a single noun, verb, or adjective (the input), and a sentence using these concepts (the output). The dataset provides several such sentences for each such concept.", "#### Splitting Criteria\n\n\nThe dev and test set were created by sampling sets of concepts of size 4 or 5 (and as many of size 3 for the dev set) present in the source captioning datasets and having crowd-workers write reference sentences using these concepts.\n\n\nConversely, the training set has more concept sets of size 3 than of size 4 and 5, and uses the original captions from the source datasets as references.\n\n\nThe authors also ensured that the training, dev and test set have different combinations of unique concepts to ensure compositionality (details in Table 1).\n\n\nDataset in GEM\n--------------", "### Rationale for Inclusion in GEM", "#### Why is the Dataset in GEM?\n\n\nCommonGen is a medium sized corpus with a unique reasoning challenge and interesting evaluation possibilities.", "#### Similar Datasets\n\n\nno", "#### Ability that the Dataset measures\n\n\nCommonsense reasoning", "### GEM-Specific Curation", "#### Modificatied for GEM?\n\n\nyes", "#### GEM Modifications\n\n\n'other'", "#### Modification Details\n\n\n4 challenge sets for CommenGen were added to the GEM evaluation suite.", "#### Additional Splits?\n\n\nyes", "#### Split Information\n\n\n1. Data Shift\n\n\nWe created subsets of the training and development sets of ~500 randomly selected inputs each.\n\n\n2. Transformations\n\n\nWe applied input scrambling on a subset of 500 randomly selected test instances; the order of the concepts was randomly reassigned.\n\n\n3. Subpopulations\n\n\nWe created a subpopulation based on input length, taking into account the number of concepts the input test structures. By comparing inputs of different lengths, we can see to what extent systems are able to handle different input sizes", "#### Split Motivation\n\n\nGeneralization and Robustness", "### Getting Started with the Task", "#### Pointers to Resources\n\n\n* Two variants of BART, Knowledge Graph augemnted-BART and Enhanced Knowledge Injection Model for Commonsense Generation, hold the top two spots on the leaderboard, followed by a fine-tuned T5 model.\n* The following script shows how to download and load the data, fine-tune, and evaluate a model using the ROUGE, BLEU, and METEOR metrics: GEM sample script.\n\n\nPrevious Results\n----------------", "### Previous Results", "#### Measured Model Abilities\n\n\nCommonsense Reasoning", "#### Metrics\n\n\n'Other: Other Metrics', 'BLEU', 'ROUGE', 'METEOR'", "#### Other Metrics\n\n\n* SPICE: An evaluation metric for image captioning that is defined over scene graphs\n* CIDEr: An n-gram overlap metric based on cosine similarity between the TF-IDF weighted ngram counts", "#### Proposed Evaluation\n\n\nThe main metrics are captioning metrics since the original concept lists were extracted from captioning datasets. A human subject study with five graduate students was conducted and they were asked to rank the \"commonsense plausibility\" of two models at a time.", "#### Previous results available?\n\n\nyes", "#### Other Evaluation Approaches\n\n\nThe currently best performing model KFCNet (URL uses the same automatic evaluation but does not conduct any human evaluation.", "#### Relevant Previous Results\n\n\nThe most relevant results can be seen on the leaderboard\n\n\nDataset Curation\n----------------", "### Original Curation", "#### Original Curation Rationale\n\n\nThe dataset creators selected sets of concepts that appeared in image and video captions (as identified by a POS tagger) to ensure that a likely real-world scenario including the set could be imagined and constructed. Section 3.1 of the paper describes a sampling scheme which encourages diversity of sets while selecting common concepts.", "#### Communicative Goal\n\n\nThe speaker is required to produce a *coherent* sentence which mentions all of the source concepts, and which describes a *likely* situation that could be captured in a picture or video.", "#### Sourced from Different Sources\n\n\nyes", "#### Source Details\n\n\n* Flickr30k\n* MSCOCO\n* Conceptual Captions\n* Video captioning datasets:\n\t+ LSMDC\n\t+ ActivityNet\n\t+ VaTeX", "### Language Data", "#### How was Language Data Obtained?\n\n\n'Crowdsourced'", "#### Where was it crowdsourced?\n\n\n'Amazon Mechanical Turk'", "#### Language Producers\n\n\nThe training data consists of concept sets and captions for the source datasets. The concept sets are the sets of labels of the images or videos, selected with a heuristic to maximize diversity while ensuring that they represent likely scenarios.\n\n\nThe dev and test set sentences were created by Amazon Mechanical Turk crowd workers. The workers were shown an example generation and a set of 4 or 5 concept names along with their part-of-speech and asked to write:\n\n\n1. One sentence mentioning all of the concepts\n2. A rationale explaining how the sentence connects the concept\n\n\nA screenshot of the interface is provided in Figure 7 of the Appendix.", "#### Topics Covered\n\n\nInformation was not provided.", "#### Data Validation\n\n\nvalidated by data curator", "#### Was Data Filtered?\n\n\nalgorithmically", "#### Filter Criteria\n\n\nDuring the data collection, workers who provided rationales that were too short, failed to have good coverage of the input in their sentences, or workers whose output had a high perplexity under a GPT-2 model were disqualified from the pool and replaced with newcomers.", "### Structured Annotations", "#### Additional Annotations?\n\n\nnone", "#### Annotation Service?\n\n\nno", "### Consent", "#### Any Consent Policy?\n\n\nno", "#### Justification for Using the Data\n\n\nThe data was sourced from Mechanical Turk which means that raters were aware that their annotations may be publicly released for research purposes.", "### Private Identifying Information (PII)", "#### Contains PII?\n\n\nno PII", "#### Justification for no PII\n\n\nThe concepts are restricted to verbs, adjectives, and common nouns, and no personal information is given in the captions.", "### Maintenance", "#### Any Maintenance Plan?\n\n\nno\n\n\nBroader Social Context\n----------------------", "### Previous Work on the Social Impact of the Dataset", "#### Usage of Models based on the Data\n\n\nno", "### Impact on Under-Served Communities", "#### Addresses needs of underserved Communities?\n\n\nno", "### Discussion of Biases", "#### Any Documented Social Biases?\n\n\nno", "#### Are the Language Producers Representative of the Language?\n\n\nThe dataset is created using data from image captioning systems and might inherit some of the social biases represented therein (see e.g. Tang et al. 2020).\n\n\nAnother related concern is the exposure bias introduced by the initial selection of pictures and video, which are likely to over-represent situations that are common in the US at the expense of other parts of the world (Flickr, for example, is a US-based company founded in Canada). For more discussion of the potential impacts of exposure bias, see e.g. The Social Impact of Natural Language Processing.\n\n\nConsiderations for Using the Data\n---------------------------------", "### PII Risks and Liability", "#### Potential PII Risk\n\n\nThe concepts are restricted to verbs, adjectives, and common nouns, and no personal information is given in the captions.", "### Licenses", "#### Copyright Restrictions on the Dataset\n\n\n'open license - commercial use allowed'", "#### Copyright Restrictions on the Language Data\n\n\n'open license - commercial use allowed'", "### Known Technical Limitations", "#### Technical Limitations\n\n\nThe dataset is in English, a language with an abundance of existing resources.\n\n\nThe use of GPT-2 to validate development ant test sentences might be cause for similar concern, but we do note that the authors only use the model to discount very high perplexity sequences which is less likely to surface those biases.\n\n\nThe language in the development and test set is crowdsourced, which means that it was written by workers whose main goal was speed. This is likely to impact the quality and variety of the targets. The population of crowdsource workers is also not identically distributed as the the base population of the locations the workers come from, which may lead to different representation of situations or underlying expectations of what these situations are.", "#### Unsuited Applications\n\n\nDue to the overrepresentation of US-situations, the system may not work for users across the world. Moreover, only limited information on the dataset quality are provided and the system may fail as a result of unknown issues.", "#### Discouraged Use Cases\n\n\nAny system needs to be evaluated on a broader set of unseen concepts then provided in the dataset. Since the references for the test set are private, it is not known how well findings generalize beyond the collection methodology." ]

[ "task_categories:table-to-text", "annotations_creators:none", "language_creators:unknown", "multilinguality:unknown", "size_categories:unknown", "source_datasets:original", "language:en", "license:cc-by-nc-4.0", "data-to-text", "region:us" ]

[]

[ "en" ]

[ "# Dataset Card for GEM/conversational_weather", "## Dataset Description\n\n- Homepage: \n- Repository: URL\n- Paper: URL\n- Leaderboard: N/A\n- Point of Contact: Kartikeya Upasani", "### Link to Main Data Card\n\nYou can find the main data card on the GEM Website.", "### Dataset Summary \n\nThe purpose of this dataset is to assess how well a model can learn a template-like structure in a very low data setting. The task here is to produce a response to a weather-related query. The reply is further specified through the data attributes and discourse structure in the input. The output contains both the lexicalized text and discourse markers for attributes (e.g., '_ARG_TEMP_ 34'). \n\nYou can load the dataset via:\n\nThe data loader can be found here.", "#### paper\nACL Anthology", "#### authors\nAnusha Balakrishnan, Jinfeng Rao, Kartikeya Upasani, Michael White, Rajen Subba (Facebook Conversational AI)", "## Dataset Overview", "### Where to find the Data and its Documentation", "#### Download\n\n\n\nGithub", "#### Paper\n\n\n\nACL Anthology", "#### BibTex", "#### Contact Name\n\n\n\n\nKartikeya Upasani", "#### Contact Email\n\n\n\nkart@URL", "#### Has a Leaderboard?\n\n\n\nno", "### Languages and Intended Use", "#### Multilingual?\n\n\n\n\nno", "#### Covered Languages\n\n\n\n\n'English'", "#### License\n\n\n\n\ncc-by-nc-4.0: Creative Commons Attribution Non Commercial 4.0 International", "#### Intended Use\n\n\n\nThis dataset is intended to help develop conversational agents that exhibit human-like properties such as matching the framing of the response with the query or contrasting relevant data attributes.", "#### Primary Task\n\n\n\nData-to-Text", "#### Communicative Goal\n\n\n\n\nProducing a text that is a response to a weather query as per the discourse structure and data attributes specified in the input meaning representation.", "### Credit", "#### Curation Organization Type(s)\n\n\n\n'industry'", "#### Curation Organization(s)\n\n\n\nFacebook", "#### Dataset Creators\n\n\n\nAnusha Balakrishnan, Jinfeng Rao, Kartikeya Upasani, Michael White, Rajen Subba (Facebook Conversational AI)", "#### Funding\n\n\n\nFacebook", "#### Who added the Dataset to GEM?\n\n\n\nVipul Raheja (Grammarly)", "### Dataset Structure", "#### Data Fields\n\n\n\n- 'gem_id': (string): GEM-formatted row id\n- 'id': (string): Row id in the original data\n- 'user_query': (string): Natural language weather query from humans\n- 'tree_str_mr': (string): Synthetically-added user context (datetime and location) in the form of a tree-structured MR\n- 'response': (string): A tree-structured annotation of the response.", "#### Example Instance", "#### Data Splits\n\n\n\n- Standard Splits: Train/Validation/Test\n- Additional Split: Disc_Test (a more challenging subset of the test set that contains discourse relations)", "#### Splitting Criteria\n\n\n\nThe test set contains 3,121 examples, of which 1.1K (35%) have unique MRs that have never been seen in the training set.", "####", "## Dataset in GEM", "### Rationale for Inclusion in GEM", "#### Why is the Dataset in GEM?\n\n\n\nThe dataset was curated to develop a weather bot that exhibits human-like properties such as matching the framing of the response with the query or contrasting relevant data attributes. \n\nThe dataset offers rich tree-based meaning representations that offer fine-grained control over the response, e.g. by specifying which two attributes are to be contrasted. The natural language input queries are also provided to model the coherence of the response based on the input. The output response is annotated with the input meaning components using special bracketing tokens, which enables developing new techniques such as constrained decoding to improve quality of output responses", "#### Similar Datasets\n\n\n\nno", "#### Ability that the Dataset measures\n\n\n\nAdequately expressing CONTRAST and JUSTIFY discourse relations with appropriate grouping of arguments; adequately generalizing to many combinations of arguments.", "### GEM-Specific Curation", "#### Modificatied for GEM?\n\n\n\nyes", "#### GEM Modifications\n\n\n\n'data points removed'", "#### Modification Details\n\n\n\nThe original repo contained a challenge set disc_test.tsv, which is a subset of the test set consisting of discourse relations (CONTRAST and JUSTIFY) , but also contained JOIN relations.\nThis discrepancy has been rectified in the GEM version. The rectified version has been added in the 'challenge_sets'", "#### Additional Splits?\n\n\n\nno", "### Getting Started with the Task", "## Previous Results", "### Previous Results", "#### Measured Model Abilities\n\n\n\nAdequately expressing CONTRAST and JUSTIFY discourse relations with appropriate grouping of arguments; adequately generalizing to many combinations of arguments.", "#### Metrics\n\n\n\n'BLEU', 'Other: Other Metrics'", "#### Other Metrics\n\n\n\nTree accuracy: It measures whether the tree structure in the prediction matches that of the input MR exactly (modulo repeated arguments that need only appear once).", "#### Proposed Evaluation\n\n\n\nAutomatic metrics are evaluated on the raw model predictions (which have de-lexicalized fields):\n* Tree accuracy: Measures whether the tree structure in the prediction matches that of the input MR exactly. \n* BLEU-4: A word overlap metric commonly used for evaluating NLG systems.\n\nAuthors also performed human evaluation studies by asking annotators to evaluate the quality of responses produced by different models. Annotators provided binary ratings on the following dimensions:\n• Grammaticality: Measures fluency of the responses. \n• Correctness: Measures semantic correctness of the responses.", "#### Previous results available?\n\n\n\nno", "## Dataset Curation", "### Original Curation", "#### Original Curation Rationale\n\n\n\nThe dataset was curated to develop a weather bot that exhibits human-like properties such as matching the framing of the response with the query or contrasting relevant data attributes. To achieve this, the dataset contains rich tree-structured meaning representations that are specified using several data arguments and discourse acts, the input natural language queries, and annotations for the responses.", "#### Communicative Goal\n\n\n\nProducing a text that is a response to a weather query as per the discourse structure and data attributes specified in the input meaning representation.", "#### Sourced from Different Sources\n\n\n\nno", "### Language Data", "#### How was Language Data Obtained?\n\n\n\n'Crowdsourced', 'Machine-generated'", "#### Where was it crowdsourced?\n\n\n\n'Other crowdworker platform'", "#### Topics Covered\n\n\n\nThe dataset is focused on the weather domain: Weather was the first successful case of NLG put into production back in the 80s (Reiter & Dale, 1997). This domain offers significant complexity for NLG. Weather forecast summaries in particular can be very long, and require reasoning over several disjoint pieces of information.", "#### Data Validation\n\n\n\nvalidated by crowdworker", "#### Data Preprocessing\n\n\n\nPlease refer to Appendix D of the original paper for details.", "#### Was Data Filtered?\n\n\n\nhybrid", "#### Filter Criteria\n\n\n\nPlease refer to Appendix C of the original paper for details.", "### Structured Annotations", "#### Additional Annotations?\n\n\n\n\nnone", "#### Annotation Service?\n\n\n\nno", "### Consent", "#### Any Consent Policy?\n\n\n\nno", "#### Justification for Using the Data\n\n\n\nAnnotation was done as work for hire and contains no PII.", "### Private Identifying Information (PII)", "#### Contains PII?\n\n\n\n\nno PII", "#### Justification for no PII\n\n\n\nData is simulated and not specific to annotator.", "### Maintenance", "#### Any Maintenance Plan?\n\n\n\nno", "## Broader Social Context", "### Previous Work on the Social Impact of the Dataset", "#### Usage of Models based on the Data\n\n\n\nno", "### Impact on Under-Served Communities", "#### Addresses needs of underserved Communities?\n\n\n\nno", "### Discussion of Biases", "#### Any Documented Social Biases?\n\n\n\nunsure", "#### Are the Language Producers Representative of the Language?\n\n\n\nGrammatical evaluations performed with the data to date have used norms from informal Standard American English. These prescriptive notions of grammaticality potentially serve to perpetuate systemic power imbalances as they’re conveyed by language. \n\nSince the data only contains informal Standard American English, its use to train a model may not be appropriate depending on the potential use case.", "## Considerations for Using the Data", "### PII Risks and Liability", "#### Potential PII Risk\n\n\n\nAnnotation was done as work for hire and contains no PII. Annotated data is simulated and not specific to annotator.", "### Licenses", "### Known Technical Limitations", "#### Unsuited Applications\n\n\n\nAn imperfect model used to convey actual weather data could mislead users about weather conditions?" ]

[ "TAGS\n#task_categories-table-to-text #annotations_creators-none #language_creators-unknown #multilinguality-unknown #size_categories-unknown #source_datasets-original #language-English #license-cc-by-nc-4.0 #data-to-text #region-us \n", "# Dataset Card for GEM/conversational_weather", "## Dataset Description\n\n- Homepage: \n- Repository: URL\n- Paper: URL\n- Leaderboard: N/A\n- Point of Contact: Kartikeya Upasani", "### Link to Main Data Card\n\nYou can find the main data card on the GEM Website.", "### Dataset Summary \n\nThe purpose of this dataset is to assess how well a model can learn a template-like structure in a very low data setting. The task here is to produce a response to a weather-related query. The reply is further specified through the data attributes and discourse structure in the input. The output contains both the lexicalized text and discourse markers for attributes (e.g., '_ARG_TEMP_ 34'). \n\nYou can load the dataset via:\n\nThe data loader can be found here.", "#### paper\nACL Anthology", "#### authors\nAnusha Balakrishnan, Jinfeng Rao, Kartikeya Upasani, Michael White, Rajen Subba (Facebook Conversational AI)", "## Dataset Overview", "### Where to find the Data and its Documentation", "#### Download\n\n\n\nGithub", "#### Paper\n\n\n\nACL Anthology", "#### BibTex", "#### Contact Name\n\n\n\n\nKartikeya Upasani", "#### Contact Email\n\n\n\nkart@URL", "#### Has a Leaderboard?\n\n\n\nno", "### Languages and Intended Use", "#### Multilingual?\n\n\n\n\nno", "#### Covered Languages\n\n\n\n\n'English'", "#### License\n\n\n\n\ncc-by-nc-4.0: Creative Commons Attribution Non Commercial 4.0 International", "#### Intended Use\n\n\n\nThis dataset is intended to help develop conversational agents that exhibit human-like properties such as matching the framing of the response with the query or contrasting relevant data attributes.", "#### Primary Task\n\n\n\nData-to-Text", "#### Communicative Goal\n\n\n\n\nProducing a text that is a response to a weather query as per the discourse structure and data attributes specified in the input meaning representation.", "### Credit", "#### Curation Organization Type(s)\n\n\n\n'industry'", "#### Curation Organization(s)\n\n\n\nFacebook", "#### Dataset Creators\n\n\n\nAnusha Balakrishnan, Jinfeng Rao, Kartikeya Upasani, Michael White, Rajen Subba (Facebook Conversational AI)", "#### Funding\n\n\n\nFacebook", "#### Who added the Dataset to GEM?\n\n\n\nVipul Raheja (Grammarly)", "### Dataset Structure", "#### Data Fields\n\n\n\n- 'gem_id': (string): GEM-formatted row id\n- 'id': (string): Row id in the original data\n- 'user_query': (string): Natural language weather query from humans\n- 'tree_str_mr': (string): Synthetically-added user context (datetime and location) in the form of a tree-structured MR\n- 'response': (string): A tree-structured annotation of the response.", "#### Example Instance", "#### Data Splits\n\n\n\n- Standard Splits: Train/Validation/Test\n- Additional Split: Disc_Test (a more challenging subset of the test set that contains discourse relations)", "#### Splitting Criteria\n\n\n\nThe test set contains 3,121 examples, of which 1.1K (35%) have unique MRs that have never been seen in the training set.", "####", "## Dataset in GEM", "### Rationale for Inclusion in GEM", "#### Why is the Dataset in GEM?\n\n\n\nThe dataset was curated to develop a weather bot that exhibits human-like properties such as matching the framing of the response with the query or contrasting relevant data attributes. \n\nThe dataset offers rich tree-based meaning representations that offer fine-grained control over the response, e.g. by specifying which two attributes are to be contrasted. The natural language input queries are also provided to model the coherence of the response based on the input. The output response is annotated with the input meaning components using special bracketing tokens, which enables developing new techniques such as constrained decoding to improve quality of output responses", "#### Similar Datasets\n\n\n\nno", "#### Ability that the Dataset measures\n\n\n\nAdequately expressing CONTRAST and JUSTIFY discourse relations with appropriate grouping of arguments; adequately generalizing to many combinations of arguments.", "### GEM-Specific Curation", "#### Modificatied for GEM?\n\n\n\nyes", "#### GEM Modifications\n\n\n\n'data points removed'", "#### Modification Details\n\n\n\nThe original repo contained a challenge set disc_test.tsv, which is a subset of the test set consisting of discourse relations (CONTRAST and JUSTIFY) , but also contained JOIN relations.\nThis discrepancy has been rectified in the GEM version. The rectified version has been added in the 'challenge_sets'", "#### Additional Splits?\n\n\n\nno", "### Getting Started with the Task", "## Previous Results", "### Previous Results", "#### Measured Model Abilities\n\n\n\nAdequately expressing CONTRAST and JUSTIFY discourse relations with appropriate grouping of arguments; adequately generalizing to many combinations of arguments.", "#### Metrics\n\n\n\n'BLEU', 'Other: Other Metrics'", "#### Other Metrics\n\n\n\nTree accuracy: It measures whether the tree structure in the prediction matches that of the input MR exactly (modulo repeated arguments that need only appear once).", "#### Proposed Evaluation\n\n\n\nAutomatic metrics are evaluated on the raw model predictions (which have de-lexicalized fields):\n* Tree accuracy: Measures whether the tree structure in the prediction matches that of the input MR exactly. \n* BLEU-4: A word overlap metric commonly used for evaluating NLG systems.\n\nAuthors also performed human evaluation studies by asking annotators to evaluate the quality of responses produced by different models. Annotators provided binary ratings on the following dimensions:\n• Grammaticality: Measures fluency of the responses. \n• Correctness: Measures semantic correctness of the responses.", "#### Previous results available?\n\n\n\nno", "## Dataset Curation", "### Original Curation", "#### Original Curation Rationale\n\n\n\nThe dataset was curated to develop a weather bot that exhibits human-like properties such as matching the framing of the response with the query or contrasting relevant data attributes. To achieve this, the dataset contains rich tree-structured meaning representations that are specified using several data arguments and discourse acts, the input natural language queries, and annotations for the responses.", "#### Communicative Goal\n\n\n\nProducing a text that is a response to a weather query as per the discourse structure and data attributes specified in the input meaning representation.", "#### Sourced from Different Sources\n\n\n\nno", "### Language Data", "#### How was Language Data Obtained?\n\n\n\n'Crowdsourced', 'Machine-generated'", "#### Where was it crowdsourced?\n\n\n\n'Other crowdworker platform'", "#### Topics Covered\n\n\n\nThe dataset is focused on the weather domain: Weather was the first successful case of NLG put into production back in the 80s (Reiter & Dale, 1997). This domain offers significant complexity for NLG. Weather forecast summaries in particular can be very long, and require reasoning over several disjoint pieces of information.", "#### Data Validation\n\n\n\nvalidated by crowdworker", "#### Data Preprocessing\n\n\n\nPlease refer to Appendix D of the original paper for details.", "#### Was Data Filtered?\n\n\n\nhybrid", "#### Filter Criteria\n\n\n\nPlease refer to Appendix C of the original paper for details.", "### Structured Annotations", "#### Additional Annotations?\n\n\n\n\nnone", "#### Annotation Service?\n\n\n\nno", "### Consent", "#### Any Consent Policy?\n\n\n\nno", "#### Justification for Using the Data\n\n\n\nAnnotation was done as work for hire and contains no PII.", "### Private Identifying Information (PII)", "#### Contains PII?\n\n\n\n\nno PII", "#### Justification for no PII\n\n\n\nData is simulated and not specific to annotator.", "### Maintenance", "#### Any Maintenance Plan?\n\n\n\nno", "## Broader Social Context", "### Previous Work on the Social Impact of the Dataset", "#### Usage of Models based on the Data\n\n\n\nno", "### Impact on Under-Served Communities", "#### Addresses needs of underserved Communities?\n\n\n\nno", "### Discussion of Biases", "#### Any Documented Social Biases?\n\n\n\nunsure", "#### Are the Language Producers Representative of the Language?\n\n\n\nGrammatical evaluations performed with the data to date have used norms from informal Standard American English. These prescriptive notions of grammaticality potentially serve to perpetuate systemic power imbalances as they’re conveyed by language. \n\nSince the data only contains informal Standard American English, its use to train a model may not be appropriate depending on the potential use case.", "## Considerations for Using the Data", "### PII Risks and Liability", "#### Potential PII Risk\n\n\n\nAnnotation was done as work for hire and contains no PII. Annotated data is simulated and not specific to annotator.", "### Licenses", "### Known Technical Limitations", "#### Unsuited Applications\n\n\n\nAn imperfect model used to convey actual weather data could mislead users about weather conditions?" ]

[ "task_categories:conversational", "annotations_creators:none", "language_creators:unknown", "multilinguality:unknown", "size_categories:unknown", "source_datasets:original", "language:cs", "license:cc-by-sa-4.0", "dialog-response-generation", "region:us" ]

[]

[ "cs" ]

[ "### Link to Main Data Card\n\n\nYou can find the main data card on the GEM Website.", "### Dataset Summary\n\n\nThe Czech Restaurants dataset is a task oriented dialog dataset in which a model needs to verbalize a response that a service agent could provide which is specified through a series of dialog acts. The dataset originated as a translation of an English dataset to test the generation capabilities of an NLG system on a highly morphologically rich language like Czech.\n\n\nYou can load the dataset via:\n\n\nThe data loader can be found here.", "#### website\n\n\nn/a", "#### paper\n\n\nGithub", "#### authors\n\n\nOndrej Dusek and Filip Jurcicek\n\n\nDataset Overview\n----------------", "### Where to find the Data and its Documentation", "#### Download\n\n\nGithub", "#### Paper\n\n\nGithub", "#### BibTex", "#### Contact Name\n\n\nOndrej Dusek", "#### Contact Email\n\n\nodusek@URL", "#### Has a Leaderboard?\n\n\nno", "### Languages and Intended Use", "#### Multilingual?\n\n\nno", "#### Covered Dialects\n\n\nNo breakdown of dialects is provided.", "#### Covered Languages\n\n\n'Czech'", "#### Whose Language?\n\n\nSix professional translators produced the outputs", "#### License\n\n\ncc-by-sa-4.0: Creative Commons Attribution Share Alike 4.0 International", "#### Intended Use\n\n\nThe dataset was created to test neural NLG systems in Czech and their ability to deal with rich morphology.", "#### Primary Task\n\n\nDialog Response Generation", "#### Communicative Goal\n\n\nProducing a text expressing the given intent/dialogue act and all and only the attributes specified in the input meaning representation.", "### Credit", "#### Curation Organization Type(s)\n\n\n'academic'", "#### Curation Organization(s)\n\n\nCharles University, Prague", "#### Dataset Creators\n\n\nOndrej Dusek and Filip Jurcicek", "#### Funding\n\n\nThis research was supported by the Charles University project PRIMUS/19/SCI/10 and by the Ministry of Education, Youth and Sports of the Czech Republic under the grant agreement LK11221. This work used using language resources distributed by the LINDAT/CLARIN project of the Ministry of Education, Youth and Sports of the Czech Republic (project LM2015071).", "#### Who added the Dataset to GEM?\n\n\nSimon Mille wrote the initial data card and Yacine Jernite the data loader. Sebastian Gehrmann migrated the data card and loader to the v2 format.", "### Dataset Structure", "#### Data Fields\n\n\nThe data is stored in a JSON or CSV format, with identical contents. The data has 4 fields:\n\n\n* 'da': the input meaning representation/dialogue act (MR)\n* 'delex\\_da': the input MR, delexicalized -- all slot values are replaced with placeholders, such as 'X-name'\n* 'text': the corresponding target natural language text (reference)\n* 'delex\\_text': the target text, delexicalized (delexicalization is applied regardless of inflection)\n\n\nIn addition, the data contains a JSON file with all possible inflected forms for all slot values in the dataset ('surface\\_forms.json').\nEach slot -> value entry contains a list of inflected forms for the given value, with the base form (lemma), the inflected form, and\na morphological tag.\n\n\nThe same MR is often repeated multiple times with different synonymous reference texts.", "#### Reason for Structure\n\n\nThe data originated as a translation and localization of Wen et al.'s SF restaurant NLG dataset.", "#### How were labels chosen?\n\n\nThe input MRs were collected from Wen et al.'s SF restaurant NLG data\nand localized by randomly replacing slot values (using a list of Prague restaurant names, neighborhoods etc.).\n\n\nThe generated slot values were then automatically replaced in reference texts in the data.", "#### Example Instance", "#### Data Splits\n\n\n\nThe data is split in a roughly 3:1:1 proportion into training, development and test sections, making sure no delexicalized MR\nappears in two different parts. On the other hand, most DA types/intents are represented in all data parts.", "#### Splitting Criteria\n\n\nThe creators ensured that after delexicalization of the meaning representation there was no overlap between training and test.\n\n\nThe data is split at a 3:1:1 rate between training, validation, and test.\n\n\nDataset in GEM\n--------------", "### Rationale for Inclusion in GEM", "#### Why is the Dataset in GEM?\n\n\nThis is one of a few non-English data-to-text datasets, in a well-known domain, but covering a morphologically rich language that is harder to generate since named entities need to be inflected. This makes it harder to apply common techniques such as delexicalization or copy mechanisms.", "#### Similar Datasets\n\n\nyes", "#### Unique Language Coverage\n\n\nyes", "#### Difference from other GEM datasets\n\n\nThe dialog acts in this dataset are much more varied than the e2e dataset which is the closest in style.", "#### Ability that the Dataset measures\n\n\nsurface realization", "### GEM-Specific Curation", "#### Modificatied for GEM?\n\n\nyes", "#### Additional Splits?\n\n\nyes", "#### Split Information\n\n\n5 challenge sets for the Czech Restaurants dataset were added to the GEM evaluation suite.\n\n\n1. Data shift: We created subsets of the training and development sets of 500 randomly selected inputs each.\n2. Scrambling: We applied input scrambling on a subset of 500 randomly selected test instances; the order of the input dialogue acts was randomly reassigned.\n3. We identified different subsets of the test set that we could compare to each other so that we would have a better understanding of the results. There are currently two selections that we have made:\n\n\nThe first comparison is based on input size: the number of predicates differs between different inputs, ranging from 1 to 5.\nThe table below provides an indication of the distribution of inputs with a particular length.\nIt is clear from the table that this distribution is not balanced, and comparisions between items should be done with caution.\nParticularly for input size 4 and 5, there may not be enough data to draw reliable conclusions.\n\n\n\nThe second comparison is based on the type of act. Again we caution against comparing the different groups that have relatively few items.\nIt is probably OK to compare 'inform' and '?request', but the other acts are all low-frequent.", "#### Split Motivation\n\n\nGeneralization and robustness.", "### Getting Started with the Task", "#### Technical Terms\n\n\n* utterance: something a system or user may say in a turn\n* meaning representation: a representation of meaning that the system should be in accordance with. The specific type of MR in this dataset are dialog acts which describe what a dialog system should do, e.g., inform a user about a value.\n\n\nPrevious Results\n----------------", "### Previous Results", "#### Measured Model Abilities\n\n\nSurface realization", "#### Metrics\n\n\n'BLEU', 'ROUGE', 'METEOR'", "#### Proposed Evaluation\n\n\nThis dataset uses the suite of word-overlap-based automatic metrics from the E2E NLG Challenge (BLEU, NIST, ROUGE-L, METEOR, and CIDEr). In addition, the slot error rate is measured.", "#### Previous results available?\n\n\nno\n\n\nDataset Curation\n----------------", "### Original Curation", "#### Original Curation Rationale\n\n\nThe dataset was created to test neural NLG systems in Czech and their ability to deal with rich morphology.", "#### Communicative Goal\n\n\nProducing a text expressing the given intent/dialogue act and all and only the attributes specified in the input MR.", "#### Sourced from Different Sources\n\n\nno", "### Language Data", "#### How was Language Data Obtained?\n\n\n'Created for the dataset'", "#### Creation Process\n\n\nSix professional translators translated the underlying dataset with the following instructions:\n\n\n* Each utterance should be translated by itself\n* fluent spoken-style Czech should be produced\n* Facts should be preserved\n* If possible, synonyms should be varied to create diverse utterances\n* Entity names should be inflected as necessary\n* the reader of the generated text should be addressed using formal form and self-references should use the female form.\n\n\nThe translators did not have access to the meaning representation.", "#### Data Validation\n\n\nvalidated by data curator", "#### Was Data Filtered?\n\n\nnot filtered", "### Structured Annotations", "#### Additional Annotations?\n\n\nnone", "#### Annotation Service?\n\n\nno", "### Consent", "#### Any Consent Policy?\n\n\nno", "#### Justification for Using the Data\n\n\nIt was not explicitly stated but we can safely assume that the translators agreed to this use of their data.", "### Private Identifying Information (PII)", "#### Contains PII?\n\n\nno PII", "#### Justification for no PII\n\n\nThis dataset does not include any information about individuals.", "### Maintenance", "#### Any Maintenance Plan?\n\n\nno\n\n\nBroader Social Context\n----------------------", "### Previous Work on the Social Impact of the Dataset", "#### Usage of Models based on the Data\n\n\nno", "### Impact on Under-Served Communities", "#### Addresses needs of underserved Communities?\n\n\nyes", "#### Details on how Dataset Addresses the Needs\n\n\nThe dataset may help improve NLG methods for morphologically rich languages beyond Czech.", "### Discussion of Biases", "#### Any Documented Social Biases?\n\n\nyes", "#### Links and Summaries of Analysis Work\n\n\nTo ensure consistency of translation, the data always uses formal/polite address for the user, and uses the female form for first-person self-references (as if the dialogue agent producing the sentences was female). This prevents data sparsity and ensures consistent results for systems trained on the dataset, but does not represent all potential situations arising in Czech.\n\n\nConsiderations for Using the Data\n---------------------------------", "### PII Risks and Liability", "### Licenses", "#### Copyright Restrictions on the Dataset\n\n\n'open license - commercial use allowed'", "#### Copyright Restrictions on the Language Data\n\n\n'open license - commercial use allowed'", "### Known Technical Limitations", "#### Technical Limitations\n\n\nThe test set may lead users to over-estimate the performance of their NLG systems with respect to their generalisability, because there are no unseen restaurants or addresses in the test set. This is something we will look into for future editions of the GEM shared task." ]

[ "TAGS\n#task_categories-conversational #annotations_creators-none #language_creators-unknown #multilinguality-unknown #size_categories-unknown #source_datasets-original #language-Czech #license-cc-by-sa-4.0 #dialog-response-generation #region-us \n", "### Link to Main Data Card\n\n\nYou can find the main data card on the GEM Website.", "### Dataset Summary\n\n\nThe Czech Restaurants dataset is a task oriented dialog dataset in which a model needs to verbalize a response that a service agent could provide which is specified through a series of dialog acts. The dataset originated as a translation of an English dataset to test the generation capabilities of an NLG system on a highly morphologically rich language like Czech.\n\n\nYou can load the dataset via:\n\n\nThe data loader can be found here.", "#### website\n\n\nn/a", "#### paper\n\n\nGithub", "#### authors\n\n\nOndrej Dusek and Filip Jurcicek\n\n\nDataset Overview\n----------------", "### Where to find the Data and its Documentation", "#### Download\n\n\nGithub", "#### Paper\n\n\nGithub", "#### BibTex", "#### Contact Name\n\n\nOndrej Dusek", "#### Contact Email\n\n\nodusek@URL", "#### Has a Leaderboard?\n\n\nno", "### Languages and Intended Use", "#### Multilingual?\n\n\nno", "#### Covered Dialects\n\n\nNo breakdown of dialects is provided.", "#### Covered Languages\n\n\n'Czech'", "#### Whose Language?\n\n\nSix professional translators produced the outputs", "#### License\n\n\ncc-by-sa-4.0: Creative Commons Attribution Share Alike 4.0 International", "#### Intended Use\n\n\nThe dataset was created to test neural NLG systems in Czech and their ability to deal with rich morphology.", "#### Primary Task\n\n\nDialog Response Generation", "#### Communicative Goal\n\n\nProducing a text expressing the given intent/dialogue act and all and only the attributes specified in the input meaning representation.", "### Credit", "#### Curation Organization Type(s)\n\n\n'academic'", "#### Curation Organization(s)\n\n\nCharles University, Prague", "#### Dataset Creators\n\n\nOndrej Dusek and Filip Jurcicek", "#### Funding\n\n\nThis research was supported by the Charles University project PRIMUS/19/SCI/10 and by the Ministry of Education, Youth and Sports of the Czech Republic under the grant agreement LK11221. This work used using language resources distributed by the LINDAT/CLARIN project of the Ministry of Education, Youth and Sports of the Czech Republic (project LM2015071).", "#### Who added the Dataset to GEM?\n\n\nSimon Mille wrote the initial data card and Yacine Jernite the data loader. Sebastian Gehrmann migrated the data card and loader to the v2 format.", "### Dataset Structure", "#### Data Fields\n\n\nThe data is stored in a JSON or CSV format, with identical contents. The data has 4 fields:\n\n\n* 'da': the input meaning representation/dialogue act (MR)\n* 'delex\\_da': the input MR, delexicalized -- all slot values are replaced with placeholders, such as 'X-name'\n* 'text': the corresponding target natural language text (reference)\n* 'delex\\_text': the target text, delexicalized (delexicalization is applied regardless of inflection)\n\n\nIn addition, the data contains a JSON file with all possible inflected forms for all slot values in the dataset ('surface\\_forms.json').\nEach slot -> value entry contains a list of inflected forms for the given value, with the base form (lemma), the inflected form, and\na morphological tag.\n\n\nThe same MR is often repeated multiple times with different synonymous reference texts.", "#### Reason for Structure\n\n\nThe data originated as a translation and localization of Wen et al.'s SF restaurant NLG dataset.", "#### How were labels chosen?\n\n\nThe input MRs were collected from Wen et al.'s SF restaurant NLG data\nand localized by randomly replacing slot values (using a list of Prague restaurant names, neighborhoods etc.).\n\n\nThe generated slot values were then automatically replaced in reference texts in the data.", "#### Example Instance", "#### Data Splits\n\n\n\nThe data is split in a roughly 3:1:1 proportion into training, development and test sections, making sure no delexicalized MR\nappears in two different parts. On the other hand, most DA types/intents are represented in all data parts.", "#### Splitting Criteria\n\n\nThe creators ensured that after delexicalization of the meaning representation there was no overlap between training and test.\n\n\nThe data is split at a 3:1:1 rate between training, validation, and test.\n\n\nDataset in GEM\n--------------", "### Rationale for Inclusion in GEM", "#### Why is the Dataset in GEM?\n\n\nThis is one of a few non-English data-to-text datasets, in a well-known domain, but covering a morphologically rich language that is harder to generate since named entities need to be inflected. This makes it harder to apply common techniques such as delexicalization or copy mechanisms.", "#### Similar Datasets\n\n\nyes", "#### Unique Language Coverage\n\n\nyes", "#### Difference from other GEM datasets\n\n\nThe dialog acts in this dataset are much more varied than the e2e dataset which is the closest in style.", "#### Ability that the Dataset measures\n\n\nsurface realization", "### GEM-Specific Curation", "#### Modificatied for GEM?\n\n\nyes", "#### Additional Splits?\n\n\nyes", "#### Split Information\n\n\n5 challenge sets for the Czech Restaurants dataset were added to the GEM evaluation suite.\n\n\n1. Data shift: We created subsets of the training and development sets of 500 randomly selected inputs each.\n2. Scrambling: We applied input scrambling on a subset of 500 randomly selected test instances; the order of the input dialogue acts was randomly reassigned.\n3. We identified different subsets of the test set that we could compare to each other so that we would have a better understanding of the results. There are currently two selections that we have made:\n\n\nThe first comparison is based on input size: the number of predicates differs between different inputs, ranging from 1 to 5.\nThe table below provides an indication of the distribution of inputs with a particular length.\nIt is clear from the table that this distribution is not balanced, and comparisions between items should be done with caution.\nParticularly for input size 4 and 5, there may not be enough data to draw reliable conclusions.\n\n\n\nThe second comparison is based on the type of act. Again we caution against comparing the different groups that have relatively few items.\nIt is probably OK to compare 'inform' and '?request', but the other acts are all low-frequent.", "#### Split Motivation\n\n\nGeneralization and robustness.", "### Getting Started with the Task", "#### Technical Terms\n\n\n* utterance: something a system or user may say in a turn\n* meaning representation: a representation of meaning that the system should be in accordance with. The specific type of MR in this dataset are dialog acts which describe what a dialog system should do, e.g., inform a user about a value.\n\n\nPrevious Results\n----------------", "### Previous Results", "#### Measured Model Abilities\n\n\nSurface realization", "#### Metrics\n\n\n'BLEU', 'ROUGE', 'METEOR'", "#### Proposed Evaluation\n\n\nThis dataset uses the suite of word-overlap-based automatic metrics from the E2E NLG Challenge (BLEU, NIST, ROUGE-L, METEOR, and CIDEr). In addition, the slot error rate is measured.", "#### Previous results available?\n\n\nno\n\n\nDataset Curation\n----------------", "### Original Curation", "#### Original Curation Rationale\n\n\nThe dataset was created to test neural NLG systems in Czech and their ability to deal with rich morphology.", "#### Communicative Goal\n\n\nProducing a text expressing the given intent/dialogue act and all and only the attributes specified in the input MR.", "#### Sourced from Different Sources\n\n\nno", "### Language Data", "#### How was Language Data Obtained?\n\n\n'Created for the dataset'", "#### Creation Process\n\n\nSix professional translators translated the underlying dataset with the following instructions:\n\n\n* Each utterance should be translated by itself\n* fluent spoken-style Czech should be produced\n* Facts should be preserved\n* If possible, synonyms should be varied to create diverse utterances\n* Entity names should be inflected as necessary\n* the reader of the generated text should be addressed using formal form and self-references should use the female form.\n\n\nThe translators did not have access to the meaning representation.", "#### Data Validation\n\n\nvalidated by data curator", "#### Was Data Filtered?\n\n\nnot filtered", "### Structured Annotations", "#### Additional Annotations?\n\n\nnone", "#### Annotation Service?\n\n\nno", "### Consent", "#### Any Consent Policy?\n\n\nno", "#### Justification for Using the Data\n\n\nIt was not explicitly stated but we can safely assume that the translators agreed to this use of their data.", "### Private Identifying Information (PII)", "#### Contains PII?\n\n\nno PII", "#### Justification for no PII\n\n\nThis dataset does not include any information about individuals.", "### Maintenance", "#### Any Maintenance Plan?\n\n\nno\n\n\nBroader Social Context\n----------------------", "### Previous Work on the Social Impact of the Dataset", "#### Usage of Models based on the Data\n\n\nno", "### Impact on Under-Served Communities", "#### Addresses needs of underserved Communities?\n\n\nyes", "#### Details on how Dataset Addresses the Needs\n\n\nThe dataset may help improve NLG methods for morphologically rich languages beyond Czech.", "### Discussion of Biases", "#### Any Documented Social Biases?\n\n\nyes", "#### Links and Summaries of Analysis Work\n\n\nTo ensure consistency of translation, the data always uses formal/polite address for the user, and uses the female form for first-person self-references (as if the dialogue agent producing the sentences was female). This prevents data sparsity and ensures consistent results for systems trained on the dataset, but does not represent all potential situations arising in Czech.\n\n\nConsiderations for Using the Data\n---------------------------------", "### PII Risks and Liability", "### Licenses", "#### Copyright Restrictions on the Dataset\n\n\n'open license - commercial use allowed'", "#### Copyright Restrictions on the Language Data\n\n\n'open license - commercial use allowed'", "### Known Technical Limitations", "#### Technical Limitations\n\n\nThe test set may lead users to over-estimate the performance of their NLG systems with respect to their generalisability, because there are no unseen restaurants or addresses in the test set. This is something we will look into for future editions of the GEM shared task." ]

[ "task_categories:table-to-text", "annotations_creators:none", "language_creators:unknown", "multilinguality:unknown", "size_categories:unknown", "source_datasets:original", "language:en", "license:mit", "data-to-text", "arxiv:1910.13461", "arxiv:1908.09022", "arxiv:2007.02871", "arxiv:1709.00103", "arxiv:1706.09254", "arxiv:1810.01170", "region:us" ]

[ "1910.13461", "1908.09022", "2007.02871", "1709.00103", "1706.09254", "1810.01170" ]

[ "en" ]

[ "# Dataset Card for GEM/dart", "## Dataset Description\n\n- Homepage: n/a\n- Repository: URL\n- Paper: URL\n- Leaderboard: URL\n- Point of Contact: Dragomir Radev, Rui Zhang, Nazneen Rajani", "### Link to Main Data Card\n\nYou can find the main data card on the GEM Website.", "### Dataset Summary \n\nDART is an English dataset aggregating multiple other data-to-text dataset in a common triple-based format. The new format is completely flat, thus not requiring a model to learn hierarchical structures, while still retaining the full information. \n\nYou can load the dataset via:\n\nThe data loader can be found here.", "#### website\nn/a", "#### paper\nACL Anthology", "#### authors\nLinyong Nan, Dragomir Radev, Rui Zhang, Amrit Rau, Abhinand Sivaprasad, Chiachun Hsieh, Xiangru Tang, Aadit Vyas, Neha Verma, Pranav Krishna, Yangxiaokang Liu, Nadia Irwanto, Jessica Pan, Faiaz Rahman, Ahmad Zaidi, Mutethia Mutuma, Yasin Tarabar, Ankit Gupta, Tao Yu, Yi Chern Tan, Xi Victoria Lin, Caiming Xiong, Richard Socher, Nazneen Fatema Rajani", "## Dataset Overview", "### Where to find the Data and its Documentation", "#### Download\n\n\n\nGithub", "#### Paper\n\n\n\nACL Anthology", "#### BibTex", "#### Contact Name\n\n\n\n\nDragomir Radev, Rui Zhang, Nazneen Rajani", "#### Contact Email\n\n\n\n{URL, r.zhang}@URL, {URL}@URL", "#### Has a Leaderboard?\n\n\n\nyes", "#### Leaderboard Link\n\n\n\nLeaderboard", "#### Leaderboard Details\n\n\n\nSeveral state-of-the-art table-to-text models were evaluated on DART, such as BART (Lewis et al., 2020), Seq2Seq-Att (MELBOURNE) and End-to-End Transformer (Castro Ferreira et al., 2019).\nThe leaderboard reports BLEU, METEOR, TER, MoverScore, BERTScore and BLEURT scores.", "### Languages and Intended Use", "#### Multilingual?\n\n\n\n\nno", "#### Covered Dialects\n\n\n\nIt is an aggregated from multiple other datasets that use general US-American or British English without differentiation between dialects.", "#### Covered Languages\n\n\n\n\n'English'", "#### Whose Language?\n\n\n\nThe dataset is aggregated from multiple others that were crowdsourced on different platforms.", "#### License\n\n\n\n\nmit: MIT License", "#### Intended Use\n\n\n\nThe dataset is aimed to further research in natural language generation from semantic data.", "#### Primary Task\n\n\n\nData-to-Text", "#### Communicative Goal\n\n\n\n\nThe speaker is required to produce coherent sentences and construct a trees structured ontology of the column headers.", "### Credit", "#### Curation Organization Type(s)\n\n\n\n'academic', 'industry'", "#### Curation Organization(s)\n\n\n\nYale University, Salesforce Research, Penn State University, The University of Hong Kong, MIT", "#### Dataset Creators\n\n\n\nLinyong Nan, Dragomir Radev, Rui Zhang, Amrit Rau, Abhinand Sivaprasad, Chiachun Hsieh, Xiangru Tang, Aadit Vyas, Neha Verma, Pranav Krishna, Yangxiaokang Liu, Nadia Irwanto, Jessica Pan, Faiaz Rahman, Ahmad Zaidi, Mutethia Mutuma, Yasin Tarabar, Ankit Gupta, Tao Yu, Yi Chern Tan, Xi Victoria Lin, Caiming Xiong, Richard Socher, Nazneen Fatema Rajani", "#### Who added the Dataset to GEM?\n\n\n\nMiruna Clinciu contributed the original data card and Yacine Jernite wrote the initial data loader. Sebastian Gehrmann migrated the data card and the loader to the new format.", "### Dataset Structure", "#### Data Fields\n\n\n\n-'tripleset': a list of tuples, each tuple has 3 items\n-'subtree_was_extended': a boolean variable (true or false)\n-'annotations': a list of dict, each with source and text keys.\n-'source': a string mentioning the name of the source table.\n-'text': a sentence string.", "#### Reason for Structure\n\n\n\nThe structure is supposed to be able more complex structures beyond \"flat\" attribute-value pairs, instead encoding hierarchical relationships.", "#### How were labels chosen?\n\n\n\nThey are a combination of those from existing datasets and new annotations that take advantage of the hierarchical structure", "#### Example Instance", "#### Data Splits\n\n\n\n|Input Unit | Examples | Vocab Size | Words per SR | Sents per SR | Tables |\n| ------------- | ------------- || ------------- || ------------- || ------------- || ------------- |\n|Triple Set | 82,191 | 33.2K | 21.6 | 1.5 | 5,623 |\n\n| Train | Dev | Test|\n| ------------- | ------------- || ------------- |\n| 62,659 | 6,980 | 12,552|\n\n\nStatistics of DART decomposed by different collection methods. DART exhibits a great deal of topical variety in terms of the number of unique predicates, the number of unique triples, and the vocabulary size. These statistics are computed from DART v1.1.1; the number of unique predicates reported is post-unification (see Section 3.4). SR: Surface Realization.\n(details in Table 1 and 2).", "#### Splitting Criteria\n\n\n\nFor WebNLG 2017 and Cleaned E2E, DART use the original data splits. For the new annotation on WikiTableQuestions and WikiSQL, random splitting will make train, dev, and test splits contain similar tables and similar <triple-set, sentence> examples. They are thus split based on Jaccard similarity such that no training examples has a similarity with a test example of over 0.5", "## Dataset in GEM", "### Rationale for Inclusion in GEM", "#### Why is the Dataset in GEM?\n\n\n\nDART is a large and open-domain structured DAta Record to Text generation corpus with high-quality sentence annotations with each input being a set of entity-relation triples following a tree-structured ontology.", "#### Similar Datasets\n\n\n\nyes", "#### Unique Language Coverage\n\n\n\nno", "#### Difference from other GEM datasets\n\n\n\nThe tree structure is unique among GEM datasets", "#### Ability that the Dataset measures\n\n\n\nReasoning, surface realization", "### GEM-Specific Curation", "#### Modificatied for GEM?\n\n\n\nno", "#### Additional Splits?\n\n\n\nno", "### Getting Started with the Task", "#### Pointers to Resources\n\n\n\nExperimental results on DART shows that BART model as the highest performance among three models with a BLEU score of 37.06. This is attributed to BART’s generalization ability due to pretraining (Table 4).", "## Previous Results", "### Previous Results", "#### Measured Model Abilities\n\n\n\nReasoning, surface realization", "#### Metrics\n\n\n\n'BLEU', 'MoverScore', 'BERT-Score', 'BLEURT'", "#### Proposed Evaluation\n\n\n\nThe leaderboard uses the combination of BLEU, METEOR, TER, MoverScore, BERTScore, PARENT and BLEURT to overcome the limitations of the n-gram overlap metrics. \nA small scale human annotation of 100 data points was conducted along the dimensions of (1) fluency - a sentence is natural and grammatical, and (2) semantic faithfulness - a sentence is supported by the input triples.", "#### Previous results available?\n\n\n\nyes", "#### Other Evaluation Approaches\n\n\n\nn/a", "#### Relevant Previous Results\n\n\n\nBART currently achieves the best performance according to the leaderboard.", "## Dataset Curation", "### Original Curation", "#### Original Curation Rationale\n\n\n\nThe dataset creators encourage through DART further research in natural language generation from semantic data. DART provides high-quality sentence annotations with each input being a set of entity-relation triples in a tree structure.", "#### Communicative Goal\n\n\n\nThe speaker is required to produce coherent sentences and construct a trees structured ontology of the column headers.", "#### Sourced from Different Sources\n\n\n\nyes", "#### Source Details\n\n\n\n- human annotation on open-domain Wikipedia tables from WikiTableQuestions (Pasupat and Liang,\n2015) and WikiSQL (Zhong et al., 2017)\n- automatic conversion of questions in WikiSQL to declarative sentences\n- incorporation of existing datasets including WebNLG 2017 (Gardent et al., 2017a,b; Shimorina and Gardent, 2018) and Cleaned E2E (Novikova et al., 2017b; Dušek et al., 2018, 2019)", "### Language Data", "#### How was Language Data Obtained?\n\n\n\n'Found', 'Created for the dataset'", "#### Where was it found?\n\n\n\n'Offline media collection'", "#### Creation Process\n\n\n\nCreators proposed a two-stage annotation process for constructing triple set sentence pairs based on a tree-structured ontology of each table. First, internal skilled annotators denote the parent column for each column header. Then, a larger number of annotators provide a sentential description of an automatically-chosen subset of table cells in a row. To form a triple set sentence pair, the highlighted cells can be converted to a connected triple set automatically according to the column ontology for the given table.", "#### Language Producers\n\n\n\nNo further information about the MTurk workers has been provided.", "#### Topics Covered\n\n\n\nThe sub-datasets are from Wikipedia, DBPedia, and artificially created restaurant data.", "#### Data Validation\n\n\n\nvalidated by crowdworker", "#### Was Data Filtered?\n\n\n\nnot filtered", "### Structured Annotations", "#### Additional Annotations?\n\n\n\n\nnone", "#### Annotation Service?\n\n\n\nno", "### Consent", "#### Any Consent Policy?\n\n\n\nno", "#### Justification for Using the Data\n\n\n\nThe new annotations are based on Wikipedia which is in the public domain and the other two datasets permit reuse (with attribution)", "### Private Identifying Information (PII)", "#### Contains PII?\n\n\n\n\nno PII", "#### Justification for no PII\n\n\n\nNone of the datasets talk about individuals", "### Maintenance", "#### Any Maintenance Plan?\n\n\n\nno", "## Broader Social Context", "### Previous Work on the Social Impact of the Dataset", "#### Usage of Models based on the Data\n\n\n\nno", "### Impact on Under-Served Communities", "#### Addresses needs of underserved Communities?\n\n\n\nno", "### Discussion of Biases", "#### Any Documented Social Biases?\n\n\n\nno", "#### Are the Language Producers Representative of the Language?\n\n\n\nNo, the annotators are raters on crowdworking platforms and thus only represent their demographics.", "## Considerations for Using the Data", "### PII Risks and Liability", "### Licenses", "#### Copyright Restrictions on the Dataset\n\n\n\n'open license - commercial use allowed'", "#### Copyright Restrictions on the Language Data\n\n\n\n'open license - commercial use allowed'", "### Known Technical Limitations", "#### Technical Limitations\n\n\n\nThe dataset may contain some social biases, as the input sentences are based on Wikipedia (WikiTableQuestions, WikiSQL, WebNLG). Studies have shown that the English Wikipedia contains gender biases(Dinan et al., 2020), racial biases(Papakyriakopoulos et al., 2020 (URL and geographical bias([Livingstone et al., 2010). More info.", "#### Unsuited Applications\n\n\n\nThe end-to-end transformer has the lowest performance since the transformer model needs intermediate pipeline planning steps to have higher performance. Similar findings can be found in Castro Ferreira et al., 2019." ]

[ "TAGS\n#task_categories-table-to-text #annotations_creators-none #language_creators-unknown #multilinguality-unknown #size_categories-unknown #source_datasets-original #language-English #license-mit #data-to-text #arxiv-1910.13461 #arxiv-1908.09022 #arxiv-2007.02871 #arxiv-1709.00103 #arxiv-1706.09254 #arxiv-1810.01170 #region-us \n", "# Dataset Card for GEM/dart", "## Dataset Description\n\n- Homepage: n/a\n- Repository: URL\n- Paper: URL\n- Leaderboard: URL\n- Point of Contact: Dragomir Radev, Rui Zhang, Nazneen Rajani", "### Link to Main Data Card\n\nYou can find the main data card on the GEM Website.", "### Dataset Summary \n\nDART is an English dataset aggregating multiple other data-to-text dataset in a common triple-based format. The new format is completely flat, thus not requiring a model to learn hierarchical structures, while still retaining the full information. \n\nYou can load the dataset via:\n\nThe data loader can be found here.", "#### website\nn/a", "#### paper\nACL Anthology", "#### authors\nLinyong Nan, Dragomir Radev, Rui Zhang, Amrit Rau, Abhinand Sivaprasad, Chiachun Hsieh, Xiangru Tang, Aadit Vyas, Neha Verma, Pranav Krishna, Yangxiaokang Liu, Nadia Irwanto, Jessica Pan, Faiaz Rahman, Ahmad Zaidi, Mutethia Mutuma, Yasin Tarabar, Ankit Gupta, Tao Yu, Yi Chern Tan, Xi Victoria Lin, Caiming Xiong, Richard Socher, Nazneen Fatema Rajani", "## Dataset Overview", "### Where to find the Data and its Documentation", "#### Download\n\n\n\nGithub", "#### Paper\n\n\n\nACL Anthology", "#### BibTex", "#### Contact Name\n\n\n\n\nDragomir Radev, Rui Zhang, Nazneen Rajani", "#### Contact Email\n\n\n\n{URL, r.zhang}@URL, {URL}@URL", "#### Has a Leaderboard?\n\n\n\nyes", "#### Leaderboard Link\n\n\n\nLeaderboard", "#### Leaderboard Details\n\n\n\nSeveral state-of-the-art table-to-text models were evaluated on DART, such as BART (Lewis et al., 2020), Seq2Seq-Att (MELBOURNE) and End-to-End Transformer (Castro Ferreira et al., 2019).\nThe leaderboard reports BLEU, METEOR, TER, MoverScore, BERTScore and BLEURT scores.", "### Languages and Intended Use", "#### Multilingual?\n\n\n\n\nno", "#### Covered Dialects\n\n\n\nIt is an aggregated from multiple other datasets that use general US-American or British English without differentiation between dialects.", "#### Covered Languages\n\n\n\n\n'English'", "#### Whose Language?\n\n\n\nThe dataset is aggregated from multiple others that were crowdsourced on different platforms.", "#### License\n\n\n\n\nmit: MIT License", "#### Intended Use\n\n\n\nThe dataset is aimed to further research in natural language generation from semantic data.", "#### Primary Task\n\n\n\nData-to-Text", "#### Communicative Goal\n\n\n\n\nThe speaker is required to produce coherent sentences and construct a trees structured ontology of the column headers.", "### Credit", "#### Curation Organization Type(s)\n\n\n\n'academic', 'industry'", "#### Curation Organization(s)\n\n\n\nYale University, Salesforce Research, Penn State University, The University of Hong Kong, MIT", "#### Dataset Creators\n\n\n\nLinyong Nan, Dragomir Radev, Rui Zhang, Amrit Rau, Abhinand Sivaprasad, Chiachun Hsieh, Xiangru Tang, Aadit Vyas, Neha Verma, Pranav Krishna, Yangxiaokang Liu, Nadia Irwanto, Jessica Pan, Faiaz Rahman, Ahmad Zaidi, Mutethia Mutuma, Yasin Tarabar, Ankit Gupta, Tao Yu, Yi Chern Tan, Xi Victoria Lin, Caiming Xiong, Richard Socher, Nazneen Fatema Rajani", "#### Who added the Dataset to GEM?\n\n\n\nMiruna Clinciu contributed the original data card and Yacine Jernite wrote the initial data loader. Sebastian Gehrmann migrated the data card and the loader to the new format.", "### Dataset Structure", "#### Data Fields\n\n\n\n-'tripleset': a list of tuples, each tuple has 3 items\n-'subtree_was_extended': a boolean variable (true or false)\n-'annotations': a list of dict, each with source and text keys.\n-'source': a string mentioning the name of the source table.\n-'text': a sentence string.", "#### Reason for Structure\n\n\n\nThe structure is supposed to be able more complex structures beyond \"flat\" attribute-value pairs, instead encoding hierarchical relationships.", "#### How were labels chosen?\n\n\n\nThey are a combination of those from existing datasets and new annotations that take advantage of the hierarchical structure", "#### Example Instance", "#### Data Splits\n\n\n\n|Input Unit | Examples | Vocab Size | Words per SR | Sents per SR | Tables |\n| ------------- | ------------- || ------------- || ------------- || ------------- || ------------- |\n|Triple Set | 82,191 | 33.2K | 21.6 | 1.5 | 5,623 |\n\n| Train | Dev | Test|\n| ------------- | ------------- || ------------- |\n| 62,659 | 6,980 | 12,552|\n\n\nStatistics of DART decomposed by different collection methods. DART exhibits a great deal of topical variety in terms of the number of unique predicates, the number of unique triples, and the vocabulary size. These statistics are computed from DART v1.1.1; the number of unique predicates reported is post-unification (see Section 3.4). SR: Surface Realization.\n(details in Table 1 and 2).", "#### Splitting Criteria\n\n\n\nFor WebNLG 2017 and Cleaned E2E, DART use the original data splits. For the new annotation on WikiTableQuestions and WikiSQL, random splitting will make train, dev, and test splits contain similar tables and similar <triple-set, sentence> examples. They are thus split based on Jaccard similarity such that no training examples has a similarity with a test example of over 0.5", "## Dataset in GEM", "### Rationale for Inclusion in GEM", "#### Why is the Dataset in GEM?\n\n\n\nDART is a large and open-domain structured DAta Record to Text generation corpus with high-quality sentence annotations with each input being a set of entity-relation triples following a tree-structured ontology.", "#### Similar Datasets\n\n\n\nyes", "#### Unique Language Coverage\n\n\n\nno", "#### Difference from other GEM datasets\n\n\n\nThe tree structure is unique among GEM datasets", "#### Ability that the Dataset measures\n\n\n\nReasoning, surface realization", "### GEM-Specific Curation", "#### Modificatied for GEM?\n\n\n\nno", "#### Additional Splits?\n\n\n\nno", "### Getting Started with the Task", "#### Pointers to Resources\n\n\n\nExperimental results on DART shows that BART model as the highest performance among three models with a BLEU score of 37.06. This is attributed to BART’s generalization ability due to pretraining (Table 4).", "## Previous Results", "### Previous Results", "#### Measured Model Abilities\n\n\n\nReasoning, surface realization", "#### Metrics\n\n\n\n'BLEU', 'MoverScore', 'BERT-Score', 'BLEURT'", "#### Proposed Evaluation\n\n\n\nThe leaderboard uses the combination of BLEU, METEOR, TER, MoverScore, BERTScore, PARENT and BLEURT to overcome the limitations of the n-gram overlap metrics. \nA small scale human annotation of 100 data points was conducted along the dimensions of (1) fluency - a sentence is natural and grammatical, and (2) semantic faithfulness - a sentence is supported by the input triples.", "#### Previous results available?\n\n\n\nyes", "#### Other Evaluation Approaches\n\n\n\nn/a", "#### Relevant Previous Results\n\n\n\nBART currently achieves the best performance according to the leaderboard.", "## Dataset Curation", "### Original Curation", "#### Original Curation Rationale\n\n\n\nThe dataset creators encourage through DART further research in natural language generation from semantic data. DART provides high-quality sentence annotations with each input being a set of entity-relation triples in a tree structure.", "#### Communicative Goal\n\n\n\nThe speaker is required to produce coherent sentences and construct a trees structured ontology of the column headers.", "#### Sourced from Different Sources\n\n\n\nyes", "#### Source Details\n\n\n\n- human annotation on open-domain Wikipedia tables from WikiTableQuestions (Pasupat and Liang,\n2015) and WikiSQL (Zhong et al., 2017)\n- automatic conversion of questions in WikiSQL to declarative sentences\n- incorporation of existing datasets including WebNLG 2017 (Gardent et al., 2017a,b; Shimorina and Gardent, 2018) and Cleaned E2E (Novikova et al., 2017b; Dušek et al., 2018, 2019)", "### Language Data", "#### How was Language Data Obtained?\n\n\n\n'Found', 'Created for the dataset'", "#### Where was it found?\n\n\n\n'Offline media collection'", "#### Creation Process\n\n\n\nCreators proposed a two-stage annotation process for constructing triple set sentence pairs based on a tree-structured ontology of each table. First, internal skilled annotators denote the parent column for each column header. Then, a larger number of annotators provide a sentential description of an automatically-chosen subset of table cells in a row. To form a triple set sentence pair, the highlighted cells can be converted to a connected triple set automatically according to the column ontology for the given table.", "#### Language Producers\n\n\n\nNo further information about the MTurk workers has been provided.", "#### Topics Covered\n\n\n\nThe sub-datasets are from Wikipedia, DBPedia, and artificially created restaurant data.", "#### Data Validation\n\n\n\nvalidated by crowdworker", "#### Was Data Filtered?\n\n\n\nnot filtered", "### Structured Annotations", "#### Additional Annotations?\n\n\n\n\nnone", "#### Annotation Service?\n\n\n\nno", "### Consent", "#### Any Consent Policy?\n\n\n\nno", "#### Justification for Using the Data\n\n\n\nThe new annotations are based on Wikipedia which is in the public domain and the other two datasets permit reuse (with attribution)", "### Private Identifying Information (PII)", "#### Contains PII?\n\n\n\n\nno PII", "#### Justification for no PII\n\n\n\nNone of the datasets talk about individuals", "### Maintenance", "#### Any Maintenance Plan?\n\n\n\nno", "## Broader Social Context", "### Previous Work on the Social Impact of the Dataset", "#### Usage of Models based on the Data\n\n\n\nno", "### Impact on Under-Served Communities", "#### Addresses needs of underserved Communities?\n\n\n\nno", "### Discussion of Biases", "#### Any Documented Social Biases?\n\n\n\nno", "#### Are the Language Producers Representative of the Language?\n\n\n\nNo, the annotators are raters on crowdworking platforms and thus only represent their demographics.", "## Considerations for Using the Data", "### PII Risks and Liability", "### Licenses", "#### Copyright Restrictions on the Dataset\n\n\n\n'open license - commercial use allowed'", "#### Copyright Restrictions on the Language Data\n\n\n\n'open license - commercial use allowed'", "### Known Technical Limitations", "#### Technical Limitations\n\n\n\nThe dataset may contain some social biases, as the input sentences are based on Wikipedia (WikiTableQuestions, WikiSQL, WebNLG). Studies have shown that the English Wikipedia contains gender biases(Dinan et al., 2020), racial biases(Papakyriakopoulos et al., 2020 (URL and geographical bias([Livingstone et al., 2010). More info.", "#### Unsuited Applications\n\n\n\nThe end-to-end transformer has the lowest performance since the transformer model needs intermediate pipeline planning steps to have higher performance. Similar findings can be found in Castro Ferreira et al., 2019." ]

[ "task_categories:conversational", "annotations_creators:none", "language_creators:unknown", "multilinguality:unknown", "size_categories:unknown", "source_datasets:original", "language:en", "license:apache-2.0", "dialog-response-generation", "region:us" ]

[]

[ "en" ]

[ "# Dataset Card for GEM/dstc10_track2_task2", "## Dataset Description\n\n- Homepage: URL\n- Repository: URL\n- Paper: URL.science/54/a1/5282d47044179737b4289622c824/how-robust-are-you-evaluating-URL\n- Leaderboard: URL\n- Point of Contact: Seokhwan Kim", "### Link to Main Data Card\n\nYou can find the main data card on the GEM Website.", "### Dataset Summary \n\nThe DSTC10 Track2 Task 2 follows the DSTC9 Track1 task, where participants have to implement knowledge-grounded dialog systems.\nThe training dataset is inherited from the DSTC9 challenge and is in the written domain, while the test set is newly collected and consists of noisy ASR transcripts.\nHence, the dataset facilitates building models for grounded dialog response generation.\n\nYou can load the dataset via:\n\nThe data loader can be found here.", "#### website\nURL", "#### paper\nURL.science/54/a1/5282d47044179737b4289622c824/how-robust-are-you-evaluating-URL", "#### authors\nSeokhwan Kim, Yang Liu, Di Jin, Alexandros Papangelis, Karthik Gopalakrishnan, Behnam Hedayatnia, Dilek Hakkani-Tur (Amazon Alexa AI)", "## Dataset Overview", "### Where to find the Data and its Documentation", "#### Webpage\n\n\n\nURL", "#### Download\n\n\n\nURL", "#### Paper\n\n\n\nURL.science/54/a1/5282d47044179737b4289622c824/how-robust-are-you-evaluating-URL", "#### BibTex\n\n\n\n@inproceedings{kim2021robust,\n title={\" How Robust ru?\": Evaluating Task-Oriented Dialogue Systems on Spoken Conversations},\n author={Kim, Seokhwan and Liu, Yang and Jin, Di and Papangelis, Alexandros and Gopalakrishnan, Karthik and Hedayatnia, Behnam and Hakkani-Tur, Dilek},\n journal={IEEE Automatic Speech Recognition and Understanding Workshop},\n year={2021}\n}", "#### Contact Name\n\n\n\n\nSeokhwan Kim", "#### Contact Email\n\n\n\nseokhwk@URL", "#### Has a Leaderboard?\n\n\n\nyes", "#### Leaderboard Link\n\n\n\nURL", "#### Leaderboard Details\n\n\n\nIt evaluates the models based on the automatic metrics defined in the task paper for the three tasks of detection, selection and generation.", "### Languages and Intended Use", "#### Multilingual?\n\n\n\n\nno", "#### Covered Languages\n\n\n\n\n'En'", "#### License\n\n\n\n\napache-2.0: Apache License 2.0", "#### Intended Use\n\n\n\nTo conduct research on dialogue state tracking and knowledge-grounded response generation.", "#### Primary Task\n\n\n\nDialog Response Generation", "#### Communicative Goal\n\n\n\n\nThis dataset aims to explore the robustness of conversational models when trained on spoken data. It has two aspects, multi-domain dialogue state tracking and conversation modeling with access to unstructured knowledge.", "### Credit", "#### Curation Organization Type(s)\n\n\n\n'industry'", "#### Curation Organization(s)\n\n\n\nAmazon", "#### Dataset Creators\n\n\n\nSeokhwan Kim, Yang Liu, Di Jin, Alexandros Papangelis, Karthik Gopalakrishnan, Behnam Hedayatnia, Dilek Hakkani-Tur (Amazon Alexa AI)", "#### Funding\n\n\n\nAmazon", "#### Who added the Dataset to GEM?\n\n\n\nAlexandros Papangelis (Amazon Alexa AI), Di Jin (Amazon Alexa AI), Nico Daheim (RWTH Aachen University)", "### Dataset Structure", "#### Data Fields\n\n\n\n features = datasets.Features(\n {\n \"id\": datasets.Value(\"string\"),\n \"gem_id\": datasets.Value(\"string\"),\n \"turns\": [\n {\n \"speaker\": datasets.Value(\"string\"),\n \"text\": datasets.Value(\"string\"),\n \"nbest\": [\n {\n \"hyp\": datasets.Value(\"string\"),\n \"score\": datasets.Value(\"float\"),\n }\n ],\n }\n ],\n \"knowledge\": {\n \"domain\": datasets.Value(\"string\"),\n \"entity_name\": datasets.Value(\"string\"),\n \"title\": datasets.Value(\"string\"),\n \"body\": datasets.Value(\"string\"),\n },\n \"response\": datasets.Value(\"string\"),\n \"source\": datasets.Value(\"string\"),\n \"linearized_input\": datasets.Value(\"string\"),\n \"target\": datasets.Value(\"string\"),\n \"references\": [datasets.Value(\"string\")],\n }\n )\n\nnbest contains an nbest list of outputs generated by an ASR system along with their scores.\n\nknowledge defines the annotated grounding as well as its metadata", "#### Reason for Structure\n\n\n\nIt was kept compatible with MultiWox 2.X data.", "#### Example Instance\n\n\n\n{'id': '0',\n 'gem_id': 'GEM-dstc10_track2_task2-test-0',\n 'turns': [{'speaker': 'U',\n 'text': \"hi uh i'm looking for restaurant in lower ha\",\n 'nbest': [{'hyp': \"hi uh i'm looking for restaurant in lower ha\",\n 'score': -25.625450134277344},\n {'hyp': \"hi uh i'm looking for restaurant in lower hai\",\n 'score': -25.969446182250977},\n {'hyp': \"hi uh i'm looking for restaurant in lower haig\",\n 'score': -32.816890716552734},\n {'hyp': \"hi uh i'm looking for restaurant in lower haigh\",\n 'score': -32.84316635131836},\n {'hyp': \"hi uh i'm looking for restaurant in lower hag\",\n 'score': -32.8637580871582},\n {'hyp': \"hi uh i'm looking for restaurant in lower hah\",\n 'score': -33.1048698425293},\n {'hyp': \"hi uh i'm looking for restaurant in lower hait\",\n 'score': -33.96509552001953},\n {'hyp': \"hi um i'm looking for restaurant in lower hai\",\n 'score': -33.97885513305664},\n {'hyp': \"hi um i'm looking for restaurant in lower haig\",\n 'score': -34.56083679199219},\n {'hyp': \"hi um i'm looking for restaurant in lower haigh\",\n 'score': -34.58711242675781}]},\n {'speaker': 'S',\n 'text': 'yeah definitely i can go ahead and help you with that ummm what kind of option in a restaurant are you looking for',\n 'nbest': []},\n {'speaker': 'U',\n 'text': 'yeah umm am looking for an expensive restaurant',\n 'nbest': [{'hyp': 'yeah umm am looking for an expensive restaurant',\n 'score': -21.272899627685547},\n {'hyp': 'yeah umm m looking for an expensive restaurant',\n 'score': -21.444047927856445},\n {'hyp': 'yeah umm a m looking for an expensive restaurant',\n 'score': -21.565458297729492},\n {'hyp': 'yeah ummm am looking for an expensive restaurant',\n 'score': -21.68832778930664},\n {'hyp': 'yeah ummm m looking for an expensive restaurant',\n 'score': -21.85947608947754},\n {'hyp': 'yeah ummm a m looking for an expensive restaurant',\n 'score': -21.980886459350586},\n {'hyp': \"yeah umm a'm looking for an expensive restaurant\",\n 'score': -22.613924026489258},\n {'hyp': \"yeah ummm a'm looking for an expensive restaurant\",\n 'score': -23.02935218811035},\n {'hyp': 'yeah um am looking for an expensive restaurant',\n 'score': -23.11180305480957},\n {'hyp': 'yeah um m looking for an expensive restaurant',\n 'score': -23.28295135498047}]},\n {'speaker': 'S',\n 'text': \"lemme go ahead and see what i can find for you ok great so i do ummm actually no i'm sorry is there something else i can help you find i don't see anything expensive\",\n 'nbest': []},\n {'speaker': 'U',\n 'text': \"sure ummm maybe if you don't have anything expensive how about something in the moderate price range\",\n 'nbest': [{'hyp': \"sure ummm maybe if you don't have anything expensive how about something in the moderate price range\",\n 'score': -27.492507934570312},\n {'hyp': \"sure umm maybe if you don't have anything expensive how about something in the moderate price range\",\n 'score': -27.75853729248047},\n {'hyp': \"sure ummm maybe if you don't have anything expensive how about something in the moderate price rang\",\n 'score': -29.44410514831543},\n {'hyp': \"sure umm maybe if you don't have anything expensive how about something in the moderate price rang\",\n 'score': -29.710134506225586},\n {'hyp': \"sure um maybe if you don't have anything expensive how about something in the moderate price range\",\n 'score': -31.136560440063477},\n {'hyp': \"sure um maybe if you don't have anything expensive how about something in the moderate price rang\",\n 'score': -33.088157653808594},\n {'hyp': \"sure ummm maybe i you don't have anything expensive how about something in the moderate price range\",\n 'score': -36.127620697021484},\n {'hyp': \"sure umm maybe i you don't have anything expensive how about something in the moderate price range\",\n 'score': -36.39365005493164},\n {'hyp': \"sure ummm maybe if yo don't have anything expensive how about something in the moderate price range\",\n 'score': -36.43605041503906},\n {'hyp': \"sure umm maybe if yo don't have anything expensive how about something in the moderate price range\",\n 'score': -36.70207977294922}]},\n {'speaker': 'S',\n 'text': 'ok moderate lemme go ahead and check to see what i can find for moderate ok great i do have several options coming up how does the view lounge sound',\n 'nbest': []},\n {'speaker': 'U',\n 'text': 'that sounds good ummm do they have any sort of happy hour special',\n 'nbest': [{'hyp': 'that sounds good ummm do they have any sort of happy hour special',\n 'score': -30.316478729248047},\n {'hyp': 'that sounds good umm do they have any sort of happy hour special',\n 'score': -30.958009719848633},\n {'hyp': 'that sounds good um do they have any sort of happy hour special',\n 'score': -34.463165283203125},\n {'hyp': 'that sounds good ummm do they have any sirt of happy hour special',\n 'score': -34.48350143432617},\n {'hyp': 'that sounds good umm do they have any sirt of happy hour special',\n 'score': -35.12503433227539},\n {'hyp': 'that sounds good ummm do they have any sord of happy hour special',\n 'score': -35.61939239501953},\n {'hyp': 'that sounds good umm do they have any sord of happy hour special',\n 'score': -36.26092529296875},\n {'hyp': 'that sounds good ummm do they have any sont of happy hour special',\n 'score': -37.697105407714844},\n {'hyp': 'that sounds good umm do they have any sont of happy hour special',\n 'score': -38.33863830566406},\n {'hyp': 'that sounds good um do they have any sirt of happy hour special',\n 'score': -38.630191802978516}]}],\n 'knowledge': {'domain': 'restaurant',\n 'entity_name': 'The View Lounge',\n 'title': 'Does The View Lounge offer happy hour?',\n 'body': 'The View Lounge offers happy hour.'},\n 'response': 'uhhh great question lemme go ahead and check that out for you ok fantastic so it looks like they do offer happy hour',\n 'source': 'sf_spoken',\n 'linearized_input': \"<U> hi uh i'm looking for restaurant in lower ha <S> yeah definitely i can go ahead and help you with that ummm what kind of option in a restaurant are you looking for <U> yeah umm am looking for an expensive restaurant <S> lemme go ahead and see what i can find for you ok great so i do ummm actually no i'm sorry is there something else i can help you find i don't see anything expensive <U> sure ummm maybe if you don't have anything expensive how about something in the moderate price range <S> ok moderate lemme go ahead and check to see what i can find for moderate ok great i do have several options coming up how does the view lounge sound <U> that sounds good ummm do they have any sort of happy hour special || knowledge domain: restaurant, entity: The View Lounge, title: Does The View Lounge offer happy hour?, information: The View Lounge offers happy hour.\",\n 'target': 'uhhh great question lemme go ahead and check that out for you ok fantastic so it looks like they do offer happy hour',\n 'references': ['uhhh great question lemme go ahead and check that out for you ok fantastic so it looks like they do offer happy hour']}", "#### Data Splits\n\n\n\ntrain: training set, val: validation set, test: test set", "#### Splitting Criteria\n\n\n\nThe track dataset originally only consists of a validation and test set in the spoken domain with noisy ASR transcripts.\nThe training set is taken from the predecessor task DSTC9 Track 1 and contains written conversations.", "## Dataset in GEM", "### Rationale for Inclusion in GEM", "#### Why is the Dataset in GEM?\n\n\n\nThis dataset can be used to evaluate conversational models on spoken inputs (using ASR hypotheses). In particular, we can evaluate the models’ ability to understand language by tracking the dialogue state, and their ability to generate knowledge-grounded responses.", "#### Similar Datasets\n\n\n\nyes", "#### Unique Language Coverage\n\n\n\nno", "#### Difference from other GEM datasets\n\n\n\nThis dataset contains transcribed spoken interactions.", "#### Ability that the Dataset measures\n\n\n\nWe can measure the model’s ability to understand language and to generate knowledge-grounded responses.", "### GEM-Specific Curation", "#### Modificatied for GEM?\n\n\n\nno", "#### Additional Splits?\n\n\n\nno", "### Getting Started with the Task", "## Previous Results", "### Previous Results", "#### Measured Model Abilities\n\n\n\nThis dataset can be used to evaluate conversational models on spoken inputs (using ASR hypotheses). In particular, we can evaluate the models’ ability to generate knowledge-grounded responses.", "#### Metrics\n\n\n\n'Other: Other Metrics'", "#### Other Metrics\n\n\n\nBLEU-1, BLEU-2, BLEU-3, BLEU-4, METEOR, ROGUE-1, ROGUE-2, ROGUE-L", "#### Previous results available?\n\n\n\nno", "## Dataset Curation", "### Original Curation", "#### Original Curation Rationale\n\n\n\nWe want to explore how conversational models perform on spoken data.", "#### Communicative Goal\n\n\n\nThis dataset aims to explore the robustness of conversational models when evaluated on spoken data. It has two aspects, multi-domain dialogue state tracking and conversation modeling with access to unstructured knowledge.", "#### Sourced from Different Sources\n\n\n\nno", "### Language Data", "#### How was Language Data Obtained?\n\n\n\n'Other'", "#### Topics Covered\n\n\n\nThe conversations revolve around 5 domains (or topics): hotels, restaurants, attractions, taxi, train.", "#### Data Validation\n\n\n\nnot validated", "#### Was Data Filtered?\n\n\n\nnot filtered", "### Structured Annotations", "#### Additional Annotations?\n\n\n\n\nnone", "#### Annotation Service?\n\n\n\nno", "### Consent", "#### Any Consent Policy?\n\n\n\nyes", "### Private Identifying Information (PII)", "#### Contains PII?\n\n\n\n\nno PII", "#### Justification for no PII\n\n\n\nThe subjects were instructed to conduct fictional conversations about booking restaurants or requesting fictional information.", "### Maintenance", "#### Any Maintenance Plan?\n\n\n\nno", "## Broader Social Context", "### Previous Work on the Social Impact of the Dataset", "#### Usage of Models based on the Data\n\n\n\nno", "### Impact on Under-Served Communities", "#### Addresses needs of underserved Communities?\n\n\n\nno", "### Discussion of Biases", "#### Any Documented Social Biases?\n\n\n\nunsure", "## Considerations for Using the Data", "### PII Risks and Liability", "#### Potential PII Risk\n\n\n\nThere should be no risk related to PII as the subjects conduct fictional conversations.", "### Licenses", "#### Copyright Restrictions on the Dataset\n\n\n\n'open license - commercial use allowed'", "#### Copyright Restrictions on the Language Data\n\n\n\n'open license - commercial use allowed'", "### Known Technical Limitations" ]

[ "TAGS\n#task_categories-conversational #annotations_creators-none #language_creators-unknown #multilinguality-unknown #size_categories-unknown #source_datasets-original #language-English #license-apache-2.0 #dialog-response-generation #region-us \n", "# Dataset Card for GEM/dstc10_track2_task2", "## Dataset Description\n\n- Homepage: URL\n- Repository: URL\n- Paper: URL.science/54/a1/5282d47044179737b4289622c824/how-robust-are-you-evaluating-URL\n- Leaderboard: URL\n- Point of Contact: Seokhwan Kim", "### Link to Main Data Card\n\nYou can find the main data card on the GEM Website.", "### Dataset Summary \n\nThe DSTC10 Track2 Task 2 follows the DSTC9 Track1 task, where participants have to implement knowledge-grounded dialog systems.\nThe training dataset is inherited from the DSTC9 challenge and is in the written domain, while the test set is newly collected and consists of noisy ASR transcripts.\nHence, the dataset facilitates building models for grounded dialog response generation.\n\nYou can load the dataset via:\n\nThe data loader can be found here.", "#### website\nURL", "#### paper\nURL.science/54/a1/5282d47044179737b4289622c824/how-robust-are-you-evaluating-URL", "#### authors\nSeokhwan Kim, Yang Liu, Di Jin, Alexandros Papangelis, Karthik Gopalakrishnan, Behnam Hedayatnia, Dilek Hakkani-Tur (Amazon Alexa AI)", "## Dataset Overview", "### Where to find the Data and its Documentation", "#### Webpage\n\n\n\nURL", "#### Download\n\n\n\nURL", "#### Paper\n\n\n\nURL.science/54/a1/5282d47044179737b4289622c824/how-robust-are-you-evaluating-URL", "#### BibTex\n\n\n\n@inproceedings{kim2021robust,\n title={\" How Robust ru?\": Evaluating Task-Oriented Dialogue Systems on Spoken Conversations},\n author={Kim, Seokhwan and Liu, Yang and Jin, Di and Papangelis, Alexandros and Gopalakrishnan, Karthik and Hedayatnia, Behnam and Hakkani-Tur, Dilek},\n journal={IEEE Automatic Speech Recognition and Understanding Workshop},\n year={2021}\n}", "#### Contact Name\n\n\n\n\nSeokhwan Kim", "#### Contact Email\n\n\n\nseokhwk@URL", "#### Has a Leaderboard?\n\n\n\nyes", "#### Leaderboard Link\n\n\n\nURL", "#### Leaderboard Details\n\n\n\nIt evaluates the models based on the automatic metrics defined in the task paper for the three tasks of detection, selection and generation.", "### Languages and Intended Use", "#### Multilingual?\n\n\n\n\nno", "#### Covered Languages\n\n\n\n\n'En'", "#### License\n\n\n\n\napache-2.0: Apache License 2.0", "#### Intended Use\n\n\n\nTo conduct research on dialogue state tracking and knowledge-grounded response generation.", "#### Primary Task\n\n\n\nDialog Response Generation", "#### Communicative Goal\n\n\n\n\nThis dataset aims to explore the robustness of conversational models when trained on spoken data. It has two aspects, multi-domain dialogue state tracking and conversation modeling with access to unstructured knowledge.", "### Credit", "#### Curation Organization Type(s)\n\n\n\n'industry'", "#### Curation Organization(s)\n\n\n\nAmazon", "#### Dataset Creators\n\n\n\nSeokhwan Kim, Yang Liu, Di Jin, Alexandros Papangelis, Karthik Gopalakrishnan, Behnam Hedayatnia, Dilek Hakkani-Tur (Amazon Alexa AI)", "#### Funding\n\n\n\nAmazon", "#### Who added the Dataset to GEM?\n\n\n\nAlexandros Papangelis (Amazon Alexa AI), Di Jin (Amazon Alexa AI), Nico Daheim (RWTH Aachen University)", "### Dataset Structure", "#### Data Fields\n\n\n\n features = datasets.Features(\n {\n \"id\": datasets.Value(\"string\"),\n \"gem_id\": datasets.Value(\"string\"),\n \"turns\": [\n {\n \"speaker\": datasets.Value(\"string\"),\n \"text\": datasets.Value(\"string\"),\n \"nbest\": [\n {\n \"hyp\": datasets.Value(\"string\"),\n \"score\": datasets.Value(\"float\"),\n }\n ],\n }\n ],\n \"knowledge\": {\n \"domain\": datasets.Value(\"string\"),\n \"entity_name\": datasets.Value(\"string\"),\n \"title\": datasets.Value(\"string\"),\n \"body\": datasets.Value(\"string\"),\n },\n \"response\": datasets.Value(\"string\"),\n \"source\": datasets.Value(\"string\"),\n \"linearized_input\": datasets.Value(\"string\"),\n \"target\": datasets.Value(\"string\"),\n \"references\": [datasets.Value(\"string\")],\n }\n )\n\nnbest contains an nbest list of outputs generated by an ASR system along with their scores.\n\nknowledge defines the annotated grounding as well as its metadata", "#### Reason for Structure\n\n\n\nIt was kept compatible with MultiWox 2.X data.", "#### Example Instance\n\n\n\n{'id': '0',\n 'gem_id': 'GEM-dstc10_track2_task2-test-0',\n 'turns': [{'speaker': 'U',\n 'text': \"hi uh i'm looking for restaurant in lower ha\",\n 'nbest': [{'hyp': \"hi uh i'm looking for restaurant in lower ha\",\n 'score': -25.625450134277344},\n {'hyp': \"hi uh i'm looking for restaurant in lower hai\",\n 'score': -25.969446182250977},\n {'hyp': \"hi uh i'm looking for restaurant in lower haig\",\n 'score': -32.816890716552734},\n {'hyp': \"hi uh i'm looking for restaurant in lower haigh\",\n 'score': -32.84316635131836},\n {'hyp': \"hi uh i'm looking for restaurant in lower hag\",\n 'score': -32.8637580871582},\n {'hyp': \"hi uh i'm looking for restaurant in lower hah\",\n 'score': -33.1048698425293},\n {'hyp': \"hi uh i'm looking for restaurant in lower hait\",\n 'score': -33.96509552001953},\n {'hyp': \"hi um i'm looking for restaurant in lower hai\",\n 'score': -33.97885513305664},\n {'hyp': \"hi um i'm looking for restaurant in lower haig\",\n 'score': -34.56083679199219},\n {'hyp': \"hi um i'm looking for restaurant in lower haigh\",\n 'score': -34.58711242675781}]},\n {'speaker': 'S',\n 'text': 'yeah definitely i can go ahead and help you with that ummm what kind of option in a restaurant are you looking for',\n 'nbest': []},\n {'speaker': 'U',\n 'text': 'yeah umm am looking for an expensive restaurant',\n 'nbest': [{'hyp': 'yeah umm am looking for an expensive restaurant',\n 'score': -21.272899627685547},\n {'hyp': 'yeah umm m looking for an expensive restaurant',\n 'score': -21.444047927856445},\n {'hyp': 'yeah umm a m looking for an expensive restaurant',\n 'score': -21.565458297729492},\n {'hyp': 'yeah ummm am looking for an expensive restaurant',\n 'score': -21.68832778930664},\n {'hyp': 'yeah ummm m looking for an expensive restaurant',\n 'score': -21.85947608947754},\n {'hyp': 'yeah ummm a m looking for an expensive restaurant',\n 'score': -21.980886459350586},\n {'hyp': \"yeah umm a'm looking for an expensive restaurant\",\n 'score': -22.613924026489258},\n {'hyp': \"yeah ummm a'm looking for an expensive restaurant\",\n 'score': -23.02935218811035},\n {'hyp': 'yeah um am looking for an expensive restaurant',\n 'score': -23.11180305480957},\n {'hyp': 'yeah um m looking for an expensive restaurant',\n 'score': -23.28295135498047}]},\n {'speaker': 'S',\n 'text': \"lemme go ahead and see what i can find for you ok great so i do ummm actually no i'm sorry is there something else i can help you find i don't see anything expensive\",\n 'nbest': []},\n {'speaker': 'U',\n 'text': \"sure ummm maybe if you don't have anything expensive how about something in the moderate price range\",\n 'nbest': [{'hyp': \"sure ummm maybe if you don't have anything expensive how about something in the moderate price range\",\n 'score': -27.492507934570312},\n {'hyp': \"sure umm maybe if you don't have anything expensive how about something in the moderate price range\",\n 'score': -27.75853729248047},\n {'hyp': \"sure ummm maybe if you don't have anything expensive how about something in the moderate price rang\",\n 'score': -29.44410514831543},\n {'hyp': \"sure umm maybe if you don't have anything expensive how about something in the moderate price rang\",\n 'score': -29.710134506225586},\n {'hyp': \"sure um maybe if you don't have anything expensive how about something in the moderate price range\",\n 'score': -31.136560440063477},\n {'hyp': \"sure um maybe if you don't have anything expensive how about something in the moderate price rang\",\n 'score': -33.088157653808594},\n {'hyp': \"sure ummm maybe i you don't have anything expensive how about something in the moderate price range\",\n 'score': -36.127620697021484},\n {'hyp': \"sure umm maybe i you don't have anything expensive how about something in the moderate price range\",\n 'score': -36.39365005493164},\n {'hyp': \"sure ummm maybe if yo don't have anything expensive how about something in the moderate price range\",\n 'score': -36.43605041503906},\n {'hyp': \"sure umm maybe if yo don't have anything expensive how about something in the moderate price range\",\n 'score': -36.70207977294922}]},\n {'speaker': 'S',\n 'text': 'ok moderate lemme go ahead and check to see what i can find for moderate ok great i do have several options coming up how does the view lounge sound',\n 'nbest': []},\n {'speaker': 'U',\n 'text': 'that sounds good ummm do they have any sort of happy hour special',\n 'nbest': [{'hyp': 'that sounds good ummm do they have any sort of happy hour special',\n 'score': -30.316478729248047},\n {'hyp': 'that sounds good umm do they have any sort of happy hour special',\n 'score': -30.958009719848633},\n {'hyp': 'that sounds good um do they have any sort of happy hour special',\n 'score': -34.463165283203125},\n {'hyp': 'that sounds good ummm do they have any sirt of happy hour special',\n 'score': -34.48350143432617},\n {'hyp': 'that sounds good umm do they have any sirt of happy hour special',\n 'score': -35.12503433227539},\n {'hyp': 'that sounds good ummm do they have any sord of happy hour special',\n 'score': -35.61939239501953},\n {'hyp': 'that sounds good umm do they have any sord of happy hour special',\n 'score': -36.26092529296875},\n {'hyp': 'that sounds good ummm do they have any sont of happy hour special',\n 'score': -37.697105407714844},\n {'hyp': 'that sounds good umm do they have any sont of happy hour special',\n 'score': -38.33863830566406},\n {'hyp': 'that sounds good um do they have any sirt of happy hour special',\n 'score': -38.630191802978516}]}],\n 'knowledge': {'domain': 'restaurant',\n 'entity_name': 'The View Lounge',\n 'title': 'Does The View Lounge offer happy hour?',\n 'body': 'The View Lounge offers happy hour.'},\n 'response': 'uhhh great question lemme go ahead and check that out for you ok fantastic so it looks like they do offer happy hour',\n 'source': 'sf_spoken',\n 'linearized_input': \"<U> hi uh i'm looking for restaurant in lower ha <S> yeah definitely i can go ahead and help you with that ummm what kind of option in a restaurant are you looking for <U> yeah umm am looking for an expensive restaurant <S> lemme go ahead and see what i can find for you ok great so i do ummm actually no i'm sorry is there something else i can help you find i don't see anything expensive <U> sure ummm maybe if you don't have anything expensive how about something in the moderate price range <S> ok moderate lemme go ahead and check to see what i can find for moderate ok great i do have several options coming up how does the view lounge sound <U> that sounds good ummm do they have any sort of happy hour special || knowledge domain: restaurant, entity: The View Lounge, title: Does The View Lounge offer happy hour?, information: The View Lounge offers happy hour.\",\n 'target': 'uhhh great question lemme go ahead and check that out for you ok fantastic so it looks like they do offer happy hour',\n 'references': ['uhhh great question lemme go ahead and check that out for you ok fantastic so it looks like they do offer happy hour']}", "#### Data Splits\n\n\n\ntrain: training set, val: validation set, test: test set", "#### Splitting Criteria\n\n\n\nThe track dataset originally only consists of a validation and test set in the spoken domain with noisy ASR transcripts.\nThe training set is taken from the predecessor task DSTC9 Track 1 and contains written conversations.", "## Dataset in GEM", "### Rationale for Inclusion in GEM", "#### Why is the Dataset in GEM?\n\n\n\nThis dataset can be used to evaluate conversational models on spoken inputs (using ASR hypotheses). In particular, we can evaluate the models’ ability to understand language by tracking the dialogue state, and their ability to generate knowledge-grounded responses.", "#### Similar Datasets\n\n\n\nyes", "#### Unique Language Coverage\n\n\n\nno", "#### Difference from other GEM datasets\n\n\n\nThis dataset contains transcribed spoken interactions.", "#### Ability that the Dataset measures\n\n\n\nWe can measure the model’s ability to understand language and to generate knowledge-grounded responses.", "### GEM-Specific Curation", "#### Modificatied for GEM?\n\n\n\nno", "#### Additional Splits?\n\n\n\nno", "### Getting Started with the Task", "## Previous Results", "### Previous Results", "#### Measured Model Abilities\n\n\n\nThis dataset can be used to evaluate conversational models on spoken inputs (using ASR hypotheses). In particular, we can evaluate the models’ ability to generate knowledge-grounded responses.", "#### Metrics\n\n\n\n'Other: Other Metrics'", "#### Other Metrics\n\n\n\nBLEU-1, BLEU-2, BLEU-3, BLEU-4, METEOR, ROGUE-1, ROGUE-2, ROGUE-L", "#### Previous results available?\n\n\n\nno", "## Dataset Curation", "### Original Curation", "#### Original Curation Rationale\n\n\n\nWe want to explore how conversational models perform on spoken data.", "#### Communicative Goal\n\n\n\nThis dataset aims to explore the robustness of conversational models when evaluated on spoken data. It has two aspects, multi-domain dialogue state tracking and conversation modeling with access to unstructured knowledge.", "#### Sourced from Different Sources\n\n\n\nno", "### Language Data", "#### How was Language Data Obtained?\n\n\n\n'Other'", "#### Topics Covered\n\n\n\nThe conversations revolve around 5 domains (or topics): hotels, restaurants, attractions, taxi, train.", "#### Data Validation\n\n\n\nnot validated", "#### Was Data Filtered?\n\n\n\nnot filtered", "### Structured Annotations", "#### Additional Annotations?\n\n\n\n\nnone", "#### Annotation Service?\n\n\n\nno", "### Consent", "#### Any Consent Policy?\n\n\n\nyes", "### Private Identifying Information (PII)", "#### Contains PII?\n\n\n\n\nno PII", "#### Justification for no PII\n\n\n\nThe subjects were instructed to conduct fictional conversations about booking restaurants or requesting fictional information.", "### Maintenance", "#### Any Maintenance Plan?\n\n\n\nno", "## Broader Social Context", "### Previous Work on the Social Impact of the Dataset", "#### Usage of Models based on the Data\n\n\n\nno", "### Impact on Under-Served Communities", "#### Addresses needs of underserved Communities?\n\n\n\nno", "### Discussion of Biases", "#### Any Documented Social Biases?\n\n\n\nunsure", "## Considerations for Using the Data", "### PII Risks and Liability", "#### Potential PII Risk\n\n\n\nThere should be no risk related to PII as the subjects conduct fictional conversations.", "### Licenses", "#### Copyright Restrictions on the Dataset\n\n\n\n'open license - commercial use allowed'", "#### Copyright Restrictions on the Language Data\n\n\n\n'open license - commercial use allowed'", "### Known Technical Limitations" ]

[ "task_categories:table-to-text", "annotations_creators:none", "language_creators:unknown", "multilinguality:unknown", "size_categories:unknown", "source_datasets:original", "language:en", "license:cc-by-sa-4.0", "data-to-text", "region:us" ]

[]

[ "en" ]

[ "### Link to Main Data Card\n\n\nYou can find the main data card on the [GEM Website.", "### Dataset Summary\n\n\nThe E2E NLG dataset is an English benchmark dataset for data-to-text models that verbalize a set of 2-9 key-value attribute pairs in the restaurant domain. The version used for GEM is the cleaned E2E NLG dataset, which filters examples with hallucinations and outputs that don't fully cover all input attributes.\n\n\nYou can load the dataset via:\n\n\nThe data loader can be found here.", "#### website\n\n\nWebsite", "#### paper\n\n\nFirst data release, Detailed E2E Challenge writeup, Cleaned E2E version", "#### authors\n\n\nJekaterina Novikova, Ondrej Dusek and Verena Rieser\n\n\nDataset Overview\n----------------", "### Where to find the Data and its Documentation", "#### Webpage\n\n\nWebsite", "#### Download\n\n\nGithub", "#### Paper\n\n\nFirst data release, Detailed E2E Challenge writeup, Cleaned E2E version", "#### BibTex", "#### Contact Name\n\n\nOndrej Dusek", "#### Contact Email\n\n\nodusek@URL", "#### Has a Leaderboard?\n\n\nno", "### Languages and Intended Use", "#### Multilingual?\n\n\nno", "#### Covered Dialects\n\n\nDialect-specific data was not collected and the language is general British English.", "#### Covered Languages\n\n\n'English'", "#### Whose Language?\n\n\nThe original dataset was collected using the CrowdFlower (now Appen) platform using native English speakers (self-reported). No demographic information was provided, but the collection was geographically limited to English-speaking countries.", "#### License\n\n\ncc-by-sa-4.0: Creative Commons Attribution Share Alike 4.0 International", "#### Intended Use\n\n\nThe dataset was collected to test neural model on a very well specified realization task.", "#### Primary Task\n\n\nData-to-Text", "#### Communicative Goal\n\n\nProducing a text informing/recommending a restaurant, given all and only the attributes specified on the input.", "### Credit", "#### Curation Organization Type(s)\n\n\n'academic'", "#### Curation Organization(s)\n\n\nHeriot-Watt University", "#### Dataset Creators\n\n\nJekaterina Novikova, Ondrej Dusek and Verena Rieser", "#### Funding\n\n\nThis research received funding from the EPSRC projects DILiGENt (EP/M005429/1) and MaDrIgAL (EP/N017536/1).", "#### Who added the Dataset to GEM?\n\n\nSimon Mille wrote the initial data card and Yacine Jernite the data loader. Sebastian Gehrmann migrated the data card to the v2 format and moved the data loader to the hub.", "### Dataset Structure", "#### Data Fields\n\n\nThe data is in a CSV format, with the following fields:\n\n\n* 'mr' -- the meaning representation (MR, input)\n* 'ref' -- reference, i.e. the corresponding natural-language description (output)\n\n\nThere are additional fields ('fixed', 'orig\\_mr') indicating whether the data was modified in the\ncleaning process and what was the original MR before cleaning, but these aren't used for NLG.\n\n\nThe MR has a flat structure -- attribute-value pairs are comma separated, with values\nenclosed in brackets (see example above). There are 8 attributes:\n\n\n* 'name' -- restaurant name\n* 'near' -- a landmark close to the restaurant\n* 'area' -- location (riverside, city centre)\n* 'food' -- food type / cuisine (e.g. Japanese, Indian, English etc.)\n* 'eatType' -- restaurant type (restaurant, coffee shop, pub)\n* 'priceRange' -- price range (low, medium, high, <£20, £20-30, >£30)\n* 'rating' -- customer rating (low, medium, high, 1/5, 3/5, 5/5)\n* 'familyFriendly' -- is the restaurant family-friendly (yes/no)\n\n\nThe same MR is often repeated multiple times with different synonymous references.", "#### How were labels chosen?\n\n\nThe source MRs were generated automatically at random from a set of valid attribute values. The labels were crowdsourced and are natural language", "#### Example Instance", "#### Data Splits\n\n\n\n“Distinct MRs” are MRs that remain distinct even if restaurant/place names (attributes 'name', 'near')\nare delexicalized, i.e., replaced with a placeholder.", "#### Splitting Criteria\n\n\nThe data are divided so that MRs in different splits do not overlap.\n\n\nDataset in GEM\n--------------", "### Rationale for Inclusion in GEM", "#### Why is the Dataset in GEM?\n\n\nThe E2E dataset is one of the largest limited-domain NLG datasets and is frequently used as a data-to-text generation benchmark. The E2E Challenge included 20 systems of very different architectures, with system outputs available for download.", "#### Similar Datasets\n\n\nyes", "#### Unique Language Coverage\n\n\nno", "#### Difference from other GEM datasets\n\n\nThe dataset is much cleaner than comparable datasets, and it is also a relatively easy task, making for a straightforward evaluation.", "#### Ability that the Dataset measures\n\n\nsurface realization.", "### GEM-Specific Curation", "#### Modificatied for GEM?\n\n\nyes", "#### Additional Splits?\n\n\nyes", "#### Split Information\n\n\n4 special test sets for E2E were added to the GEM evaluation suite.\n\n\n1. We created subsets of the training and development sets of ~500 randomly selected inputs each.\n2. We applied input scrambling on a subset of 500 randomly selected test instances; the order of the input properties was randomly reassigned.\n3. For the input size, we created subpopulations based on the number of restaurant properties in the input.", "#### Split Motivation\n\n\nGeneralization and robustness", "### Getting Started with the Task\n\n\nPrevious Results\n----------------", "### Previous Results", "#### Measured Model Abilities\n\n\nSurface realization.", "#### Metrics\n\n\n'BLEU', 'METEOR', 'ROUGE'", "#### Proposed Evaluation\n\n\nThe official evaluation script combines the MT-Eval and COCO Captioning libraries with the following metrics.\n\n\n* BLEU\n* CIDEr\n* NIST\n* METEOR\n* ROUGE-L", "#### Previous results available?\n\n\nyes", "#### Other Evaluation Approaches\n\n\nMost previous results, including the shared task results, used the library provided by the dataset creators. The shared task also conducted a human evaluation using the following two criteria:\n\n\n* 'Quality': When collecting quality ratings, system outputs were presented to crowd workers together with the corresponding meaning representation, which implies that correctness of the NL utterance relative to the MR should also influence this ranking. The crowd workers were asked: “How do you judge the overall quality of the utterance in terms of its grammatical correctness, fluency, adequacy and other important factors?”\n* 'Naturalness': When collecting naturalness ratings, system outputs were presented to crowd workers without the corresponding meaning representation. The crowd workers were asked: “Could the utterance have been produced by a native speaker?”", "#### Relevant Previous Results\n\n\nThe shared task writeup has in-depth evaluations of systems (URL\n\n\nDataset Curation\n----------------", "### Original Curation", "#### Original Curation Rationale\n\n\nThe dataset was collected to showcase/test neural NLG models. It is larger and contains more lexical richness and syntactic variation than previous closed-domain NLG datasets.", "#### Communicative Goal\n\n\nProducing a text informing/recommending a restaurant, given all and only the attributes specified on the input.", "#### Sourced from Different Sources\n\n\nno", "### Language Data", "#### How was Language Data Obtained?\n\n\n'Crowdsourced'", "#### Where was it crowdsourced?\n\n\n'Other crowdworker platform'", "#### Language Producers\n\n\nHuman references describing the MRs were collected by crowdsourcing on the CrowdFlower (now Appen) platform,\nwith either textual or pictorial MRs as a baseline.\nThe pictorial MRs were used in 20% of cases -- these yield higher lexical variation but introduce noise.", "#### Topics Covered\n\n\nThe dataset is focused on descriptions of restaurants.", "#### Data Validation\n\n\nvalidated by data curator", "#### Data Preprocessing\n\n\nThere were basic checks (length, valid characters, repetition).", "#### Was Data Filtered?\n\n\nalgorithmically", "#### Filter Criteria\n\n\nThe cleaned version of the dataset which we are using in GEM was algorithmically filtered. They used regular expressions to match all human-generated references with a more accurate input when attributes were hallucinated or dropped. Additionally, train-test overlap stemming from the transformation was removed. As a result, this data is much cleaner than the original dataset but not perfect (about 20% of instances may have misaligned slots, compared to 40% of the original data.", "### Structured Annotations", "#### Additional Annotations?\n\n\nnone", "#### Annotation Service?\n\n\nno", "### Consent", "#### Any Consent Policy?\n\n\nyes", "#### Consent Policy Details\n\n\nSince a crowdsourcing platform was used, the involved raters waived their rights to the data and are aware that the produced annotations can be publicly released.", "### Private Identifying Information (PII)", "#### Contains PII?\n\n\nno PII", "#### Justification for no PII\n\n\nThe dataset is artificial and does not contain any description of people.", "### Maintenance", "#### Any Maintenance Plan?\n\n\nno\n\n\nBroader Social Context\n----------------------", "### Previous Work on the Social Impact of the Dataset", "#### Usage of Models based on the Data\n\n\nno", "### Impact on Under-Served Communities", "#### Addresses needs of underserved Communities?\n\n\nno", "### Discussion of Biases", "#### Any Documented Social Biases?\n\n\nno", "#### Are the Language Producers Representative of the Language?\n\n\nThe source data is generated randomly, so it should not contain biases. The human references may be biased by the workers' demographic, but that was not investigated upon data collection.\n\n\nConsiderations for Using the Data\n---------------------------------", "### PII Risks and Liability", "### Licenses", "#### Copyright Restrictions on the Dataset\n\n\n'open license - commercial use allowed'", "#### Copyright Restrictions on the Language Data\n\n\n'open license - commercial use allowed'", "### Known Technical Limitations", "#### Technical Limitations\n\n\nThe cleaned version still has data points with hallucinated or omitted attributes.", "#### Unsuited Applications\n\n\nThe data only pertains to the restaurant domain and the included attributes. A model cannot be expected to handle other domains or attributes." ]

[ "TAGS\n#task_categories-table-to-text #annotations_creators-none #language_creators-unknown #multilinguality-unknown #size_categories-unknown #source_datasets-original #language-English #license-cc-by-sa-4.0 #data-to-text #region-us \n", "### Link to Main Data Card\n\n\nYou can find the main data card on the [GEM Website.", "### Dataset Summary\n\n\nThe E2E NLG dataset is an English benchmark dataset for data-to-text models that verbalize a set of 2-9 key-value attribute pairs in the restaurant domain. The version used for GEM is the cleaned E2E NLG dataset, which filters examples with hallucinations and outputs that don't fully cover all input attributes.\n\n\nYou can load the dataset via:\n\n\nThe data loader can be found here.", "#### website\n\n\nWebsite", "#### paper\n\n\nFirst data release, Detailed E2E Challenge writeup, Cleaned E2E version", "#### authors\n\n\nJekaterina Novikova, Ondrej Dusek and Verena Rieser\n\n\nDataset Overview\n----------------", "### Where to find the Data and its Documentation", "#### Webpage\n\n\nWebsite", "#### Download\n\n\nGithub", "#### Paper\n\n\nFirst data release, Detailed E2E Challenge writeup, Cleaned E2E version", "#### BibTex", "#### Contact Name\n\n\nOndrej Dusek", "#### Contact Email\n\n\nodusek@URL", "#### Has a Leaderboard?\n\n\nno", "### Languages and Intended Use", "#### Multilingual?\n\n\nno", "#### Covered Dialects\n\n\nDialect-specific data was not collected and the language is general British English.", "#### Covered Languages\n\n\n'English'", "#### Whose Language?\n\n\nThe original dataset was collected using the CrowdFlower (now Appen) platform using native English speakers (self-reported). No demographic information was provided, but the collection was geographically limited to English-speaking countries.", "#### License\n\n\ncc-by-sa-4.0: Creative Commons Attribution Share Alike 4.0 International", "#### Intended Use\n\n\nThe dataset was collected to test neural model on a very well specified realization task.", "#### Primary Task\n\n\nData-to-Text", "#### Communicative Goal\n\n\nProducing a text informing/recommending a restaurant, given all and only the attributes specified on the input.", "### Credit", "#### Curation Organization Type(s)\n\n\n'academic'", "#### Curation Organization(s)\n\n\nHeriot-Watt University", "#### Dataset Creators\n\n\nJekaterina Novikova, Ondrej Dusek and Verena Rieser", "#### Funding\n\n\nThis research received funding from the EPSRC projects DILiGENt (EP/M005429/1) and MaDrIgAL (EP/N017536/1).", "#### Who added the Dataset to GEM?\n\n\nSimon Mille wrote the initial data card and Yacine Jernite the data loader. Sebastian Gehrmann migrated the data card to the v2 format and moved the data loader to the hub.", "### Dataset Structure", "#### Data Fields\n\n\nThe data is in a CSV format, with the following fields:\n\n\n* 'mr' -- the meaning representation (MR, input)\n* 'ref' -- reference, i.e. the corresponding natural-language description (output)\n\n\nThere are additional fields ('fixed', 'orig\\_mr') indicating whether the data was modified in the\ncleaning process and what was the original MR before cleaning, but these aren't used for NLG.\n\n\nThe MR has a flat structure -- attribute-value pairs are comma separated, with values\nenclosed in brackets (see example above). There are 8 attributes:\n\n\n* 'name' -- restaurant name\n* 'near' -- a landmark close to the restaurant\n* 'area' -- location (riverside, city centre)\n* 'food' -- food type / cuisine (e.g. Japanese, Indian, English etc.)\n* 'eatType' -- restaurant type (restaurant, coffee shop, pub)\n* 'priceRange' -- price range (low, medium, high, <£20, £20-30, >£30)\n* 'rating' -- customer rating (low, medium, high, 1/5, 3/5, 5/5)\n* 'familyFriendly' -- is the restaurant family-friendly (yes/no)\n\n\nThe same MR is often repeated multiple times with different synonymous references.", "#### How were labels chosen?\n\n\nThe source MRs were generated automatically at random from a set of valid attribute values. The labels were crowdsourced and are natural language", "#### Example Instance", "#### Data Splits\n\n\n\n“Distinct MRs” are MRs that remain distinct even if restaurant/place names (attributes 'name', 'near')\nare delexicalized, i.e., replaced with a placeholder.", "#### Splitting Criteria\n\n\nThe data are divided so that MRs in different splits do not overlap.\n\n\nDataset in GEM\n--------------", "### Rationale for Inclusion in GEM", "#### Why is the Dataset in GEM?\n\n\nThe E2E dataset is one of the largest limited-domain NLG datasets and is frequently used as a data-to-text generation benchmark. The E2E Challenge included 20 systems of very different architectures, with system outputs available for download.", "#### Similar Datasets\n\n\nyes", "#### Unique Language Coverage\n\n\nno", "#### Difference from other GEM datasets\n\n\nThe dataset is much cleaner than comparable datasets, and it is also a relatively easy task, making for a straightforward evaluation.", "#### Ability that the Dataset measures\n\n\nsurface realization.", "### GEM-Specific Curation", "#### Modificatied for GEM?\n\n\nyes", "#### Additional Splits?\n\n\nyes", "#### Split Information\n\n\n4 special test sets for E2E were added to the GEM evaluation suite.\n\n\n1. We created subsets of the training and development sets of ~500 randomly selected inputs each.\n2. We applied input scrambling on a subset of 500 randomly selected test instances; the order of the input properties was randomly reassigned.\n3. For the input size, we created subpopulations based on the number of restaurant properties in the input.", "#### Split Motivation\n\n\nGeneralization and robustness", "### Getting Started with the Task\n\n\nPrevious Results\n----------------", "### Previous Results", "#### Measured Model Abilities\n\n\nSurface realization.", "#### Metrics\n\n\n'BLEU', 'METEOR', 'ROUGE'", "#### Proposed Evaluation\n\n\nThe official evaluation script combines the MT-Eval and COCO Captioning libraries with the following metrics.\n\n\n* BLEU\n* CIDEr\n* NIST\n* METEOR\n* ROUGE-L", "#### Previous results available?\n\n\nyes", "#### Other Evaluation Approaches\n\n\nMost previous results, including the shared task results, used the library provided by the dataset creators. The shared task also conducted a human evaluation using the following two criteria:\n\n\n* 'Quality': When collecting quality ratings, system outputs were presented to crowd workers together with the corresponding meaning representation, which implies that correctness of the NL utterance relative to the MR should also influence this ranking. The crowd workers were asked: “How do you judge the overall quality of the utterance in terms of its grammatical correctness, fluency, adequacy and other important factors?”\n* 'Naturalness': When collecting naturalness ratings, system outputs were presented to crowd workers without the corresponding meaning representation. The crowd workers were asked: “Could the utterance have been produced by a native speaker?”", "#### Relevant Previous Results\n\n\nThe shared task writeup has in-depth evaluations of systems (URL\n\n\nDataset Curation\n----------------", "### Original Curation", "#### Original Curation Rationale\n\n\nThe dataset was collected to showcase/test neural NLG models. It is larger and contains more lexical richness and syntactic variation than previous closed-domain NLG datasets.", "#### Communicative Goal\n\n\nProducing a text informing/recommending a restaurant, given all and only the attributes specified on the input.", "#### Sourced from Different Sources\n\n\nno", "### Language Data", "#### How was Language Data Obtained?\n\n\n'Crowdsourced'", "#### Where was it crowdsourced?\n\n\n'Other crowdworker platform'", "#### Language Producers\n\n\nHuman references describing the MRs were collected by crowdsourcing on the CrowdFlower (now Appen) platform,\nwith either textual or pictorial MRs as a baseline.\nThe pictorial MRs were used in 20% of cases -- these yield higher lexical variation but introduce noise.", "#### Topics Covered\n\n\nThe dataset is focused on descriptions of restaurants.", "#### Data Validation\n\n\nvalidated by data curator", "#### Data Preprocessing\n\n\nThere were basic checks (length, valid characters, repetition).", "#### Was Data Filtered?\n\n\nalgorithmically", "#### Filter Criteria\n\n\nThe cleaned version of the dataset which we are using in GEM was algorithmically filtered. They used regular expressions to match all human-generated references with a more accurate input when attributes were hallucinated or dropped. Additionally, train-test overlap stemming from the transformation was removed. As a result, this data is much cleaner than the original dataset but not perfect (about 20% of instances may have misaligned slots, compared to 40% of the original data.", "### Structured Annotations", "#### Additional Annotations?\n\n\nnone", "#### Annotation Service?\n\n\nno", "### Consent", "#### Any Consent Policy?\n\n\nyes", "#### Consent Policy Details\n\n\nSince a crowdsourcing platform was used, the involved raters waived their rights to the data and are aware that the produced annotations can be publicly released.", "### Private Identifying Information (PII)", "#### Contains PII?\n\n\nno PII", "#### Justification for no PII\n\n\nThe dataset is artificial and does not contain any description of people.", "### Maintenance", "#### Any Maintenance Plan?\n\n\nno\n\n\nBroader Social Context\n----------------------", "### Previous Work on the Social Impact of the Dataset", "#### Usage of Models based on the Data\n\n\nno", "### Impact on Under-Served Communities", "#### Addresses needs of underserved Communities?\n\n\nno", "### Discussion of Biases", "#### Any Documented Social Biases?\n\n\nno", "#### Are the Language Producers Representative of the Language?\n\n\nThe source data is generated randomly, so it should not contain biases. The human references may be biased by the workers' demographic, but that was not investigated upon data collection.\n\n\nConsiderations for Using the Data\n---------------------------------", "### PII Risks and Liability", "### Licenses", "#### Copyright Restrictions on the Dataset\n\n\n'open license - commercial use allowed'", "#### Copyright Restrictions on the Language Data\n\n\n'open license - commercial use allowed'", "### Known Technical Limitations", "#### Technical Limitations\n\n\nThe cleaned version still has data points with hallucinated or omitted attributes.", "#### Unsuited Applications\n\n\nThe data only pertains to the restaurant domain and the included attributes. A model cannot be expected to handle other domains or attributes." ]

[ "task_categories:summarization", "annotations_creators:none", "language_creators:unknown", "multilinguality:unknown", "size_categories:unknown", "source_datasets:original", "region:us" ]

[]

[]

[ "# Dataset Card for GEM/indonlg", "## Dataset Description\n\n- Homepage: URL\n- Repository: URL\n- Paper: URL\n- Leaderboard: N/A\n- Point of Contact: Genta Indra Winata", "### Link to Main Data Card\n\nYou can find the main data card on the GEM Website.", "### Dataset Summary \n\nIndoNLG is a collection of various Indonesian, Javanese, and Sundanese NLG tasks including summarization, question answering, chit-chat, and three different pairs of machine translation (MT) tasks.\n\nYou can load the dataset via:\n\nThe data loader can be found here.", "#### website\nGithub", "#### paper\nACL Anthology", "#### authors\nSamuel Cahyawijaya, Genta Indra Winata, Bryan Wilie, Karissa Vincentio, Xiaohong Li, Adhiguna Kuncoro, Sebastian Ruder, Zhi Yuan Lim, Syafri Bahar, Masayu Leylia Khodra, Ayu Purwarianti, Pascale Fung", "## Dataset Overview", "### Where to find the Data and its Documentation", "#### Webpage\n\n\n\nGithub", "#### Download\n\n\n\nGithub", "#### Paper\n\n\n\nACL Anthology", "#### BibTex", "#### Contact Name\n\n\n\n\nGenta Indra Winata", "#### Contact Email\n\n\n\ngentaindrawinata@URL", "#### Has a Leaderboard?\n\n\n\nno", "### Languages and Intended Use", "#### Multilingual?\n\n\n\n\nyes", "#### Covered Languages\n\n\n\n\n'Indonesian', 'Javanese', 'Sundanese'", "#### License\n\n\n\n\nmit: MIT License", "#### Intended Use\n\n\n\nIndoNLG is a collection of Natural Language Generation (NLG) resources for Bahasa Indonesia with 10 downstream tasks.", "#### Primary Task\n\n\n\nSummarization", "#### Communicative Goal\n\n\n\n\nGenerate a response according to the context and text.", "### Credit", "#### Curation Organization Type(s)\n\n\n\n'academic', 'industry'", "#### Curation Organization(s)\n\n\n\nThe Hong Kong University of Science and Technology, Gojek, Institut Teknologi Bandung, Universitas Multimedia Nusantara, DeepMind, URL", "#### Dataset Creators\n\n\n\nSamuel Cahyawijaya, Genta Indra Winata, Bryan Wilie, Karissa Vincentio, Xiaohong Li, Adhiguna Kuncoro, Sebastian Ruder, Zhi Yuan Lim, Syafri Bahar, Masayu Leylia Khodra, Ayu Purwarianti, Pascale Fung", "#### Funding\n\n\n\nThe Hong Kong University of Science and Technology, Gojek, Institut Teknologi Bandung, Universitas Multimedia Nusantara, DeepMind, URL", "#### Who added the Dataset to GEM?\n\n\n\nGenta Indra Winata (The Hong Kong University of Science and Technology)", "### Dataset Structure", "## Dataset in GEM", "### Rationale for Inclusion in GEM", "#### Similar Datasets\n\n\n\nyes", "#### Unique Language Coverage\n\n\n\nno", "### GEM-Specific Curation", "#### Modificatied for GEM?\n\n\n\nyes", "#### GEM Modifications\n\n\n\n'other'", "#### Additional Splits?\n\n\n\nno", "### Getting Started with the Task", "## Previous Results", "### Previous Results", "#### Measured Model Abilities\n\n\n\nDialog understanding, summarization, translation", "#### Metrics\n\n\n\n'BLEU'", "#### Proposed Evaluation\n\n\n\nBLEU evaluates the generation quality.", "#### Previous results available?\n\n\n\nyes", "#### Other Evaluation Approaches\n\n\n\nBLEU", "## Dataset Curation", "### Original Curation", "#### Sourced from Different Sources\n\n\n\nno", "### Language Data", "#### How was Language Data Obtained?\n\n\n\n'Crowdsourced'", "#### Where was it crowdsourced?\n\n\n\n'Participatory experiment'", "#### Data Validation\n\n\n\nvalidated by data curator", "#### Was Data Filtered?\n\n\n\nnot filtered", "### Structured Annotations", "#### Additional Annotations?\n\n\n\n\nnone", "#### Annotation Service?\n\n\n\nno", "### Consent", "#### Any Consent Policy?\n\n\n\nyes", "#### Consent Policy Details\n\n\n\nAnnotators agree using the dataset for research purpose.", "#### Other Consented Downstream Use\n\n\n\nAny", "### Private Identifying Information (PII)", "#### Contains PII?\n\n\n\n\nunlikely", "#### Categories of PII\n\n\n\n''", "### Maintenance", "#### Any Maintenance Plan?\n\n\n\nno", "## Broader Social Context", "### Previous Work on the Social Impact of the Dataset", "#### Usage of Models based on the Data\n\n\n\nno", "### Impact on Under-Served Communities", "#### Addresses needs of underserved Communities?\n\n\n\nyes", "### Discussion of Biases", "#### Any Documented Social Biases?\n\n\n\nno", "## Considerations for Using the Data", "### PII Risks and Liability", "#### Potential PII Risk\n\n\n\nNo", "### Licenses", "#### Copyright Restrictions on the Dataset\n\n\n\n'open license'", "#### Copyright Restrictions on the Language Data\n\n\n\n'open license'", "### Known Technical Limitations" ]

[ "TAGS\n#task_categories-summarization #annotations_creators-none #language_creators-unknown #multilinguality-unknown #size_categories-unknown #source_datasets-original #region-us \n", "# Dataset Card for GEM/indonlg", "## Dataset Description\n\n- Homepage: URL\n- Repository: URL\n- Paper: URL\n- Leaderboard: N/A\n- Point of Contact: Genta Indra Winata", "### Link to Main Data Card\n\nYou can find the main data card on the GEM Website.", "### Dataset Summary \n\nIndoNLG is a collection of various Indonesian, Javanese, and Sundanese NLG tasks including summarization, question answering, chit-chat, and three different pairs of machine translation (MT) tasks.\n\nYou can load the dataset via:\n\nThe data loader can be found here.", "#### website\nGithub", "#### paper\nACL Anthology", "#### authors\nSamuel Cahyawijaya, Genta Indra Winata, Bryan Wilie, Karissa Vincentio, Xiaohong Li, Adhiguna Kuncoro, Sebastian Ruder, Zhi Yuan Lim, Syafri Bahar, Masayu Leylia Khodra, Ayu Purwarianti, Pascale Fung", "## Dataset Overview", "### Where to find the Data and its Documentation", "#### Webpage\n\n\n\nGithub", "#### Download\n\n\n\nGithub", "#### Paper\n\n\n\nACL Anthology", "#### BibTex", "#### Contact Name\n\n\n\n\nGenta Indra Winata", "#### Contact Email\n\n\n\ngentaindrawinata@URL", "#### Has a Leaderboard?\n\n\n\nno", "### Languages and Intended Use", "#### Multilingual?\n\n\n\n\nyes", "#### Covered Languages\n\n\n\n\n'Indonesian', 'Javanese', 'Sundanese'", "#### License\n\n\n\n\nmit: MIT License", "#### Intended Use\n\n\n\nIndoNLG is a collection of Natural Language Generation (NLG) resources for Bahasa Indonesia with 10 downstream tasks.", "#### Primary Task\n\n\n\nSummarization", "#### Communicative Goal\n\n\n\n\nGenerate a response according to the context and text.", "### Credit", "#### Curation Organization Type(s)\n\n\n\n'academic', 'industry'", "#### Curation Organization(s)\n\n\n\nThe Hong Kong University of Science and Technology, Gojek, Institut Teknologi Bandung, Universitas Multimedia Nusantara, DeepMind, URL", "#### Dataset Creators\n\n\n\nSamuel Cahyawijaya, Genta Indra Winata, Bryan Wilie, Karissa Vincentio, Xiaohong Li, Adhiguna Kuncoro, Sebastian Ruder, Zhi Yuan Lim, Syafri Bahar, Masayu Leylia Khodra, Ayu Purwarianti, Pascale Fung", "#### Funding\n\n\n\nThe Hong Kong University of Science and Technology, Gojek, Institut Teknologi Bandung, Universitas Multimedia Nusantara, DeepMind, URL", "#### Who added the Dataset to GEM?\n\n\n\nGenta Indra Winata (The Hong Kong University of Science and Technology)", "### Dataset Structure", "## Dataset in GEM", "### Rationale for Inclusion in GEM", "#### Similar Datasets\n\n\n\nyes", "#### Unique Language Coverage\n\n\n\nno", "### GEM-Specific Curation", "#### Modificatied for GEM?\n\n\n\nyes", "#### GEM Modifications\n\n\n\n'other'", "#### Additional Splits?\n\n\n\nno", "### Getting Started with the Task", "## Previous Results", "### Previous Results", "#### Measured Model Abilities\n\n\n\nDialog understanding, summarization, translation", "#### Metrics\n\n\n\n'BLEU'", "#### Proposed Evaluation\n\n\n\nBLEU evaluates the generation quality.", "#### Previous results available?\n\n\n\nyes", "#### Other Evaluation Approaches\n\n\n\nBLEU", "## Dataset Curation", "### Original Curation", "#### Sourced from Different Sources\n\n\n\nno", "### Language Data", "#### How was Language Data Obtained?\n\n\n\n'Crowdsourced'", "#### Where was it crowdsourced?\n\n\n\n'Participatory experiment'", "#### Data Validation\n\n\n\nvalidated by data curator", "#### Was Data Filtered?\n\n\n\nnot filtered", "### Structured Annotations", "#### Additional Annotations?\n\n\n\n\nnone", "#### Annotation Service?\n\n\n\nno", "### Consent", "#### Any Consent Policy?\n\n\n\nyes", "#### Consent Policy Details\n\n\n\nAnnotators agree using the dataset for research purpose.", "#### Other Consented Downstream Use\n\n\n\nAny", "### Private Identifying Information (PII)", "#### Contains PII?\n\n\n\n\nunlikely", "#### Categories of PII\n\n\n\n''", "### Maintenance", "#### Any Maintenance Plan?\n\n\n\nno", "## Broader Social Context", "### Previous Work on the Social Impact of the Dataset", "#### Usage of Models based on the Data\n\n\n\nno", "### Impact on Under-Served Communities", "#### Addresses needs of underserved Communities?\n\n\n\nyes", "### Discussion of Biases", "#### Any Documented Social Biases?\n\n\n\nno", "## Considerations for Using the Data", "### PII Risks and Liability", "#### Potential PII Risk\n\n\n\nNo", "### Licenses", "#### Copyright Restrictions on the Dataset\n\n\n\n'open license'", "#### Copyright Restrictions on the Language Data\n\n\n\n'open license'", "### Known Technical Limitations" ]

[ "task_categories:table-to-text", "annotations_creators:none", "language_creators:unknown", "multilinguality:unknown", "size_categories:unknown", "source_datasets:original", "language:en", "license:other", "data-to-text", "region:us" ]

[]

[ "en" ]

[ "# Dataset Card for GEM/mlb_data_to_text", "## Dataset Description\n\n- Homepage: URL\n- Repository: URL\n- Paper: URL\n- Leaderboard: N/A\n- Point of Contact: Ratish Puduppully", "### Link to Main Data Card\n\nYou can find the main data card on the GEM Website.", "### Dataset Summary \n\nThe MLB dataset is an English sport-related data-to-text dataset in the baseball domain. The input is a large table with results of a game and the output is a description of the game.\n\nYou can load the dataset via:\n\nThe data loader can be found here.", "#### website\nGithub", "#### paper\nACL Anthology", "#### authors\nRatish Puduppully, Li Dong, Mirella Lapata", "## Dataset Overview", "### Where to find the Data and its Documentation", "#### Webpage\n\n\n\nGithub", "#### Download\n\n\n\nGithub", "#### Paper\n\n\n\nACL Anthology", "#### BibTex", "#### Contact Name\n\n\n\n\nRatish Puduppully", "#### Contact Email\n\n\n\nratishpuduppully@URL", "#### Has a Leaderboard?\n\n\n\nno", "### Languages and Intended Use", "#### Multilingual?\n\n\n\n\nno", "#### Covered Languages\n\n\n\n\n'English'", "#### License\n\n\n\n\nother: Other license", "#### Intended Use\n\n\n\nThe dataset can be used to study data-to-text generation. The dataset is in sports domain. It pairs statistics of Major League Baseball (MLB) game with its summary. The summary is in the form of a document containing an average of 540 tokens. Thus it is useful to study long document generation.", "#### Add. License Info\n\n\n\nRestricted to non-commercial research purposes.", "#### Primary Task\n\n\n\nData-to-Text", "#### Communicative Goal\n\n\n\n\nProduce a summary of MLB game from its statistics.", "### Credit", "#### Curation Organization Type(s)\n\n\n\n'academic'", "#### Curation Organization(s)\n\n\n\nUniversity of Edinburgh", "#### Dataset Creators\n\n\n\nRatish Puduppully, Li Dong, Mirella Lapata", "### Dataset Structure", "#### Data Fields", "#### Reason for Structure\n\n\n\nThe high level structure contains the following attributes: home_name, vis_name, home_city, vis_city, summary, summary_eval, day, gem_id, box_score, play_by_play, home_line, vis_line.\nThe attributes home_name, vis_name, home_city, vis_city and day are string values.\nThe attribute \"summary\" contains the summary in the form of a list of tokens.\nThe attribute \"summary_eval\" contains the summary in the form of a string of tokens. The difference from \"summary\" field is that \"summary_eval\" doesn't contain \"*NEWPARAGRAPH*\" delimiters to separate the paragraphs. \"summary_eval\" field should be used to evaluate model outputs. \"summary\" field may be used during the training process.\nbox_score contains the box score statistics of the players in the game. It is in the form of a list (of average size 90), with each element describing the statistics of a player. The box score statistics contain 53 attributes. \nThe description of the attributes is given below. The descriptions of most of the attributes is obtained from URL. \n\n- r : Runs scored by a player in the game. \n- rbi Runs Batted In (RBI): action of a batter results in a run scored by other players in the team. \n- pos Position of the player. \n- avg Batting Average. It is an indicator of the hits in the players' career. \n- bb A walk occurs when a pitcher throws four pitches out of the strike zone, none of which are swung at by the hitter. \n- hr Batter hits the ball in the air over the outfield fence. \n- p_r Runs given by a pitcher in the game. \n- p_bb Walks allowed by pitcher in a game. \n- p_h Hits allowed by pitcher in a game. \n- p_hr Home runs allowed by pitcher in a game. \n- p_er Earned Run (ER): An earned run is any run that scores against a pitcher. \n- p_era Earned Run Average (ERA): Earned run average represents the number of earned runs a pitcher allows per nine innings. \n- p_np Number of Pitches: A pitcher's total number of pitches is determined by all the pitches he throws in game. \n- p_ip1 Innings Pitched (IP1): Innings pitched measures the number of innings a pitcher remains in a game. Because there are three outs in an inning, each out recorded represents one-third of an inning pitched. \n- p_ip2 Innings Pitched (IP2): Innings pitched measures the number of innings a pitcher remains in a game. Because there are three outs in an inning, each out recorded represents one-third of an inning pitched. \n- p_w A pitcher receives a win when he is the pitcher of record when his team takes the lead for good. \n- p_l A pitcher receives a loss when a run that is charged to him proves to be the go-ahead run in the game, giving the opposing team a lead it never gives up. \n- p_so A strikeout occurs when a pitcher throws any combination of three swinging or looking strikes to a hitter. \n- p_save Save: A save is awarded to the relief pitcher who finishes a game for the winning team. A pitcher cannot receive a save and a win in the same game. \n- p_sv Saves: The count of saves recorded by a pitcher in his career. \n- sac A sacrifice fly occurs when a batter hits a fly-ball out to the outfield or foul territory that allows a runner to score. \n- p_bf Batters faced is simply a count of the number of total plate appearances against a certain pitcher or team. In a perfect game -- with 27 outs -- a pitcher will record 27 batters faced. \n- cs A caught stealing occurs when a runner attempts to steal but is tagged out before reaching second base, third base or home plate. \n- hbp A hit-by-pitch occurs when a batter is struck by a pitched ball without swinging at it. He is awarded first base as a result. \n- ab An official at-bat comes when a batter reaches base via a fielder's choice, hit or an error (not including catcher's interference) or when a batter is put out on a non-sacrifice. \n- p_bs A blown save occurs when a relief pitcher enters a game in a save situation, but allows the tying run to score. \n- p_s The count of strikes thrown by a pitcher \n- lob Left on base can be viewed as both an individual statistic or as a team statistic. In an individual batter's case, it refers to how many men remain on base after that batter makes an out at the plate, as the batter has failed to do his job to score those runners -- or at least put himself in a position to score. In a team's case or in an individual pitcher's case, it refers to the number of men who remain on base at the end of an inning. \n- po A fielder is credited with a putout when he is the fielder who physically records the act of completing an out -- whether it be by stepping on the base for a forceout, tagging a runner, catching a batted ball, or catching a third strike \n- ops OPS adds on-base percentage and slugging percentage to get one number that unites the two. It's meant to combine how well a hitter can reach base, with how well he can hit for average and for power. \n- p_hld A hold occurs when a relief pitcher enters the game in a save situation and maintains his team's lead for the next relief pitcher, while recording at least one out. \n- p_loss True/False- Indicates losing pitcher \n- e A fielder is given an error if, in the judgment of the official scorer, he fails to convert an out on a play that an average fielder should have made. \n- p_win True/False- Indicates winning pitcher \n- a An assist is awarded to a fielder who touches the ball before a putout is recorded by another fielder. \n- h A hit occurs when a batter strikes the baseball into fair territory and reaches base without doing so via an error or a fielder's choice. \n- so A strikeout of a batter \n- team Team of the player \n- sb A stolen base occurs when a baserunner advances by taking a base to which he isn't entitled. \n- slg Slugging percentage represents the total number of bases a player records per at-bat. Unlike on-base percentage, slugging percentage deals only with hits and does not include walks and hit-by-pitches in its equation. \n- sf A sacrifice fly occurs when a batter hits a fly-ball out to the outfield or foul territory that allows a runner to score. \n- obp OBP refers to how frequently a batter reaches base per plate appearance. Times on base include hits, walks and hit-by-pitches, but do not include errors, times reached on a fielder's choice or a dropped third strike. \n\nThe description of attributes in play-by-play is below:\n\n- batter Batter in the play. \n- pitcher Pitcher in play. \n- b1 Player/s at first base position. \n- b2 Player/s at second base position. \n- b3 Player/s at third base position. \n- scorers Player/s scored in the play. \n- fielder_error Player committed field error. \n- event Event of the play such as single, double, home run etc. \n- event2 Second event of the play such as wild pitch, error etc. \n- inning Inning of the play. \n- top/ bottom If home team is batting it is bottom and if away team is batting it is top. \n- o Count of outs \n- b Count of balls \n- s Count of strikes \n- r Count of runs \n- rbi Count of runs batted in (rbi) \n- error_runs Runs due to error \n- home_team_runs Score of home team \n- vis_team_runs Score of visiting team \n\n'home_line' and 'vis_line' contain string value pairs for 'team_name', 'team_city', 'team_runs', 'team_hits', 'team_error', 'result', and a list of runs scored in each inning.", "#### Data Splits\n\n\n\nThere are three splits in the dataset: train, validation and test", "#### Splitting Criteria\n\n\n\nThe splits are random.", "## Dataset in GEM", "### Rationale for Inclusion in GEM", "#### Why is the Dataset in GEM?\n\n\n\nThis dataset can verify if models are capable of long document generation. The challenges in long document generation conditioned on input tables include ensuring coherent output, staying faithful to the input, ensuring fluent output and avoiding repetition of text. Such aspects can be verified on models trained on this dataset", "#### Similar Datasets\n\n\n\nyes", "#### Unique Language Coverage\n\n\n\nno", "#### Difference from other GEM datasets\n\n\n\nCompared to the existing RotoWire (Wiseman et al. 2017) dataset, MLB summaries are longer (approximately by 50%) and the input records are richer and more structured (with the addition of play-by-play). Moreover, the MLB dataset is five times larger in terms of data size (i.e., pairs of tables and game summaries).", "#### Ability that the Dataset measures\n\n\n\nLong document generation, coherent ordering of information, faithfulness to the input statistics, fluency in generation and avoiding repetition of text.", "### GEM-Specific Curation", "#### Modificatied for GEM?\n\n\n\nyes", "#### GEM Modifications\n\n\n\n'data points removed'", "#### Modification Details\n\n\n\nSome examples have been removed from training dataset which satisfied the below criteria:\n1. The examples in training dataset which overlapped with validation/test.\n2. Some examples which described washed out games.", "#### Additional Splits?\n\n\n\nno", "### Getting Started with the Task", "#### Pointers to Resources\n\n\n\nThe research paper is a good resource", "## Previous Results", "### Previous Results", "#### Measured Model Abilities\n\n\n\nAutomatic evaluation measure can evaluate the factuality, content selection, content ordering and the fluency of the model output. The factuality, content selection and content ordering is measured using an Information Extraction based evaluation approach introduced by Wiseman et al (2017). The fluency is measured using BLEU.", "#### Metrics\n\n\n\n'Other: Other Metrics'", "#### Other Metrics\n\n\n\nWiseman et al. (2017) define three metrics induced from the outputs of an Information Extraction model which is run on the model/human-written game summaries . Let ŷ be the gold summary and y the model output.\n• Relation Generation (RG) measures the precision and count of relations extracted from y that also appear in records r.\n• Content Selection (CS) measures the precision and recall of relations extracted from y that are also extracted from ŷ.\n• Content Ordering (CO) measures the complement of the normalized Damerau-Levenshtein distance (Brill and Moore, 2000) between the sequences of relations extracted from y and ŷ", "#### Proposed Evaluation\n\n\n\nWe have reused the automatic metrics based on Information Extraction evaluation introduced by Wiseman et al (2017). For human evaluation, we conducted studies to evaluate the factuality, coherence, grammaticality and conciseness.", "#### Previous results available?\n\n\n\nyes", "#### Relevant Previous Results\n\n\n\nThe most relevant previous results for dataset are in the TACL 2021 paper on Data-to-text Generation with Macro Planning", "## Dataset Curation", "### Original Curation", "#### Original Curation Rationale\n\n\n\nThis dataset was curated to complement an existing data-to-text generation dataset (RotoWire by Wiseman et al. 2017) which focuses on long document generation. Compared to RotoWire , MLB summaries are longer (approximately by 50%) and the input records are richer and more structured (with the addition of play-by-play). Moreover, the MLB dataset is five times larger in terms of data size (i.e., pairs of tables and game summaries)", "#### Communicative Goal\n\n\n\nThe goal is to study automatic generation of long documents in a data-to-text setting. The generated summaries should exhibit coherent ordering of content, be faithful to the input statistics, be fluent and avoid repetition of text.", "#### Sourced from Different Sources\n\n\n\nno", "### Language Data", "#### How was Language Data Obtained?\n\n\n\n'Found'", "#### Where was it found?\n\n\n\n'Single website'", "#### Language Producers\n\n\n\nThe game summaries are produced by professional writers.", "#### Topics Covered\n\n\n\nThe language focuses on the sports domain.", "#### Data Validation\n\n\n\nnot validated", "#### Data Preprocessing\n\n\n\nGame summaries were tokenized using NLTK (Bird et al., 2009) and hyphenated words were separated. Sentences containing quotes were removed as they included opinions and non-factual statements unrelated to the input tables. Sometimes MLB summaries contain a \"Game notes\" section with incidental information which was also removed.", "#### Was Data Filtered?\n\n\n\nnot filtered", "### Structured Annotations", "#### Additional Annotations?\n\n\n\n\nnone", "#### Annotation Service?\n\n\n\nno", "### Consent", "#### Any Consent Policy?\n\n\n\nno", "#### Justification for Using the Data\n\n\n\nThe copyright remains with the original data creators and the usage permission is restricted to non-commercial uses.", "### Private Identifying Information (PII)", "#### Contains PII?\n\n\n\n\nyes/very likely", "#### Categories of PII\n\n\n\n'sensitive information', 'generic PII'", "#### Any PII Identification?\n\n\n\nno identification", "### Maintenance", "#### Any Maintenance Plan?\n\n\n\nno", "## Broader Social Context", "### Previous Work on the Social Impact of the Dataset", "#### Usage of Models based on the Data\n\n\n\nno", "### Impact on Under-Served Communities", "#### Addresses needs of underserved Communities?\n\n\n\nno", "### Discussion of Biases", "#### Any Documented Social Biases?\n\n\n\nunsure", "## Considerations for Using the Data", "### PII Risks and Liability", "### Licenses", "#### Copyright Restrictions on the Dataset\n\n\n\n'research use only'", "#### Copyright Restrictions on the Language Data\n\n\n\n'research use only'", "### Known Technical Limitations" ]

[ "TAGS\n#task_categories-table-to-text #annotations_creators-none #language_creators-unknown #multilinguality-unknown #size_categories-unknown #source_datasets-original #language-English #license-other #data-to-text #region-us \n", "# Dataset Card for GEM/mlb_data_to_text", "## Dataset Description\n\n- Homepage: URL\n- Repository: URL\n- Paper: URL\n- Leaderboard: N/A\n- Point of Contact: Ratish Puduppully", "### Link to Main Data Card\n\nYou can find the main data card on the GEM Website.", "### Dataset Summary \n\nThe MLB dataset is an English sport-related data-to-text dataset in the baseball domain. The input is a large table with results of a game and the output is a description of the game.\n\nYou can load the dataset via:\n\nThe data loader can be found here.", "#### website\nGithub", "#### paper\nACL Anthology", "#### authors\nRatish Puduppully, Li Dong, Mirella Lapata", "## Dataset Overview", "### Where to find the Data and its Documentation", "#### Webpage\n\n\n\nGithub", "#### Download\n\n\n\nGithub", "#### Paper\n\n\n\nACL Anthology", "#### BibTex", "#### Contact Name\n\n\n\n\nRatish Puduppully", "#### Contact Email\n\n\n\nratishpuduppully@URL", "#### Has a Leaderboard?\n\n\n\nno", "### Languages and Intended Use", "#### Multilingual?\n\n\n\n\nno", "#### Covered Languages\n\n\n\n\n'English'", "#### License\n\n\n\n\nother: Other license", "#### Intended Use\n\n\n\nThe dataset can be used to study data-to-text generation. The dataset is in sports domain. It pairs statistics of Major League Baseball (MLB) game with its summary. The summary is in the form of a document containing an average of 540 tokens. Thus it is useful to study long document generation.", "#### Add. License Info\n\n\n\nRestricted to non-commercial research purposes.", "#### Primary Task\n\n\n\nData-to-Text", "#### Communicative Goal\n\n\n\n\nProduce a summary of MLB game from its statistics.", "### Credit", "#### Curation Organization Type(s)\n\n\n\n'academic'", "#### Curation Organization(s)\n\n\n\nUniversity of Edinburgh", "#### Dataset Creators\n\n\n\nRatish Puduppully, Li Dong, Mirella Lapata", "### Dataset Structure", "#### Data Fields", "#### Reason for Structure\n\n\n\nThe high level structure contains the following attributes: home_name, vis_name, home_city, vis_city, summary, summary_eval, day, gem_id, box_score, play_by_play, home_line, vis_line.\nThe attributes home_name, vis_name, home_city, vis_city and day are string values.\nThe attribute \"summary\" contains the summary in the form of a list of tokens.\nThe attribute \"summary_eval\" contains the summary in the form of a string of tokens. The difference from \"summary\" field is that \"summary_eval\" doesn't contain \"*NEWPARAGRAPH*\" delimiters to separate the paragraphs. \"summary_eval\" field should be used to evaluate model outputs. \"summary\" field may be used during the training process.\nbox_score contains the box score statistics of the players in the game. It is in the form of a list (of average size 90), with each element describing the statistics of a player. The box score statistics contain 53 attributes. \nThe description of the attributes is given below. The descriptions of most of the attributes is obtained from URL. \n\n- r : Runs scored by a player in the game. \n- rbi Runs Batted In (RBI): action of a batter results in a run scored by other players in the team. \n- pos Position of the player. \n- avg Batting Average. It is an indicator of the hits in the players' career. \n- bb A walk occurs when a pitcher throws four pitches out of the strike zone, none of which are swung at by the hitter. \n- hr Batter hits the ball in the air over the outfield fence. \n- p_r Runs given by a pitcher in the game. \n- p_bb Walks allowed by pitcher in a game. \n- p_h Hits allowed by pitcher in a game. \n- p_hr Home runs allowed by pitcher in a game. \n- p_er Earned Run (ER): An earned run is any run that scores against a pitcher. \n- p_era Earned Run Average (ERA): Earned run average represents the number of earned runs a pitcher allows per nine innings. \n- p_np Number of Pitches: A pitcher's total number of pitches is determined by all the pitches he throws in game. \n- p_ip1 Innings Pitched (IP1): Innings pitched measures the number of innings a pitcher remains in a game. Because there are three outs in an inning, each out recorded represents one-third of an inning pitched. \n- p_ip2 Innings Pitched (IP2): Innings pitched measures the number of innings a pitcher remains in a game. Because there are three outs in an inning, each out recorded represents one-third of an inning pitched. \n- p_w A pitcher receives a win when he is the pitcher of record when his team takes the lead for good. \n- p_l A pitcher receives a loss when a run that is charged to him proves to be the go-ahead run in the game, giving the opposing team a lead it never gives up. \n- p_so A strikeout occurs when a pitcher throws any combination of three swinging or looking strikes to a hitter. \n- p_save Save: A save is awarded to the relief pitcher who finishes a game for the winning team. A pitcher cannot receive a save and a win in the same game. \n- p_sv Saves: The count of saves recorded by a pitcher in his career. \n- sac A sacrifice fly occurs when a batter hits a fly-ball out to the outfield or foul territory that allows a runner to score. \n- p_bf Batters faced is simply a count of the number of total plate appearances against a certain pitcher or team. In a perfect game -- with 27 outs -- a pitcher will record 27 batters faced. \n- cs A caught stealing occurs when a runner attempts to steal but is tagged out before reaching second base, third base or home plate. \n- hbp A hit-by-pitch occurs when a batter is struck by a pitched ball without swinging at it. He is awarded first base as a result. \n- ab An official at-bat comes when a batter reaches base via a fielder's choice, hit or an error (not including catcher's interference) or when a batter is put out on a non-sacrifice. \n- p_bs A blown save occurs when a relief pitcher enters a game in a save situation, but allows the tying run to score. \n- p_s The count of strikes thrown by a pitcher \n- lob Left on base can be viewed as both an individual statistic or as a team statistic. In an individual batter's case, it refers to how many men remain on base after that batter makes an out at the plate, as the batter has failed to do his job to score those runners -- or at least put himself in a position to score. In a team's case or in an individual pitcher's case, it refers to the number of men who remain on base at the end of an inning. \n- po A fielder is credited with a putout when he is the fielder who physically records the act of completing an out -- whether it be by stepping on the base for a forceout, tagging a runner, catching a batted ball, or catching a third strike \n- ops OPS adds on-base percentage and slugging percentage to get one number that unites the two. It's meant to combine how well a hitter can reach base, with how well he can hit for average and for power. \n- p_hld A hold occurs when a relief pitcher enters the game in a save situation and maintains his team's lead for the next relief pitcher, while recording at least one out. \n- p_loss True/False- Indicates losing pitcher \n- e A fielder is given an error if, in the judgment of the official scorer, he fails to convert an out on a play that an average fielder should have made. \n- p_win True/False- Indicates winning pitcher \n- a An assist is awarded to a fielder who touches the ball before a putout is recorded by another fielder. \n- h A hit occurs when a batter strikes the baseball into fair territory and reaches base without doing so via an error or a fielder's choice. \n- so A strikeout of a batter \n- team Team of the player \n- sb A stolen base occurs when a baserunner advances by taking a base to which he isn't entitled. \n- slg Slugging percentage represents the total number of bases a player records per at-bat. Unlike on-base percentage, slugging percentage deals only with hits and does not include walks and hit-by-pitches in its equation. \n- sf A sacrifice fly occurs when a batter hits a fly-ball out to the outfield or foul territory that allows a runner to score. \n- obp OBP refers to how frequently a batter reaches base per plate appearance. Times on base include hits, walks and hit-by-pitches, but do not include errors, times reached on a fielder's choice or a dropped third strike. \n\nThe description of attributes in play-by-play is below:\n\n- batter Batter in the play. \n- pitcher Pitcher in play. \n- b1 Player/s at first base position. \n- b2 Player/s at second base position. \n- b3 Player/s at third base position. \n- scorers Player/s scored in the play. \n- fielder_error Player committed field error. \n- event Event of the play such as single, double, home run etc. \n- event2 Second event of the play such as wild pitch, error etc. \n- inning Inning of the play. \n- top/ bottom If home team is batting it is bottom and if away team is batting it is top. \n- o Count of outs \n- b Count of balls \n- s Count of strikes \n- r Count of runs \n- rbi Count of runs batted in (rbi) \n- error_runs Runs due to error \n- home_team_runs Score of home team \n- vis_team_runs Score of visiting team \n\n'home_line' and 'vis_line' contain string value pairs for 'team_name', 'team_city', 'team_runs', 'team_hits', 'team_error', 'result', and a list of runs scored in each inning.", "#### Data Splits\n\n\n\nThere are three splits in the dataset: train, validation and test", "#### Splitting Criteria\n\n\n\nThe splits are random.", "## Dataset in GEM", "### Rationale for Inclusion in GEM", "#### Why is the Dataset in GEM?\n\n\n\nThis dataset can verify if models are capable of long document generation. The challenges in long document generation conditioned on input tables include ensuring coherent output, staying faithful to the input, ensuring fluent output and avoiding repetition of text. Such aspects can be verified on models trained on this dataset", "#### Similar Datasets\n\n\n\nyes", "#### Unique Language Coverage\n\n\n\nno", "#### Difference from other GEM datasets\n\n\n\nCompared to the existing RotoWire (Wiseman et al. 2017) dataset, MLB summaries are longer (approximately by 50%) and the input records are richer and more structured (with the addition of play-by-play). Moreover, the MLB dataset is five times larger in terms of data size (i.e., pairs of tables and game summaries).", "#### Ability that the Dataset measures\n\n\n\nLong document generation, coherent ordering of information, faithfulness to the input statistics, fluency in generation and avoiding repetition of text.", "### GEM-Specific Curation", "#### Modificatied for GEM?\n\n\n\nyes", "#### GEM Modifications\n\n\n\n'data points removed'", "#### Modification Details\n\n\n\nSome examples have been removed from training dataset which satisfied the below criteria:\n1. The examples in training dataset which overlapped with validation/test.\n2. Some examples which described washed out games.", "#### Additional Splits?\n\n\n\nno", "### Getting Started with the Task", "#### Pointers to Resources\n\n\n\nThe research paper is a good resource", "## Previous Results", "### Previous Results", "#### Measured Model Abilities\n\n\n\nAutomatic evaluation measure can evaluate the factuality, content selection, content ordering and the fluency of the model output. The factuality, content selection and content ordering is measured using an Information Extraction based evaluation approach introduced by Wiseman et al (2017). The fluency is measured using BLEU.", "#### Metrics\n\n\n\n'Other: Other Metrics'", "#### Other Metrics\n\n\n\nWiseman et al. (2017) define three metrics induced from the outputs of an Information Extraction model which is run on the model/human-written game summaries . Let ŷ be the gold summary and y the model output.\n• Relation Generation (RG) measures the precision and count of relations extracted from y that also appear in records r.\n• Content Selection (CS) measures the precision and recall of relations extracted from y that are also extracted from ŷ.\n• Content Ordering (CO) measures the complement of the normalized Damerau-Levenshtein distance (Brill and Moore, 2000) between the sequences of relations extracted from y and ŷ", "#### Proposed Evaluation\n\n\n\nWe have reused the automatic metrics based on Information Extraction evaluation introduced by Wiseman et al (2017). For human evaluation, we conducted studies to evaluate the factuality, coherence, grammaticality and conciseness.", "#### Previous results available?\n\n\n\nyes", "#### Relevant Previous Results\n\n\n\nThe most relevant previous results for dataset are in the TACL 2021 paper on Data-to-text Generation with Macro Planning", "## Dataset Curation", "### Original Curation", "#### Original Curation Rationale\n\n\n\nThis dataset was curated to complement an existing data-to-text generation dataset (RotoWire by Wiseman et al. 2017) which focuses on long document generation. Compared to RotoWire , MLB summaries are longer (approximately by 50%) and the input records are richer and more structured (with the addition of play-by-play). Moreover, the MLB dataset is five times larger in terms of data size (i.e., pairs of tables and game summaries)", "#### Communicative Goal\n\n\n\nThe goal is to study automatic generation of long documents in a data-to-text setting. The generated summaries should exhibit coherent ordering of content, be faithful to the input statistics, be fluent and avoid repetition of text.", "#### Sourced from Different Sources\n\n\n\nno", "### Language Data", "#### How was Language Data Obtained?\n\n\n\n'Found'", "#### Where was it found?\n\n\n\n'Single website'", "#### Language Producers\n\n\n\nThe game summaries are produced by professional writers.", "#### Topics Covered\n\n\n\nThe language focuses on the sports domain.", "#### Data Validation\n\n\n\nnot validated", "#### Data Preprocessing\n\n\n\nGame summaries were tokenized using NLTK (Bird et al., 2009) and hyphenated words were separated. Sentences containing quotes were removed as they included opinions and non-factual statements unrelated to the input tables. Sometimes MLB summaries contain a \"Game notes\" section with incidental information which was also removed.", "#### Was Data Filtered?\n\n\n\nnot filtered", "### Structured Annotations", "#### Additional Annotations?\n\n\n\n\nnone", "#### Annotation Service?\n\n\n\nno", "### Consent", "#### Any Consent Policy?\n\n\n\nno", "#### Justification for Using the Data\n\n\n\nThe copyright remains with the original data creators and the usage permission is restricted to non-commercial uses.", "### Private Identifying Information (PII)", "#### Contains PII?\n\n\n\n\nyes/very likely", "#### Categories of PII\n\n\n\n'sensitive information', 'generic PII'", "#### Any PII Identification?\n\n\n\nno identification", "### Maintenance", "#### Any Maintenance Plan?\n\n\n\nno", "## Broader Social Context", "### Previous Work on the Social Impact of the Dataset", "#### Usage of Models based on the Data\n\n\n\nno", "### Impact on Under-Served Communities", "#### Addresses needs of underserved Communities?\n\n\n\nno", "### Discussion of Biases", "#### Any Documented Social Biases?\n\n\n\nunsure", "## Considerations for Using the Data", "### PII Risks and Liability", "### Licenses", "#### Copyright Restrictions on the Dataset\n\n\n\n'research use only'", "#### Copyright Restrictions on the Language Data\n\n\n\n'research use only'", "### Known Technical Limitations" ]

[ "task_categories:summarization", "annotations_creators:none", "language_creators:unknown", "multilinguality:unknown", "size_categories:unknown", "source_datasets:original", "language:de", "language:es", "license:other", "region:us" ]

[]

[ "de", "es" ]

[ "### Link to Main Data Card\n\n\nYou can find the main data card on the GEM Website.", "### Dataset Summary\n\n\nMLSum is a multilingual summarization dataset crawled from different news websites. The GEM version supports the German and Spanish subset alongside specifically collected challenge sets for COVID-related articles to test out-of-domain generalization.\n\n\nYou can load the dataset via:\n\n\nThe data loader can be found here.", "#### website\n\n\nN/A", "#### paper\n\n\nACL Anthology", "#### authors\n\n\nThomas Scialom, Paul-Alexis Dray, Sylvain Lamprier, Benjamin Piwowarski, Jacopo Staiano\n\n\nDataset Overview\n----------------", "### Where to find the Data and its Documentation", "#### Download\n\n\nGitlab", "#### Paper\n\n\nACL Anthology", "#### BibTex", "#### Contact Name\n\n\nThomas Scialom", "#### Contact Email\n\n\n{thomas,paul-alexis,jacopo}@URL, {sylvain.lamprier,benjamin.piwowarski}@URL", "#### Has a Leaderboard?\n\n\nno", "### Languages and Intended Use", "#### Multilingual?\n\n\nyes", "#### Covered Dialects\n\n\nThere is only one dialect per language, Hochdeutsch for German and Castilian Spanish for Spanish.", "#### Covered Languages\n\n\n'German', 'Spanish, Castilian'", "#### Whose Language?\n\n\nThe German articles are crawled from Süddeutsche Zeitung and the Spanish ones from El Pais.", "#### License\n\n\nother: Other license", "#### Intended Use\n\n\nThe intended use of this dataset is to augment existing datasets for English news summarization with additional languages.", "#### Add. License Info\n\n\nRestricted to non-commercial research purposes.", "#### Primary Task\n\n\nSummarization", "#### Communicative Goal\n\n\nThe speaker is required to produce a high quality summary of news articles in the same language as the input article.", "### Credit", "#### Curation Organization Type(s)\n\n\n'other'", "#### Curation Organization(s)\n\n\nCNRS, Sorbonne Université, reciTAL", "#### Dataset Creators\n\n\nThomas Scialom, Paul-Alexis Dray, Sylvain Lamprier, Benjamin Piwowarski, Jacopo Staiano", "#### Funding\n\n\nFunding information is not specified.", "#### Who added the Dataset to GEM?\n\n\nThe original data card was written by Pedro Henrique Martins (Instituto de Telecomunicações) and Sebastian Gehrmann (Google Research) extended and updated it to the v2 format. The COVID challenge set was created by Laura Perez-Beltrachini (University of Edinburgh). Data cleaning was done by Juan Diego Rodriguez (UT Austin).", "### Dataset Structure", "#### Data Fields\n\n\nThe data fields are:\n\n\n* 'text': the source article ('string').\n* 'summary': the output summary ('string').\n* 'topic': the topic of the article ('string').\n* 'url': the article's url ('string').\n* 'title': the article's title ('string').\n* 'date': the article's date ('string').", "#### Reason for Structure\n\n\nThe structure follows previously released datasets. The 'topic' and 'title' fields were added to enable additional tasks like title generation and topic detection.", "#### How were labels chosen?\n\n\nThey are human written highlights or summaries scraped from the same website.", "#### Example Instance", "#### Data Splits\n\n\nThe statistics of the original dataset are:\n\n\n\nThe statistics of the cleaned version of the dataset are:\n\n\n\nThe COVID challenge sets have 5058 (de) and 1938 (es) examples.", "#### Splitting Criteria\n\n\nThe training set contains data from 2010 to 2018. Data from 2019 (~10% of the dataset) is used for validation (up to May) and testing(May-December 2019).", "#### \n\n\nSome topics are less represented within the dataset (e.g., Financial news in German and Television in Spanish).\n\n\nDataset in GEM\n--------------", "### Rationale for Inclusion in GEM", "#### Why is the Dataset in GEM?\n\n\nAs the first large-scale multilingual summarization dataset, it enables evaluation of summarization models beyond English.", "#### Similar Datasets\n\n\nyes", "#### Unique Language Coverage\n\n\nyes", "#### Difference from other GEM datasets\n\n\nIn our configuration, the dataset is fully non-English.", "#### Ability that the Dataset measures\n\n\nContent Selection, Content Planning, Realization", "### GEM-Specific Curation", "#### Modificatied for GEM?\n\n\nyes", "#### GEM Modifications\n\n\n'data points removed', 'data points added'", "#### Modification Details\n\n\nThe modifications done to the original dataset are the following:\n\n\n* Selection of 2 languages (Spanish and German) out of the dataset 5 languages due to copyright restrictions.\n* Removal of duplicate articles.\n* Manually removal of article-summary pairs for which the summary is not related to the article.\n* Removal of article-summary pairs written in a different language (detected using the langdetect library).", "#### Additional Splits?\n\n\nyes", "#### Split Information\n\n\nFor both selected languages (German and Spanish), we compiled time-shifted test data in the form of new articles for the second semester of 2020 with Covid19-related keywords. We collected articles from the same German and Spanish outlets as the original MLSUM datasets (El Pais and Süddeutsche Zeitung). We used the scripts provided for the re-creation of the MLSUM datasets. The new challenge test set for German contains 5058 instances and the Spanish one contains 1938.\n\n\nWe additionally sample 500 training and validation points as additional challenge sets to measure overfitting.", "#### Split Motivation\n\n\nGeneralization to unseen topics.", "### Getting Started with the Task\n\n\nPrevious Results\n----------------", "### Previous Results", "#### Measured Model Abilities\n\n\nContent Selection, Content Planning, Realization", "#### Metrics\n\n\n'METEOR', 'ROUGE', 'Other: Other Metrics'", "#### Other Metrics\n\n\nNovelty: Number of generated n-grams not included in the source articles.", "#### Proposed Evaluation\n\n\nROUGE and METEOR both measure n-gram overlap with a focus on recall and are standard summarization metrics. Novelty is often reported alongside them to characterize how much a model diverges from its inputs.", "#### Previous results available?\n\n\nyes", "#### Other Evaluation Approaches\n\n\nThe GEM benchmark results (URL report a wide range of metrics include lexical overlap metrics but also semantic ones like BLEURT and BERT-Score.\n\n\nDataset Curation\n----------------", "### Original Curation", "#### Original Curation Rationale\n\n\nThe rationale was to create a multilingual news summarization dataset that mirrors the format of popular English datasets like XSum or CNN/DM.", "#### Communicative Goal\n\n\nThe speaker is required to produce a high quality summary of news articles in the same language as the input article.", "#### Sourced from Different Sources\n\n\nyes", "#### Source Details\n\n\nURL\nURL\nURL\nURL\nURL", "### Language Data", "#### How was Language Data Obtained?\n\n\n'Found'", "#### Where was it found?\n\n\n'Multiple websites'", "#### Language Producers\n\n\nThe language producers are professional journalists.", "#### Topics Covered\n\n\n4/5 of the original languages report their topics (except Turkish) and the distributions differ between sources. The dominant topics in German are Politik, Sport, Wirtschaft (economy). The dominant topics in Spanish are actualidad (current news) and opinion. French and Russian are different as well but we omit these languages in the GEM version.", "#### Data Validation\n\n\nnot validated", "#### Was Data Filtered?\n\n\nalgorithmically", "#### Filter Criteria\n\n\nIn the original dataset, only one filter was applied: all the articles shorter than 50 words or summaries shorter than 10 words are discarded.\n\n\nThe GEM version additionally applies langID filter to ensure that articles are in the correct language.", "### Structured Annotations", "#### Additional Annotations?\n\n\nnone", "#### Annotation Service?\n\n\nno", "### Consent", "#### Any Consent Policy?\n\n\nno", "#### Justification for Using the Data\n\n\nThe copyright remains with the original data creators and the usage permission is restricted to non-commercial uses.", "### Private Identifying Information (PII)", "#### Contains PII?\n\n\nyes/very likely", "#### Categories of PII\n\n\n'sensitive information', 'generic PII'", "#### Any PII Identification?\n\n\nno identification", "### Maintenance", "#### Any Maintenance Plan?\n\n\nno\n\n\nBroader Social Context\n----------------------", "### Previous Work on the Social Impact of the Dataset", "#### Usage of Models based on the Data\n\n\nno", "### Impact on Under-Served Communities", "#### Addresses needs of underserved Communities?\n\n\nno", "### Discussion of Biases", "#### Any Documented Social Biases?\n\n\nno" ]

[ "TAGS\n#task_categories-summarization #annotations_creators-none #language_creators-unknown #multilinguality-unknown #size_categories-unknown #source_datasets-original #language-German #language-Spanish #license-other #region-us \n", "### Link to Main Data Card\n\n\nYou can find the main data card on the GEM Website.", "### Dataset Summary\n\n\nMLSum is a multilingual summarization dataset crawled from different news websites. The GEM version supports the German and Spanish subset alongside specifically collected challenge sets for COVID-related articles to test out-of-domain generalization.\n\n\nYou can load the dataset via:\n\n\nThe data loader can be found here.", "#### website\n\n\nN/A", "#### paper\n\n\nACL Anthology", "#### authors\n\n\nThomas Scialom, Paul-Alexis Dray, Sylvain Lamprier, Benjamin Piwowarski, Jacopo Staiano\n\n\nDataset Overview\n----------------", "### Where to find the Data and its Documentation", "#### Download\n\n\nGitlab", "#### Paper\n\n\nACL Anthology", "#### BibTex", "#### Contact Name\n\n\nThomas Scialom", "#### Contact Email\n\n\n{thomas,paul-alexis,jacopo}@URL, {sylvain.lamprier,benjamin.piwowarski}@URL", "#### Has a Leaderboard?\n\n\nno", "### Languages and Intended Use", "#### Multilingual?\n\n\nyes", "#### Covered Dialects\n\n\nThere is only one dialect per language, Hochdeutsch for German and Castilian Spanish for Spanish.", "#### Covered Languages\n\n\n'German', 'Spanish, Castilian'", "#### Whose Language?\n\n\nThe German articles are crawled from Süddeutsche Zeitung and the Spanish ones from El Pais.", "#### License\n\n\nother: Other license", "#### Intended Use\n\n\nThe intended use of this dataset is to augment existing datasets for English news summarization with additional languages.", "#### Add. License Info\n\n\nRestricted to non-commercial research purposes.", "#### Primary Task\n\n\nSummarization", "#### Communicative Goal\n\n\nThe speaker is required to produce a high quality summary of news articles in the same language as the input article.", "### Credit", "#### Curation Organization Type(s)\n\n\n'other'", "#### Curation Organization(s)\n\n\nCNRS, Sorbonne Université, reciTAL", "#### Dataset Creators\n\n\nThomas Scialom, Paul-Alexis Dray, Sylvain Lamprier, Benjamin Piwowarski, Jacopo Staiano", "#### Funding\n\n\nFunding information is not specified.", "#### Who added the Dataset to GEM?\n\n\nThe original data card was written by Pedro Henrique Martins (Instituto de Telecomunicações) and Sebastian Gehrmann (Google Research) extended and updated it to the v2 format. The COVID challenge set was created by Laura Perez-Beltrachini (University of Edinburgh). Data cleaning was done by Juan Diego Rodriguez (UT Austin).", "### Dataset Structure", "#### Data Fields\n\n\nThe data fields are:\n\n\n* 'text': the source article ('string').\n* 'summary': the output summary ('string').\n* 'topic': the topic of the article ('string').\n* 'url': the article's url ('string').\n* 'title': the article's title ('string').\n* 'date': the article's date ('string').", "#### Reason for Structure\n\n\nThe structure follows previously released datasets. The 'topic' and 'title' fields were added to enable additional tasks like title generation and topic detection.", "#### How were labels chosen?\n\n\nThey are human written highlights or summaries scraped from the same website.", "#### Example Instance", "#### Data Splits\n\n\nThe statistics of the original dataset are:\n\n\n\nThe statistics of the cleaned version of the dataset are:\n\n\n\nThe COVID challenge sets have 5058 (de) and 1938 (es) examples.", "#### Splitting Criteria\n\n\nThe training set contains data from 2010 to 2018. Data from 2019 (~10% of the dataset) is used for validation (up to May) and testing(May-December 2019).", "#### \n\n\nSome topics are less represented within the dataset (e.g., Financial news in German and Television in Spanish).\n\n\nDataset in GEM\n--------------", "### Rationale for Inclusion in GEM", "#### Why is the Dataset in GEM?\n\n\nAs the first large-scale multilingual summarization dataset, it enables evaluation of summarization models beyond English.", "#### Similar Datasets\n\n\nyes", "#### Unique Language Coverage\n\n\nyes", "#### Difference from other GEM datasets\n\n\nIn our configuration, the dataset is fully non-English.", "#### Ability that the Dataset measures\n\n\nContent Selection, Content Planning, Realization", "### GEM-Specific Curation", "#### Modificatied for GEM?\n\n\nyes", "#### GEM Modifications\n\n\n'data points removed', 'data points added'", "#### Modification Details\n\n\nThe modifications done to the original dataset are the following:\n\n\n* Selection of 2 languages (Spanish and German) out of the dataset 5 languages due to copyright restrictions.\n* Removal of duplicate articles.\n* Manually removal of article-summary pairs for which the summary is not related to the article.\n* Removal of article-summary pairs written in a different language (detected using the langdetect library).", "#### Additional Splits?\n\n\nyes", "#### Split Information\n\n\nFor both selected languages (German and Spanish), we compiled time-shifted test data in the form of new articles for the second semester of 2020 with Covid19-related keywords. We collected articles from the same German and Spanish outlets as the original MLSUM datasets (El Pais and Süddeutsche Zeitung). We used the scripts provided for the re-creation of the MLSUM datasets. The new challenge test set for German contains 5058 instances and the Spanish one contains 1938.\n\n\nWe additionally sample 500 training and validation points as additional challenge sets to measure overfitting.", "#### Split Motivation\n\n\nGeneralization to unseen topics.", "### Getting Started with the Task\n\n\nPrevious Results\n----------------", "### Previous Results", "#### Measured Model Abilities\n\n\nContent Selection, Content Planning, Realization", "#### Metrics\n\n\n'METEOR', 'ROUGE', 'Other: Other Metrics'", "#### Other Metrics\n\n\nNovelty: Number of generated n-grams not included in the source articles.", "#### Proposed Evaluation\n\n\nROUGE and METEOR both measure n-gram overlap with a focus on recall and are standard summarization metrics. Novelty is often reported alongside them to characterize how much a model diverges from its inputs.", "#### Previous results available?\n\n\nyes", "#### Other Evaluation Approaches\n\n\nThe GEM benchmark results (URL report a wide range of metrics include lexical overlap metrics but also semantic ones like BLEURT and BERT-Score.\n\n\nDataset Curation\n----------------", "### Original Curation", "#### Original Curation Rationale\n\n\nThe rationale was to create a multilingual news summarization dataset that mirrors the format of popular English datasets like XSum or CNN/DM.", "#### Communicative Goal\n\n\nThe speaker is required to produce a high quality summary of news articles in the same language as the input article.", "#### Sourced from Different Sources\n\n\nyes", "#### Source Details\n\n\nURL\nURL\nURL\nURL\nURL", "### Language Data", "#### How was Language Data Obtained?\n\n\n'Found'", "#### Where was it found?\n\n\n'Multiple websites'", "#### Language Producers\n\n\nThe language producers are professional journalists.", "#### Topics Covered\n\n\n4/5 of the original languages report their topics (except Turkish) and the distributions differ between sources. The dominant topics in German are Politik, Sport, Wirtschaft (economy). The dominant topics in Spanish are actualidad (current news) and opinion. French and Russian are different as well but we omit these languages in the GEM version.", "#### Data Validation\n\n\nnot validated", "#### Was Data Filtered?\n\n\nalgorithmically", "#### Filter Criteria\n\n\nIn the original dataset, only one filter was applied: all the articles shorter than 50 words or summaries shorter than 10 words are discarded.\n\n\nThe GEM version additionally applies langID filter to ensure that articles are in the correct language.", "### Structured Annotations", "#### Additional Annotations?\n\n\nnone", "#### Annotation Service?\n\n\nno", "### Consent", "#### Any Consent Policy?\n\n\nno", "#### Justification for Using the Data\n\n\nThe copyright remains with the original data creators and the usage permission is restricted to non-commercial uses.", "### Private Identifying Information (PII)", "#### Contains PII?\n\n\nyes/very likely", "#### Categories of PII\n\n\n'sensitive information', 'generic PII'", "#### Any PII Identification?\n\n\nno identification", "### Maintenance", "#### Any Maintenance Plan?\n\n\nno\n\n\nBroader Social Context\n----------------------", "### Previous Work on the Social Impact of the Dataset", "#### Usage of Models based on the Data\n\n\nno", "### Impact on Under-Served Communities", "#### Addresses needs of underserved Communities?\n\n\nno", "### Discussion of Biases", "#### Any Documented Social Biases?\n\n\nno" ]

[ "task_categories:other", "annotations_creators:expert-created", "language_creators:unknown", "multilinguality:unknown", "size_categories:unknown", "source_datasets:original", "language:de", "language:en", "language:fi", "language:fr", "language:ru", "language:sv", "license:cc-by-nc-4.0", "paraphrasing", "region:us" ]

[]

[ "de", "en", "fi", "fr", "ru", "sv" ]

[ "### Link to Main Data Card\n\n\nYou can find the main data card on the GEM Website.", "### Dataset Summary\n\n\nOpusparcus is a paraphrase corpus for six European language: German, English, Finnish, French, Russian, and Swedish. The paraphrases consist of subtitles from movies and TV shows.\n\n\nYou can load the dataset via:\n\n\nThe data loader can be found here.", "#### website\n\n\nWebsite", "#### paper\n\n\nLREC\n\n\nDataset Overview\n----------------", "### Where to find the Data and its Documentation", "#### Webpage\n\n\nWebsite", "#### Download\n\n\nWebsite", "#### Paper\n\n\nLREC", "#### BibTex", "#### Contact Name\n\n\nMathias Creutz", "#### Contact Email\n\n\nfirstname dot lastname at helsinki dot fi", "#### Has a Leaderboard?\n\n\nno", "### Languages and Intended Use", "#### Multilingual?\n\n\nyes", "#### Covered Languages\n\n\n'German', 'English', 'Finnish', 'French', 'Russian', 'Swedish'", "#### Whose Language?\n\n\nOpusparcus is a paraphrase corpus for six European language: German, English, Finnish, French, Russian, and Swedish. The paraphrases consist of subtitles from movies and TV shows.\n\n\nThe data in Opusparcus has been extracted from OpenSubtitles2016, which is in turn based on data from OpenSubtitles.", "#### License\n\n\ncc-by-nc-4.0: Creative Commons Attribution Non Commercial 4.0 International", "#### Intended Use\n\n\nOpusparcus is a sentential paraphrase corpus for multiple languages containing colloquial language.", "#### Primary Task\n\n\nParaphrasing", "#### Communicative Goal\n\n\nModels can be trained, e.g., for paraphrase detection and generation, that is, determining whether two given sentences mean the same thing or generating new paraphrases for a given sentence.", "### Credit", "#### Who added the Dataset to GEM?\n\n\nMathias Creutz (University of Helsinki)", "### Dataset Structure", "#### Data Fields\n\n\n* 'sent1': a tokenized sentence\n* 'sent2': another tokenized sentence, which is potentially a paraphrase of 'sent1'.\n* 'annot\\_score': a value between 1.0 and 4.0 indicating how good an example of paraphrases 'sent1' and 'sent2' are. (For the training sets, the value is 0.0, which indicates that no manual annotation has taken place.)\n* 'lang': language of this dataset\n* 'gem\\_id': unique identifier of this entry\n\n\nAll fields are strings except 'annot\\_score', which is a float.", "#### Reason for Structure\n\n\nFor each target language, the Opusparcus data have been partitioned into three types of data sets: training, validation and test sets. The training sets are large, consisting of millions of sentence pairs, and have been compiled automatically, with the help of probabilistic ranking functions. The development and test sets consist of sentence pairs that have been annotated manually; each set contains approximately 1000 sentence pairs that have been verified to be acceptable paraphrases by two independent annotators.\n\n\nWhen you download Opusparcus, you must always indicate the language you want to retrieve, for instance:\n\n\nThe above command will download the validation and test sets for German. If additionally, you want to retrieve training data, you need to specify the level of quality you desire, such as \"French, with 90% quality of the training data\":\n\n\nThe entries in the training sets have been ranked automatically by how likely they are paraphrases, best first, worst last. The quality parameter indicates the estimated proportion (in percent) of true\nparaphrases in the training set. Allowed quality values range between 60 and 100, in increments of 5 (60, 65, 70, ..., 100). A value of 60 means that 60% of the sentence pairs in the training set are estimated to be true paraphrases (and the remaining 40% are not). A higher value produces a smaller but cleaner set. The smaller sets are subsets of the larger sets, such that the 'quality=95' set is a subset of 'quality=90', which is a subset of 'quality=85', and so on.\n\n\nThe default 'quality' value, if omitted, is 100. This matches no training data at all, which can be convenient, if you are only interested in the validation and test sets, which are considerably\nsmaller, but manually annotated.\n\n\nNote that an alternative to typing the parameter values explicitly, you can use configuration names instead. The following commands are equivalent to the ones above:", "#### How were labels chosen?\n\n\nAnnotators have used the following scores to label sentence pairs in the test and validation sets:\n\n\n4: Good example of paraphrases (Dark green button in the annotation tool): The two sentences can be used in the same situation and essentially \"mean the same thing\".\n\n\n3: Mostly good example of paraphrases (Light green button in the annotation tool): It is acceptable to think that the two sentences refer to the same thing, although one sentence might be more specific\nthan the other one, or there are differences in style, such as polite form versus familiar form.\n\n\n2: Mostly bad example of paraphrases (Yellow button in the annotation tool): There is some connection between the sentences that explains why they occur together, but one would not really consider them to mean the same thing.\n\n\n1: Bad example of paraphrases (Red button in the annotation tool): There is no obvious connection. The sentences mean different things.\n\n\nIf the two annotators fully agreed on the category, the value in the 'annot\\_score' field is 4.0, 3.0, 2.0 or 1.0. If the two annotators chose adjacent categories, the value in this field will be 3.5, 2.5 or\n1.5. For instance, a value of 2.5 means that one annotator gave a score of 3 (\"mostly good\"), indicating a possible paraphrase pair, whereas the other annotator scored this as a 2 (\"mostly bad\"), that is, unlikely to be a paraphrase pair. If the annotators disagreed by more than one category, the sentence pair was discarded and won't show up in the datasets.\n\n\nThe training sets were not annotated manually. This is indicated by\nthe value 0.0 in the 'annot\\_score' field.\n\n\nFor an assessment of of inter-annotator agreement, see Aulamo et al. (2019). Annotation of subtitle paraphrases using a new web tool. In *Proceedings of the\nDigital Humanities in the Nordic Countries 4th Conference*, Copenhagen, Denmark.", "#### Example Instance", "#### Data Splits\n\n\nThe data is split into training, validation and test sets. The validation and test sets come in two versions, the regular validation and test sets and the full sets, called URL and URL. The full sets contain all sentence pairs successfully annotated by the annotators, including the sentence pairs that were rejected as paraphrases. The annotation scores of the full sets thus range between 1.0 and 4.0. The regular validation and test sets only contain sentence pairs that qualify as paraphrases, scored between 3.0 and 4.0 by the annotators.\n\n\nThe number of sentence pairs in the data splits are as follows for each of the languages. The range between the smallest ('quality=95') and largest ('quality=60') train configuration have been shown.\n\n\n\nAs a concrete example, loading the English data requesting 95% quality of the train split produces the following:", "#### Splitting Criteria\n\n\nThe validation and test sets have been annotated manually, but the training sets have been produced using automatic scoring and come in different size configurations depending on the desired quality level. (See above descriptions and examples for more details.)\n\n\nPlease note that previous work suggests that a larger and noisier training set is better than a\nsmaller and clean set. See Sjöblom et al. (2018). Paraphrase Detection on Noisy Subtitles in Six\nLanguages. In *Proceedings of the 2018 EMNLP Workshop W-NUT: The 4th Workshop on Noisy User-generated Text*, and Vahtola et al. (2021). Coping with Noisy Training Data Labels in Paraphrase Detection. In *Proceedings of the 7th Workshop on Noisy User-generated Text*.\n\n\nDataset in GEM\n--------------", "### Rationale for Inclusion in GEM", "#### Why is the Dataset in GEM?\n\n\nOpusparcus provides examples of sentences that mean the same thing or have very similar meaning. Sentences are available in six languages and the style is colloquial language.", "#### Similar Datasets\n\n\nyes", "#### Unique Language Coverage\n\n\nyes", "#### Difference from other GEM datasets\n\n\nThere is another data set containing manually labeled Finnish paraphrases.", "#### Ability that the Dataset measures\n\n\nSentence meaning", "### GEM-Specific Curation", "#### Modificatied for GEM?\n\n\nyes", "#### GEM Modifications\n\n\n'other'", "#### Modification Details\n\n\nTraining sets have been prepared for each the \"quality levels\" 60% – 95%.\n\n\nIn the original release, this task was left to the user of the data.", "#### Additional Splits?\n\n\nyes", "#### Split Information\n\n\nThere are two versions of the validations and test sets: the regular sets which only contain positive examples of paraphrases and the full sets containing all examples.", "#### Split Motivation\n\n\nIn the original release, only the full validation and test sets were supplied. The \"regular sets\" have been added in order to make it easier to test on true parapahrases only.", "### Getting Started with the Task", "#### Pointers to Resources\n\n\nCreutz (2018). Open Subtitles Paraphrase Corpus for Six Languages, Proceedings of the 11th edition of the Language Resources and Evaluation Conference (LREC 2018).\n\n\nSjöblom et al. (2018). Paraphrase Detection on Noisy Subtitles in Six Languages. In Proceedings of the 2018 EMNLP Workshop W-NUT: The 4th Workshop on Noisy User-generated Text.\n\n\nAulamo et al. (2019). Annotation of subtitle paraphrases using a new web tool. In Proceedings of the Digital Humanities in the Nordic Countries 4th Conference.\n\n\nSjöblom et al. (2020). Paraphrase Generation and Evaluation on Colloquial-Style Sentences, Proceedings of the 12th Language Resources and Evaluation Conference (LREC).\n\n\nVahtola et al. (2021). Coping with Noisy Training Data Labels in Paraphrase Detection. In Proceedings of the 7th Workshop on Noisy User-generated Text.\n\n\nPrevious Results\n----------------", "### Previous Results", "#### Measured Model Abilities\n\n\nSentence meaning\n\n\nIn a scenario of paraphrase detection, the model determines whether two given sentences carry approximately the same meaning.\n\n\nIn a scenario of paraphrase generation, the model generates a potential paraphrase of a given sentence.", "#### Metrics\n\n\n'BLEU', 'BERT-Score', 'Other: Other Metrics'", "#### Other Metrics\n\n\nPINC", "#### Proposed Evaluation\n\n\nThe metrics mentioned above can be used to assess how well a generated paraphrase corresponds to a given reference sentence. The PINC score additionally assesses how different the surface forms are.", "#### Previous results available?\n\n\nyes", "#### Other Evaluation Approaches\n\n\nSee publications on using Opusparcus", "#### Relevant Previous Results\n\n\nSjöblom et al. (2020). Paraphrase Generation and Evaluation on Colloquial-Style Sentences, Proceedings of the 12th Language Resources and Evaluation Conference (LREC).\n\n\nDataset Curation\n----------------", "### Original Curation", "#### Original Curation Rationale\n\n\nOpusparcus was created in order to produce a *sentential* paraphrase corpus for multiple languages containing *colloquial* language (as opposed to news or religious text, for instance).", "#### Communicative Goal\n\n\nOpusparcus provides labeled examples of pairs of sentences that have similar (or dissimilar) meanings.", "#### Sourced from Different Sources\n\n\nno", "### Language Data", "#### How was Language Data Obtained?\n\n\n'Crowdsourced'", "#### Where was it crowdsourced?\n\n\n'Other crowdworker platform'", "#### Language Producers\n\n\nThe data in Opusparcus has been extracted from OpenSubtitles2016, which is in turn based on data from URL.\n\n\nThe texts consists of subtitles that have been produced using crowdsourcing.", "#### Topics Covered\n\n\nThe language is representative of movies and TV shows. Domains covered include comedy, drama, relationships, suspense, etc.", "#### Data Validation\n\n\nvalidated by data curator", "#### Data Preprocessing\n\n\nSentence and word tokenization was performed.", "#### Was Data Filtered?\n\n\nalgorithmically", "#### Filter Criteria\n\n\nThe sentence pairs in the training sets were ordered automatically based on the estimated likelihood that the sentences were paraphrases, most likely paraphrases on the top, and least likely paraphrases on the bottom.\n\n\nThe validation and test sets were checked and annotated manually, but the sentence pairs selected for annotation had to be different enough in terms of minimum edit distance (Levenshtein distance). This ensured that annotators would not spend their time annotating pairs of more or less identical sentences.", "### Structured Annotations", "#### Additional Annotations?\n\n\nexpert created", "#### Number of Raters\n\n\n11<n<50", "#### Rater Qualifications\n\n\nStudents and staff at the University of Helsinki (native or very proficient speakers of the target languages)", "#### Raters per Training Example\n\n\n0", "#### Raters per Test Example\n\n\n2", "#### Annotation Service?\n\n\nno", "#### Annotation Values\n\n\nThe development and test sets consist of sentence pairs that have been annotated manually; each set contains approximately 1000 sentence pairs that have been verified to be acceptable paraphrases by two independent annotators.\n\n\nThe 'annot\\_score' field reflects the judgments made by the annotators. If the annnotators fully agreed on the category (4.0: dark green, 3.0: light green, 2.0: yellow, 1.0: red), the value of 'annot\\_score' is 4.0, 3.0, 2.0 or 1.0. If the annotators chose adjacent categories, the value in this field will be 3.5, 2.5 or 1.5. For instance, a value of 2.5 means that one annotator gave a score of 3 (\"mostly good\"), indicating a possible paraphrase pair, whereas the other annotator scored this as a 2 (\"mostly bad\"), that is, unlikely to be a paraphrase pair. If the annotators disagreed by more than one category, the sentence pair was discarded and won't show up in the datasets.\n\n\nAnnotators could also reject a sentence pair as being corrupted data.", "#### Any Quality Control?\n\n\nvalidated by another rater", "#### Quality Control Details\n\n\nIf the annotators disagreed by more than one category, the sentence pair was discarded and is not part of the final dataset.", "### Consent", "#### Any Consent Policy?\n\n\nno", "### Private Identifying Information (PII)", "#### Contains PII?\n\n\nyes/very likely", "#### Any PII Identification?\n\n\nno identification", "### Maintenance", "#### Any Maintenance Plan?\n\n\nno\n\n\nBroader Social Context\n----------------------", "### Previous Work on the Social Impact of the Dataset", "#### Usage of Models based on the Data\n\n\nno", "### Impact on Under-Served Communities", "#### Addresses needs of underserved Communities?\n\n\nno", "### Discussion of Biases", "#### Any Documented Social Biases?\n\n\nno", "#### Are the Language Producers Representative of the Language?\n\n\nWhat social bias there may be in the subtitles in this dataset has not been studied.\n\n\nConsiderations for Using the Data\n---------------------------------", "### PII Risks and Liability", "#### Potential PII Risk\n\n\nThe data only contains subtitles of publicly available movies and TV shows.", "### Licenses", "#### Copyright Restrictions on the Dataset\n\n\n'non-commercial use only'", "#### Copyright Restrictions on the Language Data\n\n\n'non-commercial use only'", "### Known Technical Limitations", "#### Technical Limitations\n\n\nSome subtitles contain typos that are caused by inaccurate OCR.", "#### Unsuited Applications\n\n\nThe models might memorize individual subtitles of existing movies and TV shows, but there is no context across sentence boundaries in the data.", "#### Discouraged Use Cases\n\n\nA general issue with paraphrasing is that very small modifications in the surface form might produce valid paraphrases, which are however rather uninteresting. It is more valuable to produce paraphrases with clearly different surface realizations (e.g., measured using minimum edit distance)." ]

[ "TAGS\n#task_categories-other #annotations_creators-expert-created #language_creators-unknown #multilinguality-unknown #size_categories-unknown #source_datasets-original #language-German #language-English #language-Finnish #language-French #language-Russian #language-Swedish #license-cc-by-nc-4.0 #paraphrasing #region-us \n", "### Link to Main Data Card\n\n\nYou can find the main data card on the GEM Website.", "### Dataset Summary\n\n\nOpusparcus is a paraphrase corpus for six European language: German, English, Finnish, French, Russian, and Swedish. The paraphrases consist of subtitles from movies and TV shows.\n\n\nYou can load the dataset via:\n\n\nThe data loader can be found here.", "#### website\n\n\nWebsite", "#### paper\n\n\nLREC\n\n\nDataset Overview\n----------------", "### Where to find the Data and its Documentation", "#### Webpage\n\n\nWebsite", "#### Download\n\n\nWebsite", "#### Paper\n\n\nLREC", "#### BibTex", "#### Contact Name\n\n\nMathias Creutz", "#### Contact Email\n\n\nfirstname dot lastname at helsinki dot fi", "#### Has a Leaderboard?\n\n\nno", "### Languages and Intended Use", "#### Multilingual?\n\n\nyes", "#### Covered Languages\n\n\n'German', 'English', 'Finnish', 'French', 'Russian', 'Swedish'", "#### Whose Language?\n\n\nOpusparcus is a paraphrase corpus for six European language: German, English, Finnish, French, Russian, and Swedish. The paraphrases consist of subtitles from movies and TV shows.\n\n\nThe data in Opusparcus has been extracted from OpenSubtitles2016, which is in turn based on data from OpenSubtitles.", "#### License\n\n\ncc-by-nc-4.0: Creative Commons Attribution Non Commercial 4.0 International", "#### Intended Use\n\n\nOpusparcus is a sentential paraphrase corpus for multiple languages containing colloquial language.", "#### Primary Task\n\n\nParaphrasing", "#### Communicative Goal\n\n\nModels can be trained, e.g., for paraphrase detection and generation, that is, determining whether two given sentences mean the same thing or generating new paraphrases for a given sentence.", "### Credit", "#### Who added the Dataset to GEM?\n\n\nMathias Creutz (University of Helsinki)", "### Dataset Structure", "#### Data Fields\n\n\n* 'sent1': a tokenized sentence\n* 'sent2': another tokenized sentence, which is potentially a paraphrase of 'sent1'.\n* 'annot\\_score': a value between 1.0 and 4.0 indicating how good an example of paraphrases 'sent1' and 'sent2' are. (For the training sets, the value is 0.0, which indicates that no manual annotation has taken place.)\n* 'lang': language of this dataset\n* 'gem\\_id': unique identifier of this entry\n\n\nAll fields are strings except 'annot\\_score', which is a float.", "#### Reason for Structure\n\n\nFor each target language, the Opusparcus data have been partitioned into three types of data sets: training, validation and test sets. The training sets are large, consisting of millions of sentence pairs, and have been compiled automatically, with the help of probabilistic ranking functions. The development and test sets consist of sentence pairs that have been annotated manually; each set contains approximately 1000 sentence pairs that have been verified to be acceptable paraphrases by two independent annotators.\n\n\nWhen you download Opusparcus, you must always indicate the language you want to retrieve, for instance:\n\n\nThe above command will download the validation and test sets for German. If additionally, you want to retrieve training data, you need to specify the level of quality you desire, such as \"French, with 90% quality of the training data\":\n\n\nThe entries in the training sets have been ranked automatically by how likely they are paraphrases, best first, worst last. The quality parameter indicates the estimated proportion (in percent) of true\nparaphrases in the training set. Allowed quality values range between 60 and 100, in increments of 5 (60, 65, 70, ..., 100). A value of 60 means that 60% of the sentence pairs in the training set are estimated to be true paraphrases (and the remaining 40% are not). A higher value produces a smaller but cleaner set. The smaller sets are subsets of the larger sets, such that the 'quality=95' set is a subset of 'quality=90', which is a subset of 'quality=85', and so on.\n\n\nThe default 'quality' value, if omitted, is 100. This matches no training data at all, which can be convenient, if you are only interested in the validation and test sets, which are considerably\nsmaller, but manually annotated.\n\n\nNote that an alternative to typing the parameter values explicitly, you can use configuration names instead. The following commands are equivalent to the ones above:", "#### How were labels chosen?\n\n\nAnnotators have used the following scores to label sentence pairs in the test and validation sets:\n\n\n4: Good example of paraphrases (Dark green button in the annotation tool): The two sentences can be used in the same situation and essentially \"mean the same thing\".\n\n\n3: Mostly good example of paraphrases (Light green button in the annotation tool): It is acceptable to think that the two sentences refer to the same thing, although one sentence might be more specific\nthan the other one, or there are differences in style, such as polite form versus familiar form.\n\n\n2: Mostly bad example of paraphrases (Yellow button in the annotation tool): There is some connection between the sentences that explains why they occur together, but one would not really consider them to mean the same thing.\n\n\n1: Bad example of paraphrases (Red button in the annotation tool): There is no obvious connection. The sentences mean different things.\n\n\nIf the two annotators fully agreed on the category, the value in the 'annot\\_score' field is 4.0, 3.0, 2.0 or 1.0. If the two annotators chose adjacent categories, the value in this field will be 3.5, 2.5 or\n1.5. For instance, a value of 2.5 means that one annotator gave a score of 3 (\"mostly good\"), indicating a possible paraphrase pair, whereas the other annotator scored this as a 2 (\"mostly bad\"), that is, unlikely to be a paraphrase pair. If the annotators disagreed by more than one category, the sentence pair was discarded and won't show up in the datasets.\n\n\nThe training sets were not annotated manually. This is indicated by\nthe value 0.0 in the 'annot\\_score' field.\n\n\nFor an assessment of of inter-annotator agreement, see Aulamo et al. (2019). Annotation of subtitle paraphrases using a new web tool. In *Proceedings of the\nDigital Humanities in the Nordic Countries 4th Conference*, Copenhagen, Denmark.", "#### Example Instance", "#### Data Splits\n\n\nThe data is split into training, validation and test sets. The validation and test sets come in two versions, the regular validation and test sets and the full sets, called URL and URL. The full sets contain all sentence pairs successfully annotated by the annotators, including the sentence pairs that were rejected as paraphrases. The annotation scores of the full sets thus range between 1.0 and 4.0. The regular validation and test sets only contain sentence pairs that qualify as paraphrases, scored between 3.0 and 4.0 by the annotators.\n\n\nThe number of sentence pairs in the data splits are as follows for each of the languages. The range between the smallest ('quality=95') and largest ('quality=60') train configuration have been shown.\n\n\n\nAs a concrete example, loading the English data requesting 95% quality of the train split produces the following:", "#### Splitting Criteria\n\n\nThe validation and test sets have been annotated manually, but the training sets have been produced using automatic scoring and come in different size configurations depending on the desired quality level. (See above descriptions and examples for more details.)\n\n\nPlease note that previous work suggests that a larger and noisier training set is better than a\nsmaller and clean set. See Sjöblom et al. (2018). Paraphrase Detection on Noisy Subtitles in Six\nLanguages. In *Proceedings of the 2018 EMNLP Workshop W-NUT: The 4th Workshop on Noisy User-generated Text*, and Vahtola et al. (2021). Coping with Noisy Training Data Labels in Paraphrase Detection. In *Proceedings of the 7th Workshop on Noisy User-generated Text*.\n\n\nDataset in GEM\n--------------", "### Rationale for Inclusion in GEM", "#### Why is the Dataset in GEM?\n\n\nOpusparcus provides examples of sentences that mean the same thing or have very similar meaning. Sentences are available in six languages and the style is colloquial language.", "#### Similar Datasets\n\n\nyes", "#### Unique Language Coverage\n\n\nyes", "#### Difference from other GEM datasets\n\n\nThere is another data set containing manually labeled Finnish paraphrases.", "#### Ability that the Dataset measures\n\n\nSentence meaning", "### GEM-Specific Curation", "#### Modificatied for GEM?\n\n\nyes", "#### GEM Modifications\n\n\n'other'", "#### Modification Details\n\n\nTraining sets have been prepared for each the \"quality levels\" 60% – 95%.\n\n\nIn the original release, this task was left to the user of the data.", "#### Additional Splits?\n\n\nyes", "#### Split Information\n\n\nThere are two versions of the validations and test sets: the regular sets which only contain positive examples of paraphrases and the full sets containing all examples.", "#### Split Motivation\n\n\nIn the original release, only the full validation and test sets were supplied. The \"regular sets\" have been added in order to make it easier to test on true parapahrases only.", "### Getting Started with the Task", "#### Pointers to Resources\n\n\nCreutz (2018). Open Subtitles Paraphrase Corpus for Six Languages, Proceedings of the 11th edition of the Language Resources and Evaluation Conference (LREC 2018).\n\n\nSjöblom et al. (2018). Paraphrase Detection on Noisy Subtitles in Six Languages. In Proceedings of the 2018 EMNLP Workshop W-NUT: The 4th Workshop on Noisy User-generated Text.\n\n\nAulamo et al. (2019). Annotation of subtitle paraphrases using a new web tool. In Proceedings of the Digital Humanities in the Nordic Countries 4th Conference.\n\n\nSjöblom et al. (2020). Paraphrase Generation and Evaluation on Colloquial-Style Sentences, Proceedings of the 12th Language Resources and Evaluation Conference (LREC).\n\n\nVahtola et al. (2021). Coping with Noisy Training Data Labels in Paraphrase Detection. In Proceedings of the 7th Workshop on Noisy User-generated Text.\n\n\nPrevious Results\n----------------", "### Previous Results", "#### Measured Model Abilities\n\n\nSentence meaning\n\n\nIn a scenario of paraphrase detection, the model determines whether two given sentences carry approximately the same meaning.\n\n\nIn a scenario of paraphrase generation, the model generates a potential paraphrase of a given sentence.", "#### Metrics\n\n\n'BLEU', 'BERT-Score', 'Other: Other Metrics'", "#### Other Metrics\n\n\nPINC", "#### Proposed Evaluation\n\n\nThe metrics mentioned above can be used to assess how well a generated paraphrase corresponds to a given reference sentence. The PINC score additionally assesses how different the surface forms are.", "#### Previous results available?\n\n\nyes", "#### Other Evaluation Approaches\n\n\nSee publications on using Opusparcus", "#### Relevant Previous Results\n\n\nSjöblom et al. (2020). Paraphrase Generation and Evaluation on Colloquial-Style Sentences, Proceedings of the 12th Language Resources and Evaluation Conference (LREC).\n\n\nDataset Curation\n----------------", "### Original Curation", "#### Original Curation Rationale\n\n\nOpusparcus was created in order to produce a *sentential* paraphrase corpus for multiple languages containing *colloquial* language (as opposed to news or religious text, for instance).", "#### Communicative Goal\n\n\nOpusparcus provides labeled examples of pairs of sentences that have similar (or dissimilar) meanings.", "#### Sourced from Different Sources\n\n\nno", "### Language Data", "#### How was Language Data Obtained?\n\n\n'Crowdsourced'", "#### Where was it crowdsourced?\n\n\n'Other crowdworker platform'", "#### Language Producers\n\n\nThe data in Opusparcus has been extracted from OpenSubtitles2016, which is in turn based on data from URL.\n\n\nThe texts consists of subtitles that have been produced using crowdsourcing.", "#### Topics Covered\n\n\nThe language is representative of movies and TV shows. Domains covered include comedy, drama, relationships, suspense, etc.", "#### Data Validation\n\n\nvalidated by data curator", "#### Data Preprocessing\n\n\nSentence and word tokenization was performed.", "#### Was Data Filtered?\n\n\nalgorithmically", "#### Filter Criteria\n\n\nThe sentence pairs in the training sets were ordered automatically based on the estimated likelihood that the sentences were paraphrases, most likely paraphrases on the top, and least likely paraphrases on the bottom.\n\n\nThe validation and test sets were checked and annotated manually, but the sentence pairs selected for annotation had to be different enough in terms of minimum edit distance (Levenshtein distance). This ensured that annotators would not spend their time annotating pairs of more or less identical sentences.", "### Structured Annotations", "#### Additional Annotations?\n\n\nexpert created", "#### Number of Raters\n\n\n11<n<50", "#### Rater Qualifications\n\n\nStudents and staff at the University of Helsinki (native or very proficient speakers of the target languages)", "#### Raters per Training Example\n\n\n0", "#### Raters per Test Example\n\n\n2", "#### Annotation Service?\n\n\nno", "#### Annotation Values\n\n\nThe development and test sets consist of sentence pairs that have been annotated manually; each set contains approximately 1000 sentence pairs that have been verified to be acceptable paraphrases by two independent annotators.\n\n\nThe 'annot\\_score' field reflects the judgments made by the annotators. If the annnotators fully agreed on the category (4.0: dark green, 3.0: light green, 2.0: yellow, 1.0: red), the value of 'annot\\_score' is 4.0, 3.0, 2.0 or 1.0. If the annotators chose adjacent categories, the value in this field will be 3.5, 2.5 or 1.5. For instance, a value of 2.5 means that one annotator gave a score of 3 (\"mostly good\"), indicating a possible paraphrase pair, whereas the other annotator scored this as a 2 (\"mostly bad\"), that is, unlikely to be a paraphrase pair. If the annotators disagreed by more than one category, the sentence pair was discarded and won't show up in the datasets.\n\n\nAnnotators could also reject a sentence pair as being corrupted data.", "#### Any Quality Control?\n\n\nvalidated by another rater", "#### Quality Control Details\n\n\nIf the annotators disagreed by more than one category, the sentence pair was discarded and is not part of the final dataset.", "### Consent", "#### Any Consent Policy?\n\n\nno", "### Private Identifying Information (PII)", "#### Contains PII?\n\n\nyes/very likely", "#### Any PII Identification?\n\n\nno identification", "### Maintenance", "#### Any Maintenance Plan?\n\n\nno\n\n\nBroader Social Context\n----------------------", "### Previous Work on the Social Impact of the Dataset", "#### Usage of Models based on the Data\n\n\nno", "### Impact on Under-Served Communities", "#### Addresses needs of underserved Communities?\n\n\nno", "### Discussion of Biases", "#### Any Documented Social Biases?\n\n\nno", "#### Are the Language Producers Representative of the Language?\n\n\nWhat social bias there may be in the subtitles in this dataset has not been studied.\n\n\nConsiderations for Using the Data\n---------------------------------", "### PII Risks and Liability", "#### Potential PII Risk\n\n\nThe data only contains subtitles of publicly available movies and TV shows.", "### Licenses", "#### Copyright Restrictions on the Dataset\n\n\n'non-commercial use only'", "#### Copyright Restrictions on the Language Data\n\n\n'non-commercial use only'", "### Known Technical Limitations", "#### Technical Limitations\n\n\nSome subtitles contain typos that are caused by inaccurate OCR.", "#### Unsuited Applications\n\n\nThe models might memorize individual subtitles of existing movies and TV shows, but there is no context across sentence boundaries in the data.", "#### Discouraged Use Cases\n\n\nA general issue with paraphrasing is that very small modifications in the surface form might produce valid paraphrases, which are however rather uninteresting. It is more valuable to produce paraphrases with clearly different surface realizations (e.g., measured using minimum edit distance)." ]

[ "region:us" ]

[]

[]

[ "# GEM References", "## What is it?\n\nThis repository contains all the reference datasets that are used for running evaluation on the GEM benchmark. Some of these datasets were originally hosted as a GitHub release on the 'GEM-metrics' repository, but have been migrated to the Hugging Face Hub.", "## Converting datasets to JSON\n\nWe provide a 'convert_dataset_to_json.py' conversion script that converts the datasets in the GEM organisation to the JSON format expected by the 'GEM-metrics' library. To run the script, first install 'jq' and then install the script's Python dependencies:\n\n\n\nYou can then run the script as follows:\n\n\n\nThis script will:\n\n* Download and convert the datasets under the GEM organisation to JSON format\n* Validate that the each dataset has the expected columns of 'gem_id', 'target', and 'references'" ]

[ "TAGS\n#region-us \n", "# GEM References", "## What is it?\n\nThis repository contains all the reference datasets that are used for running evaluation on the GEM benchmark. Some of these datasets were originally hosted as a GitHub release on the 'GEM-metrics' repository, but have been migrated to the Hugging Face Hub.", "## Converting datasets to JSON\n\nWe provide a 'convert_dataset_to_json.py' conversion script that converts the datasets in the GEM organisation to the JSON format expected by the 'GEM-metrics' library. To run the script, first install 'jq' and then install the script's Python dependencies:\n\n\n\nYou can then run the script as follows:\n\n\n\nThis script will:\n\n* Download and convert the datasets under the GEM organisation to JSON format\n* Validate that the each dataset has the expected columns of 'gem_id', 'target', and 'references'" ]

[ "task_categories:conversational", "annotations_creators:crowd-sourced", "language_creators:unknown", "multilinguality:unknown", "size_categories:unknown", "source_datasets:original", "language:en", "license:cc-by-sa-4.0", "dialog-response-generation", "arxiv:1909.05855", "arxiv:2004.15006", "arxiv:2002.01359", "region:us" ]

[ "1909.05855", "2004.15006", "2002.01359" ]

[ "en" ]

[ "### Link to Main Data Card\n\n\nYou can find the main data card on the [GEM Website.", "### Dataset Summary\n\n\nThe GEM version of this dataset functions as a response generation dataset. The input specifies dialog acts that a model needs to verbalize. The Schema-Guided Dialog dataset is challenging since it comprises multiple domains from hotel and travel to restaurants, and a wide range of dialog acts. The context of each conversation is provided as well.\n\n\nYou can load the dataset via:\n\n\nThe data loader can be found here.", "#### website\n\n\nn/a", "#### paper\n\n\nArxiv", "#### authors\n\n\nAbhinav Rastogi, Xiaoxue Zang, Srinivas Sunkara, Raghav Gupta, Pranav Khaitan, Amir Fayazi, Maria Wang, and Guan-Lin Chao\n\n\nDataset Overview\n----------------", "### Where to find the Data and its Documentation", "#### Download\n\n\nGithub[(URL", "#### Paper\n\n\n[Arxiv", "#### BibTex", "#### Contact Name\n\n\nAbhinav Rastogi", "#### Contact Email\n\n\nschema-guided-dst@URL", "#### Has a Leaderboard?\n\n\nno", "### Languages and Intended Use", "#### Multilingual?\n\n\nno", "#### Covered Languages\n\n\n'English'", "#### Whose Language?\n\n\nThe language structure is machine-generated, and the language realizations are produced by crowd workers.\nThe dataset paper does not provide demographic information for the crowd workers.", "#### License\n\n\ncc-by-sa-4.0: Creative Commons Attribution Share Alike 4.0 International", "#### Intended Use\n\n\nThe Schema-Guided Dialogue (SGD) dataset contains 18K multi-domain task-oriented dialogues between a human and a virtual assistant, which covers 17 domains ranging from banks and events to media, calendar, travel, and weather.\nThe language presents in the datset is only English.\nThe SGD dataset provides a challenging testbed for a number of tasks in task-oriented dialogue, including language understanding, slot filling, dialogue state tracking and response generation.\nFor the creation of the SGD dataset, they developed a multi-domain dialogue simulator that generates dialogue outlines over an arbitrary combination of APIs, dialogue states and system actions. Then, they used a crowd-sourcing procedure to paraphrase these outlines to natural language utterances.\nThis novel crowd-sourcing procedure preserves all annotations obtained from the simulator and does not require any extra annotations after dialogue collection.", "#### Primary Task\n\n\nDialog Response Generation", "#### Communicative Goal\n\n\nThe goal of a speaker who generates the target utterance is to help users accomplish tasks including but not limited to finding flights, booking restaurants, searching for nearby events and movies.", "### Credit", "#### Curation Organization Type(s)\n\n\n'industry'", "#### Curation Organization(s)\n\n\nGoogle", "#### Dataset Creators\n\n\nAbhinav Rastogi, Xiaoxue Zang, Srinivas Sunkara, Raghav Gupta, Pranav Khaitan, Amir Fayazi, Maria Wang, and Guan-Lin Chao", "#### Funding\n\n\nGoogle", "#### Who added the Dataset to GEM?\n\n\nWanyu Du wrote the initial data card and Yacine Jernite the data loader. Simon Mille updated the data card with the additional splits. Sebastian Gehrmann migrated the data card and loader to the v2 version and extended the missing information.", "### Dataset Structure", "#### Data Fields\n\n\nEach dialog instance has the following fields:\n\n\n* 'dialogue\\_id': A unique identifier for a dialogue.\n* 'services': A list of services present in the dialogue.\n* 'turns': A list of annotated system or user utterances. Each turn consists of the following fields:\n\t+ 'speaker': The speaker for the turn, either 'USER' or 'SYSTEM'.\n\t+ 'utterance': A string containing the natural language utterance.\n\t+ 'frames': A list of frames, each frame containing annotations for a single service and consists of the following fields:\n\t\t- 'service': The name of the service corresponding to the frame. The slots and intents used in the following fields are taken from the schema of this service.\n\t\t- 'slots': A list of slot spans in the utterance, only provided for non-categorical slots. Each slot span contains the following fields:\n\t\t\t* 'slot': The name of the slot.\n\t\t\t* 'start': The index of the starting character in the utterance corresponding to the slot value.\n\t\t\t* 'exclusive\\_end': The index of the character just after the last character corresponding to the slot value in the utterance.\n\t\t- 'actions': A list of actions corresponding to the system. Each action has the following fields:\n\t\t\t* 'act': The type of action.\n\t\t\t* 'slot': (optional) A slot argument for some of the actions.\n\t\t\t* 'values': (optional) A list of values assigned to the slot. If the values list is non-empty, then the slot must be present.\n\t\t\t* 'canonical\\_values': (optional) The values in their canonicalized form as used by the service. It is a list of strings of the same length as values.\n\t\t- 'service\\_call': (system turns only, optional) The request sent to the service. It consists of the following fields:\n\t\t\t* 'method': The name of the intent or function of the service or API being executed.\n\t\t\t* 'parameters': A pair of lists of the same lengths: 'parameter\\_slot\\_name' contains slot names and 'parameter\\_canonical\\_value' contains the corresponding values in their canonicalized form.\n\t\t- 'service\\_results': (system turns only, optional) A list of entities containing the results obtained from the service. It is only available for turns in which a service call is made. Each entity is represented as a pair of lists of the same length: 'service\\_slot\\_name' contains slot names and 'service\\_canonical\\_value' contains the corresponding canonical values.\n\t\t- 'state': (user turns only) The dialogue state corresponding to the service. It consists of the following fields:\n\t\t\t* 'active\\_intent': The intent corresponding to the service of the frame which is currently being fulfilled by the system. It takes the value \"NONE\" if none of the intents are active.\n\t\t\t* 'requested\\_slots': A list of slots requested by the user in the current turn.\n\t\t\t* 'slot\\_values': A pair of lists of the same lengths: 'slot\\_name' contains slot names and 'slot\\_value\\_list' contains the corresponding lists of strings. For categorical slots, this list contains a single value assigned to the slot. For non-categorical slots, all the values in this list are spoken variations of each other and are equivalent (e.g, \"6 pm\", \"six in the evening\", \"evening at 6\" etc.).", "#### Example Instance", "#### Data Splits\n\n\nThe dataset is split into a train, validation, and test set with the following sizes:", "#### Splitting Criteria\n\n\nThe data is generally split i.i.d, but some topics only appear in training and some only for testing. For example, the domains Messaging, Payment, and Train are test-only.\n\n\nDataset in GEM\n--------------", "### Rationale for Inclusion in GEM", "#### Why is the Dataset in GEM?\n\n\nThis dataset comprises a wide range of dialog capabilities and thus enables the evaluation of many more generation capabilities of comparable datasets. Its collection methodology ensures a high diversity but also high quality of the data.", "#### Similar Datasets\n\n\nyes", "#### Unique Language Coverage\n\n\nno", "#### Difference from other GEM datasets\n\n\nThe domains a lot more diverse than other datasets.", "#### Ability that the Dataset measures\n\n\nsurface realization, compositionality.", "### GEM-Specific Curation", "#### Modificatied for GEM?\n\n\nyes", "#### GEM Modifications\n\n\n'data points modified'", "#### Modification Details\n\n\nWe are focusing on the response-generation part of the dataset and thus reformatted the dataset to treat the service agent utterances as the targets to be generated and the previous customer utterance and the agent's dialog act as the input. We additionally reformat the dialog acts to directly conform to the format described in this paper.", "#### Additional Splits?\n\n\nyes", "#### Split Information\n\n\n9 challenge sets for Schema-Guided Dialog were added to the GEM evaluation suite.\n\n\n1. We created subsets of the training and development sets of 500 randomly selected inputs each.\n2. We applied 5 transformations to respectively 5 sets of 500 randomly selected inputs: (i) back-translation, (ii)-(iii) introduction of typographical errors, using Butterfingers with two thresholds (0.02 and 0.05), resulting in two sets with different amounts of typos introduced (there are more typos with the 0.05 threshold than with the 0.02 one), (iv) removal of final punctuations (when any), and (v) input scrambling, for which the order of the dialogue acts was randomly reassigned.\n3. For the input size, we created subpopulations based on the number of dialogue acts in the input.\n\n\n\nWe also split the test data according to the type of dialogue act, represented by cardinal numbers in the dataset.", "#### Split Motivation\n\n\nGeneralization and Robustness.", "### Getting Started with the Task", "#### Pointers to Resources\n\n\n* Paper for dataset and DST baseline\n* DSTC8 overview paper\n* Code for DST baseline\n* Natural language generation baseline paper\n* Blog post announcing the dataset\n\n\nPrevious Results\n----------------", "### Previous Results", "#### Measured Model Abilities\n\n\nSurface realization and compositionally.", "#### Metrics\n\n\n'BLEURT', 'BLEU', 'ROUGE'", "#### Proposed Evaluation\n\n\nThe original paper focused on the task of dialog state prediction instead of response generation and thus did not suggest any set of metrics.", "#### Previous results available?\n\n\nno\n\n\nDataset Curation\n----------------", "### Original Curation", "#### Original Curation Rationale\n\n\nPrevious multi-domain task-oriented dialogue datsets do not sufficiently capture the real-world challenges in virtual assistants, since they cover few domains and assume a single static ontology per domain.\nThe SGD datset is created to cover 17 domains with over 16K dialogues, and contain multiple different APIs in most domains, many of which have overlapping functionalities but different interfaces, which reflects common real-world scenarios.\nThe wide range of available annotations can be used for intent prediction, slot filling, dialogue state tracking, policy imitation learning, language generation, user simulation learning, among other tasks in large-scale virtual assistants.", "#### Communicative Goal\n\n\nThe goal of a speaker who generates the target utterance is to help users accomplish tasks including but not limited to finding flights, booking restaurants, searching for nearby events and movies.", "#### Sourced from Different Sources\n\n\nno", "### Language Data", "#### How was Language Data Obtained?\n\n\n'Machine-generated'", "#### Generation Method Link\n\n\nGithub", "#### Language Producers\n\n\nThe dialogue outlines are first generated by a simulator. The dialogue simulator interacts with the services to generate dialogue outlines. It consists of two agents playing the roles of the user and the system, interacting with each other using a finite set of actions specified through dialogue acts over a probabilistic automaton designed to capture varied dialogue trajectories. It is worth noting that the simulation automaton does not include any domain-specific constraints: all domain-specific constraints are encoded in the schema and scenario.\n\n\nThe dialogue paraphrasing framework then converts the outlines generated by the simulator into a natural conversation. Users may refer to the slot values in the dialogue acts in various different ways during the conversation, e.g., “los angeles” may be referred to as “LA” or “LAX”. To introduce these natural variations in the slot values, different slot values are replaced with a randomly selected variation while being kept consistent across user turns in a dialogue. The actions are then converted to pseudo-natural language utterances using a set of manually defined action-to-text templates, and the resulting utterances for the different actions in a turn are concatenated together.", "#### Topics Covered\n\n\nThe dataset covers the following domains: Alarm, Banks, Buses, Calendar, Events, Flights, Homes, Hotels, Media, Messaging, Movies, Music, Payment, RentalCars, Restaurants, RideSharing, Services, Train, Travel, and Weather. The domain ‘Service’ includes salons, dentists, doctors etc. The ‘Alarm’, ‘Messaging’, ‘Payment’ and ‘Train’ domains are only present in the dev or test sets.\nto test generalization to new domains.", "#### Data Validation\n\n\nnot validated", "#### Was Data Filtered?\n\n\nnot filtered", "### Structured Annotations", "#### Additional Annotations?\n\n\ncrowd-sourced", "#### Number of Raters\n\n\nunknown", "#### Raters per Training Example\n\n\n0", "#### Raters per Test Example\n\n\n0", "#### Annotation Service?\n\n\nunknown", "#### Annotation Values\n\n\nThe dialogue transformed by these steps is sent to the crowd workers to be reformulated into more natural language. One crowd worker is tasked with paraphrasing all utterances of a dialogue to ensure naturalness and coherence. The crowd workers are asked to exactly repeat the slot values in their paraphrases so that the span indices for the slots can be recovered via string matching.", "#### Any Quality Control?\n\n\nnone", "### Consent", "#### Any Consent Policy?\n\n\nno", "#### Justification for Using the Data\n\n\nWhile no policy is reported, we assume that one was in place for the collection.", "### Private Identifying Information (PII)", "#### Contains PII?\n\n\nno PII", "#### Justification for no PII\n\n\nThe SGD dataset does not use identity categories and does not contain sensitive data.", "### Maintenance", "#### Any Maintenance Plan?\n\n\nno\n\n\nBroader Social Context\n----------------------", "### Previous Work on the Social Impact of the Dataset", "#### Usage of Models based on the Data\n\n\nno", "### Impact on Under-Served Communities", "#### Addresses needs of underserved Communities?\n\n\nno", "### Discussion of Biases", "#### Any Documented Social Biases?\n\n\nno", "#### Are the Language Producers Representative of the Language?\n\n\nDue to the combination of the automatic generation and crowd rater paraphasing, the language can be very formulaic. While this may be acceptable for the model part (i.e., we may actually desire an automated agent to form formulaic responses), the input utterances of the simulated customers likely do not cover the entire spectrum of the English language.\n\n\nConsiderations for Using the Data\n---------------------------------", "### PII Risks and Liability", "### Licenses", "#### Copyright Restrictions on the Dataset\n\n\n'open license - commercial use allowed'", "#### Copyright Restrictions on the Language Data\n\n\n'open license - commercial use allowed'", "### Known Technical Limitations", "#### Technical Limitations\n\n\nThe dialogues under each domain distributed unevenly, where the flights domain has 3644 dialogues while the payment domain only contains 222 dialogues.\nBesides, all dialogues are paraphrased by crowd-workers, and it is possible that crow-workers with different culture backgrounds will exhibit biased opinions.", "#### Unsuited Applications\n\n\nSince the initial data was automatically generated, the coverage of entity names is necessarily biased. An agent thus needs to be evaluated in a more realistic environment." ]

[ "TAGS\n#task_categories-conversational #annotations_creators-crowd-sourced #language_creators-unknown #multilinguality-unknown #size_categories-unknown #source_datasets-original #language-English #license-cc-by-sa-4.0 #dialog-response-generation #arxiv-1909.05855 #arxiv-2004.15006 #arxiv-2002.01359 #region-us \n", "### Link to Main Data Card\n\n\nYou can find the main data card on the [GEM Website.", "### Dataset Summary\n\n\nThe GEM version of this dataset functions as a response generation dataset. The input specifies dialog acts that a model needs to verbalize. The Schema-Guided Dialog dataset is challenging since it comprises multiple domains from hotel and travel to restaurants, and a wide range of dialog acts. The context of each conversation is provided as well.\n\n\nYou can load the dataset via:\n\n\nThe data loader can be found here.", "#### website\n\n\nn/a", "#### paper\n\n\nArxiv", "#### authors\n\n\nAbhinav Rastogi, Xiaoxue Zang, Srinivas Sunkara, Raghav Gupta, Pranav Khaitan, Amir Fayazi, Maria Wang, and Guan-Lin Chao\n\n\nDataset Overview\n----------------", "### Where to find the Data and its Documentation", "#### Download\n\n\nGithub[(URL", "#### Paper\n\n\n[Arxiv", "#### BibTex", "#### Contact Name\n\n\nAbhinav Rastogi", "#### Contact Email\n\n\nschema-guided-dst@URL", "#### Has a Leaderboard?\n\n\nno", "### Languages and Intended Use", "#### Multilingual?\n\n\nno", "#### Covered Languages\n\n\n'English'", "#### Whose Language?\n\n\nThe language structure is machine-generated, and the language realizations are produced by crowd workers.\nThe dataset paper does not provide demographic information for the crowd workers.", "#### License\n\n\ncc-by-sa-4.0: Creative Commons Attribution Share Alike 4.0 International", "#### Intended Use\n\n\nThe Schema-Guided Dialogue (SGD) dataset contains 18K multi-domain task-oriented dialogues between a human and a virtual assistant, which covers 17 domains ranging from banks and events to media, calendar, travel, and weather.\nThe language presents in the datset is only English.\nThe SGD dataset provides a challenging testbed for a number of tasks in task-oriented dialogue, including language understanding, slot filling, dialogue state tracking and response generation.\nFor the creation of the SGD dataset, they developed a multi-domain dialogue simulator that generates dialogue outlines over an arbitrary combination of APIs, dialogue states and system actions. Then, they used a crowd-sourcing procedure to paraphrase these outlines to natural language utterances.\nThis novel crowd-sourcing procedure preserves all annotations obtained from the simulator and does not require any extra annotations after dialogue collection.", "#### Primary Task\n\n\nDialog Response Generation", "#### Communicative Goal\n\n\nThe goal of a speaker who generates the target utterance is to help users accomplish tasks including but not limited to finding flights, booking restaurants, searching for nearby events and movies.", "### Credit", "#### Curation Organization Type(s)\n\n\n'industry'", "#### Curation Organization(s)\n\n\nGoogle", "#### Dataset Creators\n\n\nAbhinav Rastogi, Xiaoxue Zang, Srinivas Sunkara, Raghav Gupta, Pranav Khaitan, Amir Fayazi, Maria Wang, and Guan-Lin Chao", "#### Funding\n\n\nGoogle", "#### Who added the Dataset to GEM?\n\n\nWanyu Du wrote the initial data card and Yacine Jernite the data loader. Simon Mille updated the data card with the additional splits. Sebastian Gehrmann migrated the data card and loader to the v2 version and extended the missing information.", "### Dataset Structure", "#### Data Fields\n\n\nEach dialog instance has the following fields:\n\n\n* 'dialogue\\_id': A unique identifier for a dialogue.\n* 'services': A list of services present in the dialogue.\n* 'turns': A list of annotated system or user utterances. Each turn consists of the following fields:\n\t+ 'speaker': The speaker for the turn, either 'USER' or 'SYSTEM'.\n\t+ 'utterance': A string containing the natural language utterance.\n\t+ 'frames': A list of frames, each frame containing annotations for a single service and consists of the following fields:\n\t\t- 'service': The name of the service corresponding to the frame. The slots and intents used in the following fields are taken from the schema of this service.\n\t\t- 'slots': A list of slot spans in the utterance, only provided for non-categorical slots. Each slot span contains the following fields:\n\t\t\t* 'slot': The name of the slot.\n\t\t\t* 'start': The index of the starting character in the utterance corresponding to the slot value.\n\t\t\t* 'exclusive\\_end': The index of the character just after the last character corresponding to the slot value in the utterance.\n\t\t- 'actions': A list of actions corresponding to the system. Each action has the following fields:\n\t\t\t* 'act': The type of action.\n\t\t\t* 'slot': (optional) A slot argument for some of the actions.\n\t\t\t* 'values': (optional) A list of values assigned to the slot. If the values list is non-empty, then the slot must be present.\n\t\t\t* 'canonical\\_values': (optional) The values in their canonicalized form as used by the service. It is a list of strings of the same length as values.\n\t\t- 'service\\_call': (system turns only, optional) The request sent to the service. It consists of the following fields:\n\t\t\t* 'method': The name of the intent or function of the service or API being executed.\n\t\t\t* 'parameters': A pair of lists of the same lengths: 'parameter\\_slot\\_name' contains slot names and 'parameter\\_canonical\\_value' contains the corresponding values in their canonicalized form.\n\t\t- 'service\\_results': (system turns only, optional) A list of entities containing the results obtained from the service. It is only available for turns in which a service call is made. Each entity is represented as a pair of lists of the same length: 'service\\_slot\\_name' contains slot names and 'service\\_canonical\\_value' contains the corresponding canonical values.\n\t\t- 'state': (user turns only) The dialogue state corresponding to the service. It consists of the following fields:\n\t\t\t* 'active\\_intent': The intent corresponding to the service of the frame which is currently being fulfilled by the system. It takes the value \"NONE\" if none of the intents are active.\n\t\t\t* 'requested\\_slots': A list of slots requested by the user in the current turn.\n\t\t\t* 'slot\\_values': A pair of lists of the same lengths: 'slot\\_name' contains slot names and 'slot\\_value\\_list' contains the corresponding lists of strings. For categorical slots, this list contains a single value assigned to the slot. For non-categorical slots, all the values in this list are spoken variations of each other and are equivalent (e.g, \"6 pm\", \"six in the evening\", \"evening at 6\" etc.).", "#### Example Instance", "#### Data Splits\n\n\nThe dataset is split into a train, validation, and test set with the following sizes:", "#### Splitting Criteria\n\n\nThe data is generally split i.i.d, but some topics only appear in training and some only for testing. For example, the domains Messaging, Payment, and Train are test-only.\n\n\nDataset in GEM\n--------------", "### Rationale for Inclusion in GEM", "#### Why is the Dataset in GEM?\n\n\nThis dataset comprises a wide range of dialog capabilities and thus enables the evaluation of many more generation capabilities of comparable datasets. Its collection methodology ensures a high diversity but also high quality of the data.", "#### Similar Datasets\n\n\nyes", "#### Unique Language Coverage\n\n\nno", "#### Difference from other GEM datasets\n\n\nThe domains a lot more diverse than other datasets.", "#### Ability that the Dataset measures\n\n\nsurface realization, compositionality.", "### GEM-Specific Curation", "#### Modificatied for GEM?\n\n\nyes", "#### GEM Modifications\n\n\n'data points modified'", "#### Modification Details\n\n\nWe are focusing on the response-generation part of the dataset and thus reformatted the dataset to treat the service agent utterances as the targets to be generated and the previous customer utterance and the agent's dialog act as the input. We additionally reformat the dialog acts to directly conform to the format described in this paper.", "#### Additional Splits?\n\n\nyes", "#### Split Information\n\n\n9 challenge sets for Schema-Guided Dialog were added to the GEM evaluation suite.\n\n\n1. We created subsets of the training and development sets of 500 randomly selected inputs each.\n2. We applied 5 transformations to respectively 5 sets of 500 randomly selected inputs: (i) back-translation, (ii)-(iii) introduction of typographical errors, using Butterfingers with two thresholds (0.02 and 0.05), resulting in two sets with different amounts of typos introduced (there are more typos with the 0.05 threshold than with the 0.02 one), (iv) removal of final punctuations (when any), and (v) input scrambling, for which the order of the dialogue acts was randomly reassigned.\n3. For the input size, we created subpopulations based on the number of dialogue acts in the input.\n\n\n\nWe also split the test data according to the type of dialogue act, represented by cardinal numbers in the dataset.", "#### Split Motivation\n\n\nGeneralization and Robustness.", "### Getting Started with the Task", "#### Pointers to Resources\n\n\n* Paper for dataset and DST baseline\n* DSTC8 overview paper\n* Code for DST baseline\n* Natural language generation baseline paper\n* Blog post announcing the dataset\n\n\nPrevious Results\n----------------", "### Previous Results", "#### Measured Model Abilities\n\n\nSurface realization and compositionally.", "#### Metrics\n\n\n'BLEURT', 'BLEU', 'ROUGE'", "#### Proposed Evaluation\n\n\nThe original paper focused on the task of dialog state prediction instead of response generation and thus did not suggest any set of metrics.", "#### Previous results available?\n\n\nno\n\n\nDataset Curation\n----------------", "### Original Curation", "#### Original Curation Rationale\n\n\nPrevious multi-domain task-oriented dialogue datsets do not sufficiently capture the real-world challenges in virtual assistants, since they cover few domains and assume a single static ontology per domain.\nThe SGD datset is created to cover 17 domains with over 16K dialogues, and contain multiple different APIs in most domains, many of which have overlapping functionalities but different interfaces, which reflects common real-world scenarios.\nThe wide range of available annotations can be used for intent prediction, slot filling, dialogue state tracking, policy imitation learning, language generation, user simulation learning, among other tasks in large-scale virtual assistants.", "#### Communicative Goal\n\n\nThe goal of a speaker who generates the target utterance is to help users accomplish tasks including but not limited to finding flights, booking restaurants, searching for nearby events and movies.", "#### Sourced from Different Sources\n\n\nno", "### Language Data", "#### How was Language Data Obtained?\n\n\n'Machine-generated'", "#### Generation Method Link\n\n\nGithub", "#### Language Producers\n\n\nThe dialogue outlines are first generated by a simulator. The dialogue simulator interacts with the services to generate dialogue outlines. It consists of two agents playing the roles of the user and the system, interacting with each other using a finite set of actions specified through dialogue acts over a probabilistic automaton designed to capture varied dialogue trajectories. It is worth noting that the simulation automaton does not include any domain-specific constraints: all domain-specific constraints are encoded in the schema and scenario.\n\n\nThe dialogue paraphrasing framework then converts the outlines generated by the simulator into a natural conversation. Users may refer to the slot values in the dialogue acts in various different ways during the conversation, e.g., “los angeles” may be referred to as “LA” or “LAX”. To introduce these natural variations in the slot values, different slot values are replaced with a randomly selected variation while being kept consistent across user turns in a dialogue. The actions are then converted to pseudo-natural language utterances using a set of manually defined action-to-text templates, and the resulting utterances for the different actions in a turn are concatenated together.", "#### Topics Covered\n\n\nThe dataset covers the following domains: Alarm, Banks, Buses, Calendar, Events, Flights, Homes, Hotels, Media, Messaging, Movies, Music, Payment, RentalCars, Restaurants, RideSharing, Services, Train, Travel, and Weather. The domain ‘Service’ includes salons, dentists, doctors etc. The ‘Alarm’, ‘Messaging’, ‘Payment’ and ‘Train’ domains are only present in the dev or test sets.\nto test generalization to new domains.", "#### Data Validation\n\n\nnot validated", "#### Was Data Filtered?\n\n\nnot filtered", "### Structured Annotations", "#### Additional Annotations?\n\n\ncrowd-sourced", "#### Number of Raters\n\n\nunknown", "#### Raters per Training Example\n\n\n0", "#### Raters per Test Example\n\n\n0", "#### Annotation Service?\n\n\nunknown", "#### Annotation Values\n\n\nThe dialogue transformed by these steps is sent to the crowd workers to be reformulated into more natural language. One crowd worker is tasked with paraphrasing all utterances of a dialogue to ensure naturalness and coherence. The crowd workers are asked to exactly repeat the slot values in their paraphrases so that the span indices for the slots can be recovered via string matching.", "#### Any Quality Control?\n\n\nnone", "### Consent", "#### Any Consent Policy?\n\n\nno", "#### Justification for Using the Data\n\n\nWhile no policy is reported, we assume that one was in place for the collection.", "### Private Identifying Information (PII)", "#### Contains PII?\n\n\nno PII", "#### Justification for no PII\n\n\nThe SGD dataset does not use identity categories and does not contain sensitive data.", "### Maintenance", "#### Any Maintenance Plan?\n\n\nno\n\n\nBroader Social Context\n----------------------", "### Previous Work on the Social Impact of the Dataset", "#### Usage of Models based on the Data\n\n\nno", "### Impact on Under-Served Communities", "#### Addresses needs of underserved Communities?\n\n\nno", "### Discussion of Biases", "#### Any Documented Social Biases?\n\n\nno", "#### Are the Language Producers Representative of the Language?\n\n\nDue to the combination of the automatic generation and crowd rater paraphasing, the language can be very formulaic. While this may be acceptable for the model part (i.e., we may actually desire an automated agent to form formulaic responses), the input utterances of the simulated customers likely do not cover the entire spectrum of the English language.\n\n\nConsiderations for Using the Data\n---------------------------------", "### PII Risks and Liability", "### Licenses", "#### Copyright Restrictions on the Dataset\n\n\n'open license - commercial use allowed'", "#### Copyright Restrictions on the Language Data\n\n\n'open license - commercial use allowed'", "### Known Technical Limitations", "#### Technical Limitations\n\n\nThe dialogues under each domain distributed unevenly, where the flights domain has 3644 dialogues while the payment domain only contains 222 dialogues.\nBesides, all dialogues are paraphrased by crowd-workers, and it is possible that crow-workers with different culture backgrounds will exhibit biased opinions.", "#### Unsuited Applications\n\n\nSince the initial data was automatically generated, the coverage of entity names is necessarily biased. An agent thus needs to be evaluated in a more realistic environment." ]

[ "task_categories:table-to-text", "annotations_creators:none", "language_creators:unknown", "multilinguality:unknown", "size_categories:unknown", "source_datasets:original", "language:en", "license:mit", "data-to-text", "region:us" ]

[]

[ "en" ]

[ "# Dataset Card for GEM/sportsett_basketball", "## Dataset Description\n\n- Homepage: URL\n- Repository: URL\n- Paper: URL\n- Leaderboard: N/A\n- Point of Contact: Craig Thomson", "### Link to Main Data Card\n\nYou can find the main data card on the GEM Website.", "### Dataset Summary \n\nThe sportsett dataset is an English data-to-text dataset in the basketball domain. The inputs are statistics summarizing an NBA game and the outputs are high-quality descriptions of the game in natural language. \n\nYou can load the dataset via:\n\nThe data loader can be found here.", "#### website\nGithub", "#### paper\nACL Anthology", "#### authors\nCraig Thomson, Ashish Upadhyay", "## Dataset Overview", "### Where to find the Data and its Documentation", "#### Webpage\n\n\n\nGithub", "#### Download\n\n\n\nGithub", "#### Paper\n\n\n\nACL Anthology", "#### BibTex", "#### Contact Name\n\n\n\n\nCraig Thomson", "#### Contact Email\n\n\n\nc.thomson@URL", "#### Has a Leaderboard?\n\n\n\nno", "### Languages and Intended Use", "#### Multilingual?\n\n\n\n\nno", "#### Covered Dialects\n\n\n\nAmerican English\n\nOne dialect, one language.", "#### Covered Languages\n\n\n\n\n'English'", "#### Whose Language?\n\n\n\nAmerican sports writers", "#### License\n\n\n\n\nmit: MIT License", "#### Intended Use\n\n\n\nMaintain a robust and scalable Data-to-Text generation resource with structured data and textual summaries", "#### Primary Task\n\n\n\nData-to-Text", "#### Communicative Goal\n\n\n\n\nA model trained on this dataset should summarise the statistical and other information from a basketball game. This will be focused on a single game, although facts from prior games, or aggregate statistics over many games can and should be used for comparison where appropriate. There no single common narrative, although summaries usually start with who player, when, where, and the score. They then provide high level commentary on what the difference in the game was (why the winner won). breakdowns of statistics for prominent players follow, winning team first. Finally, the upcoming schedule for both teams is usually included. There are, however, other types of fact that can be included, and other narrative structures.", "### Credit", "#### Curation Organization Type(s)\n\n\n\n'academic'", "#### Curation Organization(s)\n\n\n\nUniversity of Aberdeen, Robert Gordon University", "#### Dataset Creators\n\n\n\nCraig Thomson, Ashish Upadhyay", "#### Funding\n\n\n\nEPSRC", "#### Who added the Dataset to GEM?\n\n\n\nCraig Thomson, Ashish Upadhyay", "### Dataset Structure", "#### Data Fields\n\n\n\nEach instance in the dataset has five fields. \n\n1. \"sportsett_id\": This is a unique id as used in the original SportSett database. It starts with '1' with the first instance in the train-set and ends with '6150' with the last instance in test-set.\n\n2. \"gem_id\": This is a unique id created as per GEM's requirement which follows the 'GEM-${DATASET_NAME}-${SPLIT-NAME}-${id}' pattern.\n\n3. \"game\": This field contains a dictionary with information about current game. It has information such as date on which the game was played alongwith the stadium, city, state where it was played.\n\n4. \"teams\": This filed is a dictionary of multiple nested dictionaries. On the highest level, it has two keys: 'home' and 'vis', which provide the stats for home team and visiting team of the game. Both are dictionaries with same structure. Each dictionary will contain team's information such as name of the team, their total wins/losses in current season, their conference standing, the SportSett ids for their current and previous games. Apart from these general information, they also have the box- and line- scores for the team in the game. Box score is the stats of players from the team at the end of the game, while line score along with the whole game stats is divided into quarters and halves as well as the extra-time (if happened in the game). After these scores, there is another field of next-game, which gives general information about team's next game such as the place and opponent's name of the next game.\n\n5. \"summaries\": This is a list of summaries for each game. Some games will have more than one summary, in that case, the list will have more than one entries. Each summary in the list is a string which can be tokenised by a space, following the practices in RotoWire-FG dataset (Wang, 2019).", "#### Reason for Structure\n\n\n\nThe structure mostly follows the original structure defined in RotoWire dataset (Wiseman et. al. 2017) with some modifications (such as game and next-game keys) address the problem of information gap between input and output data (Thomson et. al. 2020).", "#### How were labels chosen?\n\n\n\nSimilar to RotoWire dataset (Wiseman et. al. 2017)", "#### Example Instance", "#### Data Splits\n\n\n\n- Train: NBA seasons - 2014, 2015, & 2016; total instances - 3690\n- Validation: NBA seasons - 2017; total instances - 1230\n- Test: NBA seasons - 2018; total instances - 1230", "#### Splitting Criteria\n\n\n\nThe splits were created as per different NBA seasons. All the games in regular season (no play-offs) are added in the dataset", "## Dataset in GEM", "### Rationale for Inclusion in GEM", "#### Why is the Dataset in GEM?\n\n\n\nThis dataset contains a data analytics problem in the classic sense (Reiter, 2007). That is, there is a large amount of data from which insights need to be selected. Further, the insights should be both from simple shallow queries (such as dirext transcriptions of the properties of a subject, i.e., a player and their statistics), as well as aggregated (how a player has done over time). There is far more on the data side than is required to be realised, and indeed, could be practically realised. This depth of data analytics problem does not exist in other datasets.", "#### Similar Datasets\n\n\n\nno", "#### Ability that the Dataset measures\n\n\n\nMany, if not all aspects of data-to-text systems can be measured with this dataset. It has complex data analytics, meaninful document planning (10-15 sentence documents with a narrative structure), as well as microplanning and realisation requirements. Finding models to handle this volume of data, as well as methods for meaninfully evaluate generations is a very open question.", "### GEM-Specific Curation", "#### Modificatied for GEM?\n\n\n\nno", "#### Additional Splits?\n\n\n\nno", "### Getting Started with the Task", "#### Pointers to Resources\n\n\n\nFor dataset discussion see Thomson et al, 2020\n\nFor evaluation see:\n- Thomson & Reiter 2020, Thomson & Reiter (2021)\n- Kasner et al (2021)\n\nFor a system using the relational database form of SportSett, see:\n- Thomson et al (2020)\n\nFor recent systems using the Rotowire dataset, see:\n- Puduppully & Lapata (2021)\n- Rebuffel et all (2020)", "## Previous Results", "### Previous Results", "#### Measured Model Abilities\n\n\n\nMany, if not all aspects of data-to-text systems can be measured with this dataset. It has complex data analytics, meaninful document planning (10-15 sentence documents with a narrative structure), as well as microplanning and realisation requirements. Finding models to handle this volume of data, as well as methods for meaninfully evaluate generations is a very open question.", "#### Metrics\n\n\n\n'BLEU'", "#### Proposed Evaluation\n\n\n\nBLEU is the only off-the-shelf metric commonly used. Works have also used custom metrics like RG (Wiseman et al, 2017), and a recent shared task explored other metrics and their corrolation with human evaluation (Thomson & Reiter, 2021).", "#### Previous results available?\n\n\n\nyes", "#### Other Evaluation Approaches\n\n\n\nMost results from prior works use the original Rotowire dataset, which has train/validation/test contamination. For results of BLEU and RG on the relational database format of SportSett, as a guide, see Thomson et al, 2020.", "#### Relevant Previous Results\n\n\n\nThe results on this dataset are largely unexplored, as is the selection of suitable metrics that correlate with human judgment. See Thomson et al, 2021 (URL for an overview, and Kasner et al (2021) for the best performing metric at the time of writing (URL", "## Dataset Curation", "### Original Curation", "#### Original Curation Rationale\n\n\n\nThe references texts were taken from the existing dataset RotoWire-FG (Wang, 2019), which is in turn based on Rotowire (Wiseman et al, 2017). The rationale behind this dataset was to re-structure the data such that aggregate statistics over multiple games, as well as upcoming game schedules could be included, moving the dataset from snapshots of single games, to a format where almost everything that could be present in the reference texts could be found in the data.", "#### Communicative Goal\n\n\n\nCreate a summary of a basketball, with insightful facts about the game, teams, and players, both within the game, withing periods during the game, and over the course of seasons/careers where appropriate. This is a data-to-text problem in the classic sense (Reiter, 2007) in that it has a difficult data analystics state, in addition to ordering and transcription of selected facts.", "#### Sourced from Different Sources\n\n\n\nyes", "#### Source Details\n\n\n\nRotoWire-FG (URL).\nWikipedia (URL\nBasketball Reference (URL)", "### Language Data", "#### How was Language Data Obtained?\n\n\n\n'Found'", "#### Where was it found?\n\n\n\n'Multiple websites'", "#### Language Producers\n\n\n\nNone", "#### Topics Covered\n\n\n\nSummaries of basketball games (in the NBA).", "#### Data Validation\n\n\n\nnot validated", "#### Data Preprocessing\n\n\n\nIt retains the original tokenization scheme employed by Wang 2019", "#### Was Data Filtered?\n\n\n\nmanually", "#### Filter Criteria\n\n\n\nGames from the 2014 through 2018 seasons were selected. Within these seasons games are not filtered, all are present, but this was an arbitrary solution from the original RotoWirte-FG dataset.", "### Structured Annotations", "#### Additional Annotations?\n\n\n\n\nnone", "#### Annotation Service?\n\n\n\nno", "### Consent", "#### Any Consent Policy?\n\n\n\nno", "#### Justification for Using the Data\n\n\n\nThe dataset consits of a pre-existing dataset, as well as publically available facts.", "### Private Identifying Information (PII)", "#### Contains PII?\n\n\n\n\nunlikely", "#### Categories of PII\n\n\n\n'generic PII'", "#### Any PII Identification?\n\n\n\nno identification", "### Maintenance", "#### Any Maintenance Plan?\n\n\n\nno", "## Broader Social Context", "### Previous Work on the Social Impact of the Dataset", "#### Usage of Models based on the Data\n\n\n\nno", "### Impact on Under-Served Communities", "#### Addresses needs of underserved Communities?\n\n\n\nno", "### Discussion of Biases", "#### Any Documented Social Biases?\n\n\n\nyes", "#### Links and Summaries of Analysis Work\n\n\n\nUnaware of any work, but, this is a dataset considting solely of summaries of mens professional basketball games. It does not cover different levels of the sport, or different genders, and all pronouns are likely to be male unless a specific player is referred to by other pronouns in the training text. This makes it difficult to train systems where gender can be specified as an attribute, although it is an interesting, open problem that could be investigated using the dataset.", "#### Are the Language Producers Representative of the Language?\n\n\n\nNo, it is very specifically American English from the sports journalism domain.", "## Considerations for Using the Data", "### PII Risks and Liability", "#### Potential PII Risk\n\n\n\nAll information relating to persons is of public record.", "### Licenses", "#### Copyright Restrictions on the Dataset\n\n\n\n'public domain'", "#### Copyright Restrictions on the Language Data\n\n\n\n'public domain'", "### Known Technical Limitations", "#### Technical Limitations\n\n\n\nSportSett resolved the major overlap problems of RotoWire, although some overlap is unavoidable. For example, whilst it is not possible to find career totals and other historic information for all players (the data only goes back to 2014), it is possible to do so for some players. It is unavoidable that some data which is aggregated, exists in its base form in previous partitions. The season-based partition scheme heavily constrains this however.", "#### Unsuited Applications\n\n\n\nFactual accuray continues to be a problem, systems may incorrectly represent the facts of the game.", "#### Discouraged Use Cases\n\n\n\nUsing the RG metric to maximise the number of true facts in a generate summary is not nececeraly" ]

[ "TAGS\n#task_categories-table-to-text #annotations_creators-none #language_creators-unknown #multilinguality-unknown #size_categories-unknown #source_datasets-original #language-English #license-mit #data-to-text #region-us \n", "# Dataset Card for GEM/sportsett_basketball", "## Dataset Description\n\n- Homepage: URL\n- Repository: URL\n- Paper: URL\n- Leaderboard: N/A\n- Point of Contact: Craig Thomson", "### Link to Main Data Card\n\nYou can find the main data card on the GEM Website.", "### Dataset Summary \n\nThe sportsett dataset is an English data-to-text dataset in the basketball domain. The inputs are statistics summarizing an NBA game and the outputs are high-quality descriptions of the game in natural language. \n\nYou can load the dataset via:\n\nThe data loader can be found here.", "#### website\nGithub", "#### paper\nACL Anthology", "#### authors\nCraig Thomson, Ashish Upadhyay", "## Dataset Overview", "### Where to find the Data and its Documentation", "#### Webpage\n\n\n\nGithub", "#### Download\n\n\n\nGithub", "#### Paper\n\n\n\nACL Anthology", "#### BibTex", "#### Contact Name\n\n\n\n\nCraig Thomson", "#### Contact Email\n\n\n\nc.thomson@URL", "#### Has a Leaderboard?\n\n\n\nno", "### Languages and Intended Use", "#### Multilingual?\n\n\n\n\nno", "#### Covered Dialects\n\n\n\nAmerican English\n\nOne dialect, one language.", "#### Covered Languages\n\n\n\n\n'English'", "#### Whose Language?\n\n\n\nAmerican sports writers", "#### License\n\n\n\n\nmit: MIT License", "#### Intended Use\n\n\n\nMaintain a robust and scalable Data-to-Text generation resource with structured data and textual summaries", "#### Primary Task\n\n\n\nData-to-Text", "#### Communicative Goal\n\n\n\n\nA model trained on this dataset should summarise the statistical and other information from a basketball game. This will be focused on a single game, although facts from prior games, or aggregate statistics over many games can and should be used for comparison where appropriate. There no single common narrative, although summaries usually start with who player, when, where, and the score. They then provide high level commentary on what the difference in the game was (why the winner won). breakdowns of statistics for prominent players follow, winning team first. Finally, the upcoming schedule for both teams is usually included. There are, however, other types of fact that can be included, and other narrative structures.", "### Credit", "#### Curation Organization Type(s)\n\n\n\n'academic'", "#### Curation Organization(s)\n\n\n\nUniversity of Aberdeen, Robert Gordon University", "#### Dataset Creators\n\n\n\nCraig Thomson, Ashish Upadhyay", "#### Funding\n\n\n\nEPSRC", "#### Who added the Dataset to GEM?\n\n\n\nCraig Thomson, Ashish Upadhyay", "### Dataset Structure", "#### Data Fields\n\n\n\nEach instance in the dataset has five fields. \n\n1. \"sportsett_id\": This is a unique id as used in the original SportSett database. It starts with '1' with the first instance in the train-set and ends with '6150' with the last instance in test-set.\n\n2. \"gem_id\": This is a unique id created as per GEM's requirement which follows the 'GEM-${DATASET_NAME}-${SPLIT-NAME}-${id}' pattern.\n\n3. \"game\": This field contains a dictionary with information about current game. It has information such as date on which the game was played alongwith the stadium, city, state where it was played.\n\n4. \"teams\": This filed is a dictionary of multiple nested dictionaries. On the highest level, it has two keys: 'home' and 'vis', which provide the stats for home team and visiting team of the game. Both are dictionaries with same structure. Each dictionary will contain team's information such as name of the team, their total wins/losses in current season, their conference standing, the SportSett ids for their current and previous games. Apart from these general information, they also have the box- and line- scores for the team in the game. Box score is the stats of players from the team at the end of the game, while line score along with the whole game stats is divided into quarters and halves as well as the extra-time (if happened in the game). After these scores, there is another field of next-game, which gives general information about team's next game such as the place and opponent's name of the next game.\n\n5. \"summaries\": This is a list of summaries for each game. Some games will have more than one summary, in that case, the list will have more than one entries. Each summary in the list is a string which can be tokenised by a space, following the practices in RotoWire-FG dataset (Wang, 2019).", "#### Reason for Structure\n\n\n\nThe structure mostly follows the original structure defined in RotoWire dataset (Wiseman et. al. 2017) with some modifications (such as game and next-game keys) address the problem of information gap between input and output data (Thomson et. al. 2020).", "#### How were labels chosen?\n\n\n\nSimilar to RotoWire dataset (Wiseman et. al. 2017)", "#### Example Instance", "#### Data Splits\n\n\n\n- Train: NBA seasons - 2014, 2015, & 2016; total instances - 3690\n- Validation: NBA seasons - 2017; total instances - 1230\n- Test: NBA seasons - 2018; total instances - 1230", "#### Splitting Criteria\n\n\n\nThe splits were created as per different NBA seasons. All the games in regular season (no play-offs) are added in the dataset", "## Dataset in GEM", "### Rationale for Inclusion in GEM", "#### Why is the Dataset in GEM?\n\n\n\nThis dataset contains a data analytics problem in the classic sense (Reiter, 2007). That is, there is a large amount of data from which insights need to be selected. Further, the insights should be both from simple shallow queries (such as dirext transcriptions of the properties of a subject, i.e., a player and their statistics), as well as aggregated (how a player has done over time). There is far more on the data side than is required to be realised, and indeed, could be practically realised. This depth of data analytics problem does not exist in other datasets.", "#### Similar Datasets\n\n\n\nno", "#### Ability that the Dataset measures\n\n\n\nMany, if not all aspects of data-to-text systems can be measured with this dataset. It has complex data analytics, meaninful document planning (10-15 sentence documents with a narrative structure), as well as microplanning and realisation requirements. Finding models to handle this volume of data, as well as methods for meaninfully evaluate generations is a very open question.", "### GEM-Specific Curation", "#### Modificatied for GEM?\n\n\n\nno", "#### Additional Splits?\n\n\n\nno", "### Getting Started with the Task", "#### Pointers to Resources\n\n\n\nFor dataset discussion see Thomson et al, 2020\n\nFor evaluation see:\n- Thomson & Reiter 2020, Thomson & Reiter (2021)\n- Kasner et al (2021)\n\nFor a system using the relational database form of SportSett, see:\n- Thomson et al (2020)\n\nFor recent systems using the Rotowire dataset, see:\n- Puduppully & Lapata (2021)\n- Rebuffel et all (2020)", "## Previous Results", "### Previous Results", "#### Measured Model Abilities\n\n\n\nMany, if not all aspects of data-to-text systems can be measured with this dataset. It has complex data analytics, meaninful document planning (10-15 sentence documents with a narrative structure), as well as microplanning and realisation requirements. Finding models to handle this volume of data, as well as methods for meaninfully evaluate generations is a very open question.", "#### Metrics\n\n\n\n'BLEU'", "#### Proposed Evaluation\n\n\n\nBLEU is the only off-the-shelf metric commonly used. Works have also used custom metrics like RG (Wiseman et al, 2017), and a recent shared task explored other metrics and their corrolation with human evaluation (Thomson & Reiter, 2021).", "#### Previous results available?\n\n\n\nyes", "#### Other Evaluation Approaches\n\n\n\nMost results from prior works use the original Rotowire dataset, which has train/validation/test contamination. For results of BLEU and RG on the relational database format of SportSett, as a guide, see Thomson et al, 2020.", "#### Relevant Previous Results\n\n\n\nThe results on this dataset are largely unexplored, as is the selection of suitable metrics that correlate with human judgment. See Thomson et al, 2021 (URL for an overview, and Kasner et al (2021) for the best performing metric at the time of writing (URL", "## Dataset Curation", "### Original Curation", "#### Original Curation Rationale\n\n\n\nThe references texts were taken from the existing dataset RotoWire-FG (Wang, 2019), which is in turn based on Rotowire (Wiseman et al, 2017). The rationale behind this dataset was to re-structure the data such that aggregate statistics over multiple games, as well as upcoming game schedules could be included, moving the dataset from snapshots of single games, to a format where almost everything that could be present in the reference texts could be found in the data.", "#### Communicative Goal\n\n\n\nCreate a summary of a basketball, with insightful facts about the game, teams, and players, both within the game, withing periods during the game, and over the course of seasons/careers where appropriate. This is a data-to-text problem in the classic sense (Reiter, 2007) in that it has a difficult data analystics state, in addition to ordering and transcription of selected facts.", "#### Sourced from Different Sources\n\n\n\nyes", "#### Source Details\n\n\n\nRotoWire-FG (URL).\nWikipedia (URL\nBasketball Reference (URL)", "### Language Data", "#### How was Language Data Obtained?\n\n\n\n'Found'", "#### Where was it found?\n\n\n\n'Multiple websites'", "#### Language Producers\n\n\n\nNone", "#### Topics Covered\n\n\n\nSummaries of basketball games (in the NBA).", "#### Data Validation\n\n\n\nnot validated", "#### Data Preprocessing\n\n\n\nIt retains the original tokenization scheme employed by Wang 2019", "#### Was Data Filtered?\n\n\n\nmanually", "#### Filter Criteria\n\n\n\nGames from the 2014 through 2018 seasons were selected. Within these seasons games are not filtered, all are present, but this was an arbitrary solution from the original RotoWirte-FG dataset.", "### Structured Annotations", "#### Additional Annotations?\n\n\n\n\nnone", "#### Annotation Service?\n\n\n\nno", "### Consent", "#### Any Consent Policy?\n\n\n\nno", "#### Justification for Using the Data\n\n\n\nThe dataset consits of a pre-existing dataset, as well as publically available facts.", "### Private Identifying Information (PII)", "#### Contains PII?\n\n\n\n\nunlikely", "#### Categories of PII\n\n\n\n'generic PII'", "#### Any PII Identification?\n\n\n\nno identification", "### Maintenance", "#### Any Maintenance Plan?\n\n\n\nno", "## Broader Social Context", "### Previous Work on the Social Impact of the Dataset", "#### Usage of Models based on the Data\n\n\n\nno", "### Impact on Under-Served Communities", "#### Addresses needs of underserved Communities?\n\n\n\nno", "### Discussion of Biases", "#### Any Documented Social Biases?\n\n\n\nyes", "#### Links and Summaries of Analysis Work\n\n\n\nUnaware of any work, but, this is a dataset considting solely of summaries of mens professional basketball games. It does not cover different levels of the sport, or different genders, and all pronouns are likely to be male unless a specific player is referred to by other pronouns in the training text. This makes it difficult to train systems where gender can be specified as an attribute, although it is an interesting, open problem that could be investigated using the dataset.", "#### Are the Language Producers Representative of the Language?\n\n\n\nNo, it is very specifically American English from the sports journalism domain.", "## Considerations for Using the Data", "### PII Risks and Liability", "#### Potential PII Risk\n\n\n\nAll information relating to persons is of public record.", "### Licenses", "#### Copyright Restrictions on the Dataset\n\n\n\n'public domain'", "#### Copyright Restrictions on the Language Data\n\n\n\n'public domain'", "### Known Technical Limitations", "#### Technical Limitations\n\n\n\nSportSett resolved the major overlap problems of RotoWire, although some overlap is unavoidable. For example, whilst it is not possible to find career totals and other historic information for all players (the data only goes back to 2014), it is possible to do so for some players. It is unavoidable that some data which is aggregated, exists in its base form in previous partitions. The season-based partition scheme heavily constrains this however.", "#### Unsuited Applications\n\n\n\nFactual accuray continues to be a problem, systems may incorrectly represent the facts of the game.", "#### Discouraged Use Cases\n\n\n\nUsing the RG metric to maximise the number of true facts in a generate summary is not nececeraly" ]

[ "task_categories:other", "annotations_creators:crowd-sourced", "language_creators:unknown", "multilinguality:unknown", "size_categories:unknown", "source_datasets:original", "language:en", "license:cc-by-sa-4.0", "question-generation", "arxiv:1806.03822", "region:us" ]

[ "1806.03822" ]

[ "en" ]

[ "### Link to Main Data Card\n\n\nYou can find the main data card on the GEM Website.", "### Dataset Summary\n\n\nSQuAD2.0 is a dataset that tests the ability of a system to not only answer reading comprehension questions, but also abstain when presented with a question that cannot be answered based on the provided paragraph. F1 score is used to evaluate models on the leaderboard. In GEM, we are using this dataset for the question-generation task in which a model should generate squad-like questions from an input text.\n\n\nYou can load the dataset via:\n\n\nThe data loader can be found here.", "#### website\n\n\nWebsite", "#### paper\n\n\nArxiv", "#### authors\n\n\nPranav Rajpurkar, Robin Jia and Percy Liang\n\n\nDataset Overview\n----------------", "### Where to find the Data and its Documentation", "#### Webpage\n\n\nWebsite", "#### Download\n\n\nWebsite", "#### Paper\n\n\nArxiv", "#### BibTex", "#### Contact Name\n\n\nRobin Jia", "#### Contact Email\n\n\nrobinjia@URL", "#### Has a Leaderboard?\n\n\nyes", "#### Leaderboard Link\n\n\nWebsite", "#### Leaderboard Details\n\n\nSQuAD2.0 tests the ability of a system to not only answer reading comprehension questions, but also abstain when presented with a question that cannot be answered based on the provided paragraph. F1 score is used to evaluate models on the leaderboard.", "### Languages and Intended Use", "#### Multilingual?\n\n\nno", "#### Covered Languages\n\n\n'English'", "#### License\n\n\ncc-by-sa-4.0: Creative Commons Attribution Share Alike 4.0 International", "#### Intended Use\n\n\nThe idea behind SQuAD2.0 dataset is to make the models understand when a question cannot be answered given a context. This will help in building models such that they know what they don't know, and therefore make the models understand language at a deeper level. The tasks that can be supported by the dataset are machine reading comprehension, extractive QA, and question generation.", "#### Primary Task\n\n\nQuestion Generation", "#### Communicative Goal\n\n\nGiven an input passage and an answer span, the goal is to generate a question that asks for the answer.", "### Credit", "#### Curation Organization Type(s)\n\n\n'academic'", "#### Curation Organization(s)\n\n\nStanford University", "#### Dataset Creators\n\n\nPranav Rajpurkar, Robin Jia and Percy Liang", "#### Funding\n\n\nFacebook and NSF Graduate Research Fellowship under Grant No. DGE-114747", "#### Who added the Dataset to GEM?\n\n\n(Abinaya Mahendiran)[URL Manager Data Science, NEXT Labs,", "### Dataset Structure", "#### Data Fields\n\n\nThe data fields are the same among all splits.", "#### squad\\_v2\n\n\n* 'id': a 'string' feature.\n* 'gem\\_id': a 'string' feature.\n* 'title': a 'string' feature.\n* 'context': a 'string' feature.\n* 'question': a 'string' feature.\n* 'answers': a dictionary feature containing:\n\t+ 'text': a 'string' feature.\n\t+ 'answer\\_start': a 'int32' feature.", "#### Example Instance\n\n\nHere is an example of a validation data point. This example was too long and was cropped:", "#### Data Splits\n\n\nThe original SQuAD2.0 dataset has only training and dev (validation) splits. The train split is further divided into test split and added as part of the GEM datasets.\n\n\n\nDataset in GEM\n--------------", "### Rationale for Inclusion in GEM", "#### Why is the Dataset in GEM?\n\n\nSQuAD2.0 will encourage the development of new reading comprehension models\nthat know what they don’t know, and therefore understand language at a deeper level. It can also help in building better models for answer-aware question generation .", "#### Similar Datasets\n\n\nno", "#### Unique Language Coverage\n\n\nyes", "#### Ability that the Dataset measures\n\n\nReasoning capability", "### GEM-Specific Curation", "#### Modificatied for GEM?\n\n\nyes", "#### GEM Modifications\n\n\n'other'", "#### Additional Splits?\n\n\nyes", "#### Split Information\n\n\nThe train(80%) and validation(10%) split of SQuAD2.0 are made available to public whereas the test(10%) split is not available.\n\n\nAs part of GEM, the train split, 80% of the original data is split into two train split (90%) and test split (remaining 10%). The idea is to provide all three splits for the users to use.", "### Getting Started with the Task\n\n\nPrevious Results\n----------------", "### Previous Results", "#### Measured Model Abilities\n\n\nExtractive QA, Question Generation", "#### Metrics\n\n\n'Other: Other Metrics', 'METEOR', 'ROUGE', 'BLEU'", "#### Other Metrics\n\n\n* Extractive QA uses Exact Match and F1 Score\n* Question generation users METEOR, ROUGE-L, BLEU-4", "#### Previous results available?\n\n\nyes", "#### Other Evaluation Approaches\n\n\nQuestion generation users METEOR, ROUGE-L, BLEU-4", "#### Relevant Previous Results\n\n\n@article{Dong2019UnifiedLM,\ntitle={Unified Language Model Pre-training for Natural Language Understanding and Generation},\nauthor={Li Dong and Nan Yang and Wenhui Wang and Furu Wei and Xiaodong Liu and Yu Wang and Jianfeng Gao and M. Zhou and Hsiao-Wuen Hon},\njournal={ArXiv},\nyear={2019},\nvolume={abs/1905.03197}\n}\n\n\nDataset Curation\n----------------", "### Original Curation", "#### Original Curation Rationale\n\n\nThe dataset is curated in three stages:\n\n\n* Curating passages,\n* Crowdsourcing question-answers on those passages,\n* Obtaining additional answers\nAs part of SQuAD1.1, 10000 high-quality articles from English Wikipedia is extracted using Project Nayuki’s Wikipedia’s internal PageRanks, from which 536 articles are sampled uniformly at random. From each of these articles, individual paragraphs are extracted, stripping away images, figures, tables, and discarding paragraphs shorter than 500 characters.\n\n\nSQuAD2.0 combines the 100,000 questions in SQuAD1.1 with over 50,000 unanswerable questions written adversarially by crowdworkers to look similar to answerable ones.", "#### Communicative Goal\n\n\nTo build systems that not only answer questions when possible, but also determine when no\nanswer is supported by the paragraph and abstain from answering.", "#### Sourced from Different Sources\n\n\nyes", "#### Source Details\n\n\nWikipedia", "### Language Data", "#### How was Language Data Obtained?\n\n\n'Found'", "#### Where was it found?\n\n\n'Single website'", "#### Topics Covered\n\n\nThe dataset contains 536 articles covering a wide range of topics, from musical celebrities to abstract concepts.", "#### Data Validation\n\n\nvalidated by crowdworker", "#### Data Preprocessing\n\n\nFrom the sampled articles from Wikipedia, individual paragraphs are extracted, stripping\naway images, figures, tables, and discarding paragraphs shorter than 500 characters and partitioned into training(80%), development set(10%) and test set(10%).", "#### Was Data Filtered?\n\n\nalgorithmically", "#### Filter Criteria\n\n\nTo retrieve high-quality articles, Project Nayuki’s Wikipedia’s internal PageRanks was used to obtain the top 10000 articles of English Wikipedia, from which 536 articles are sampled uniformly at random.", "### Structured Annotations", "#### Additional Annotations?\n\n\ncrowd-sourced", "#### Number of Raters\n\n\nunknown", "#### Rater Qualifications\n\n\nCrowdworkers from the United States or Canada with a 97% HIT acceptance rate, a minimum of 1000 HITs, were employed to create questions.", "#### Raters per Training Example\n\n\n0", "#### Raters per Test Example\n\n\n0", "#### Annotation Service?\n\n\nyes", "#### Which Annotation Service\n\n\n'other', 'Amazon Mechanical Turk'", "#### Annotation Values\n\n\nFor SQuAD 1.1 , crowdworkers were tasked with asking and answering up to 5 questions on the\ncontent of that paragraph. The questions had to be entered in a text field, and the answers had to be\nhighlighted in the paragraph.\n\n\nFor SQuAD2.0, each task consisted of an entire article from SQuAD 1.1. For each paragraph in the article, workers were asked to pose up to five questions that were impossible to answer\nbased on the paragraph alone, while referencing entities in the paragraph and ensuring that a plausible answer is present.", "#### Any Quality Control?\n\n\nvalidated by another rater", "#### Quality Control Details\n\n\nQuestions from workers who wrote 25 or fewer questions on an article is removed; this filter\nhelped remove noise from workers who had trouble understanding the task, and therefore quit before completing the whole article. This filter to both SQuAD2.0 and the existing answerable questions from SQuAD 1.1.", "### Consent", "#### Any Consent Policy?\n\n\nno", "### Private Identifying Information (PII)", "#### Contains PII?\n\n\nunlikely", "#### Any PII Identification?\n\n\nno identification", "### Maintenance", "#### Any Maintenance Plan?\n\n\nno\n\n\nBroader Social Context\n----------------------", "### Previous Work on the Social Impact of the Dataset", "#### Usage of Models based on the Data\n\n\nno", "### Impact on Under-Served Communities", "#### Addresses needs of underserved Communities?\n\n\nno", "### Discussion of Biases", "#### Any Documented Social Biases?\n\n\nyes\n\n\nConsiderations for Using the Data\n---------------------------------", "### PII Risks and Liability", "### Licenses", "### Known Technical Limitations" ]

[ "TAGS\n#task_categories-other #annotations_creators-crowd-sourced #language_creators-unknown #multilinguality-unknown #size_categories-unknown #source_datasets-original #language-English #license-cc-by-sa-4.0 #question-generation #arxiv-1806.03822 #region-us \n", "### Link to Main Data Card\n\n\nYou can find the main data card on the GEM Website.", "### Dataset Summary\n\n\nSQuAD2.0 is a dataset that tests the ability of a system to not only answer reading comprehension questions, but also abstain when presented with a question that cannot be answered based on the provided paragraph. F1 score is used to evaluate models on the leaderboard. In GEM, we are using this dataset for the question-generation task in which a model should generate squad-like questions from an input text.\n\n\nYou can load the dataset via:\n\n\nThe data loader can be found here.", "#### website\n\n\nWebsite", "#### paper\n\n\nArxiv", "#### authors\n\n\nPranav Rajpurkar, Robin Jia and Percy Liang\n\n\nDataset Overview\n----------------", "### Where to find the Data and its Documentation", "#### Webpage\n\n\nWebsite", "#### Download\n\n\nWebsite", "#### Paper\n\n\nArxiv", "#### BibTex", "#### Contact Name\n\n\nRobin Jia", "#### Contact Email\n\n\nrobinjia@URL", "#### Has a Leaderboard?\n\n\nyes", "#### Leaderboard Link\n\n\nWebsite", "#### Leaderboard Details\n\n\nSQuAD2.0 tests the ability of a system to not only answer reading comprehension questions, but also abstain when presented with a question that cannot be answered based on the provided paragraph. F1 score is used to evaluate models on the leaderboard.", "### Languages and Intended Use", "#### Multilingual?\n\n\nno", "#### Covered Languages\n\n\n'English'", "#### License\n\n\ncc-by-sa-4.0: Creative Commons Attribution Share Alike 4.0 International", "#### Intended Use\n\n\nThe idea behind SQuAD2.0 dataset is to make the models understand when a question cannot be answered given a context. This will help in building models such that they know what they don't know, and therefore make the models understand language at a deeper level. The tasks that can be supported by the dataset are machine reading comprehension, extractive QA, and question generation.", "#### Primary Task\n\n\nQuestion Generation", "#### Communicative Goal\n\n\nGiven an input passage and an answer span, the goal is to generate a question that asks for the answer.", "### Credit", "#### Curation Organization Type(s)\n\n\n'academic'", "#### Curation Organization(s)\n\n\nStanford University", "#### Dataset Creators\n\n\nPranav Rajpurkar, Robin Jia and Percy Liang", "#### Funding\n\n\nFacebook and NSF Graduate Research Fellowship under Grant No. DGE-114747", "#### Who added the Dataset to GEM?\n\n\n(Abinaya Mahendiran)[URL Manager Data Science, NEXT Labs,", "### Dataset Structure", "#### Data Fields\n\n\nThe data fields are the same among all splits.", "#### squad\\_v2\n\n\n* 'id': a 'string' feature.\n* 'gem\\_id': a 'string' feature.\n* 'title': a 'string' feature.\n* 'context': a 'string' feature.\n* 'question': a 'string' feature.\n* 'answers': a dictionary feature containing:\n\t+ 'text': a 'string' feature.\n\t+ 'answer\\_start': a 'int32' feature.", "#### Example Instance\n\n\nHere is an example of a validation data point. This example was too long and was cropped:", "#### Data Splits\n\n\nThe original SQuAD2.0 dataset has only training and dev (validation) splits. The train split is further divided into test split and added as part of the GEM datasets.\n\n\n\nDataset in GEM\n--------------", "### Rationale for Inclusion in GEM", "#### Why is the Dataset in GEM?\n\n\nSQuAD2.0 will encourage the development of new reading comprehension models\nthat know what they don’t know, and therefore understand language at a deeper level. It can also help in building better models for answer-aware question generation .", "#### Similar Datasets\n\n\nno", "#### Unique Language Coverage\n\n\nyes", "#### Ability that the Dataset measures\n\n\nReasoning capability", "### GEM-Specific Curation", "#### Modificatied for GEM?\n\n\nyes", "#### GEM Modifications\n\n\n'other'", "#### Additional Splits?\n\n\nyes", "#### Split Information\n\n\nThe train(80%) and validation(10%) split of SQuAD2.0 are made available to public whereas the test(10%) split is not available.\n\n\nAs part of GEM, the train split, 80% of the original data is split into two train split (90%) and test split (remaining 10%). The idea is to provide all three splits for the users to use.", "### Getting Started with the Task\n\n\nPrevious Results\n----------------", "### Previous Results", "#### Measured Model Abilities\n\n\nExtractive QA, Question Generation", "#### Metrics\n\n\n'Other: Other Metrics', 'METEOR', 'ROUGE', 'BLEU'", "#### Other Metrics\n\n\n* Extractive QA uses Exact Match and F1 Score\n* Question generation users METEOR, ROUGE-L, BLEU-4", "#### Previous results available?\n\n\nyes", "#### Other Evaluation Approaches\n\n\nQuestion generation users METEOR, ROUGE-L, BLEU-4", "#### Relevant Previous Results\n\n\n@article{Dong2019UnifiedLM,\ntitle={Unified Language Model Pre-training for Natural Language Understanding and Generation},\nauthor={Li Dong and Nan Yang and Wenhui Wang and Furu Wei and Xiaodong Liu and Yu Wang and Jianfeng Gao and M. Zhou and Hsiao-Wuen Hon},\njournal={ArXiv},\nyear={2019},\nvolume={abs/1905.03197}\n}\n\n\nDataset Curation\n----------------", "### Original Curation", "#### Original Curation Rationale\n\n\nThe dataset is curated in three stages:\n\n\n* Curating passages,\n* Crowdsourcing question-answers on those passages,\n* Obtaining additional answers\nAs part of SQuAD1.1, 10000 high-quality articles from English Wikipedia is extracted using Project Nayuki’s Wikipedia’s internal PageRanks, from which 536 articles are sampled uniformly at random. From each of these articles, individual paragraphs are extracted, stripping away images, figures, tables, and discarding paragraphs shorter than 500 characters.\n\n\nSQuAD2.0 combines the 100,000 questions in SQuAD1.1 with over 50,000 unanswerable questions written adversarially by crowdworkers to look similar to answerable ones.", "#### Communicative Goal\n\n\nTo build systems that not only answer questions when possible, but also determine when no\nanswer is supported by the paragraph and abstain from answering.", "#### Sourced from Different Sources\n\n\nyes", "#### Source Details\n\n\nWikipedia", "### Language Data", "#### How was Language Data Obtained?\n\n\n'Found'", "#### Where was it found?\n\n\n'Single website'", "#### Topics Covered\n\n\nThe dataset contains 536 articles covering a wide range of topics, from musical celebrities to abstract concepts.", "#### Data Validation\n\n\nvalidated by crowdworker", "#### Data Preprocessing\n\n\nFrom the sampled articles from Wikipedia, individual paragraphs are extracted, stripping\naway images, figures, tables, and discarding paragraphs shorter than 500 characters and partitioned into training(80%), development set(10%) and test set(10%).", "#### Was Data Filtered?\n\n\nalgorithmically", "#### Filter Criteria\n\n\nTo retrieve high-quality articles, Project Nayuki’s Wikipedia’s internal PageRanks was used to obtain the top 10000 articles of English Wikipedia, from which 536 articles are sampled uniformly at random.", "### Structured Annotations", "#### Additional Annotations?\n\n\ncrowd-sourced", "#### Number of Raters\n\n\nunknown", "#### Rater Qualifications\n\n\nCrowdworkers from the United States or Canada with a 97% HIT acceptance rate, a minimum of 1000 HITs, were employed to create questions.", "#### Raters per Training Example\n\n\n0", "#### Raters per Test Example\n\n\n0", "#### Annotation Service?\n\n\nyes", "#### Which Annotation Service\n\n\n'other', 'Amazon Mechanical Turk'", "#### Annotation Values\n\n\nFor SQuAD 1.1 , crowdworkers were tasked with asking and answering up to 5 questions on the\ncontent of that paragraph. The questions had to be entered in a text field, and the answers had to be\nhighlighted in the paragraph.\n\n\nFor SQuAD2.0, each task consisted of an entire article from SQuAD 1.1. For each paragraph in the article, workers were asked to pose up to five questions that were impossible to answer\nbased on the paragraph alone, while referencing entities in the paragraph and ensuring that a plausible answer is present.", "#### Any Quality Control?\n\n\nvalidated by another rater", "#### Quality Control Details\n\n\nQuestions from workers who wrote 25 or fewer questions on an article is removed; this filter\nhelped remove noise from workers who had trouble understanding the task, and therefore quit before completing the whole article. This filter to both SQuAD2.0 and the existing answerable questions from SQuAD 1.1.", "### Consent", "#### Any Consent Policy?\n\n\nno", "### Private Identifying Information (PII)", "#### Contains PII?\n\n\nunlikely", "#### Any PII Identification?\n\n\nno identification", "### Maintenance", "#### Any Maintenance Plan?\n\n\nno\n\n\nBroader Social Context\n----------------------", "### Previous Work on the Social Impact of the Dataset", "#### Usage of Models based on the Data\n\n\nno", "### Impact on Under-Served Communities", "#### Addresses needs of underserved Communities?\n\n\nno", "### Discussion of Biases", "#### Any Documented Social Biases?\n\n\nyes\n\n\nConsiderations for Using the Data\n---------------------------------", "### PII Risks and Liability", "### Licenses", "### Known Technical Limitations" ]

[ "task_categories:table-to-text", "annotations_creators:none", "language_creators:unknown", "multilinguality:unknown", "size_categories:unknown", "source_datasets:original", "language:ar", "language:zh", "language:en", "language:fr", "language:hi", "language:id", "language:ja", "language:ko", "language:pt", "language:ru", "language:es", "license:cc-by-2.5", "data-to-text", "region:us" ]

[]

[ "ar", "zh", "en", "fr", "hi", "id", "ja", "ko", "pt", "ru", "es" ]

[ "# Dataset Card for GEM/surface_realisation_st_2020", "## Dataset Description\n\n- Homepage: URL\n- Repository: URL\n- Paper: URL\n- Leaderboard: N/A\n- Point of Contact: Simon Mille", "### Link to Main Data Card\n\nYou can find the main data card on the GEM Website.", "### Dataset Summary \n\nThis dataset was used as part of the multilingual surface realization shared task in which a model gets full or partial universal dependency structures and has to reconstruct the natural language. This dataset support 11 languages. \n\nYou can load the dataset via:\n\nThe data loader can be found here.", "#### website\nWebsite", "#### paper\nACL Anthology", "#### authors\nSimon Mille (Pompeu Fabra University); Leo Wanner (Pompeu Fabra University); Anya Belz (Brighton University); Bernd Bohnet (Google Inc.); Thiago Castro Ferreira (Federal University of Minas Gerais); Yvette Graham (ADAPT/Trinity College Dublin)", "## Dataset Overview", "### Where to find the Data and its Documentation", "#### Webpage\n\n\n\nWebsite", "#### Download\n\n\n\nWebsite", "#### Paper\n\n\n\nACL Anthology", "#### BibTex", "#### Contact Name\n\n\n\n\nSimon Mille", "#### Contact Email\n\n\n\nsfmille@URL", "#### Has a Leaderboard?\n\n\n\nno", "### Languages and Intended Use", "#### Multilingual?\n\n\n\n\nyes", "#### Covered Dialects\n\n\n\nNo multiple dialects.", "#### Covered Languages\n\n\n\n\n'Arabic', 'Chinese', 'English', 'French', 'Hindi', 'Indonesian', 'Japanese', 'Korean', 'Portuguese', 'Russian', 'Spanish, Castilian'", "#### Whose Language?\n\n\n\nUnknown", "#### License\n\n\n\n\ncc-by-2.5: Creative Commons Attribution 2.5 Generic", "#### Intended Use\n\n\n\nThe dataset is intended to be used for training models to solve several NLG subtasks, such as function word introduction, morphological agreement resolution, word order determination and inflection generation.\n\nComment about the license: the dataset has multiple licences, since each original dataset has their own type of licence. All datasets but one are CC-BY and subclasses of it, the other one is GPL (French Sequoia).", "#### Primary Task\n\n\n\nData-to-Text", "#### Communicative Goal\n\n\n\n\nThe models are able to introduce surface features (syntax, morphology, topology) from more or less abstract inputs in different, the most abstract being predicate-argument structures. The datasets cover a large variety of domains (news, blogs, forums, wikipedia pages, etc.).", "### Credit", "#### Curation Organization Type(s)\n\n\n\n'industry', 'academic'", "#### Curation Organization(s)\n\n\n\nPompeu Fabra University, Google Inc., University of Brighton, Federal University of Minas Gerais, ADAPT/Trinity College Dublin", "#### Dataset Creators\n\n\n\nSimon Mille (Pompeu Fabra University); Leo Wanner (Pompeu Fabra University); Anya Belz (Brighton University); Bernd Bohnet (Google Inc.); Thiago Castro Ferreira (Federal University of Minas Gerais); Yvette Graham (ADAPT/Trinity College Dublin)", "#### Funding\n\n\n\nMostly EU funds via H2020 projects", "#### Who added the Dataset to GEM?\n\n\n\nSimon Mille (Pompeu Fabra University)", "### Dataset Structure", "#### Data Fields\n\n\n\n'input' (string): this field contains an input tree in CoNLL-U format; the CoNLL-U format is a one-word-per-line format with the following tab-separated 10 columns (see here): [1] Position, [2] Lemma, [3] Wordform, [4] Part of Speech, [5] Fine-grained Part of Speech (if available), [6] Features (FEATS), [7] governor, [8] dependency relation, [9] additional dependency information, and [10] metadata. For the surface task, the input is a Universal Dependency tree of a given language in which the word order was scrambled and the surface forms removed (only lemmas are available); for the deep task, the input is a tree derived from the surface input, with predicate-argument relations between content words only (function words were removed) and without any morphological agreement information.\n\n'target_tokenized' (string): this field contains the target sentence to generate, in which every non-initial and non-final token is surrounded by two spaces. This output is usually used for automatic evaluations.\n\n'target' (string): this field contains the detokenised target sentence to generate. This output is usually used for human evaluations.\n\n'gem_id' (string): a unique ID.\n\n'sentence_id' (string): the original ID of a sentence in the UD dataset.", "#### Reason for Structure\n\n\n\nThe structure of the input (CoNLL-U) was chosen according to the standards in parsing, and because the original UD datasets were provided in this format.", "#### How were labels chosen?\n\n\n\nThe input labels for the surface track are the original labels in the UD treebanks; see here for the dependencies, here for the features, and here for the PoS tags.\n\nThe input labels for the deep track are a subset of the PoS tags and features of the surface track, and for the relations, universal predicate-argument relations augmented with a few specific relations to capture coordinations and named entity relations for instance.", "#### Example Instance", "#### Data Splits\n\n\n\nThere are 119 splits in the dataset:\n\n- 29 training sets, which correspond to 20 UD datasets (11 languages), 9 of which have both surface and deep inputs (3 languages);\n- 29 development set which correspond to the 29 training sets above;\n- 29 test sets for the data described above;\n- 4 out-of-domain test sets, 3 surface inputs and 1 deep one (3 languages for which PUD out-of-domain datasets were available);\n- 9 automatically parsed in-domain test sets, 6 surface inputs and 3 deep inputs (6 languages for which good UD parsers were available);\n- 9 automatically parsed out-of-domain test sets, 6 surface inputs and 3 deep inputs (6 languages for which we were able to create clean Wikipedia text and that had a good UD parser).", "#### Splitting Criteria\n\n\n\nDescribed above for more clarity.", "#### \n\n\n\nAn outlier would usually be an input that corresponds to a very long sentence (e.g. 159 words in English, when the average number of words per sentence is around 25).", "## Dataset in GEM", "### Rationale for Inclusion in GEM", "#### Why is the Dataset in GEM?\n\n\n\nThe datset includes languages from different families and some languages not often used in NLG (e.g. Arabic, Indonesian, Korean, Hindi). It proposes two tasks, which can be tackled both separately and in one shot, with different levels of difficulty: the most superficial task (T1) consits in ordering and inflecting some trees, and the deeper task (T2) includes extra tasks such as defining the syntactic structure and introducing function words and morphological agreement information. Both tasks can allow for developing modules for pipeline NLG architectures. T1 is rather straightforward to evaluate: BLEU works quite well for some languages since all the words are present in the input and few word orders only can be possible for a syntactic tree. But T2 is more challenging to evaluate, since more outputs are correct given one particular input.\n\nThere is a large variety of sizes in the datasets, both clean and noisy data, parallel data in different languages, and many already available system outputs to use as baselines.", "#### Similar Datasets\n\n\n\nyes", "#### Unique Language Coverage\n\n\n\nyes", "#### Difference from other GEM datasets\n\n\n\nThis is possibly the only dataset that starts the generation process from predicate-argument structures and from syntactic structures. It also has parallel datasets in a few languages (coming from the PUD parallel annotations).", "#### Ability that the Dataset measures\n\n\n\nSyntacticisation, functional word introduction, word order resolution, agreement resolution, morphological inflection", "### GEM-Specific Curation", "#### Modificatied for GEM?\n\n\n\nno", "#### Additional Splits?\n\n\n\nno", "### Getting Started with the Task", "#### Pointers to Resources\n\n\n\nWebsite", "#### Technical Terms\n\n\n\nSyntacticisation: prediction of the syntactic", "## Previous Results", "### Previous Results", "#### Measured Model Abilities\n\n\n\nSyntacticisation, functional word introduction, word order resolution, morphological agreement resolution, morphological inflection", "#### Metrics\n\n\n\n'BLEU', 'BERT-Score', 'Other: Other Metrics'", "#### Other Metrics\n\n\n\nNIST: n-gram similarity metric weighted in favour of less frequent n-grams which are taken to be more informative.\n\nNormalised edit distance (DIST): inverse, normalised, character-based string-edit distance that starts by computing the minimum number of character inserts, deletes and substitutions (all at cost 1) required to turn the system output into the (single) reference text.", "#### Proposed Evaluation\n\n\n\nBLEU, NIST, BERTScore and DIST simply aim at calculating in different ways the similarity between a predicted and a reference sentence.\n\nTwo additional criteria have been used for human evaluation, Readability and Meaning SImilarity. The statement to be assessed in the Readability evaluation was: \"The text reads well and is free from grammatical errors and awkward constructions.\". The corresponding statement in the Meaning Similarity evaluation, in which system outputs (‘the black text’) were compared to reference sentences (‘the gray text’), was: \"The meaning of the gray text is adequately expressed by the black text.\"", "#### Previous results available?\n\n\n\nyes", "#### Other Evaluation Approaches\n\n\n\nSame as above.", "#### Relevant Previous Results\n\n\n\n- Fast and Accurate Non-Projective Dependency Tree Linearization\n- Shape of Synth to Come: Why We Should Use Synthetic Data for English Surface Realization", "## Dataset Curation", "### Original Curation", "#### Original Curation Rationale\n\n\n\nThe datasets were created in the context of the Surface Realisation Shared Task series.", "#### Communicative Goal\n\n\n\nThe dataset's objective was to allow for training systems to perform tasks related to surface realisation (introduction of function words, syntacticisation, resolution of morphological agreements, word order resolution, inflection generation.", "#### Sourced from Different Sources\n\n\n\nyes", "#### Source Details\n\n\n\nEach of the 20 used UD datasets comes from various sources, all listed on the individual page of each UD treeebank (URL\n\nAdditional test sets were created for the task, and were obtained from Wikipedia pages for 6 languages.", "### Language Data", "#### How was Language Data Obtained?\n\n\n\n'Found'", "#### Where was it found?\n\n\n\n'Multiple websites'", "#### Language Producers\n\n\n\nThere are numerous sources of language in the multiple datasets.", "#### Topics Covered\n\n\n\nThere is a large variety of topics in the multiple datasets.", "#### Data Validation\n\n\n\nnot validated", "#### Data Preprocessing\n\n\n\nThe text data was detokenised so as to create references for automatic evaluations (several languages don't use spaces to separate words, and running metrics like BLEU would not make sense without separating all the tokens in a sentence).", "#### Was Data Filtered?\n\n\n\nhybrid", "#### Filter Criteria\n\n\n\nFor the Wikipedia test created for the shared task, extensive filtering was applied to achieve reasonably good text quality. Sentences that include special characters, contain unusual tokens (e.g. ISBN), or have unbalanced quotation marks or brackets were skipped. Furthermore, only sentences with more than 5 tokens and shorter than 50 tokens were selected. After the initial filtering, quite a few malformed sentences remained. In order to remove those, the sentences were scored with BERT and\nonly the top half scored sentences were kept. Finally, via manual inspection, patterns and expressions were identified to\nfurther reduce the number of malformed sentences.", "### Structured Annotations", "#### Additional Annotations?\n\n\n\n\nnone", "#### Annotation Service?\n\n\n\nno", "### Consent", "#### Any Consent Policy?\n\n\n\nno", "#### Justification for Using the Data\n\n\n\nThe Universal Dependency data had been previously used for shared tasks on parsing, so it made sense to reuse it for generation.", "### Private Identifying Information (PII)", "#### Contains PII?\n\n\n\n\nunlikely", "#### Any PII Identification?\n\n\n\nno identification", "### Maintenance", "#### Any Maintenance Plan?\n\n\n\nno", "## Broader Social Context", "### Previous Work on the Social Impact of the Dataset", "#### Usage of Models based on the Data\n\n\n\nno", "### Impact on Under-Served Communities", "#### Addresses needs of underserved Communities?\n\n\n\nyes", "#### Details on how Dataset Addresses the Needs\n\n\n\nThanks to the original work of the UD dataset creators, the surface realisation dataset addresses a few languages which are possibly under-served in NLG: e.g. Arabic, Hindi, Indonesian, Korean.", "### Discussion of Biases", "#### Any Documented Social Biases?\n\n\n\nno", "#### Are the Language Producers Representative of the Language?\n\n\n\nIt is very likely that the distribution of language producers is not fully represented in the datasets of each language.", "## Considerations for Using the Data", "### PII Risks and Liability", "#### Potential PII Risk\n\n\n\nNo risks foreseen.", "### Licenses", "#### Copyright Restrictions on the Dataset\n\n\n\n'multiple licenses', 'open license - commercial use allowed'", "#### Copyright Restrictions on the Language Data\n\n\n\n'multiple licenses', 'open license - commercial use allowed'", "### Known Technical Limitations", "#### Technical Limitations\n\n\n\nThe deep track inputs (predicate-argument structures) are not of perfect quality, they were derived automatically from gold or predicted syntactic parses using handcrafted grammars.", "#### Unsuited Applications\n\n\n\nThe datasets are probably not fitted to train tools to produce \"unusual\" languages (e.g. poetry, kid writing etc.).", "#### Discouraged Use Cases\n\n\n\nTo be thought of :)" ]

[ "TAGS\n#task_categories-table-to-text #annotations_creators-none #language_creators-unknown #multilinguality-unknown #size_categories-unknown #source_datasets-original #language-Arabic #language-Chinese #language-English #language-French #language-Hindi #language-Indonesian #language-Japanese #language-Korean #language-Portuguese #language-Russian #language-Spanish #license-cc-by-2.5 #data-to-text #region-us \n", "# Dataset Card for GEM/surface_realisation_st_2020", "## Dataset Description\n\n- Homepage: URL\n- Repository: URL\n- Paper: URL\n- Leaderboard: N/A\n- Point of Contact: Simon Mille", "### Link to Main Data Card\n\nYou can find the main data card on the GEM Website.", "### Dataset Summary \n\nThis dataset was used as part of the multilingual surface realization shared task in which a model gets full or partial universal dependency structures and has to reconstruct the natural language. This dataset support 11 languages. \n\nYou can load the dataset via:\n\nThe data loader can be found here.", "#### website\nWebsite", "#### paper\nACL Anthology", "#### authors\nSimon Mille (Pompeu Fabra University); Leo Wanner (Pompeu Fabra University); Anya Belz (Brighton University); Bernd Bohnet (Google Inc.); Thiago Castro Ferreira (Federal University of Minas Gerais); Yvette Graham (ADAPT/Trinity College Dublin)", "## Dataset Overview", "### Where to find the Data and its Documentation", "#### Webpage\n\n\n\nWebsite", "#### Download\n\n\n\nWebsite", "#### Paper\n\n\n\nACL Anthology", "#### BibTex", "#### Contact Name\n\n\n\n\nSimon Mille", "#### Contact Email\n\n\n\nsfmille@URL", "#### Has a Leaderboard?\n\n\n\nno", "### Languages and Intended Use", "#### Multilingual?\n\n\n\n\nyes", "#### Covered Dialects\n\n\n\nNo multiple dialects.", "#### Covered Languages\n\n\n\n\n'Arabic', 'Chinese', 'English', 'French', 'Hindi', 'Indonesian', 'Japanese', 'Korean', 'Portuguese', 'Russian', 'Spanish, Castilian'", "#### Whose Language?\n\n\n\nUnknown", "#### License\n\n\n\n\ncc-by-2.5: Creative Commons Attribution 2.5 Generic", "#### Intended Use\n\n\n\nThe dataset is intended to be used for training models to solve several NLG subtasks, such as function word introduction, morphological agreement resolution, word order determination and inflection generation.\n\nComment about the license: the dataset has multiple licences, since each original dataset has their own type of licence. All datasets but one are CC-BY and subclasses of it, the other one is GPL (French Sequoia).", "#### Primary Task\n\n\n\nData-to-Text", "#### Communicative Goal\n\n\n\n\nThe models are able to introduce surface features (syntax, morphology, topology) from more or less abstract inputs in different, the most abstract being predicate-argument structures. The datasets cover a large variety of domains (news, blogs, forums, wikipedia pages, etc.).", "### Credit", "#### Curation Organization Type(s)\n\n\n\n'industry', 'academic'", "#### Curation Organization(s)\n\n\n\nPompeu Fabra University, Google Inc., University of Brighton, Federal University of Minas Gerais, ADAPT/Trinity College Dublin", "#### Dataset Creators\n\n\n\nSimon Mille (Pompeu Fabra University); Leo Wanner (Pompeu Fabra University); Anya Belz (Brighton University); Bernd Bohnet (Google Inc.); Thiago Castro Ferreira (Federal University of Minas Gerais); Yvette Graham (ADAPT/Trinity College Dublin)", "#### Funding\n\n\n\nMostly EU funds via H2020 projects", "#### Who added the Dataset to GEM?\n\n\n\nSimon Mille (Pompeu Fabra University)", "### Dataset Structure", "#### Data Fields\n\n\n\n'input' (string): this field contains an input tree in CoNLL-U format; the CoNLL-U format is a one-word-per-line format with the following tab-separated 10 columns (see here): [1] Position, [2] Lemma, [3] Wordform, [4] Part of Speech, [5] Fine-grained Part of Speech (if available), [6] Features (FEATS), [7] governor, [8] dependency relation, [9] additional dependency information, and [10] metadata. For the surface task, the input is a Universal Dependency tree of a given language in which the word order was scrambled and the surface forms removed (only lemmas are available); for the deep task, the input is a tree derived from the surface input, with predicate-argument relations between content words only (function words were removed) and without any morphological agreement information.\n\n'target_tokenized' (string): this field contains the target sentence to generate, in which every non-initial and non-final token is surrounded by two spaces. This output is usually used for automatic evaluations.\n\n'target' (string): this field contains the detokenised target sentence to generate. This output is usually used for human evaluations.\n\n'gem_id' (string): a unique ID.\n\n'sentence_id' (string): the original ID of a sentence in the UD dataset.", "#### Reason for Structure\n\n\n\nThe structure of the input (CoNLL-U) was chosen according to the standards in parsing, and because the original UD datasets were provided in this format.", "#### How were labels chosen?\n\n\n\nThe input labels for the surface track are the original labels in the UD treebanks; see here for the dependencies, here for the features, and here for the PoS tags.\n\nThe input labels for the deep track are a subset of the PoS tags and features of the surface track, and for the relations, universal predicate-argument relations augmented with a few specific relations to capture coordinations and named entity relations for instance.", "#### Example Instance", "#### Data Splits\n\n\n\nThere are 119 splits in the dataset:\n\n- 29 training sets, which correspond to 20 UD datasets (11 languages), 9 of which have both surface and deep inputs (3 languages);\n- 29 development set which correspond to the 29 training sets above;\n- 29 test sets for the data described above;\n- 4 out-of-domain test sets, 3 surface inputs and 1 deep one (3 languages for which PUD out-of-domain datasets were available);\n- 9 automatically parsed in-domain test sets, 6 surface inputs and 3 deep inputs (6 languages for which good UD parsers were available);\n- 9 automatically parsed out-of-domain test sets, 6 surface inputs and 3 deep inputs (6 languages for which we were able to create clean Wikipedia text and that had a good UD parser).", "#### Splitting Criteria\n\n\n\nDescribed above for more clarity.", "#### \n\n\n\nAn outlier would usually be an input that corresponds to a very long sentence (e.g. 159 words in English, when the average number of words per sentence is around 25).", "## Dataset in GEM", "### Rationale for Inclusion in GEM", "#### Why is the Dataset in GEM?\n\n\n\nThe datset includes languages from different families and some languages not often used in NLG (e.g. Arabic, Indonesian, Korean, Hindi). It proposes two tasks, which can be tackled both separately and in one shot, with different levels of difficulty: the most superficial task (T1) consits in ordering and inflecting some trees, and the deeper task (T2) includes extra tasks such as defining the syntactic structure and introducing function words and morphological agreement information. Both tasks can allow for developing modules for pipeline NLG architectures. T1 is rather straightforward to evaluate: BLEU works quite well for some languages since all the words are present in the input and few word orders only can be possible for a syntactic tree. But T2 is more challenging to evaluate, since more outputs are correct given one particular input.\n\nThere is a large variety of sizes in the datasets, both clean and noisy data, parallel data in different languages, and many already available system outputs to use as baselines.", "#### Similar Datasets\n\n\n\nyes", "#### Unique Language Coverage\n\n\n\nyes", "#### Difference from other GEM datasets\n\n\n\nThis is possibly the only dataset that starts the generation process from predicate-argument structures and from syntactic structures. It also has parallel datasets in a few languages (coming from the PUD parallel annotations).", "#### Ability that the Dataset measures\n\n\n\nSyntacticisation, functional word introduction, word order resolution, agreement resolution, morphological inflection", "### GEM-Specific Curation", "#### Modificatied for GEM?\n\n\n\nno", "#### Additional Splits?\n\n\n\nno", "### Getting Started with the Task", "#### Pointers to Resources\n\n\n\nWebsite", "#### Technical Terms\n\n\n\nSyntacticisation: prediction of the syntactic", "## Previous Results", "### Previous Results", "#### Measured Model Abilities\n\n\n\nSyntacticisation, functional word introduction, word order resolution, morphological agreement resolution, morphological inflection", "#### Metrics\n\n\n\n'BLEU', 'BERT-Score', 'Other: Other Metrics'", "#### Other Metrics\n\n\n\nNIST: n-gram similarity metric weighted in favour of less frequent n-grams which are taken to be more informative.\n\nNormalised edit distance (DIST): inverse, normalised, character-based string-edit distance that starts by computing the minimum number of character inserts, deletes and substitutions (all at cost 1) required to turn the system output into the (single) reference text.", "#### Proposed Evaluation\n\n\n\nBLEU, NIST, BERTScore and DIST simply aim at calculating in different ways the similarity between a predicted and a reference sentence.\n\nTwo additional criteria have been used for human evaluation, Readability and Meaning SImilarity. The statement to be assessed in the Readability evaluation was: \"The text reads well and is free from grammatical errors and awkward constructions.\". The corresponding statement in the Meaning Similarity evaluation, in which system outputs (‘the black text’) were compared to reference sentences (‘the gray text’), was: \"The meaning of the gray text is adequately expressed by the black text.\"", "#### Previous results available?\n\n\n\nyes", "#### Other Evaluation Approaches\n\n\n\nSame as above.", "#### Relevant Previous Results\n\n\n\n- Fast and Accurate Non-Projective Dependency Tree Linearization\n- Shape of Synth to Come: Why We Should Use Synthetic Data for English Surface Realization", "## Dataset Curation", "### Original Curation", "#### Original Curation Rationale\n\n\n\nThe datasets were created in the context of the Surface Realisation Shared Task series.", "#### Communicative Goal\n\n\n\nThe dataset's objective was to allow for training systems to perform tasks related to surface realisation (introduction of function words, syntacticisation, resolution of morphological agreements, word order resolution, inflection generation.", "#### Sourced from Different Sources\n\n\n\nyes", "#### Source Details\n\n\n\nEach of the 20 used UD datasets comes from various sources, all listed on the individual page of each UD treeebank (URL\n\nAdditional test sets were created for the task, and were obtained from Wikipedia pages for 6 languages.", "### Language Data", "#### How was Language Data Obtained?\n\n\n\n'Found'", "#### Where was it found?\n\n\n\n'Multiple websites'", "#### Language Producers\n\n\n\nThere are numerous sources of language in the multiple datasets.", "#### Topics Covered\n\n\n\nThere is a large variety of topics in the multiple datasets.", "#### Data Validation\n\n\n\nnot validated", "#### Data Preprocessing\n\n\n\nThe text data was detokenised so as to create references for automatic evaluations (several languages don't use spaces to separate words, and running metrics like BLEU would not make sense without separating all the tokens in a sentence).", "#### Was Data Filtered?\n\n\n\nhybrid", "#### Filter Criteria\n\n\n\nFor the Wikipedia test created for the shared task, extensive filtering was applied to achieve reasonably good text quality. Sentences that include special characters, contain unusual tokens (e.g. ISBN), or have unbalanced quotation marks or brackets were skipped. Furthermore, only sentences with more than 5 tokens and shorter than 50 tokens were selected. After the initial filtering, quite a few malformed sentences remained. In order to remove those, the sentences were scored with BERT and\nonly the top half scored sentences were kept. Finally, via manual inspection, patterns and expressions were identified to\nfurther reduce the number of malformed sentences.", "### Structured Annotations", "#### Additional Annotations?\n\n\n\n\nnone", "#### Annotation Service?\n\n\n\nno", "### Consent", "#### Any Consent Policy?\n\n\n\nno", "#### Justification for Using the Data\n\n\n\nThe Universal Dependency data had been previously used for shared tasks on parsing, so it made sense to reuse it for generation.", "### Private Identifying Information (PII)", "#### Contains PII?\n\n\n\n\nunlikely", "#### Any PII Identification?\n\n\n\nno identification", "### Maintenance", "#### Any Maintenance Plan?\n\n\n\nno", "## Broader Social Context", "### Previous Work on the Social Impact of the Dataset", "#### Usage of Models based on the Data\n\n\n\nno", "### Impact on Under-Served Communities", "#### Addresses needs of underserved Communities?\n\n\n\nyes", "#### Details on how Dataset Addresses the Needs\n\n\n\nThanks to the original work of the UD dataset creators, the surface realisation dataset addresses a few languages which are possibly under-served in NLG: e.g. Arabic, Hindi, Indonesian, Korean.", "### Discussion of Biases", "#### Any Documented Social Biases?\n\n\n\nno", "#### Are the Language Producers Representative of the Language?\n\n\n\nIt is very likely that the distribution of language producers is not fully represented in the datasets of each language.", "## Considerations for Using the Data", "### PII Risks and Liability", "#### Potential PII Risk\n\n\n\nNo risks foreseen.", "### Licenses", "#### Copyright Restrictions on the Dataset\n\n\n\n'multiple licenses', 'open license - commercial use allowed'", "#### Copyright Restrictions on the Language Data\n\n\n\n'multiple licenses', 'open license - commercial use allowed'", "### Known Technical Limitations", "#### Technical Limitations\n\n\n\nThe deep track inputs (predicate-argument structures) are not of perfect quality, they were derived automatically from gold or predicted syntactic parses using handcrafted grammars.", "#### Unsuited Applications\n\n\n\nThe datasets are probably not fitted to train tools to produce \"unusual\" languages (e.g. poetry, kid writing etc.).", "#### Discouraged Use Cases\n\n\n\nTo be thought of :)" ]

[ "task_categories:table-to-text", "annotations_creators:none", "language_creators:unknown", "multilinguality:unknown", "size_categories:unknown", "source_datasets:original", "language:en", "license:cc-by-sa-3.0", "data-to-text", "arxiv:1603.07771", "arxiv:2007.02871", "arxiv:2005.10433", "region:us" ]

[ "1603.07771", "2007.02871", "2005.10433" ]

[ "en" ]

[ "### Link to Main Data Card\n\n\nYou can find the main data card on the [GEM Website.", "### Dataset Summary\n\n\nToTTo is a high-quality English table-to-text dataset with more than 100,000 examples in which a table from Wikipedia with highlighted cells is paired with a sentence that describes the highlighted cells. All examples in the dataset were post-edited in multiple steps to ensure that the targets are fully faithful to the input information.\n\n\nYou can load the dataset via:\n\n\nThe data loader can be found here.", "#### website\n\n\nn/a", "#### paper\n\n\nACL Anthology", "#### authors\n\n\nAnkur Parikh, Xuezhi Wang, Sebastian Gehrmann, Manaal Faruqui, Bhuwan Dhingra, Diyi Yang, Dipanjan Das\n\n\nDataset Overview\n----------------", "### Where to find the Data and its Documentation", "#### Download\n\n\nToTTo Main Repo + ToTTo Supplementary Repo", "#### Paper\n\n\nACL Anthology", "#### BibTex", "#### Contact Name\n\n\nAnkur Parikh", "#### Contact Email\n\n\ntotto@URL", "#### Has a Leaderboard?\n\n\nyes", "#### Leaderboard Link\n\n\nGithub", "#### Leaderboard Details\n\n\nThis dataset has an associated, active leaderboard maintained by the authors.\nThe test set ground truth targets / references are private, i.e they are not publicly shared or downloadable - hence, leaderboard submission is necessary for test set evaluation.\nTo evaluate your model on the dev or test set AND/OR submit to the leaderboard, you need to submit your model files through this form (The form provides an option to opt-out of going on the leaderboard).\n\n\nThe leaderboard reports three sets of BLEU, PARENT and BLEURT scores for each submission - on the overall test set, the *Overlap* subset of the test set and the *non-Overlap* subset of the test set.", "### Languages and Intended Use", "#### Multilingual?\n\n\nno", "#### Covered Dialects\n\n\nNo specific dialects. The original language is from Wikipedia and it was post-edited by crowdraters", "#### Covered Languages\n\n\n'English'", "#### Whose Language?\n\n\nThe language is post-edited English only (BCP-47: 'en') Wikipedia text. No demographic information about annotators is provided.\nSome amounts of what may be called non-English text, including characters such as French accents or Cyrillic characters, could sometimes occur, especially through fields with entity names as values in the input table cells.", "#### License\n\n\ncc-by-sa-3.0: Creative Commons Attribution Share Alike 3.0 Unported", "#### Intended Use\n\n\nToTTo is a Table-to-Text NLG task, as the paper title says. The task is as follows: Given a Wikipedia table with row names, column names and table cells, with a subset of cells highlighted, generate a natural language description for the highlighted part of the table . The table need not be exactly rectangular in that - cells can sometimes be multi-row or multi-column.\n\n\nAn earlier example of a Table-to-Text NLG task is Wikibio - here the inputs were Wikipedia infoboxes (from the top right corner of entity-related Wiki pages). In contrast, ToTTo mostly has Wikipedia tables from the main article content itself. In general, Table-To-Text NLG tasks can be seen as a subclass of Data-To-Text NLG tasks - where the task is to generate natural language descriptions of inputs which are in the form of structured or semi-structured data. In general, all Data-To-Text NLG tasks need not have an explicit table or other structure - e.g the input in WebNLG is simply a list of triples.\n\n\nImportantly, ToTTo differs from earlier examples of Table-To-Text NLG in that:\n\n\n1. It does not suffer from the problem of divergent references - where ground truth descriptions themselves have additional information not found in the table. ToTTo overcomes this by having a multi-step annotation process to edit the initial, free-form table descriptions (which are from Wikipedia) to make them faithful, unambiguous and independent of article context.\n2. Since it provides control in the form of highlighted table cells, it prevents the problem of there being a large number of valid descriptions focussing on different parts of the table.", "#### Primary Task\n\n\nData-to-Text", "#### Communicative Goal\n\n\nThe speaker is required to produce a single, coherent English sentence that describes the highlighted cells in the given table, also using metadata and any other information from the table as applicable.", "### Credit", "#### Curation Organization Type(s)\n\n\n'industry'", "#### Curation Organization(s)\n\n\nGoogle Research", "#### Dataset Creators\n\n\nAnkur Parikh, Xuezhi Wang, Sebastian Gehrmann, Manaal Faruqui, Bhuwan Dhingra, Diyi Yang, Dipanjan Das", "#### Funding\n\n\nGoogle Research", "#### Who added the Dataset to GEM?\n\n\nVarun Gangal created the initial data card and Yacine Jernite wrote the data loader. The data card was updated with new splits by Simon Mille. Sebastian Gehrmann ported the data card and loader from the v1 to the v2 version and extended it with the new fields.", "### Dataset Structure", "#### Data Fields\n\n\n* The 'table' field is a 'List[List[Dict]]' in row-major order, with outer lists representing rows and the inner lists columns.\n* Each 'Dict' has the fields 'column\\_span: int', 'is\\_header: bool', 'row\\_span: int', and 'value: str'.\n* Table metadata consists of 'table\\_page\\_title', 'table\\_section\\_title' and 'table\\_section\\_texts'\n* The 'highlighted\\_cells' are represented as 'List[[row\\_index,column\\_index]]', with each '[row\\_index,column\\_index]' indicating that 'table[row\\_index][column\\_index]' is highlighted.\n* 'example\\_id' is the unique id per example.\n* 'sentence\\_annotations[final\\_sentence]' which is the table description/generation target", "#### Reason for Structure\n\n\nThe structure is aimed to encode highlighted tables in a way that allows rows and columns to span multiple fields in width. The other fields are meta-data about the source and the annotations", "#### How were labels chosen?\n\n\nThe initial table-description pairs are tables from Wikipedia articles, extracted through heuristics such as Number Matching (tables and sentences that overlap with a non-date number of atleast 3 non-zero digits) (Refer to Section 4 of the paper for more)\n\n\n1. Table Readability: Tables which are deemed non-readable (due to foreign language, poor formatting etc - a very small fraction of 0.5%) are removed from the dataset here.\n2. Cell Highlighting: The annotator highlights the cells of the table which support the description.\n3. Deletion: The annotator removes phrases in the description which are not supported by the highlighted cells\n4. Decontextualization: Descriptions may contain pronouns or other forms of anaphora, or other phenomena which depend on the overall article topic - these are fixed by replacement (e.g replacing pronouns with the entity, provided it occurs in the table). The replacements allowed are limited to one, and annotators are also instructed to conserve fluency.\n5. Secondary Annotation: A second set of annotators is shown the output of Stage 4, and asked to fix it if required to ensure it is grammatical.", "#### Example Instance\n\n\nThe main repository's 'URL' already provides a thorough walkthrough of data instances and fields here\n\n\nBelow is the instance for a table from the wiki-page for the musical artist *Weird Al' Yankovic* , likely listing his on-television appearances.\n\n\nThe 'table' field is expanded as below:\n\n\nThe Supplementary Repo also provides browsable samples under its 'sample/' folder. It additionally provides HTML visualization scripts with their outputs located under the aforementioned folder. The instructions to access and visualize these samples can also be found here.", "#### Data Splits\n\n\nThe dataset consists of 120,000 train examples and equi-sized dev and test sets with 7700 examples.\nRefer to Table 5 in the paper for a more extensive list of properties about table size, target vocabulary etc and their aggregates.", "#### Splitting Criteria\n\n\nThe dev and test splits are further equally distributed between *Overlap* and *non-Overlap* .\nThe examples in the *Overlap* set are harder on account of the domain shift resulting from them having none of their header (row and column) names in common with those seen during training.\n\n\nRefer to Table 5 in the paper for a more extensive list of properties about table size, target vocabulary etc and their aggregates.", "#### \n\n\nThere are some very large tables in the dataset with thousands of rows. Table 7 shows some of the challenges of the dataset, showing that very few examples require access to the table description itself which makes those examples an outlier.\n\n\nDataset in GEM\n--------------", "### Rationale for Inclusion in GEM", "#### Why is the Dataset in GEM?\n\n\nToTTo is one of the two datasets representing Table-to-Text NLG in GEM, the other one being DART. Unlike DART, which combines datasets from multiple sources and furnishes them in a unified setting, ToTTo is from a homogeneous source. As explained in the Task Summary above, it also has an annotation process explicitly crafted to reduce divergent descriptions, which is not true of DART.\n\n\nFurthermore, ToTTo is also an instance of a controlled generation task - where in addition to the input (in this case the table) an additional control (in this case the highlighted cells) is given as an additional goal for the generation. The DART task formulation does not include controls.", "#### Similar Datasets\n\n\nyes", "#### Unique Language Coverage\n\n\nno", "#### Difference from other GEM datasets\n\n\nThe input is much more complex and the quality much better than that of comparable datasets. The highlighted table cells provide a unique challenge to models.", "#### Ability that the Dataset measures\n\n\nReasoning, surface realization", "### GEM-Specific Curation", "#### Modificatied for GEM?\n\n\nyes", "#### Additional Splits?\n\n\nyes", "#### Split Information\n\n\n9 challenge sets for ToTTo were added to the GEM evaluation suite, 8 created specifically for the task and 1 coming from the original data.\n\n\n1. We created subsets of the training and development sets of 500 randomly selected inputs each.\n2. We applied input scrambling on a subset of 500 randomly selected test instances; the order of the highlighted cells was randomly reassigned.\n3. For the input size, we created subpopulations based on the number of input highlighted cells in the whole table.\n\n\n\n4. We also divided the test set according to the size of the whole table, based on the idea that larger tables represent a bigger space to take into account when generating the highlighted cells; a larger table could be more challenging to generate accurate text than a smaller table. There are 693 different table sizes, ranging from 2 to 15834 cells.\n\n\n\n5. We also created three splits based on the subset of test examples in pages about people.\nWe then used the structured information in WikiData to identify the following information:\n\n\n* gender (male, and female),\n* nationality grouped by continent (Africa, Asia, Europe, North America, Oceania, and South America)\n* ethnicity (African American and all USA)\n\n\nThe categories within gender, ethnicity, and nationality were chosen based on data availability; The ToTTo dataset includes mostly tables that do not focus on people. As a result, only seven people in the original test set are marked as having a non-binary gender. Similar sparsity informed the grouping of nationalities by continent – only 19 countries are represented by more than 10 people in the test set. In case a person has citizenships across multiple continents, we may include the person in any of the included continents.\n\n\nFinally, ethnicity is very sparsely annotated in WikiData; only 150 test examples in ToTTo have this information and 128 of these are African Americans. We thus are unable to compare the performance on, e.g., Yoruba or Punjabi people, both of which have fewer than five instances. Another caveat here is that only 21 of the 128 people are female. We thus compare the African American population to results on a subset that includes all US citizens.", "#### Split Motivation\n\n\ngeneralization, fairness, robustness", "### Getting Started with the Task", "#### Pointers to Resources\n\n\n* The highest spot on the leaderboard is currently held by an anonymous method, with BLEU=49.2, PARENT=58.7 and BLEURT=0.249 on the *Overall* test set.\n* The highest scoring non-anonymous method is the T5-based method of Kale, 2020. This method uses a simple row-major linearization scheme to convert the table (it chooses only the highlighted cells and ignores the other cells - table titles and section titles are prefixed at the start of the respective section table) to a flat string. The linearized input - output description pairs from training examples are then used to finetune T5, with BLEU being used as the dev metric to pick checkpoints, and beam search with beam size 10 being the decoding method.\n\n\nThough the best numbers from this method are naturally from the largest T5-pretrained architecture (T5-3B), the paper shows improvements over the next-highest BERT-to-BERT method even when using T5-Base or T5-Small, which have the same and lesser parameters than BERT-to-BERT respectively.\n* The Supplementary Repo provides several useful modules to get started with for new approach implementation:\n\n\n\t1. Code for the particular preprocessing / linearization scheme used to linearize the tables into flat sequences for the baseline approaches described in the paper has been described and shared herein\n\t2. An evaluation script for locally scoring BLEU and PARENT system outputs on dev (or train) sets. Since BLEURT is a model-based metric, a slightly separate set of instructions is provided to evaluate on the same.\n\n\nPrevious Results\n----------------", "### Previous Results", "#### Measured Model Abilities\n\n\nReasoning, surface realization", "#### Metrics\n\n\n'BLEU', 'BLEURT', 'Other: Other Metrics'", "#### Other Metrics\n\n\nParent: a metric that measures the F-1 score of overlap between input content words and those used in references and those in generated text while ignoring the general surface form. It can thus measure the faithfulness much better than metrics that measure overlap with a reference", "#### Proposed Evaluation\n\n\nThe metrics are used as in the leaderboard. The original paper additionally conducted a human evaluation focusing on fluency, faithfulness, and coverage.\nFaithfulness was measured as whether facts in the text are not supported by the input, and coverage as the number of highlighted cells that were considered. They thus represent precision and recall of the content.", "#### Previous results available?\n\n\nyes", "#### Relevant Previous Results\n\n\nSee leaderboard.\n\n\nDataset Curation\n----------------", "### Original Curation", "#### Original Curation Rationale\n\n\nTables occurring in Wikipedia articles were chosen as the data source with the following reasons in mind:\n\n\n1. Wide coverage in terms of both vocabulary and concepts.\n2. Wikipedia tables are not confined to a regular structure, with multi-row or multi-column cells occurring with a sufficient frequency.\n3. Likely to contain reasonable-quality, natural text descriptions in the proximity of the table, which are also extractable by heuristics. (see the start of Section 4 for the heuristics used)\n\n\nTo prevent an overlap with the earlier Wikibio dataset which focussed on Infobox-first sentence pairs from Wikipedia biography articles, the authors avoid using Infoboxes as a data source.\n\n\nThe overall curation process of initially collecting free text and then annotator-revising it, was designed to combine the advantages of free-form text descriptions (which are fluent, high-quality and unhurriedly written, but also divergent and unfaithful) with annotator descriptions (which can be tailored to be faithful and to conform exactly to desired task requirements)", "#### Communicative Goal\n\n\nThe speaker is required to produce a single, coherent English sentence that describes the highlighted cells in the given table, also using metadata and any other information from the table as applicable.", "#### Sourced from Different Sources\n\n\nyes", "#### Source Details\n\n\nURL", "### Language Data", "#### How was Language Data Obtained?\n\n\n'Crowdsourced'", "#### Where was it crowdsourced?\n\n\n'Other crowdworker platform'", "#### Language Producers\n\n\nThe basic source language producers are Wikipedia authors and/or editors, since the annotation starts with the natural text description near the Wikipedia table.\nThe auxiliary source language producers are the annotators (two per example) who iteratively revise these descriptions to make them unambiguous and faithful to a subset of highlighted cells in the table.", "#### Data Validation\n\n\nvalidated by crowdworker", "#### Data Preprocessing\n\n\nThe initial table-description pairs are tables from Wikipedia articles, extracted through heuristics such as Number Matching (tables and sentences that overlap with a non-date number of atleast 3 non-zero digits) (Refer to Section 4 of the paper for more)\n\n\n1. Table Readability: Tables which are deemed non-readable (due to foreign language, poor formatting etc - a very small fraction of 0.5%) are removed from the dataset here.\n2. Cell Highlighting: The annotator highlights the cells of the table which support the description.\n3. Deletion: The annotator removes phrases in the description which are not supported by the highlighted cells\n4. Decontextualization: Descriptions may contain pronouns or other forms of anaphora, or other phenomena which depend on the overall article topic - these are fixed by replacement (e.g replacing pronouns with the entity, provided it occurs in the table). The replacements allowed are limited to one, and annotators are also instructed to conserve fluency.\n5. Secondary Annotation: A second set of annotators is shown the output of Stage 4, and asked to fix it if required to ensure it is grammatical.\n\n\nThe paper does not specifically describe the annotation platform or location profiles of the annotators.", "#### Was Data Filtered?\n\n\nalgorithmically", "#### Filter Criteria\n\n\nAfter construction of the splits, the data curators filtered training examples that had rare table header combinations (<=5 examples) and which had an overlap with the validation or test splits.", "### Structured Annotations", "#### Additional Annotations?\n\n\nnone", "#### Annotation Service?\n\n\nno", "### Consent", "#### Any Consent Policy?\n\n\nyes", "#### Consent Policy Details\n\n\nAnnotators were full time employees that were aware of the goal of the project and consented to having the data released as part of the dataset.", "### Private Identifying Information (PII)", "#### Contains PII?\n\n\nno PII", "#### Justification for no PII\n\n\nSince the source data is from wikipedia, only data in the public domain is included in the dataset.", "### Maintenance", "#### Any Maintenance Plan?\n\n\nyes", "#### Maintenance Plan Details\n\n\nFor submissions, you can delete your data by emailing totto@URL from the email account used to sign up for the submission. Deletion requests will be responded to within 60 days.", "#### Maintainer Contact Information\n\n\nAnkur Parikh (aparikh@URL)", "#### Any Contestation Mechanism?\n\n\nform submission", "#### Contestation Form Link\n\n\ntotto@URL\n\n\nBroader Social Context\n----------------------", "### Previous Work on the Social Impact of the Dataset", "#### Usage of Models based on the Data\n\n\nno", "### Impact on Under-Served Communities", "#### Addresses needs of underserved Communities?\n\n\nno", "### Discussion of Biases", "#### Any Documented Social Biases?\n\n\nyes", "#### Links and Summaries of Analysis Work\n\n\nThe original work as well as our GEM paper analyzes some biases", "#### Are the Language Producers Representative of the Language?\n\n\nThis dataset is created using tables and the table cell contents may hence naturally exhibit biases which have been found to exist in Wikipedia such as some forms of gender bias (e.g (Graells-Garido et al.,2015) notes that spouse information is more likely discussed for females than males)\n\n\nThe table descriptions (targets/references) are, as discussed earlier, collected through a two-step process.\n\n\n1. The natural text description near the table is taken as a starting point. This is Wikipedia article text as created upto that point in time by a chain of collaborative edits from Wikipedia authors.\n2. The initial description is revised by chain of two or more annotated revisions, to make it unambiguous and faithful to a set of highlighted table cells.\n\n\nFrom their origin in 1), the descriptions may exhibit biases seen in Wikipedia text as mentioned above. From their revisions in 2), the descriptions may show biases originating from annotator-authored text, such as a preference for shorter descriptions since they're faster to write, or linguistic preferences influenced by the locations dominant in the annotator distribution. (However, note that these are likely to be much reduced since the annotators here are merely revising rather than completely authoring. Moreover, each sentence goes through atleast two annotators, which acts as a check against the personal biases of a single annotator.)\n\n\nNaturally-occurring text is also known to suffer from other biases such as reporting bias (Gordon and Van Durme, 2013) - this also applies to this dataset via its origin from Wikipedia.\n\n\nConsiderations for Using the Data\n---------------------------------", "### PII Risks and Liability", "#### Potential PII Risk\n\n\nSince the source data is from wikipedia, only data in the public domain is included in the dataset.", "### Licenses", "#### Copyright Restrictions on the Dataset\n\n\n'open license - commercial use allowed'", "#### Copyright Restrictions on the Language Data\n\n\n'open license - commercial use allowed'", "### Known Technical Limitations", "#### Technical Limitations\n\n\nThe dataset is limited to topics that are present in Wikipedia, more specifically those topics that are present in articles which contain atleast one table\n*Sports* and *Countries* form 53.4% of the dataset. The remaining fraction is made up of broader topics like *Europe*, *North America*and *Politics*" ]

[ "TAGS\n#task_categories-table-to-text #annotations_creators-none #language_creators-unknown #multilinguality-unknown #size_categories-unknown #source_datasets-original #language-English #license-cc-by-sa-3.0 #data-to-text #arxiv-1603.07771 #arxiv-2007.02871 #arxiv-2005.10433 #region-us \n", "### Link to Main Data Card\n\n\nYou can find the main data card on the [GEM Website.", "### Dataset Summary\n\n\nToTTo is a high-quality English table-to-text dataset with more than 100,000 examples in which a table from Wikipedia with highlighted cells is paired with a sentence that describes the highlighted cells. All examples in the dataset were post-edited in multiple steps to ensure that the targets are fully faithful to the input information.\n\n\nYou can load the dataset via:\n\n\nThe data loader can be found here.", "#### website\n\n\nn/a", "#### paper\n\n\nACL Anthology", "#### authors\n\n\nAnkur Parikh, Xuezhi Wang, Sebastian Gehrmann, Manaal Faruqui, Bhuwan Dhingra, Diyi Yang, Dipanjan Das\n\n\nDataset Overview\n----------------", "### Where to find the Data and its Documentation", "#### Download\n\n\nToTTo Main Repo + ToTTo Supplementary Repo", "#### Paper\n\n\nACL Anthology", "#### BibTex", "#### Contact Name\n\n\nAnkur Parikh", "#### Contact Email\n\n\ntotto@URL", "#### Has a Leaderboard?\n\n\nyes", "#### Leaderboard Link\n\n\nGithub", "#### Leaderboard Details\n\n\nThis dataset has an associated, active leaderboard maintained by the authors.\nThe test set ground truth targets / references are private, i.e they are not publicly shared or downloadable - hence, leaderboard submission is necessary for test set evaluation.\nTo evaluate your model on the dev or test set AND/OR submit to the leaderboard, you need to submit your model files through this form (The form provides an option to opt-out of going on the leaderboard).\n\n\nThe leaderboard reports three sets of BLEU, PARENT and BLEURT scores for each submission - on the overall test set, the *Overlap* subset of the test set and the *non-Overlap* subset of the test set.", "### Languages and Intended Use", "#### Multilingual?\n\n\nno", "#### Covered Dialects\n\n\nNo specific dialects. The original language is from Wikipedia and it was post-edited by crowdraters", "#### Covered Languages\n\n\n'English'", "#### Whose Language?\n\n\nThe language is post-edited English only (BCP-47: 'en') Wikipedia text. No demographic information about annotators is provided.\nSome amounts of what may be called non-English text, including characters such as French accents or Cyrillic characters, could sometimes occur, especially through fields with entity names as values in the input table cells.", "#### License\n\n\ncc-by-sa-3.0: Creative Commons Attribution Share Alike 3.0 Unported", "#### Intended Use\n\n\nToTTo is a Table-to-Text NLG task, as the paper title says. The task is as follows: Given a Wikipedia table with row names, column names and table cells, with a subset of cells highlighted, generate a natural language description for the highlighted part of the table . The table need not be exactly rectangular in that - cells can sometimes be multi-row or multi-column.\n\n\nAn earlier example of a Table-to-Text NLG task is Wikibio - here the inputs were Wikipedia infoboxes (from the top right corner of entity-related Wiki pages). In contrast, ToTTo mostly has Wikipedia tables from the main article content itself. In general, Table-To-Text NLG tasks can be seen as a subclass of Data-To-Text NLG tasks - where the task is to generate natural language descriptions of inputs which are in the form of structured or semi-structured data. In general, all Data-To-Text NLG tasks need not have an explicit table or other structure - e.g the input in WebNLG is simply a list of triples.\n\n\nImportantly, ToTTo differs from earlier examples of Table-To-Text NLG in that:\n\n\n1. It does not suffer from the problem of divergent references - where ground truth descriptions themselves have additional information not found in the table. ToTTo overcomes this by having a multi-step annotation process to edit the initial, free-form table descriptions (which are from Wikipedia) to make them faithful, unambiguous and independent of article context.\n2. Since it provides control in the form of highlighted table cells, it prevents the problem of there being a large number of valid descriptions focussing on different parts of the table.", "#### Primary Task\n\n\nData-to-Text", "#### Communicative Goal\n\n\nThe speaker is required to produce a single, coherent English sentence that describes the highlighted cells in the given table, also using metadata and any other information from the table as applicable.", "### Credit", "#### Curation Organization Type(s)\n\n\n'industry'", "#### Curation Organization(s)\n\n\nGoogle Research", "#### Dataset Creators\n\n\nAnkur Parikh, Xuezhi Wang, Sebastian Gehrmann, Manaal Faruqui, Bhuwan Dhingra, Diyi Yang, Dipanjan Das", "#### Funding\n\n\nGoogle Research", "#### Who added the Dataset to GEM?\n\n\nVarun Gangal created the initial data card and Yacine Jernite wrote the data loader. The data card was updated with new splits by Simon Mille. Sebastian Gehrmann ported the data card and loader from the v1 to the v2 version and extended it with the new fields.", "### Dataset Structure", "#### Data Fields\n\n\n* The 'table' field is a 'List[List[Dict]]' in row-major order, with outer lists representing rows and the inner lists columns.\n* Each 'Dict' has the fields 'column\\_span: int', 'is\\_header: bool', 'row\\_span: int', and 'value: str'.\n* Table metadata consists of 'table\\_page\\_title', 'table\\_section\\_title' and 'table\\_section\\_texts'\n* The 'highlighted\\_cells' are represented as 'List[[row\\_index,column\\_index]]', with each '[row\\_index,column\\_index]' indicating that 'table[row\\_index][column\\_index]' is highlighted.\n* 'example\\_id' is the unique id per example.\n* 'sentence\\_annotations[final\\_sentence]' which is the table description/generation target", "#### Reason for Structure\n\n\nThe structure is aimed to encode highlighted tables in a way that allows rows and columns to span multiple fields in width. The other fields are meta-data about the source and the annotations", "#### How were labels chosen?\n\n\nThe initial table-description pairs are tables from Wikipedia articles, extracted through heuristics such as Number Matching (tables and sentences that overlap with a non-date number of atleast 3 non-zero digits) (Refer to Section 4 of the paper for more)\n\n\n1. Table Readability: Tables which are deemed non-readable (due to foreign language, poor formatting etc - a very small fraction of 0.5%) are removed from the dataset here.\n2. Cell Highlighting: The annotator highlights the cells of the table which support the description.\n3. Deletion: The annotator removes phrases in the description which are not supported by the highlighted cells\n4. Decontextualization: Descriptions may contain pronouns or other forms of anaphora, or other phenomena which depend on the overall article topic - these are fixed by replacement (e.g replacing pronouns with the entity, provided it occurs in the table). The replacements allowed are limited to one, and annotators are also instructed to conserve fluency.\n5. Secondary Annotation: A second set of annotators is shown the output of Stage 4, and asked to fix it if required to ensure it is grammatical.", "#### Example Instance\n\n\nThe main repository's 'URL' already provides a thorough walkthrough of data instances and fields here\n\n\nBelow is the instance for a table from the wiki-page for the musical artist *Weird Al' Yankovic* , likely listing his on-television appearances.\n\n\nThe 'table' field is expanded as below:\n\n\nThe Supplementary Repo also provides browsable samples under its 'sample/' folder. It additionally provides HTML visualization scripts with their outputs located under the aforementioned folder. The instructions to access and visualize these samples can also be found here.", "#### Data Splits\n\n\nThe dataset consists of 120,000 train examples and equi-sized dev and test sets with 7700 examples.\nRefer to Table 5 in the paper for a more extensive list of properties about table size, target vocabulary etc and their aggregates.", "#### Splitting Criteria\n\n\nThe dev and test splits are further equally distributed between *Overlap* and *non-Overlap* .\nThe examples in the *Overlap* set are harder on account of the domain shift resulting from them having none of their header (row and column) names in common with those seen during training.\n\n\nRefer to Table 5 in the paper for a more extensive list of properties about table size, target vocabulary etc and their aggregates.", "#### \n\n\nThere are some very large tables in the dataset with thousands of rows. Table 7 shows some of the challenges of the dataset, showing that very few examples require access to the table description itself which makes those examples an outlier.\n\n\nDataset in GEM\n--------------", "### Rationale for Inclusion in GEM", "#### Why is the Dataset in GEM?\n\n\nToTTo is one of the two datasets representing Table-to-Text NLG in GEM, the other one being DART. Unlike DART, which combines datasets from multiple sources and furnishes them in a unified setting, ToTTo is from a homogeneous source. As explained in the Task Summary above, it also has an annotation process explicitly crafted to reduce divergent descriptions, which is not true of DART.\n\n\nFurthermore, ToTTo is also an instance of a controlled generation task - where in addition to the input (in this case the table) an additional control (in this case the highlighted cells) is given as an additional goal for the generation. The DART task formulation does not include controls.", "#### Similar Datasets\n\n\nyes", "#### Unique Language Coverage\n\n\nno", "#### Difference from other GEM datasets\n\n\nThe input is much more complex and the quality much better than that of comparable datasets. The highlighted table cells provide a unique challenge to models.", "#### Ability that the Dataset measures\n\n\nReasoning, surface realization", "### GEM-Specific Curation", "#### Modificatied for GEM?\n\n\nyes", "#### Additional Splits?\n\n\nyes", "#### Split Information\n\n\n9 challenge sets for ToTTo were added to the GEM evaluation suite, 8 created specifically for the task and 1 coming from the original data.\n\n\n1. We created subsets of the training and development sets of 500 randomly selected inputs each.\n2. We applied input scrambling on a subset of 500 randomly selected test instances; the order of the highlighted cells was randomly reassigned.\n3. For the input size, we created subpopulations based on the number of input highlighted cells in the whole table.\n\n\n\n4. We also divided the test set according to the size of the whole table, based on the idea that larger tables represent a bigger space to take into account when generating the highlighted cells; a larger table could be more challenging to generate accurate text than a smaller table. There are 693 different table sizes, ranging from 2 to 15834 cells.\n\n\n\n5. We also created three splits based on the subset of test examples in pages about people.\nWe then used the structured information in WikiData to identify the following information:\n\n\n* gender (male, and female),\n* nationality grouped by continent (Africa, Asia, Europe, North America, Oceania, and South America)\n* ethnicity (African American and all USA)\n\n\nThe categories within gender, ethnicity, and nationality were chosen based on data availability; The ToTTo dataset includes mostly tables that do not focus on people. As a result, only seven people in the original test set are marked as having a non-binary gender. Similar sparsity informed the grouping of nationalities by continent – only 19 countries are represented by more than 10 people in the test set. In case a person has citizenships across multiple continents, we may include the person in any of the included continents.\n\n\nFinally, ethnicity is very sparsely annotated in WikiData; only 150 test examples in ToTTo have this information and 128 of these are African Americans. We thus are unable to compare the performance on, e.g., Yoruba or Punjabi people, both of which have fewer than five instances. Another caveat here is that only 21 of the 128 people are female. We thus compare the African American population to results on a subset that includes all US citizens.", "#### Split Motivation\n\n\ngeneralization, fairness, robustness", "### Getting Started with the Task", "#### Pointers to Resources\n\n\n* The highest spot on the leaderboard is currently held by an anonymous method, with BLEU=49.2, PARENT=58.7 and BLEURT=0.249 on the *Overall* test set.\n* The highest scoring non-anonymous method is the T5-based method of Kale, 2020. This method uses a simple row-major linearization scheme to convert the table (it chooses only the highlighted cells and ignores the other cells - table titles and section titles are prefixed at the start of the respective section table) to a flat string. The linearized input - output description pairs from training examples are then used to finetune T5, with BLEU being used as the dev metric to pick checkpoints, and beam search with beam size 10 being the decoding method.\n\n\nThough the best numbers from this method are naturally from the largest T5-pretrained architecture (T5-3B), the paper shows improvements over the next-highest BERT-to-BERT method even when using T5-Base or T5-Small, which have the same and lesser parameters than BERT-to-BERT respectively.\n* The Supplementary Repo provides several useful modules to get started with for new approach implementation:\n\n\n\t1. Code for the particular preprocessing / linearization scheme used to linearize the tables into flat sequences for the baseline approaches described in the paper has been described and shared herein\n\t2. An evaluation script for locally scoring BLEU and PARENT system outputs on dev (or train) sets. Since BLEURT is a model-based metric, a slightly separate set of instructions is provided to evaluate on the same.\n\n\nPrevious Results\n----------------", "### Previous Results", "#### Measured Model Abilities\n\n\nReasoning, surface realization", "#### Metrics\n\n\n'BLEU', 'BLEURT', 'Other: Other Metrics'", "#### Other Metrics\n\n\nParent: a metric that measures the F-1 score of overlap between input content words and those used in references and those in generated text while ignoring the general surface form. It can thus measure the faithfulness much better than metrics that measure overlap with a reference", "#### Proposed Evaluation\n\n\nThe metrics are used as in the leaderboard. The original paper additionally conducted a human evaluation focusing on fluency, faithfulness, and coverage.\nFaithfulness was measured as whether facts in the text are not supported by the input, and coverage as the number of highlighted cells that were considered. They thus represent precision and recall of the content.", "#### Previous results available?\n\n\nyes", "#### Relevant Previous Results\n\n\nSee leaderboard.\n\n\nDataset Curation\n----------------", "### Original Curation", "#### Original Curation Rationale\n\n\nTables occurring in Wikipedia articles were chosen as the data source with the following reasons in mind:\n\n\n1. Wide coverage in terms of both vocabulary and concepts.\n2. Wikipedia tables are not confined to a regular structure, with multi-row or multi-column cells occurring with a sufficient frequency.\n3. Likely to contain reasonable-quality, natural text descriptions in the proximity of the table, which are also extractable by heuristics. (see the start of Section 4 for the heuristics used)\n\n\nTo prevent an overlap with the earlier Wikibio dataset which focussed on Infobox-first sentence pairs from Wikipedia biography articles, the authors avoid using Infoboxes as a data source.\n\n\nThe overall curation process of initially collecting free text and then annotator-revising it, was designed to combine the advantages of free-form text descriptions (which are fluent, high-quality and unhurriedly written, but also divergent and unfaithful) with annotator descriptions (which can be tailored to be faithful and to conform exactly to desired task requirements)", "#### Communicative Goal\n\n\nThe speaker is required to produce a single, coherent English sentence that describes the highlighted cells in the given table, also using metadata and any other information from the table as applicable.", "#### Sourced from Different Sources\n\n\nyes", "#### Source Details\n\n\nURL", "### Language Data", "#### How was Language Data Obtained?\n\n\n'Crowdsourced'", "#### Where was it crowdsourced?\n\n\n'Other crowdworker platform'", "#### Language Producers\n\n\nThe basic source language producers are Wikipedia authors and/or editors, since the annotation starts with the natural text description near the Wikipedia table.\nThe auxiliary source language producers are the annotators (two per example) who iteratively revise these descriptions to make them unambiguous and faithful to a subset of highlighted cells in the table.", "#### Data Validation\n\n\nvalidated by crowdworker", "#### Data Preprocessing\n\n\nThe initial table-description pairs are tables from Wikipedia articles, extracted through heuristics such as Number Matching (tables and sentences that overlap with a non-date number of atleast 3 non-zero digits) (Refer to Section 4 of the paper for more)\n\n\n1. Table Readability: Tables which are deemed non-readable (due to foreign language, poor formatting etc - a very small fraction of 0.5%) are removed from the dataset here.\n2. Cell Highlighting: The annotator highlights the cells of the table which support the description.\n3. Deletion: The annotator removes phrases in the description which are not supported by the highlighted cells\n4. Decontextualization: Descriptions may contain pronouns or other forms of anaphora, or other phenomena which depend on the overall article topic - these are fixed by replacement (e.g replacing pronouns with the entity, provided it occurs in the table). The replacements allowed are limited to one, and annotators are also instructed to conserve fluency.\n5. Secondary Annotation: A second set of annotators is shown the output of Stage 4, and asked to fix it if required to ensure it is grammatical.\n\n\nThe paper does not specifically describe the annotation platform or location profiles of the annotators.", "#### Was Data Filtered?\n\n\nalgorithmically", "#### Filter Criteria\n\n\nAfter construction of the splits, the data curators filtered training examples that had rare table header combinations (<=5 examples) and which had an overlap with the validation or test splits.", "### Structured Annotations", "#### Additional Annotations?\n\n\nnone", "#### Annotation Service?\n\n\nno", "### Consent", "#### Any Consent Policy?\n\n\nyes", "#### Consent Policy Details\n\n\nAnnotators were full time employees that were aware of the goal of the project and consented to having the data released as part of the dataset.", "### Private Identifying Information (PII)", "#### Contains PII?\n\n\nno PII", "#### Justification for no PII\n\n\nSince the source data is from wikipedia, only data in the public domain is included in the dataset.", "### Maintenance", "#### Any Maintenance Plan?\n\n\nyes", "#### Maintenance Plan Details\n\n\nFor submissions, you can delete your data by emailing totto@URL from the email account used to sign up for the submission. Deletion requests will be responded to within 60 days.", "#### Maintainer Contact Information\n\n\nAnkur Parikh (aparikh@URL)", "#### Any Contestation Mechanism?\n\n\nform submission", "#### Contestation Form Link\n\n\ntotto@URL\n\n\nBroader Social Context\n----------------------", "### Previous Work on the Social Impact of the Dataset", "#### Usage of Models based on the Data\n\n\nno", "### Impact on Under-Served Communities", "#### Addresses needs of underserved Communities?\n\n\nno", "### Discussion of Biases", "#### Any Documented Social Biases?\n\n\nyes", "#### Links and Summaries of Analysis Work\n\n\nThe original work as well as our GEM paper analyzes some biases", "#### Are the Language Producers Representative of the Language?\n\n\nThis dataset is created using tables and the table cell contents may hence naturally exhibit biases which have been found to exist in Wikipedia such as some forms of gender bias (e.g (Graells-Garido et al.,2015) notes that spouse information is more likely discussed for females than males)\n\n\nThe table descriptions (targets/references) are, as discussed earlier, collected through a two-step process.\n\n\n1. The natural text description near the table is taken as a starting point. This is Wikipedia article text as created upto that point in time by a chain of collaborative edits from Wikipedia authors.\n2. The initial description is revised by chain of two or more annotated revisions, to make it unambiguous and faithful to a set of highlighted table cells.\n\n\nFrom their origin in 1), the descriptions may exhibit biases seen in Wikipedia text as mentioned above. From their revisions in 2), the descriptions may show biases originating from annotator-authored text, such as a preference for shorter descriptions since they're faster to write, or linguistic preferences influenced by the locations dominant in the annotator distribution. (However, note that these are likely to be much reduced since the annotators here are merely revising rather than completely authoring. Moreover, each sentence goes through atleast two annotators, which acts as a check against the personal biases of a single annotator.)\n\n\nNaturally-occurring text is also known to suffer from other biases such as reporting bias (Gordon and Van Durme, 2013) - this also applies to this dataset via its origin from Wikipedia.\n\n\nConsiderations for Using the Data\n---------------------------------", "### PII Risks and Liability", "#### Potential PII Risk\n\n\nSince the source data is from wikipedia, only data in the public domain is included in the dataset.", "### Licenses", "#### Copyright Restrictions on the Dataset\n\n\n'open license - commercial use allowed'", "#### Copyright Restrictions on the Language Data\n\n\n'open license - commercial use allowed'", "### Known Technical Limitations", "#### Technical Limitations\n\n\nThe dataset is limited to topics that are present in Wikipedia, more specifically those topics that are present in articles which contain atleast one table\n*Sports* and *Countries* form 53.4% of the dataset. The remaining fraction is made up of broader topics like *Europe*, *North America*and *Politics*" ]

[ "task_categories:table-to-text", "annotations_creators:expert-created", "language_creators:unknown", "multilinguality:unknown", "size_categories:unknown", "source_datasets:original", "language:fi", "license:cc-by-nc-sa-4.0", "data-to-text", "region:us" ]

[]

[ "fi" ]

[ "# Dataset Card for GEM/turku_hockey_data2text", "## Dataset Description\n\n- Homepage: URL\n- Repository: URL\n- Paper: URL\n- Leaderboard: N/A\n- Point of Contact: Jenna Kanerva, Filip Ginter", "### Link to Main Data Card\n\nYou can find the main data card on the GEM Website.", "### Dataset Summary \n\nThis is a Finnish data-to-text dataset in which the input is structured information about a hockey game and the output a description of the game.\n\nYou can load the dataset via:\n\nThe data loader can be found here.", "#### website\nWebsite", "#### paper\nACL anthology", "#### authors\nJenna Kanerva, Samuel Rönnqvist, Riina Kekki, Tapio Salakoski, Filip Ginter (TurkuNLP / University of Turku)", "## Dataset Overview", "### Where to find the Data and its Documentation", "#### Webpage\n\n\n\nWebsite", "#### Download\n\n\n\nGithub", "#### Paper\n\n\n\nACL anthology", "#### BibTex", "#### Contact Name\n\n\n\n\nJenna Kanerva, Filip Ginter", "#### Contact Email\n\n\n\njmnybl@URL, figint@URL", "#### Has a Leaderboard?\n\n\n\nno", "### Languages and Intended Use", "#### Multilingual?\n\n\n\n\nno", "#### Covered Dialects\n\n\n\nwritten standard language", "#### Covered Languages\n\n\n\n\n'Finnish'", "#### Whose Language?\n\n\n\nThe original news articles are written by professional journalists. The text passages extracted in the annotation may be slightly edited compared to the original language during the corpus annotation.", "#### License\n\n\n\n\ncc-by-nc-sa-4.0: Creative Commons Attribution Non Commercial Share Alike 4.0 International", "#### Intended Use\n\n\n\nThis dataset was developed as a benchmark for evaluating template-free, machine learning methods on Finnish news generation in the area of ice hockey reporting.", "#### Primary Task\n\n\n\nData-to-Text", "#### Communicative Goal\n\n\n\n\nDescribe an event from an ice hockey game based on the given structural data.", "### Credit", "#### Curation Organization Type(s)\n\n\n\n'academic'", "#### Curation Organization(s)\n\n\n\nUniversity of Turku", "#### Dataset Creators\n\n\n\nJenna Kanerva, Samuel Rönnqvist, Riina Kekki, Tapio Salakoski, Filip Ginter (TurkuNLP / University of Turku)", "#### Funding\n\n\n\nThe project was supported by the Google Digital News Innovation Fund.", "#### Who added the Dataset to GEM?\n\n\n\nJenna Kanerva, Filip Ginter (TurkuNLP / University of Turku)", "### Dataset Structure", "#### Data Fields\n\n\n\nThe dataset is constructed of games, where each game is a list of events. If the event was annotated (corresponding sentence was found from the news article), it includes 'text' field with value other than empty string (\"\").\n\nFor each game (dict), there are keys 'gem_id' (string), 'id' (string), 'news_article' (string), and 'events' (list).\n\nFor each event (dict), there are different, relevant keys available with non empty values depending on the event type (e.g. goal or penalty). The mandatory keys for each event are 'event_id' (string), 'event_type' (string), 'text' (string, empty string if not annotated), and 'multi_reference' (bool). The keys not relevant for the specific event type are left empty.\n\nThe relevant keys in the event dictionary are:\n\nFor each event type, the following keys are relevant:\n 'event_id': Identifier of the event, unique to the game but not globally, in chronological order (string)\n 'event_type': Type of the event, possible values are 'game result', 'goal', 'penalty', or 'saves' (string)\n 'text': Natural language description of the event, or empty string if not available (string)\n 'multi_reference': Does this event refer to a text passage describing multiple events? (bool)\n\n\nThe rest of the fields are specific to the event type. The relevant fields for each event type are:\n\ngame result:\n 'event_id': Identifier of the event, unique to the game but not globally, in chronological order (string)\n 'event_type': Type of the event (string)\n 'home_team': Name of the home team (string)\n 'guest_team': Name of the guest team (string)\n 'score': Final score of the game, in the form of home–guest (string)\n 'periods': Scores for individual periods, each in the form of home–guest score in that period (list of strings)\n 'features': Additional features, such as overtime win or shoot out (list of strings)\n 'text': Natural language description of the event, or empty string if not available (string)\n 'multi_reference': Does this event refer to a text passage describing multiple events? (bool)\n\ngoal:\n 'event_id': Identifier of the event, unique to the game but not globally, in chronological order (string)\n 'event_type': Type of the event (string)\n 'player': Name of the player scoring (string)\n 'assist': Names of the players assisting, at most two players (list of strings)\n 'team': Team scoring with possible values of 'home' or 'guest' (string)\n 'team_name': Name of the team scoring (string)\n 'score': Score after the goal, in the form of home–guest (string)\n 'time': Time of the goal, minutes and seconds from the beginning (string)\n 'features': Additional features, such as power play or short-handed goal (list of strings)\n 'text': Natural language description of the event, or empty string if not available (string)\n 'multi_reference': Does this event refer to a text passage describing multiple events? (bool)\n\npenalty:\n 'event_id': Identifier of the event, unique to the game but not globally, in chronological order (string)\n 'event_type': Type of the event (string)\n 'player': Name of the player getting the penalty (string)\n 'team': Team getting the penalty with possible values of 'home' or 'guest' (string)\n 'team_name': Name of the team getting the penalty (string)\n 'penalty_minutes': Penalty minutes (string)\n 'time': Time of the penalty, minutes and seconds from the beginning (string)\n 'text': Natural language description of the event, or empty string if not available (string)\n 'multi_reference': Does this event refer to a text passage describing multiple events? (bool)\n\nsaves:\n 'event_id': Identifier of the event, unique to the game but not globally, in chronological order (string)\n 'event_type': Type of the event (string)\n 'player': Name of the goalkeeper (string)\n 'team': Team of the goalkeeper with possible values of 'home' or 'guest' (string)\n 'team_name': Name of the team (string)\n 'saves': Number of saves in the game (string)\n 'text': Natural language description of the event, or empty string if not available (string)\n 'multi_reference': Does this event refer to a text passage describing multiple events? (bool)\n\n\nText passages describing multiple events (multi_reference):\n\nSome text passages refer to multiple events in such way that separating them to individual statements is not adequate (e.g. \"The home team received two penalties towards the end of the first period.\"). In these cases, multiple events are aligned to the same text passage so that the first event (in chronological order) include the annotated text passage, while the rest of the events referring to the same text passage include the identifier of the first event in the annotated text field (e.g. 'text': \"E4\").", "#### Example Instance", "#### Data Splits\n\n\n\nThe corpus include 3 splits: train, validation, and test.", "## Dataset in GEM", "### Rationale for Inclusion in GEM", "#### Why is the Dataset in GEM?\n\n\n\nThe dataset was created to develop machine learned text generation models for Finnish ice hockey news, where the generation would reflect the natural language variation found from the game reports written by professional journalists. While the original game reports often include additional information not derivable from the game statistics, the corpus was fully manually curated to remove all such information from the natural language descriptions. The rationale of such curation was to prevent model 'hallucinating' additional facts.", "#### Similar Datasets\n\n\n\nyes", "#### Unique Language Coverage\n\n\n\nyes", "#### Difference from other GEM datasets\n\n\n\nThis is the only data2text corpus for Finnish in GEM.", "#### Ability that the Dataset measures\n\n\n\nmorphological inflection, language variation", "### GEM-Specific Curation", "#### Modificatied for GEM?\n\n\n\nyes", "#### GEM Modifications\n\n\n\n'data points modified'", "#### Modification Details\n\n\n\nStructural data was translated into English.", "#### Additional Splits?\n\n\n\nno", "### Getting Started with the Task", "## Previous Results", "### Previous Results", "#### Metrics\n\n\n\n'BLEU', 'METEOR', 'ROUGE', 'WER'", "#### Proposed Evaluation\n\n\n\nAutomatic evaluation: BLEU, NIST, METEOR, ROUGE-L, CIDEr\nManual evaluation: factual mistakes, grammatical errors, minimum edit distance to an acceptable game report (using WER)", "#### Previous results available?\n\n\n\nyes", "## Dataset Curation", "### Original Curation", "#### Original Curation Rationale\n\n\n\nThe dataset is designed for text generation (data2text), where the original source of natural language descriptions is news articles written by journalists. While the link between structural data (ice hockey game statistics) and the news articles describing the game was quite weak (news articles including a lot of information not derivable from the statistics, while leaving many events unmentioned), the corpus includes full manual annotation aligning the events extracted from game statistics and the corresponding natural language passages extracted from the news articles.\n\nEach event is manually aligned into a sentence-like passage, and in case a suitable passage was not found, the annotation is left empty (with value 'None'). The extracted passages were manually modified not to include additional information not derivable from the game statistics, or not considered as world knowledge. The manual curation of passages is designed to prevent model hallucination, i.e. model learning to generate facts not derivable from the input data.", "#### Communicative Goal\n\n\n\nDescribing the given events (structural data) in natural language, and therefore generating ice hockey game reports.", "#### Sourced from Different Sources\n\n\n\nno", "### Language Data", "#### How was Language Data Obtained?\n\n\n\n'Other'", "#### Language Producers\n\n\n\nThe initial data, both game statistics and news articles, were obtained from the Finnish News Agency STT news archives released for academic use (URL The original news articles are written by professional journalists.\n\nWe (TurkuNLP) gratefully acknowledge the collaboration of Maija Paikkala, Salla Salmela and Pihla Lehmusjoki from the Finnish News Agency STT while creating the corpus.", "#### Topics Covered\n\n\n\nIce hockey, news", "#### Data Validation\n\n\n\nnot validated", "#### Was Data Filtered?\n\n\n\nalgorithmically", "#### Filter Criteria\n\n\n\nInclude only games, where both game statistics and a news article describing the game were available (based on timestamps and team names).", "### Structured Annotations", "#### Additional Annotations?\n\n\n\n\nexpert created", "#### Number of Raters\n\n\n\n1", "#### Rater Qualifications\n\n\n\nMembers of the TurkuNLP research group, native speakers of Finnish.", "#### Raters per Training Example\n\n\n\n1", "#### Raters per Test Example\n\n\n\n1", "#### Annotation Service?\n\n\n\nno", "#### Annotation Values\n\n\n\nManual alignment of events and their natural language descriptions. Removing information not derivable from the input data or world knowledge in order to prevent the model 'hallucination'.", "#### Any Quality Control?\n\n\n\nvalidated by data curators", "#### Quality Control Details\n\n\n\nManual inspection of examples during the initial annotation training phrase.", "### Consent", "#### Any Consent Policy?\n\n\n\nyes", "#### Consent Policy Details\n\n\n\nThe corpus license was agreed with the providers of the source material.", "### Private Identifying Information (PII)", "#### Contains PII?\n\n\n\n\nyes/very likely", "#### Categories of PII\n\n\n\n'generic PII'", "#### Any PII Identification?\n\n\n\nno identification", "### Maintenance", "#### Any Maintenance Plan?\n\n\n\nno", "## Broader Social Context", "### Previous Work on the Social Impact of the Dataset", "#### Usage of Models based on the Data\n\n\n\nno", "### Impact on Under-Served Communities", "#### Addresses needs of underserved Communities?\n\n\n\nno", "### Discussion of Biases", "#### Any Documented Social Biases?\n\n\n\nno", "#### Are the Language Producers Representative of the Language?\n\n\n\nThe dataset represents only written standard language.", "## Considerations for Using the Data", "### PII Risks and Liability", "#### Potential PII Risk\n\n\n\nNone", "### Licenses", "#### Copyright Restrictions on the Dataset\n\n\n\n'non-commercial use only'", "#### Copyright Restrictions on the Language Data\n\n\n\n'non-commercial use only'", "### Known Technical Limitations" ]

[ "TAGS\n#task_categories-table-to-text #annotations_creators-expert-created #language_creators-unknown #multilinguality-unknown #size_categories-unknown #source_datasets-original #language-Finnish #license-cc-by-nc-sa-4.0 #data-to-text #region-us \n", "# Dataset Card for GEM/turku_hockey_data2text", "## Dataset Description\n\n- Homepage: URL\n- Repository: URL\n- Paper: URL\n- Leaderboard: N/A\n- Point of Contact: Jenna Kanerva, Filip Ginter", "### Link to Main Data Card\n\nYou can find the main data card on the GEM Website.", "### Dataset Summary \n\nThis is a Finnish data-to-text dataset in which the input is structured information about a hockey game and the output a description of the game.\n\nYou can load the dataset via:\n\nThe data loader can be found here.", "#### website\nWebsite", "#### paper\nACL anthology", "#### authors\nJenna Kanerva, Samuel Rönnqvist, Riina Kekki, Tapio Salakoski, Filip Ginter (TurkuNLP / University of Turku)", "## Dataset Overview", "### Where to find the Data and its Documentation", "#### Webpage\n\n\n\nWebsite", "#### Download\n\n\n\nGithub", "#### Paper\n\n\n\nACL anthology", "#### BibTex", "#### Contact Name\n\n\n\n\nJenna Kanerva, Filip Ginter", "#### Contact Email\n\n\n\njmnybl@URL, figint@URL", "#### Has a Leaderboard?\n\n\n\nno", "### Languages and Intended Use", "#### Multilingual?\n\n\n\n\nno", "#### Covered Dialects\n\n\n\nwritten standard language", "#### Covered Languages\n\n\n\n\n'Finnish'", "#### Whose Language?\n\n\n\nThe original news articles are written by professional journalists. The text passages extracted in the annotation may be slightly edited compared to the original language during the corpus annotation.", "#### License\n\n\n\n\ncc-by-nc-sa-4.0: Creative Commons Attribution Non Commercial Share Alike 4.0 International", "#### Intended Use\n\n\n\nThis dataset was developed as a benchmark for evaluating template-free, machine learning methods on Finnish news generation in the area of ice hockey reporting.", "#### Primary Task\n\n\n\nData-to-Text", "#### Communicative Goal\n\n\n\n\nDescribe an event from an ice hockey game based on the given structural data.", "### Credit", "#### Curation Organization Type(s)\n\n\n\n'academic'", "#### Curation Organization(s)\n\n\n\nUniversity of Turku", "#### Dataset Creators\n\n\n\nJenna Kanerva, Samuel Rönnqvist, Riina Kekki, Tapio Salakoski, Filip Ginter (TurkuNLP / University of Turku)", "#### Funding\n\n\n\nThe project was supported by the Google Digital News Innovation Fund.", "#### Who added the Dataset to GEM?\n\n\n\nJenna Kanerva, Filip Ginter (TurkuNLP / University of Turku)", "### Dataset Structure", "#### Data Fields\n\n\n\nThe dataset is constructed of games, where each game is a list of events. If the event was annotated (corresponding sentence was found from the news article), it includes 'text' field with value other than empty string (\"\").\n\nFor each game (dict), there are keys 'gem_id' (string), 'id' (string), 'news_article' (string), and 'events' (list).\n\nFor each event (dict), there are different, relevant keys available with non empty values depending on the event type (e.g. goal or penalty). The mandatory keys for each event are 'event_id' (string), 'event_type' (string), 'text' (string, empty string if not annotated), and 'multi_reference' (bool). The keys not relevant for the specific event type are left empty.\n\nThe relevant keys in the event dictionary are:\n\nFor each event type, the following keys are relevant:\n 'event_id': Identifier of the event, unique to the game but not globally, in chronological order (string)\n 'event_type': Type of the event, possible values are 'game result', 'goal', 'penalty', or 'saves' (string)\n 'text': Natural language description of the event, or empty string if not available (string)\n 'multi_reference': Does this event refer to a text passage describing multiple events? (bool)\n\n\nThe rest of the fields are specific to the event type. The relevant fields for each event type are:\n\ngame result:\n 'event_id': Identifier of the event, unique to the game but not globally, in chronological order (string)\n 'event_type': Type of the event (string)\n 'home_team': Name of the home team (string)\n 'guest_team': Name of the guest team (string)\n 'score': Final score of the game, in the form of home–guest (string)\n 'periods': Scores for individual periods, each in the form of home–guest score in that period (list of strings)\n 'features': Additional features, such as overtime win or shoot out (list of strings)\n 'text': Natural language description of the event, or empty string if not available (string)\n 'multi_reference': Does this event refer to a text passage describing multiple events? (bool)\n\ngoal:\n 'event_id': Identifier of the event, unique to the game but not globally, in chronological order (string)\n 'event_type': Type of the event (string)\n 'player': Name of the player scoring (string)\n 'assist': Names of the players assisting, at most two players (list of strings)\n 'team': Team scoring with possible values of 'home' or 'guest' (string)\n 'team_name': Name of the team scoring (string)\n 'score': Score after the goal, in the form of home–guest (string)\n 'time': Time of the goal, minutes and seconds from the beginning (string)\n 'features': Additional features, such as power play or short-handed goal (list of strings)\n 'text': Natural language description of the event, or empty string if not available (string)\n 'multi_reference': Does this event refer to a text passage describing multiple events? (bool)\n\npenalty:\n 'event_id': Identifier of the event, unique to the game but not globally, in chronological order (string)\n 'event_type': Type of the event (string)\n 'player': Name of the player getting the penalty (string)\n 'team': Team getting the penalty with possible values of 'home' or 'guest' (string)\n 'team_name': Name of the team getting the penalty (string)\n 'penalty_minutes': Penalty minutes (string)\n 'time': Time of the penalty, minutes and seconds from the beginning (string)\n 'text': Natural language description of the event, or empty string if not available (string)\n 'multi_reference': Does this event refer to a text passage describing multiple events? (bool)\n\nsaves:\n 'event_id': Identifier of the event, unique to the game but not globally, in chronological order (string)\n 'event_type': Type of the event (string)\n 'player': Name of the goalkeeper (string)\n 'team': Team of the goalkeeper with possible values of 'home' or 'guest' (string)\n 'team_name': Name of the team (string)\n 'saves': Number of saves in the game (string)\n 'text': Natural language description of the event, or empty string if not available (string)\n 'multi_reference': Does this event refer to a text passage describing multiple events? (bool)\n\n\nText passages describing multiple events (multi_reference):\n\nSome text passages refer to multiple events in such way that separating them to individual statements is not adequate (e.g. \"The home team received two penalties towards the end of the first period.\"). In these cases, multiple events are aligned to the same text passage so that the first event (in chronological order) include the annotated text passage, while the rest of the events referring to the same text passage include the identifier of the first event in the annotated text field (e.g. 'text': \"E4\").", "#### Example Instance", "#### Data Splits\n\n\n\nThe corpus include 3 splits: train, validation, and test.", "## Dataset in GEM", "### Rationale for Inclusion in GEM", "#### Why is the Dataset in GEM?\n\n\n\nThe dataset was created to develop machine learned text generation models for Finnish ice hockey news, where the generation would reflect the natural language variation found from the game reports written by professional journalists. While the original game reports often include additional information not derivable from the game statistics, the corpus was fully manually curated to remove all such information from the natural language descriptions. The rationale of such curation was to prevent model 'hallucinating' additional facts.", "#### Similar Datasets\n\n\n\nyes", "#### Unique Language Coverage\n\n\n\nyes", "#### Difference from other GEM datasets\n\n\n\nThis is the only data2text corpus for Finnish in GEM.", "#### Ability that the Dataset measures\n\n\n\nmorphological inflection, language variation", "### GEM-Specific Curation", "#### Modificatied for GEM?\n\n\n\nyes", "#### GEM Modifications\n\n\n\n'data points modified'", "#### Modification Details\n\n\n\nStructural data was translated into English.", "#### Additional Splits?\n\n\n\nno", "### Getting Started with the Task", "## Previous Results", "### Previous Results", "#### Metrics\n\n\n\n'BLEU', 'METEOR', 'ROUGE', 'WER'", "#### Proposed Evaluation\n\n\n\nAutomatic evaluation: BLEU, NIST, METEOR, ROUGE-L, CIDEr\nManual evaluation: factual mistakes, grammatical errors, minimum edit distance to an acceptable game report (using WER)", "#### Previous results available?\n\n\n\nyes", "## Dataset Curation", "### Original Curation", "#### Original Curation Rationale\n\n\n\nThe dataset is designed for text generation (data2text), where the original source of natural language descriptions is news articles written by journalists. While the link between structural data (ice hockey game statistics) and the news articles describing the game was quite weak (news articles including a lot of information not derivable from the statistics, while leaving many events unmentioned), the corpus includes full manual annotation aligning the events extracted from game statistics and the corresponding natural language passages extracted from the news articles.\n\nEach event is manually aligned into a sentence-like passage, and in case a suitable passage was not found, the annotation is left empty (with value 'None'). The extracted passages were manually modified not to include additional information not derivable from the game statistics, or not considered as world knowledge. The manual curation of passages is designed to prevent model hallucination, i.e. model learning to generate facts not derivable from the input data.", "#### Communicative Goal\n\n\n\nDescribing the given events (structural data) in natural language, and therefore generating ice hockey game reports.", "#### Sourced from Different Sources\n\n\n\nno", "### Language Data", "#### How was Language Data Obtained?\n\n\n\n'Other'", "#### Language Producers\n\n\n\nThe initial data, both game statistics and news articles, were obtained from the Finnish News Agency STT news archives released for academic use (URL The original news articles are written by professional journalists.\n\nWe (TurkuNLP) gratefully acknowledge the collaboration of Maija Paikkala, Salla Salmela and Pihla Lehmusjoki from the Finnish News Agency STT while creating the corpus.", "#### Topics Covered\n\n\n\nIce hockey, news", "#### Data Validation\n\n\n\nnot validated", "#### Was Data Filtered?\n\n\n\nalgorithmically", "#### Filter Criteria\n\n\n\nInclude only games, where both game statistics and a news article describing the game were available (based on timestamps and team names).", "### Structured Annotations", "#### Additional Annotations?\n\n\n\n\nexpert created", "#### Number of Raters\n\n\n\n1", "#### Rater Qualifications\n\n\n\nMembers of the TurkuNLP research group, native speakers of Finnish.", "#### Raters per Training Example\n\n\n\n1", "#### Raters per Test Example\n\n\n\n1", "#### Annotation Service?\n\n\n\nno", "#### Annotation Values\n\n\n\nManual alignment of events and their natural language descriptions. Removing information not derivable from the input data or world knowledge in order to prevent the model 'hallucination'.", "#### Any Quality Control?\n\n\n\nvalidated by data curators", "#### Quality Control Details\n\n\n\nManual inspection of examples during the initial annotation training phrase.", "### Consent", "#### Any Consent Policy?\n\n\n\nyes", "#### Consent Policy Details\n\n\n\nThe corpus license was agreed with the providers of the source material.", "### Private Identifying Information (PII)", "#### Contains PII?\n\n\n\n\nyes/very likely", "#### Categories of PII\n\n\n\n'generic PII'", "#### Any PII Identification?\n\n\n\nno identification", "### Maintenance", "#### Any Maintenance Plan?\n\n\n\nno", "## Broader Social Context", "### Previous Work on the Social Impact of the Dataset", "#### Usage of Models based on the Data\n\n\n\nno", "### Impact on Under-Served Communities", "#### Addresses needs of underserved Communities?\n\n\n\nno", "### Discussion of Biases", "#### Any Documented Social Biases?\n\n\n\nno", "#### Are the Language Producers Representative of the Language?\n\n\n\nThe dataset represents only written standard language.", "## Considerations for Using the Data", "### PII Risks and Liability", "#### Potential PII Risk\n\n\n\nNone", "### Licenses", "#### Copyright Restrictions on the Dataset\n\n\n\n'non-commercial use only'", "#### Copyright Restrictions on the Language Data\n\n\n\n'non-commercial use only'", "### Known Technical Limitations" ]

[ "task_categories:other", "annotations_creators:expert-created", "language_creators:unknown", "multilinguality:unknown", "size_categories:unknown", "source_datasets:original", "language:fi", "license:cc-by-sa-4.0", "paraphrasing", "region:us" ]

[]

[ "fi" ]

[ "# Dataset Card for GEM/turku_paraphrase_corpus", "## Dataset Description\n\n- Homepage: URL\n- Repository: URL\n- Paper: URL\n- Leaderboard: N/A\n- Point of Contact: Jenna Kanerva, Filip Ginter", "### Link to Main Data Card\n\nYou can find the main data card on the GEM Website.", "### Dataset Summary \n\nThis is a Finnish paraphrase corpus which consists of pairs of text passages, where a typical passage is about a sentence long. It can be used to either identify or generate paraphrases.\n\nYou can load the dataset via:\n\nThe data loader can be found here.", "#### website\nWebsite", "#### paper\nACL Anthology", "#### authors\nJenna Kanerva, Filip Ginter, Li-Hsin Chang, Iiro Rastas, Valtteri Skantsi, Jemina Kilpeläinen, Hanna-Mari Kupari, Aurora Piirto, Jenna Saarni, Maija Sevón, Otto Tarkka (TurkuNLP / University of Turku)", "## Dataset Overview", "### Where to find the Data and its Documentation", "#### Webpage\n\n\n\nWebsite", "#### Download\n\n\n\nGithub", "#### Paper\n\n\n\nACL Anthology", "#### BibTex", "#### Contact Name\n\n\n\n\nJenna Kanerva, Filip Ginter", "#### Contact Email\n\n\n\njmnybl@URL, figint@URL", "#### Has a Leaderboard?\n\n\n\nno", "### Languages and Intended Use", "#### Multilingual?\n\n\n\n\nno", "#### Covered Dialects\n\n\n\nwritten standard language, spoken language", "#### Covered Languages\n\n\n\n\n'Finnish'", "#### License\n\n\n\n\ncc-by-sa-4.0: Creative Commons Attribution Share Alike 4.0 International", "#### Intended Use\n\n\n\nParaphrase classification, paraphrase generation", "#### Primary Task\n\n\n\nParaphrasing", "#### Communicative Goal\n\n\n\n\nThe corpus provides naturally occurring Finnish paraphrases striving for low lexical overlap, thus supporting many different downstream applications requiring language understanding.", "### Credit", "#### Curation Organization Type(s)\n\n\n\n'academic'", "#### Curation Organization(s)\n\n\n\nUniversity of Turku", "#### Dataset Creators\n\n\n\nJenna Kanerva, Filip Ginter, Li-Hsin Chang, Iiro Rastas, Valtteri Skantsi, Jemina Kilpeläinen, Hanna-Mari Kupari, Aurora Piirto, Jenna Saarni, Maija Sevón, Otto Tarkka (TurkuNLP / University of Turku)", "#### Funding\n\n\n\nThe Turku paraphrase corpus project was funded by the Academy of Finland, as well as the European Language Grid project through its open call for pilot projects. The European Language Grid project has received funding from the European Union’s Horizon 2020 Research and Innovation programme under Grant Agreement no. 825627 (ELG).", "#### Who added the Dataset to GEM?\n\n\n\nJenna Kanerva, Filip Ginter (TurkuNLP / University of Turku)", "### Dataset Structure", "#### Data Fields\n\n\n\nThe dataset consist of pairs of text passages, where a typical passage is about a sentence long, however, a passage may also be longer or shorter than a sentence. Thus, each example include two text passages (string), a manually annotated label to indicate the paraphrase type (string), and additional metadata.\n\nThe dataset include three different 'modes', plain, classification, and generation. The 'plain' mode loads the original data without any additional preprocessing or transformations, while the 'classification' mode directly builds the data in a form suitable for training a paraphrase classifier, where each example is doubled in the data with different directions (text1, text2, label) --> (text2, text1, label) taking care of the label flipping as well if needed (paraphrases with directionality flag < or >). In the 'generation' mode, the examples are preprocessed to be directly suitable for paraphrase generation task. In here, paraphrases not suitable for generation are discarded (negative, and highly context-dependent paraphrases), and directional paraphrases are provided so that the generation goes from more detailed passage to the more general one in order to prevent model hallucination (i.e. model learning to introduce new information). The rest of the paraphrases are provided in both directions (text1, text2, label) --> (text2, text1, label).\n\n\nEach pair in 'plain' and 'classification' mode will include fields:\n\n'gem_id': Identifier of the paraphrase pair (string)\n'goeswith': Identifier of the document from which the paraphrase was extracted, can be 'not available' in case the source of the paraphrase is not from document-structured data (string)\n'fold': 0-99, data split into 100 parts respecting document boundaries, you can use this e.g. to implement crossvalidation safely as all paraphrases from one document are in one fold (int)\n'text1': First paraphrase passage (string)\n'text2': Second paraphrase passage (string)\n'label': Manually annotated labels (string)\n'binary_label': Label turned into binary with values 'positive' (paraphrase) and 'negative' (not-paraphrase) (string)\n'is_rewrite': Indicator whether the example is human produced rewrite or naturally occurring paraphrase (bool)\n\nEach pair in 'generation' mode will include the same fields expect 'text1' and 'text2' are renamed to 'input' and 'output' in order to indicate the generation direction. Thus the fields are:\n\n'gem_id': Identifier of the paraphrase pair (string)\n'goeswith': Identifier of the document from which the paraphrase was extracted, can be 'not available' in case the source of the paraphrase is not from document-structured data (string)\n'fold': 0-99, data split into 100 parts respecting document boundaries, you can use this e.g. to implement crossvalidation safely as all paraphrases from one document are in one fold (int)\n'input': The input paraphrase passage for generation (string)\n'output': The output paraphrase passage for generation (string)\n'label': Manually annotated labels (string)\n'binary_label': Label turned into binary with values 'positive' (paraphrase) and 'negative' (not-paraphrase) (string)\n'is_rewrite': Indicator whether the example is human produced rewrite or naturally occurring paraphrase (bool)", "#### Example Instance", "#### Data Splits\n\n\n\nThe corpus include 3 splits: train, validation, and test.", "#### Splitting Criteria\n\n\n\nThe data is split randomly into the three section with a restriction of all paraphrases from the same document (movie, TV episode, news article, student translation, or exam question) being in the same section. All splits are manually annotated.", "## Dataset in GEM", "### Rationale for Inclusion in GEM", "#### Why is the Dataset in GEM?\n\n\n\nThis dataset provides a large amount of high quality (manually collected and verified) paraphrases for Finnish.", "#### Similar Datasets\n\n\n\nyes", "#### Unique Language Coverage\n\n\n\nno", "#### Ability that the Dataset measures\n\n\n\nnatural language understanding, language variation", "### GEM-Specific Curation", "#### Modificatied for GEM?\n\n\n\nyes", "#### GEM Modifications\n\n\n\n'data points modified'", "#### Modification Details\n\n\n\nData structure is slightly simplified, and the release provides ready made transformations into two tasks (paraphrase classification and generation), where some data instances are doubled with different direction, and some are discarded as not being suitable for generation (e.g. negatives).", "#### Additional Splits?\n\n\n\nno", "### Getting Started with the Task", "## Previous Results", "### Previous Results", "#### Measured Model Abilities\n\n\n\nnatural language understanding, language variation", "#### Previous results available?\n\n\n\nyes", "#### Other Evaluation Approaches\n\n\n\nF-score in paraphrase classification", "## Dataset Curation", "### Original Curation", "#### Original Curation Rationale\n\n\n\nThe dataset is fully manually annotated. The dataset strives for interesting paraphrases with low lexical overlap, thus the annotation is two fold. First the paraphrases are manually extracted from two related documents, where the annotators are instructed to extract only interesting paraphrases. In the second phrase, all extracted paraphrases are manually labeled given the annotation scheme.\n\nThe annotation scheme is:\n4 : paraphrase in all reasonably possible contexts\n3 : paraphrase in the given document contexts, but not in general\n2 : related but not paraphrase\nDuring annotation also labels 1 (unrelated) and x (skip, e.g. wrong language) were used, however, the insignificant amount of examples annotated with these labels were discarded from the released corpus.\n\nThe following flags are annotated to label 4 paraphrases:\n< : txt1 is more general than txt2; txt2 is more specific than txt1 (directional paraphrase where txt2 can be replaced with txt1 in all contexts but not to the other direction)\n> : txt2 is more general than txt1; txt1 is more specific than txt2 (directional paraphrase where txt1 can be replaced with txt2 in all contexts but not to the other direction)\ni : minor traceable difference (differing in terms of grammatical number or case, 'this' vs 'that', etc.)\ns : style or strength difference (e.g. equivalent meaning, but one of the statements substantially more colloquial than the other)\n\nFor paraphrases where the annotated label was something else than label 4 without any flags, the annotators had an option to rewrite the text passages so that the rewritten paraphrase pair formed label 4 (universal) paraphrase. This was used for cases where simple edit would turn e.g. contextual or directional paraphrase into universal one. For the rewritten examples, both the original and the rewritten pairs are available with corresponding labels annotated.", "#### Communicative Goal\n\n\n\nRepresenting text passages with identical meaning but different surface realization.", "#### Sourced from Different Sources\n\n\n\nyes", "#### Source Details\n\n\n\nmovie and TV series subtitles (82%)\nnews articles (9%)\ndiscussion forum messages (8%)\nuniversity translation exercises (1%)\nuniversity course essays and exams (<1%)", "### Language Data", "#### How was Language Data Obtained?\n\n\n\n'Found', 'Other'", "#### Where was it found?\n\n\n\n'Multiple websites', 'Offline media collection', 'Other'", "#### Language Producers\n\n\n\nThe movie and TV series subtitles are extracted from OPUS OpenSubtitles2018 collection, which is based on data from OpenSubtitles.\nThe news articles are collected from two Finnish news sites, YLE and HS, during years 2017-2020.\nDiscussion forum messages are obtained from the Finnish Suomi24 discussion forum released for academic use (URL\nUniversity translation exercises, essays and exams are collected during the project.", "#### Data Validation\n\n\n\nvalidated by data curator", "#### Was Data Filtered?\n\n\n\nnot filtered", "### Structured Annotations", "#### Additional Annotations?\n\n\n\n\nexpert created", "#### Number of Raters\n\n\n\n2<n<10", "#### Rater Qualifications\n\n\n\nMembers of the TurkuNLP research group, native speakers of Finnish, each annotator has a strong background in language studies by having an academic degree or ongoing studies in a field related to languages or linguistics.", "#### Raters per Training Example\n\n\n\n1", "#### Raters per Test Example\n\n\n\n1", "#### Annotation Service?\n\n\n\nno", "#### Annotation Values\n\n\n\n1. Manual extraction of interesting paraphrases from two related documents.\n2. Manual labeling of each extracted paraphrase based on the given annotation scheme, e.g. distinguishing contextual and universal paraphrases, marking style or strength differences, etc.", "#### Any Quality Control?\n\n\n\nvalidated by another rater", "#### Quality Control Details\n\n\n\nPartial double annotation, double annotation batches are assigned regularly in order to monitor annotation consistency. In double annotation, one annotator first extracts the candidate paraphrases, and these candidates are assigned to two different annotators, who does the label annotation independently from each other. Afterwards, the label annotations are merged, and conflicting labels are resolved together with the whole annotation team.", "### Consent", "#### Any Consent Policy?\n\n\n\nyes", "#### Consent Policy Details\n\n\n\nThe corpus is mostly based on public/open data. For other data sources (student material), the licensing was agreed with the data providers during the collection.", "### Private Identifying Information (PII)", "#### Contains PII?\n\n\n\n\nlikely", "#### Categories of PII\n\n\n\n'generic PII'", "#### Any PII Identification?\n\n\n\nno identification", "### Maintenance", "#### Any Maintenance Plan?\n\n\n\nno", "## Broader Social Context", "### Previous Work on the Social Impact of the Dataset", "#### Usage of Models based on the Data\n\n\n\nno", "### Impact on Under-Served Communities", "#### Addresses needs of underserved Communities?\n\n\n\nno", "### Discussion of Biases", "#### Any Documented Social Biases?\n\n\n\nno", "## Considerations for Using the Data", "### PII Risks and Liability", "#### Potential PII Risk\n\n\n\nNone", "### Licenses", "#### Copyright Restrictions on the Dataset\n\n\n\n'open license - commercial use allowed'", "#### Copyright Restrictions on the Language Data\n\n\n\n'open license - commercial use allowed'", "### Known Technical Limitations" ]

[ "TAGS\n#task_categories-other #annotations_creators-expert-created #language_creators-unknown #multilinguality-unknown #size_categories-unknown #source_datasets-original #language-Finnish #license-cc-by-sa-4.0 #paraphrasing #region-us \n", "# Dataset Card for GEM/turku_paraphrase_corpus", "## Dataset Description\n\n- Homepage: URL\n- Repository: URL\n- Paper: URL\n- Leaderboard: N/A\n- Point of Contact: Jenna Kanerva, Filip Ginter", "### Link to Main Data Card\n\nYou can find the main data card on the GEM Website.", "### Dataset Summary \n\nThis is a Finnish paraphrase corpus which consists of pairs of text passages, where a typical passage is about a sentence long. It can be used to either identify or generate paraphrases.\n\nYou can load the dataset via:\n\nThe data loader can be found here.", "#### website\nWebsite", "#### paper\nACL Anthology", "#### authors\nJenna Kanerva, Filip Ginter, Li-Hsin Chang, Iiro Rastas, Valtteri Skantsi, Jemina Kilpeläinen, Hanna-Mari Kupari, Aurora Piirto, Jenna Saarni, Maija Sevón, Otto Tarkka (TurkuNLP / University of Turku)", "## Dataset Overview", "### Where to find the Data and its Documentation", "#### Webpage\n\n\n\nWebsite", "#### Download\n\n\n\nGithub", "#### Paper\n\n\n\nACL Anthology", "#### BibTex", "#### Contact Name\n\n\n\n\nJenna Kanerva, Filip Ginter", "#### Contact Email\n\n\n\njmnybl@URL, figint@URL", "#### Has a Leaderboard?\n\n\n\nno", "### Languages and Intended Use", "#### Multilingual?\n\n\n\n\nno", "#### Covered Dialects\n\n\n\nwritten standard language, spoken language", "#### Covered Languages\n\n\n\n\n'Finnish'", "#### License\n\n\n\n\ncc-by-sa-4.0: Creative Commons Attribution Share Alike 4.0 International", "#### Intended Use\n\n\n\nParaphrase classification, paraphrase generation", "#### Primary Task\n\n\n\nParaphrasing", "#### Communicative Goal\n\n\n\n\nThe corpus provides naturally occurring Finnish paraphrases striving for low lexical overlap, thus supporting many different downstream applications requiring language understanding.", "### Credit", "#### Curation Organization Type(s)\n\n\n\n'academic'", "#### Curation Organization(s)\n\n\n\nUniversity of Turku", "#### Dataset Creators\n\n\n\nJenna Kanerva, Filip Ginter, Li-Hsin Chang, Iiro Rastas, Valtteri Skantsi, Jemina Kilpeläinen, Hanna-Mari Kupari, Aurora Piirto, Jenna Saarni, Maija Sevón, Otto Tarkka (TurkuNLP / University of Turku)", "#### Funding\n\n\n\nThe Turku paraphrase corpus project was funded by the Academy of Finland, as well as the European Language Grid project through its open call for pilot projects. The European Language Grid project has received funding from the European Union’s Horizon 2020 Research and Innovation programme under Grant Agreement no. 825627 (ELG).", "#### Who added the Dataset to GEM?\n\n\n\nJenna Kanerva, Filip Ginter (TurkuNLP / University of Turku)", "### Dataset Structure", "#### Data Fields\n\n\n\nThe dataset consist of pairs of text passages, where a typical passage is about a sentence long, however, a passage may also be longer or shorter than a sentence. Thus, each example include two text passages (string), a manually annotated label to indicate the paraphrase type (string), and additional metadata.\n\nThe dataset include three different 'modes', plain, classification, and generation. The 'plain' mode loads the original data without any additional preprocessing or transformations, while the 'classification' mode directly builds the data in a form suitable for training a paraphrase classifier, where each example is doubled in the data with different directions (text1, text2, label) --> (text2, text1, label) taking care of the label flipping as well if needed (paraphrases with directionality flag < or >). In the 'generation' mode, the examples are preprocessed to be directly suitable for paraphrase generation task. In here, paraphrases not suitable for generation are discarded (negative, and highly context-dependent paraphrases), and directional paraphrases are provided so that the generation goes from more detailed passage to the more general one in order to prevent model hallucination (i.e. model learning to introduce new information). The rest of the paraphrases are provided in both directions (text1, text2, label) --> (text2, text1, label).\n\n\nEach pair in 'plain' and 'classification' mode will include fields:\n\n'gem_id': Identifier of the paraphrase pair (string)\n'goeswith': Identifier of the document from which the paraphrase was extracted, can be 'not available' in case the source of the paraphrase is not from document-structured data (string)\n'fold': 0-99, data split into 100 parts respecting document boundaries, you can use this e.g. to implement crossvalidation safely as all paraphrases from one document are in one fold (int)\n'text1': First paraphrase passage (string)\n'text2': Second paraphrase passage (string)\n'label': Manually annotated labels (string)\n'binary_label': Label turned into binary with values 'positive' (paraphrase) and 'negative' (not-paraphrase) (string)\n'is_rewrite': Indicator whether the example is human produced rewrite or naturally occurring paraphrase (bool)\n\nEach pair in 'generation' mode will include the same fields expect 'text1' and 'text2' are renamed to 'input' and 'output' in order to indicate the generation direction. Thus the fields are:\n\n'gem_id': Identifier of the paraphrase pair (string)\n'goeswith': Identifier of the document from which the paraphrase was extracted, can be 'not available' in case the source of the paraphrase is not from document-structured data (string)\n'fold': 0-99, data split into 100 parts respecting document boundaries, you can use this e.g. to implement crossvalidation safely as all paraphrases from one document are in one fold (int)\n'input': The input paraphrase passage for generation (string)\n'output': The output paraphrase passage for generation (string)\n'label': Manually annotated labels (string)\n'binary_label': Label turned into binary with values 'positive' (paraphrase) and 'negative' (not-paraphrase) (string)\n'is_rewrite': Indicator whether the example is human produced rewrite or naturally occurring paraphrase (bool)", "#### Example Instance", "#### Data Splits\n\n\n\nThe corpus include 3 splits: train, validation, and test.", "#### Splitting Criteria\n\n\n\nThe data is split randomly into the three section with a restriction of all paraphrases from the same document (movie, TV episode, news article, student translation, or exam question) being in the same section. All splits are manually annotated.", "## Dataset in GEM", "### Rationale for Inclusion in GEM", "#### Why is the Dataset in GEM?\n\n\n\nThis dataset provides a large amount of high quality (manually collected and verified) paraphrases for Finnish.", "#### Similar Datasets\n\n\n\nyes", "#### Unique Language Coverage\n\n\n\nno", "#### Ability that the Dataset measures\n\n\n\nnatural language understanding, language variation", "### GEM-Specific Curation", "#### Modificatied for GEM?\n\n\n\nyes", "#### GEM Modifications\n\n\n\n'data points modified'", "#### Modification Details\n\n\n\nData structure is slightly simplified, and the release provides ready made transformations into two tasks (paraphrase classification and generation), where some data instances are doubled with different direction, and some are discarded as not being suitable for generation (e.g. negatives).", "#### Additional Splits?\n\n\n\nno", "### Getting Started with the Task", "## Previous Results", "### Previous Results", "#### Measured Model Abilities\n\n\n\nnatural language understanding, language variation", "#### Previous results available?\n\n\n\nyes", "#### Other Evaluation Approaches\n\n\n\nF-score in paraphrase classification", "## Dataset Curation", "### Original Curation", "#### Original Curation Rationale\n\n\n\nThe dataset is fully manually annotated. The dataset strives for interesting paraphrases with low lexical overlap, thus the annotation is two fold. First the paraphrases are manually extracted from two related documents, where the annotators are instructed to extract only interesting paraphrases. In the second phrase, all extracted paraphrases are manually labeled given the annotation scheme.\n\nThe annotation scheme is:\n4 : paraphrase in all reasonably possible contexts\n3 : paraphrase in the given document contexts, but not in general\n2 : related but not paraphrase\nDuring annotation also labels 1 (unrelated) and x (skip, e.g. wrong language) were used, however, the insignificant amount of examples annotated with these labels were discarded from the released corpus.\n\nThe following flags are annotated to label 4 paraphrases:\n< : txt1 is more general than txt2; txt2 is more specific than txt1 (directional paraphrase where txt2 can be replaced with txt1 in all contexts but not to the other direction)\n> : txt2 is more general than txt1; txt1 is more specific than txt2 (directional paraphrase where txt1 can be replaced with txt2 in all contexts but not to the other direction)\ni : minor traceable difference (differing in terms of grammatical number or case, 'this' vs 'that', etc.)\ns : style or strength difference (e.g. equivalent meaning, but one of the statements substantially more colloquial than the other)\n\nFor paraphrases where the annotated label was something else than label 4 without any flags, the annotators had an option to rewrite the text passages so that the rewritten paraphrase pair formed label 4 (universal) paraphrase. This was used for cases where simple edit would turn e.g. contextual or directional paraphrase into universal one. For the rewritten examples, both the original and the rewritten pairs are available with corresponding labels annotated.", "#### Communicative Goal\n\n\n\nRepresenting text passages with identical meaning but different surface realization.", "#### Sourced from Different Sources\n\n\n\nyes", "#### Source Details\n\n\n\nmovie and TV series subtitles (82%)\nnews articles (9%)\ndiscussion forum messages (8%)\nuniversity translation exercises (1%)\nuniversity course essays and exams (<1%)", "### Language Data", "#### How was Language Data Obtained?\n\n\n\n'Found', 'Other'", "#### Where was it found?\n\n\n\n'Multiple websites', 'Offline media collection', 'Other'", "#### Language Producers\n\n\n\nThe movie and TV series subtitles are extracted from OPUS OpenSubtitles2018 collection, which is based on data from OpenSubtitles.\nThe news articles are collected from two Finnish news sites, YLE and HS, during years 2017-2020.\nDiscussion forum messages are obtained from the Finnish Suomi24 discussion forum released for academic use (URL\nUniversity translation exercises, essays and exams are collected during the project.", "#### Data Validation\n\n\n\nvalidated by data curator", "#### Was Data Filtered?\n\n\n\nnot filtered", "### Structured Annotations", "#### Additional Annotations?\n\n\n\n\nexpert created", "#### Number of Raters\n\n\n\n2<n<10", "#### Rater Qualifications\n\n\n\nMembers of the TurkuNLP research group, native speakers of Finnish, each annotator has a strong background in language studies by having an academic degree or ongoing studies in a field related to languages or linguistics.", "#### Raters per Training Example\n\n\n\n1", "#### Raters per Test Example\n\n\n\n1", "#### Annotation Service?\n\n\n\nno", "#### Annotation Values\n\n\n\n1. Manual extraction of interesting paraphrases from two related documents.\n2. Manual labeling of each extracted paraphrase based on the given annotation scheme, e.g. distinguishing contextual and universal paraphrases, marking style or strength differences, etc.", "#### Any Quality Control?\n\n\n\nvalidated by another rater", "#### Quality Control Details\n\n\n\nPartial double annotation, double annotation batches are assigned regularly in order to monitor annotation consistency. In double annotation, one annotator first extracts the candidate paraphrases, and these candidates are assigned to two different annotators, who does the label annotation independently from each other. Afterwards, the label annotations are merged, and conflicting labels are resolved together with the whole annotation team.", "### Consent", "#### Any Consent Policy?\n\n\n\nyes", "#### Consent Policy Details\n\n\n\nThe corpus is mostly based on public/open data. For other data sources (student material), the licensing was agreed with the data providers during the collection.", "### Private Identifying Information (PII)", "#### Contains PII?\n\n\n\n\nlikely", "#### Categories of PII\n\n\n\n'generic PII'", "#### Any PII Identification?\n\n\n\nno identification", "### Maintenance", "#### Any Maintenance Plan?\n\n\n\nno", "## Broader Social Context", "### Previous Work on the Social Impact of the Dataset", "#### Usage of Models based on the Data\n\n\n\nno", "### Impact on Under-Served Communities", "#### Addresses needs of underserved Communities?\n\n\n\nno", "### Discussion of Biases", "#### Any Documented Social Biases?\n\n\n\nno", "## Considerations for Using the Data", "### PII Risks and Liability", "#### Potential PII Risk\n\n\n\nNone", "### Licenses", "#### Copyright Restrictions on the Dataset\n\n\n\n'open license - commercial use allowed'", "#### Copyright Restrictions on the Language Data\n\n\n\n'open license - commercial use allowed'", "### Known Technical Limitations" ]

[ "task_categories:table-to-text", "annotations_creators:none", "language_creators:unknown", "multilinguality:unknown", "size_categories:unknown", "source_datasets:original", "language:en", "license:cc-by-sa-4.0", "data-to-text", "region:us" ]

[]

[ "en" ]

[ "# Dataset Card for GEM/viggo", "## Dataset Description\n\n- Homepage: URL\n- Repository: \n- Paper: URL\n- Leaderboard: N/A\n- Point of Contact: Juraj Juraska", "### Link to Main Data Card\n\nYou can find the main data card on the GEM Website.", "### Dataset Summary \n\nViGGO is an English data-to-text generation dataset in the video game domain, with target responses being more conversational than information-seeking, yet constrained to the information presented in a meaning representation. The dataset is relatively small with about 5,000 datasets but very clean, and can thus serve for evaluating transfer learning, low-resource, or few-shot capabilities of neural models.\n\nYou can load the dataset via:\n\nThe data loader can be found here.", "#### website\nWesbite", "#### paper\nACL Anthology", "#### authors\nJuraj Juraska, Kevin K. Bowden, Marilyn Walker", "## Dataset Overview", "### Where to find the Data and its Documentation", "#### Webpage\n\n\n\nWesbite", "#### Paper\n\n\n\nACL Anthology", "#### BibTex", "#### Contact Name\n\n\n\n\nJuraj Juraska", "#### Contact Email\n\n\n\njjuraska@URL", "#### Has a Leaderboard?\n\n\n\nno", "### Languages and Intended Use", "#### Multilingual?\n\n\n\n\nno", "#### Covered Languages\n\n\n\n\n'English'", "#### License\n\n\n\n\ncc-by-sa-4.0: Creative Commons Attribution Share Alike 4.0 International", "#### Intended Use\n\n\n\nViGGO was designed for the task of data-to-text generation in chatbots (as opposed to task-oriented dialogue systems), with target responses being more conversational than information-seeking, yet constrained to the information presented in a meaning representation. The dataset, being relatively small and clean, can also serve for demonstrating transfer learning capabilities of neural models.", "#### Primary Task\n\n\n\nData-to-Text", "### Credit", "#### Curation Organization Type(s)\n\n\n\n'academic'", "#### Curation Organization(s)\n\n\n\nUniversity of California, Santa Cruz", "#### Dataset Creators\n\n\n\nJuraj Juraska, Kevin K. Bowden, Marilyn Walker", "#### Who added the Dataset to GEM?\n\n\n\nJuraj Juraska", "### Dataset Structure", "#### Data Fields\n\n\n\nEach example in the dataset has the following two fields:\n\n- 'mr': A meaning representation (MR) that, in a structured format, provides the information to convey, as well as the desired dialogue act (DA) type.\n- 'ref': A reference output, i.e., a corresponding utterance realizing all the information in the MR.\n\nEach MR is a flattened dictionary of attribute-and-value pairs, \"wrapped\" in the dialogue act type indication. This format was chosen primarily for its compactness, but also to allow for easy concatenation of multiple DAs (each with potentially different attributes) in a single MR.\n\nFollowing is the list of all possible attributes (which are also refered to as \"slots\") in ViGGO along with their types/possible values:\n\n- 'name': The name of a video game (e.g., Rise of the Tomb Raider).\n- 'release_year': The year a video game was released in (e.g., 2015).\n- 'exp_release_date': For a not-yet-released game, the date when it is expected to be released (e.g., February 22, 2019). *Note: This slot cannot appear together with 'release_year' in the same dialogue act.*\n- 'developer': The name of the studio/person that created the game (e.g., Crystal Dynamics).\n- 'genres': A list of one or more genre labels from a set of possible values (e.g., action-adventure, shooter).\n- 'player_perspective': A list of one or more perspectives from which the game is/can be played (possible values: first person, third person, side view, bird view).\n- 'platforms': A list of one or more gaming platforms the game was officially released for (possible values: PC, PlayStation, Xbox, Nintendo, Nintendo Switch).\n- 'esrb': A game's content rating as determined by the ESRB (possible values: E (for Everyone), E 10+ (for Everyone 10 and Older), T (for Teen), M (for Mature)).\n- 'rating': Depending on the dialogue act this slot is used with, it is a categorical representation of either the game's average rating or the game's liking (possible values: excellent, good, average, poor).\n- 'has_multiplayer': Indicates whether a game supports multiplayer or can only be played in single-player mode (possible values: yes, no).\n- 'available_on_steam': Indicates whether a game can be purchased through the Steam digital distribution service (possible values: yes, no).\n- 'has_linux_release': Indicates whether a game is supported on Linux operating systems (possible values: yes, no).\n- 'has_mac_release': Indicates whether a game is supported on macOS (possible values: yes, no).\n- 'specifier': A game specifier used by the 'request' DA, typically an adjective (e.g., addictive, easiest, overrated, visually impressive).\n\nEach MR in the dataset has 3 distinct reference utterances, which are represented as 3 separate examples with the same MR.", "#### Reason for Structure\n\n\n\nThe dataset structure mostly follows the format of the popular E2E dataset, however, with added dialogue act type indications, new list-type attributes introduced, and unified naming convention for multi-word attribute names.", "#### Example Instance", "#### Data Splits\n\n\n\nViGGO is split into 3 partitions, with no MRs in common between the training set and either of the validation and the test set (and that *after* delexicalizing the 'name' and 'developer' slots). The ratio of examples in the partitions is approximately 7.5 : 1 : 1.5, with their exact sizes listed below:\n\n- Train: 5,103 (1,675 unique MRs)\n- Validation: 714 (238 unique MRs)\n- Test: 1,083 (359 unique MRs)\n- TOTAL: 6,900 (2,253 unique MRs)\n\n*Note: The reason why the number of unique MRs is not exactly one third of all examples is that for each 'request_attribute' DA (which only has one slot, and that without a value) 12 reference utterances were collected instead of 3.*", "#### Splitting Criteria\n\n\n\nA similar MR length and slot distribution was preserved across the partitions. The distribution of DA types, on the other hand, is skewed slightly toward fewer 'inform' DA instances (the most prevalent DA type) and a higher proportion of the less prevalent DAs in the validation and the test set.", "####", "## Dataset in GEM", "### Rationale for Inclusion in GEM", "#### Why is the Dataset in GEM?\n\n\n\nViGGO is a fairly small dataset but includes a greater variety of utterance types than most other datasets for NLG from structured meaning representations. This makes it more interesting from the perspective of model evaluation, since models have to learn to differentiate between various dialogue act types that share the same slots.", "#### Similar Datasets\n\n\n\nyes", "#### Unique Language Coverage\n\n\n\nno", "#### Difference from other GEM datasets\n\n\n\nViGGO's language is more casual and conversational -- as opposed to information-seeking -- which differentiates it from the majority of popular datasets for the same type of data-to-text task. Moreover, the video game domain is a rather uncommon one in the NLG community, despite being very well-suited for data-to-text generation, considering it offers entities with many attributes to talk about, which can be described in a structured format.", "### GEM-Specific Curation", "#### Modificatied for GEM?\n\n\n\nno", "#### Additional Splits?\n\n\n\nno", "### Getting Started with the Task", "#### Pointers to Resources\n\n\n\n- E2E NLG Challenge", "#### Technical Terms\n\n\n\n- MR = meaning representation\n- DA = dialogue act", "## Previous Results", "### Previous Results", "#### Metrics\n\n\n\n'BLEU', 'METEOR', 'ROUGE', 'BERT-Score', 'BLEURT', 'Other: Other Metrics'", "#### Other Metrics\n\n\n\nSER (slot error rate): Indicates the proportion of missing/incorrect/duplicate/hallucinated slot mentions in the utterances across a test set. The closer to zero a model scores in this metric, the more semantically accurate its outputs are. This metric is typically calculated either manually on a small sample of generated outputs, or heuristically using domain-specific regex rules and gazetteers.", "#### Previous results available?\n\n\n\nyes", "#### Relevant Previous Results\n\n\n\n- Juraska et al., 2019. ViGGO: A Video Game Corpus for Data-To-Text Generation in Open-Domain Conversation.\n- Harkous et al., 2020. Have Your Text and Use It Too! End-to-End Neural Data-to-Text Generation with Semantic Fidelity.\n- Kedzie and McKeown, 2020. Controllable Meaning Representation to Text Generation: Linearization and Data Augmentation Strategies.\n- Juraska and Walker, 2021. Attention Is Indeed All You Need: Semantically Attention-Guided Decoding for Data-to-Text NLG.", "## Dataset Curation", "### Original Curation", "#### Original Curation Rationale\n\n\n\nThe primary motivation behind ViGGO was to create a data-to-text corpus in a new but conversational domain, and intended for use in open-domain chatbots rather than task-oriented dialogue systems. To this end, the dataset contains utterances of 9 generalizable and conversational dialogue act types, revolving around various aspects of video games. The idea is that similar, relatively small datasets could fairly easily be collected for other conversational domains -- especially other entertainment domains (such as music or books), but perhaps also topics like animals or food -- to support an open-domain conversational agent with controllable neural NLG.\n\nAnother desired quality of the ViGGO dataset was cleanliness (no typos and grammatical errors) and semantic accuracy, which has often not been the case with other crowdsourced data-to-text corpora. In general, for the data-to-text generation task, there is arguably no need to put the burden on the generation model to figure out the noise, since the noise would not be expected to be there in a real-world system whose dialogue manager that creates the input for the NLG module is usually configurable and tightly controlled.", "#### Communicative Goal\n\n\n\nProduce a response from a structured meaning representation in the context of a conversation about video games. It can be a brief opinion or a description of a game, as well as a request for attribute (e.g., genre, player perspective, or platform) preference/confirmation or an inquiry about liking a particular type of games.", "#### Sourced from Different Sources\n\n\n\nno", "### Language Data", "#### How was Language Data Obtained?\n\n\n\n'Crowdsourced'", "#### Where was it crowdsourced?\n\n\n\n'Amazon Mechanical Turk'", "#### Language Producers\n\n\n\nThe paid crowdworkers who produced the reference utterances were from English-speaking countries, and they had at least 1,000 HITs approved and a HIT approval rate of 98% or more. Furthermore, in the instructions, crowdworkers were discouraged from taking on the task unless they considered themselves a gamer.", "#### Topics Covered\n\n\n\nThe dataset focuses on video games and their various aspects, and hence the language of the utterances may contain video game-specific jargon.", "#### Data Validation\n\n\n\nvalidated by data curator", "#### Data Preprocessing\n\n\n\nFirst, regular expressions were used to enforce several standardization policies regarding special characters, punctuation, and the correction of undesired abbreviations/misspellings of standard domain-specific terms (e.g., terms like \"Play station\" or \"PS4\" would be changed to the uniform \"PlayStation\"). At the same time, hyphens were removed or enforced uniformly in certain terms, for example, \"single-player\". Although phrases such as \"first person\" should correctly have a hyphen when used as adjective, the crowdworkers used this rule very inconsistently. In order to avoid model outputs being penalized during the evaluation by the arbitrary choice of a hyphen presence or absence in the reference utterances, the hyphen was removed in all such phrases regardless of the noun vs. adjective use.\n\nSecond, an extensive set of heuristics was developed to identify slot-related errors. This process revealed the vast majority of missing or incorrect slot mentions, which were subsequently fixed according to the corresponding MRs. This eventually led to the development of a robust, cross-domain, heuristic slot aligner that can be used for automatic slot error rate evaluation. For details, see the appendix in Juraska and Walker, 2021.\n\nCrowdworkers would sometimes also inject a piece of information which was not present in the MR, some of which is not even represented by any of the slots, e.g., plot or main characters. This unsolicited information was removed from the utterances so as to avoid confusing the neural model. Finally, any remaining typos and grammatical errors were resolved.", "#### Was Data Filtered?\n\n\n\nmanually", "#### Filter Criteria\n\n\n\nCompliance with the indicated dialogue act type, semantic accuracy (i.e., all information in the corresponding MR mentioned and that correctly), and minimal extraneous information (e.g., personal experience/opinion). Whenever it was within a reasonable amount of effort, the utterances were manually fixed instead of being discarded/crowdsourced anew.", "### Structured Annotations", "#### Additional Annotations?\n\n\n\n\nnone", "#### Annotation Service?\n\n\n\nno", "### Consent", "#### Any Consent Policy?\n\n\n\nno", "### Private Identifying Information (PII)", "#### Contains PII?\n\n\n\n\nno PII", "#### Justification for no PII\n\n\n\nCrowdworkers were instructed to only express the information in the provided meaning representation, which never prompted them to mention anything about themselves. Occasionally, they would still include a bit of personal experience (e.g., \"I used to like the game as a kid.\") or opinion, but these would be too general to be considered PII.", "### Maintenance", "#### Any Maintenance Plan?\n\n\n\nno", "## Broader Social Context", "### Previous Work on the Social Impact of the Dataset", "#### Usage of Models based on the Data\n\n\n\nno", "### Impact on Under-Served Communities", "#### Addresses needs of underserved Communities?\n\n\n\nno", "### Discussion of Biases", "#### Any Documented Social Biases?\n\n\n\nno", "## Considerations for Using the Data", "### PII Risks and Liability", "### Licenses", "### Known Technical Limitations", "#### Technical Limitations\n\n\n\nThe dataset is limited to a single domain: video games. One caveat of using a language generator trained on this dataset in a dialogue system as-is is that multiple subsequent turns discussing the same video game would be repeating its full name. ViGGO was designed for generation without context, and therefore it is up to the dialogue manager to ensure that pronouns are substituted for the names whenever it would sound more natural in a dialogue. Alternately, the dataset can easily be augmented with automatically constructed samples which omit the 'name' slot in the MR and replace the name with a pronoun in the reference utterance." ]

[ "TAGS\n#task_categories-table-to-text #annotations_creators-none #language_creators-unknown #multilinguality-unknown #size_categories-unknown #source_datasets-original #language-English #license-cc-by-sa-4.0 #data-to-text #region-us \n", "# Dataset Card for GEM/viggo", "## Dataset Description\n\n- Homepage: URL\n- Repository: \n- Paper: URL\n- Leaderboard: N/A\n- Point of Contact: Juraj Juraska", "### Link to Main Data Card\n\nYou can find the main data card on the GEM Website.", "### Dataset Summary \n\nViGGO is an English data-to-text generation dataset in the video game domain, with target responses being more conversational than information-seeking, yet constrained to the information presented in a meaning representation. The dataset is relatively small with about 5,000 datasets but very clean, and can thus serve for evaluating transfer learning, low-resource, or few-shot capabilities of neural models.\n\nYou can load the dataset via:\n\nThe data loader can be found here.", "#### website\nWesbite", "#### paper\nACL Anthology", "#### authors\nJuraj Juraska, Kevin K. Bowden, Marilyn Walker", "## Dataset Overview", "### Where to find the Data and its Documentation", "#### Webpage\n\n\n\nWesbite", "#### Paper\n\n\n\nACL Anthology", "#### BibTex", "#### Contact Name\n\n\n\n\nJuraj Juraska", "#### Contact Email\n\n\n\njjuraska@URL", "#### Has a Leaderboard?\n\n\n\nno", "### Languages and Intended Use", "#### Multilingual?\n\n\n\n\nno", "#### Covered Languages\n\n\n\n\n'English'", "#### License\n\n\n\n\ncc-by-sa-4.0: Creative Commons Attribution Share Alike 4.0 International", "#### Intended Use\n\n\n\nViGGO was designed for the task of data-to-text generation in chatbots (as opposed to task-oriented dialogue systems), with target responses being more conversational than information-seeking, yet constrained to the information presented in a meaning representation. The dataset, being relatively small and clean, can also serve for demonstrating transfer learning capabilities of neural models.", "#### Primary Task\n\n\n\nData-to-Text", "### Credit", "#### Curation Organization Type(s)\n\n\n\n'academic'", "#### Curation Organization(s)\n\n\n\nUniversity of California, Santa Cruz", "#### Dataset Creators\n\n\n\nJuraj Juraska, Kevin K. Bowden, Marilyn Walker", "#### Who added the Dataset to GEM?\n\n\n\nJuraj Juraska", "### Dataset Structure", "#### Data Fields\n\n\n\nEach example in the dataset has the following two fields:\n\n- 'mr': A meaning representation (MR) that, in a structured format, provides the information to convey, as well as the desired dialogue act (DA) type.\n- 'ref': A reference output, i.e., a corresponding utterance realizing all the information in the MR.\n\nEach MR is a flattened dictionary of attribute-and-value pairs, \"wrapped\" in the dialogue act type indication. This format was chosen primarily for its compactness, but also to allow for easy concatenation of multiple DAs (each with potentially different attributes) in a single MR.\n\nFollowing is the list of all possible attributes (which are also refered to as \"slots\") in ViGGO along with their types/possible values:\n\n- 'name': The name of a video game (e.g., Rise of the Tomb Raider).\n- 'release_year': The year a video game was released in (e.g., 2015).\n- 'exp_release_date': For a not-yet-released game, the date when it is expected to be released (e.g., February 22, 2019). *Note: This slot cannot appear together with 'release_year' in the same dialogue act.*\n- 'developer': The name of the studio/person that created the game (e.g., Crystal Dynamics).\n- 'genres': A list of one or more genre labels from a set of possible values (e.g., action-adventure, shooter).\n- 'player_perspective': A list of one or more perspectives from which the game is/can be played (possible values: first person, third person, side view, bird view).\n- 'platforms': A list of one or more gaming platforms the game was officially released for (possible values: PC, PlayStation, Xbox, Nintendo, Nintendo Switch).\n- 'esrb': A game's content rating as determined by the ESRB (possible values: E (for Everyone), E 10+ (for Everyone 10 and Older), T (for Teen), M (for Mature)).\n- 'rating': Depending on the dialogue act this slot is used with, it is a categorical representation of either the game's average rating or the game's liking (possible values: excellent, good, average, poor).\n- 'has_multiplayer': Indicates whether a game supports multiplayer or can only be played in single-player mode (possible values: yes, no).\n- 'available_on_steam': Indicates whether a game can be purchased through the Steam digital distribution service (possible values: yes, no).\n- 'has_linux_release': Indicates whether a game is supported on Linux operating systems (possible values: yes, no).\n- 'has_mac_release': Indicates whether a game is supported on macOS (possible values: yes, no).\n- 'specifier': A game specifier used by the 'request' DA, typically an adjective (e.g., addictive, easiest, overrated, visually impressive).\n\nEach MR in the dataset has 3 distinct reference utterances, which are represented as 3 separate examples with the same MR.", "#### Reason for Structure\n\n\n\nThe dataset structure mostly follows the format of the popular E2E dataset, however, with added dialogue act type indications, new list-type attributes introduced, and unified naming convention for multi-word attribute names.", "#### Example Instance", "#### Data Splits\n\n\n\nViGGO is split into 3 partitions, with no MRs in common between the training set and either of the validation and the test set (and that *after* delexicalizing the 'name' and 'developer' slots). The ratio of examples in the partitions is approximately 7.5 : 1 : 1.5, with their exact sizes listed below:\n\n- Train: 5,103 (1,675 unique MRs)\n- Validation: 714 (238 unique MRs)\n- Test: 1,083 (359 unique MRs)\n- TOTAL: 6,900 (2,253 unique MRs)\n\n*Note: The reason why the number of unique MRs is not exactly one third of all examples is that for each 'request_attribute' DA (which only has one slot, and that without a value) 12 reference utterances were collected instead of 3.*", "#### Splitting Criteria\n\n\n\nA similar MR length and slot distribution was preserved across the partitions. The distribution of DA types, on the other hand, is skewed slightly toward fewer 'inform' DA instances (the most prevalent DA type) and a higher proportion of the less prevalent DAs in the validation and the test set.", "####", "## Dataset in GEM", "### Rationale for Inclusion in GEM", "#### Why is the Dataset in GEM?\n\n\n\nViGGO is a fairly small dataset but includes a greater variety of utterance types than most other datasets for NLG from structured meaning representations. This makes it more interesting from the perspective of model evaluation, since models have to learn to differentiate between various dialogue act types that share the same slots.", "#### Similar Datasets\n\n\n\nyes", "#### Unique Language Coverage\n\n\n\nno", "#### Difference from other GEM datasets\n\n\n\nViGGO's language is more casual and conversational -- as opposed to information-seeking -- which differentiates it from the majority of popular datasets for the same type of data-to-text task. Moreover, the video game domain is a rather uncommon one in the NLG community, despite being very well-suited for data-to-text generation, considering it offers entities with many attributes to talk about, which can be described in a structured format.", "### GEM-Specific Curation", "#### Modificatied for GEM?\n\n\n\nno", "#### Additional Splits?\n\n\n\nno", "### Getting Started with the Task", "#### Pointers to Resources\n\n\n\n- E2E NLG Challenge", "#### Technical Terms\n\n\n\n- MR = meaning representation\n- DA = dialogue act", "## Previous Results", "### Previous Results", "#### Metrics\n\n\n\n'BLEU', 'METEOR', 'ROUGE', 'BERT-Score', 'BLEURT', 'Other: Other Metrics'", "#### Other Metrics\n\n\n\nSER (slot error rate): Indicates the proportion of missing/incorrect/duplicate/hallucinated slot mentions in the utterances across a test set. The closer to zero a model scores in this metric, the more semantically accurate its outputs are. This metric is typically calculated either manually on a small sample of generated outputs, or heuristically using domain-specific regex rules and gazetteers.", "#### Previous results available?\n\n\n\nyes", "#### Relevant Previous Results\n\n\n\n- Juraska et al., 2019. ViGGO: A Video Game Corpus for Data-To-Text Generation in Open-Domain Conversation.\n- Harkous et al., 2020. Have Your Text and Use It Too! End-to-End Neural Data-to-Text Generation with Semantic Fidelity.\n- Kedzie and McKeown, 2020. Controllable Meaning Representation to Text Generation: Linearization and Data Augmentation Strategies.\n- Juraska and Walker, 2021. Attention Is Indeed All You Need: Semantically Attention-Guided Decoding for Data-to-Text NLG.", "## Dataset Curation", "### Original Curation", "#### Original Curation Rationale\n\n\n\nThe primary motivation behind ViGGO was to create a data-to-text corpus in a new but conversational domain, and intended for use in open-domain chatbots rather than task-oriented dialogue systems. To this end, the dataset contains utterances of 9 generalizable and conversational dialogue act types, revolving around various aspects of video games. The idea is that similar, relatively small datasets could fairly easily be collected for other conversational domains -- especially other entertainment domains (such as music or books), but perhaps also topics like animals or food -- to support an open-domain conversational agent with controllable neural NLG.\n\nAnother desired quality of the ViGGO dataset was cleanliness (no typos and grammatical errors) and semantic accuracy, which has often not been the case with other crowdsourced data-to-text corpora. In general, for the data-to-text generation task, there is arguably no need to put the burden on the generation model to figure out the noise, since the noise would not be expected to be there in a real-world system whose dialogue manager that creates the input for the NLG module is usually configurable and tightly controlled.", "#### Communicative Goal\n\n\n\nProduce a response from a structured meaning representation in the context of a conversation about video games. It can be a brief opinion or a description of a game, as well as a request for attribute (e.g., genre, player perspective, or platform) preference/confirmation or an inquiry about liking a particular type of games.", "#### Sourced from Different Sources\n\n\n\nno", "### Language Data", "#### How was Language Data Obtained?\n\n\n\n'Crowdsourced'", "#### Where was it crowdsourced?\n\n\n\n'Amazon Mechanical Turk'", "#### Language Producers\n\n\n\nThe paid crowdworkers who produced the reference utterances were from English-speaking countries, and they had at least 1,000 HITs approved and a HIT approval rate of 98% or more. Furthermore, in the instructions, crowdworkers were discouraged from taking on the task unless they considered themselves a gamer.", "#### Topics Covered\n\n\n\nThe dataset focuses on video games and their various aspects, and hence the language of the utterances may contain video game-specific jargon.", "#### Data Validation\n\n\n\nvalidated by data curator", "#### Data Preprocessing\n\n\n\nFirst, regular expressions were used to enforce several standardization policies regarding special characters, punctuation, and the correction of undesired abbreviations/misspellings of standard domain-specific terms (e.g., terms like \"Play station\" or \"PS4\" would be changed to the uniform \"PlayStation\"). At the same time, hyphens were removed or enforced uniformly in certain terms, for example, \"single-player\". Although phrases such as \"first person\" should correctly have a hyphen when used as adjective, the crowdworkers used this rule very inconsistently. In order to avoid model outputs being penalized during the evaluation by the arbitrary choice of a hyphen presence or absence in the reference utterances, the hyphen was removed in all such phrases regardless of the noun vs. adjective use.\n\nSecond, an extensive set of heuristics was developed to identify slot-related errors. This process revealed the vast majority of missing or incorrect slot mentions, which were subsequently fixed according to the corresponding MRs. This eventually led to the development of a robust, cross-domain, heuristic slot aligner that can be used for automatic slot error rate evaluation. For details, see the appendix in Juraska and Walker, 2021.\n\nCrowdworkers would sometimes also inject a piece of information which was not present in the MR, some of which is not even represented by any of the slots, e.g., plot or main characters. This unsolicited information was removed from the utterances so as to avoid confusing the neural model. Finally, any remaining typos and grammatical errors were resolved.", "#### Was Data Filtered?\n\n\n\nmanually", "#### Filter Criteria\n\n\n\nCompliance with the indicated dialogue act type, semantic accuracy (i.e., all information in the corresponding MR mentioned and that correctly), and minimal extraneous information (e.g., personal experience/opinion). Whenever it was within a reasonable amount of effort, the utterances were manually fixed instead of being discarded/crowdsourced anew.", "### Structured Annotations", "#### Additional Annotations?\n\n\n\n\nnone", "#### Annotation Service?\n\n\n\nno", "### Consent", "#### Any Consent Policy?\n\n\n\nno", "### Private Identifying Information (PII)", "#### Contains PII?\n\n\n\n\nno PII", "#### Justification for no PII\n\n\n\nCrowdworkers were instructed to only express the information in the provided meaning representation, which never prompted them to mention anything about themselves. Occasionally, they would still include a bit of personal experience (e.g., \"I used to like the game as a kid.\") or opinion, but these would be too general to be considered PII.", "### Maintenance", "#### Any Maintenance Plan?\n\n\n\nno", "## Broader Social Context", "### Previous Work on the Social Impact of the Dataset", "#### Usage of Models based on the Data\n\n\n\nno", "### Impact on Under-Served Communities", "#### Addresses needs of underserved Communities?\n\n\n\nno", "### Discussion of Biases", "#### Any Documented Social Biases?\n\n\n\nno", "## Considerations for Using the Data", "### PII Risks and Liability", "### Licenses", "### Known Technical Limitations", "#### Technical Limitations\n\n\n\nThe dataset is limited to a single domain: video games. One caveat of using a language generator trained on this dataset in a dialogue system as-is is that multiple subsequent turns discussing the same video game would be repeating its full name. ViGGO was designed for generation without context, and therefore it is up to the dialogue manager to ensure that pronouns are substituted for the names whenever it would sound more natural in a dialogue. Alternately, the dataset can easily be augmented with automatically constructed samples which omit the 'name' slot in the MR and replace the name with a pronoun in the reference utterance." ]

[ "task_categories:table-to-text", "annotations_creators:unknown", "language_creators:unknown", "multilinguality:unknown", "size_categories:unknown", "source_datasets:original", "language:en", "license:cc-by-nc-4.0", "data-to-text", "region:us" ]

[]

[ "en" ]

[ "### Link to Main Data Card\n\n\nYou can find the main data card on the GEM Website.", "### Dataset Summary\n\n\nWebNLG is a bi-lingual dataset (English, Russian) of parallel DBpedia triple sets and short texts that cover about 450 different DBpedia properties. The WebNLG data was originally created to promote the development of RDF verbalisers able to generate short text and to handle micro-planning (i.e., sentence segmentation and ordering, referring expression generation, aggregation); the goal of the task is to generate texts starting from 1 to 7 input triples which have entities in common (so the input is actually a connected Knowledge Graph). The dataset contains about 17,000 triple sets and 45,000 crowdsourced texts in English, and 7,000 triples sets and 19,000 crowdsourced texts in Russian. A challenging test set section with entities and/or properties that have not been seen at training time is available.\n\n\nYou can load the dataset via:\n\n\nThe data loader can be found here.", "#### website\n\n\nWebsite", "#### paper\n\n\nFirst Dataset Release, WebNLG Challenge 2017 Report, WebNLG Challenge 2020 Report", "#### authors\n\n\nThe principle curator of the dataset is Anastasia Shimorina (Université de Lorraine / LORIA, France). Throughout the WebNLG releases, several people contributed to their construction: Claire Gardent (CNRS / LORIA, France), Shashi Narayan (Google, UK), Laura Perez-Beltrachini (University of Edinburgh, UK), Elena Khasanova, and Thiago Castro Ferreira (Federal University of Minas Gerais, Brazil).\n\n\nDataset Overview\n----------------", "### Where to find the Data and its Documentation", "#### Webpage\n\n\nWebsite", "#### Download\n\n\nGitlab", "#### Paper\n\n\nFirst Dataset Release, WebNLG Challenge 2017 Report, WebNLG Challenge 2020 Report", "#### BibTex\n\n\nInitial release of the dataset:\n\n\nThe latest version 3.0:", "#### Contact Email\n\n\nwebnlg-challenge@URL", "#### Has a Leaderboard?\n\n\nyes", "#### Leaderboard Link\n\n\nWebsite", "#### Leaderboard Details\n\n\nThe model outputs are evaluated against the crowdsourced references; the leaderboard reports BLEU-4, METEOR, chrF++, TER, BERTScore and BLEURT scores.", "### Languages and Intended Use", "#### Multilingual?\n\n\nyes", "#### Covered Languages\n\n\n'Russian', 'English'", "#### License\n\n\ncc-by-nc-4.0: Creative Commons Attribution Non Commercial 4.0 International", "#### Intended Use\n\n\nThe WebNLG dataset was created to promote the development (*i*) of RDF verbalisers and (*ii*) of microplanners able to handle a wide range of linguistic constructions. The dataset aims at covering knowledge in different domains (\"categories\"). The same properties and entities can appear in several categories.", "#### Primary Task\n\n\nData-to-Text", "#### Communicative Goal\n\n\nA model should verbalize all and only the provided input triples in natural language.", "### Credit", "#### Curation Organization Type(s)\n\n\n'academic'", "#### Curation Organization(s)\n\n\nUniversité de Lorraine / LORIA, France, CNRS / LORIA, France, University of Edinburgh, UK, Federal University of Minas Gerais, Brazil", "#### Dataset Creators\n\n\nThe principle curator of the dataset is Anastasia Shimorina (Université de Lorraine / LORIA, France). Throughout the WebNLG releases, several people contributed to their construction: Claire Gardent (CNRS / LORIA, France), Shashi Narayan (Google, UK), Laura Perez-Beltrachini (University of Edinburgh, UK), Elena Khasanova, and Thiago Castro Ferreira (Federal University of Minas Gerais, Brazil).", "#### Funding\n\n\nThe dataset construction was funded by the French National Research Agency (ANR).", "#### Who added the Dataset to GEM?\n\n\nSimon Mille and Sebastian Gehrmann added the dataset and wrote the data card.", "### Dataset Structure", "#### Data Fields\n\n\nSee official documentation.\n\n\n'entry': a data instance of the benchmark. Each entry has five attributes: a DBpedia category ('category'), entry ID ('eid'), shape, shape type, and triple set size ('size').\n\n\n* 'shape': a string representation of the RDF tree with nested parentheses where 'X' is a node (see Newick tree format).\n* 'shape\\_type': a type of the tree shape. We identify three types of tree shapes:\n\n\n\t+ 'chain' (the object of one triple is the subject of the other);\n\t+ 'sibling' (triples with a shared subject);\n\t+ 'mixed' (both 'chain' and 'sibling' types present).\n* 'eid': an entry ID. It is unique only within a category and a size.\n* 'category': a DBpedia category (Astronaut, City, MusicalWork, Politician, etc.).\n* 'size': the number of RDF triples in a set. Ranges from 1 to 7.\n\n\nEach 'entry' has three fields: 'originaltripleset', 'modifiedtripleset', and 'lexs'.\n\n\n'originaltripleset': a set of RDF triples as extracted from DBpedia. Each set of RDF triples is a tree. Triples have the subject-predicate-object structure.\n\n\n'modifiedtripleset': a set of RDF triples as presented to crowdworkers (for more details on modifications, see below).\n\n\nOriginal and modified triples serve different purposes: the original triples — to link data to a knowledge base (DBpedia), whereas the modified triples — to ensure consistency and homogeneity throughout the data. To train models, the modified triples should be used.\n\n\n'lexs' (shortened for lexicalisations): a natural language text verbalising the triples. Each lexicalisation has two attributes: a comment ('comment'), and a lexicalisation ID ('lid'). By default, comments have the value 'good', except rare cases when they were manually marked as 'toFix'. That was done during the corpus creation, when it was seen that a lexicalisation did not exactly match a triple set.\n\n\nRussian data has additional optional fields comparing to English:\n\n\n'': RDF triples extracted from DBpedia between English and Russian entities by means of the property 'sameAs'.\n\n\n'': RDF triples created manually for some entities to serve as pointers to translators. There are two types of them:\n\n\n* with 'sameAs' ('Spaniards | sameAs | испанцы')\n* with 'includes' ('Tomatoes, guanciale, cheese, olive oil | includes | гуанчиале'). Those were mostly created for string literals to translate some parts of them.\n\n\nLexicalisations in the Russian WebNLG have a new parameter 'lang' (values: 'en', 'ru') because original English texts were kept in the Russian version (see the example above).", "#### Example Instance\n\n\nThe XML-formatted example is here.", "#### Data Splits\n\n\n\n\nDataset in GEM\n--------------", "### Rationale for Inclusion in GEM", "#### Why is the Dataset in GEM?\n\n\nDue to the constrained generation task, this dataset can be used to evaluate very specific and narrow generation capabilities.", "#### Similar Datasets\n\n\nyes", "#### Unique Language Coverage\n\n\nyes", "#### Difference from other GEM datasets\n\n\nThe RDF-triple format is unique to WebNLG.", "#### Ability that the Dataset measures\n\n\nsurface realization", "### GEM-Specific Curation", "#### Modificatied for GEM?\n\n\nyes", "#### GEM Modifications\n\n\n'other'", "#### Modification Details\n\n\nNo changes to the main content of the dataset. The version 3.0 of the dataset is used.", "#### Additional Splits?\n\n\nyes", "#### Split Information\n\n\n23 special test sets for WebNLG were added to the GEM evaluation suite, 12 for English and 11 for Russian.\nFor both languages, we created subsets of the training and development sets of ~500 randomly selected inputs each. The inputs were sampled proportionally from each category.\n\n\nTwo types of transformations have been applied to WebNLG: (i) input scrambling (English and Russian) and (ii) numerical value replacements (English); in both cases, a subset of about 500 inputs was randomly selected. For (i), the order of the triples was randomly reassigned (each triple kept the same Subject-Property-Object internal order). For (ii), the change was performed respecting the format of the current cardinal value (e.g., alpha, integer, or floating-point) and replacing it with a new random value. The new number is lower-bounded between zero and upper bounded to be within to the highest power of 10 unit for the given value (e.g., replacing 54 would result in a random value between 0-100). Floating values maintain the degree of precision.\n\n\nFor both languages, we did identify different subsets of the test set that we could compare to each other so that we would have a better understanding of the results. There are currently 8 selections that we have made:\n\n\nSelection 1 (size): input length. This selection corresponds to the number of predicates in the input. By comparing inputs of different lengths, we can see to what extent NLG systems are able to handle different input sizes. The table below provides the relevant frequencies. Please be aware that comparing selections with fewer than 100 items may result in unreliable comparisons.\n\n\nInput length: 1, Frequency English: 369, Frequency Russian: 254\nInput length: 2, Frequency English: 349, Frequency Russian: 200\nInput length: 3, Frequency English: 350, Frequency Russian: 214\nInput length: 4, Frequency English: 305, Frequency Russian: 214\nInput length: 5, Frequency English: 213, Frequency Russian: 159\nInput length: 6, Frequency English: 114, Frequency Russian: 32\nInput length: 7, Frequency English: 79, Frequency Russian: 29\n\n\nSelection 2 (frequency): seen/unseen single predicates. This selection corresponds to the inputs with only one predicate. We compare which predicates are seen/unseen in the training data. The table below provides the relevant frequencies. Note that the comparison is only valid for English. Not for Russian, since there is only one example of unseen single predicates.\n\n\n\\_ in training: Seen, Frequency English: 297, Frequency Russian: 253\n\\_ in training: Unseen, Frequency English: 72, Frequency Russian: 1\n\n\nSelection 3 (frequency): seen/unseen combinations of predicates. This selection checks for all combinations of predicates whether that combination has been seen in the training data. For example: if the combination of predicates A and B is seen, that means that there is an input in the training data consisting of two triples, where one triple uses predicate A and the other uses predicate B. If the combination is unseen, then the converse is true. The table below provides the relevant frequencies.\n\n\n\\_ in training: unseen, Frequency English: 1295, Frequency Russian: 354\n\\_ in training: seen, Frequency English: 115, Frequency Russian: 494\n\n\nSelection 4 (frequency): seen/unseen arguments. This selection checks for all input whether or not all arg1s and arg2s in the input have been seen during the training phase. For this selection, *Seen* is the default. Only if all arg1 instances for a particular input are unseen, do we count the arg1s of the input as unseen. The same holds for arg2. So \"seen\" here really means that at least some of the arg1s or arg2s are seen in the input. The table below provides the relevant frequencies. Note that the comparison is only valid for English. Not for Russian, since there are very few examples of unseen combinations of predicates.\n\n\nArguments seen in training?: both\\_seen, Frequency English: 518, Frequency Russian: 1075\nArguments seen in training?: both\\_unseen, Frequency English: 1177, Frequency Russian: 4\nArguments seen in training?: arg1\\_unseen, Frequency English: 56, Frequency Russian: 19\nArguments seen in training?: arg2\\_unseen, Frequency English: 28, Frequency Russian: 4\n\n\nSelection 5 (shape): repeated subjects. For this selection, the subsets are based on the times a subject is repeated in the input; it only takes into account the maximum number of times a subject is repeated, that is, if in one input a subject appears 3 times and a different subject 2 times, this input will be in the \"3\\_subjects\\_same' split. Unique\\_subjects means all subjects are different.\n\n\nMax num. of repeated subjects: unique\\_subjects, Frequency English: 453, Frequency Russian: 339\nMax num. of repeated subjects: 2\\_subjects\\_same, Frequency English: 414, Frequency Russian: 316\nMax num. of repeated subjects: 3\\_subjects\\_same, Frequency English: 382, Frequency Russian: 217\nMax num. of repeated subjects: 4\\_subjects\\_same, Frequency English: 251, Frequency Russian: 143\nMax num. of repeated subjects: 5\\_subjects\\_same, Frequency English: 158, Frequency Russian: 56\nMax num. of repeated subjects: 6\\_subjects\\_same, Frequency English: 80, Frequency Russian: 19\nMax num. of repeated subjects: 7\\_subjects\\_same, Frequency English: 41, Frequency Russian: 12\n\n\nSelection 6 (shape): repeated objects. Same as for subjects above, but for objects. There are much less cases of repeated objects, so there are only two categories for this selection, unique\\_objects and some\\_objects\\_repeated; for the latter, we have up to 3 coreferring objects in English, and XXX in Russian.\n\n\nMax num. of repeated objects: unique\\_objects, Frequency English: 1654, Frequency Russian: 1099\nMax num. of repeated objects: some\\_objects\\_same, Frequency English: 125, Frequency Russian: 3\n\n\nSelection 7 (shape): repeated properties. Same as for objects above, but for properties; up to two properties can be the same in English, up to XXX in Russian.\n\n\nMax num. of repeated properties: unique\\_properties, Frequency English: 1510, Frequency Russian: 986\nMax num. of repeated properties: some\\_properties\\_same, Frequency English: 269, Frequency Russian: 116\n\n\nSelection 8 (shape): entities that appear both as subject and object. For this selection, we grouped together the inputs in which no entity is found as both subject and object, and on the other side inputs in which one or more entity/ies appear both as subject and as object. We found up to two such entities per input in English, and up to XXX in Russian.\n\n\nMax num. of objects and subjects in common: unique\\_properties, Frequency English: 1322, Frequency Russian: 642\nMax num. of objects and subjects in common: some\\_properties\\_same, Frequency English: 457, Frequency Russian: 460", "#### Split Motivation\n\n\nRobustness", "### Getting Started with the Task", "#### Pointers to Resources\n\n\nDataset construction: main dataset paper, RDF triple extraction, Russian translation\n\n\nWebNLG Challenge 2017: webpage, paper\n\n\nWebNLG Challenge 2020: webpage, paper\n\n\nEnriched version of WebNLG: repository, paper\n\n\nRelated research papers: webpage\n\n\nPrevious Results\n----------------", "### Previous Results", "#### Proposed Evaluation\n\n\nFor both languages, the participating systems are automatically evaluated in a multi-reference scenario. Each English hypothesis is compared to a maximum of 5 references, and each Russian one to a maximum of 7 references. On average, English data has 2.89 references per test instance, and Russian data has 2.52 references per instance.\n\n\nIn a human evaluation, example are uniformly sampled across size of triple sets and the following dimensions are assessed (on MTurk and Yandex.Toloka):\n\n\n1. Data Coverage: Does the text include descriptions of all predicates presented in the data?\n2. Relevance: Does the text describe only such predicates (with related subjects and objects), which are found in the data?\n3. Correctness: When describing predicates which are found in the data, does the text mention correct the objects and adequately introduces the subject for this specific predicate?\n4. Text Structure: Is the text grammatical, well-structured, written in acceptable English language?\n5. Fluency: Is it possible to say that the text progresses naturally, forms a coherent whole and it is easy to understand the text?\n\n\nFor additional information like the instructions, we refer to the original paper.", "#### Previous results available?\n\n\nyes", "#### Other Evaluation Approaches\n\n\nWe evaluated a wide range of models as part of the GEM benchmark.", "#### Relevant Previous Results\n\n\nResults can be found on the GEM website.\n\n\nBroader Social Context\n----------------------", "### Previous Work on the Social Impact of the Dataset", "#### Usage of Models based on the Data\n\n\nyes - related tasks", "#### Social Impact Observations\n\n\nWe do not foresee any negative social impact in particular from this dataset or task.\n\n\nPositive outlooks: Being able to generate good quality text from RDF data would permit, e.g., making this data more accessible to lay users, enriching existing text with information drawn from knowledge bases such as DBpedia or describing, comparing and relating entities present in these knowledge bases.", "### Impact on Under-Served Communities", "#### Addresses needs of underserved Communities?\n\n\nno", "### Discussion of Biases", "#### Any Documented Social Biases?\n\n\nyes", "#### Links and Summaries of Analysis Work\n\n\nThis dataset is created using DBpedia RDF triples which naturally exhibit biases that have been found to exist in Wikipedia such as some forms of, e.g., gender bias.\n\n\nThe choice of entities, described by RDF trees, was not controlled. As such, they may contain gender biases; for instance, all the astronauts described by RDF triples are male. Hence, in texts, pronouns *he/him/his* occur more often. Similarly, entities can be related to the Western culture more often than to other cultures.", "#### Are the Language Producers Representative of the Language?\n\n\nIn English, the dataset is limited to the language that crowdraters speak. In Russian, the language is heavily biased by the translationese of the translation system that is post-edited.\n\n\nConsiderations for Using the Data\n---------------------------------", "### PII Risks and Liability", "#### Potential PII Risk\n\n\nThere is no PII in this dataset.", "### Licenses", "#### Copyright Restrictions on the Dataset\n\n\n'non-commercial use only'", "#### Copyright Restrictions on the Language Data\n\n\n'public domain'", "### Known Technical Limitations", "#### Technical Limitations\n\n\nThe quality of the crowdsourced references is limited, in particular in terms of fluency/naturalness of the collected texts.\n\n\nRussian data was machine-translated and then post-edited by crowdworkers, so some examples may still exhibit issues related to bad translations.", "#### Unsuited Applications\n\n\nOnly a limited number of domains are covered in this dataset. As a result, it cannot be used as a general-purpose realizer." ]

[ "TAGS\n#task_categories-table-to-text #annotations_creators-unknown #language_creators-unknown #multilinguality-unknown #size_categories-unknown #source_datasets-original #language-English #license-cc-by-nc-4.0 #data-to-text #region-us \n", "### Link to Main Data Card\n\n\nYou can find the main data card on the GEM Website.", "### Dataset Summary\n\n\nWebNLG is a bi-lingual dataset (English, Russian) of parallel DBpedia triple sets and short texts that cover about 450 different DBpedia properties. The WebNLG data was originally created to promote the development of RDF verbalisers able to generate short text and to handle micro-planning (i.e., sentence segmentation and ordering, referring expression generation, aggregation); the goal of the task is to generate texts starting from 1 to 7 input triples which have entities in common (so the input is actually a connected Knowledge Graph). The dataset contains about 17,000 triple sets and 45,000 crowdsourced texts in English, and 7,000 triples sets and 19,000 crowdsourced texts in Russian. A challenging test set section with entities and/or properties that have not been seen at training time is available.\n\n\nYou can load the dataset via:\n\n\nThe data loader can be found here.", "#### website\n\n\nWebsite", "#### paper\n\n\nFirst Dataset Release, WebNLG Challenge 2017 Report, WebNLG Challenge 2020 Report", "#### authors\n\n\nThe principle curator of the dataset is Anastasia Shimorina (Université de Lorraine / LORIA, France). Throughout the WebNLG releases, several people contributed to their construction: Claire Gardent (CNRS / LORIA, France), Shashi Narayan (Google, UK), Laura Perez-Beltrachini (University of Edinburgh, UK), Elena Khasanova, and Thiago Castro Ferreira (Federal University of Minas Gerais, Brazil).\n\n\nDataset Overview\n----------------", "### Where to find the Data and its Documentation", "#### Webpage\n\n\nWebsite", "#### Download\n\n\nGitlab", "#### Paper\n\n\nFirst Dataset Release, WebNLG Challenge 2017 Report, WebNLG Challenge 2020 Report", "#### BibTex\n\n\nInitial release of the dataset:\n\n\nThe latest version 3.0:", "#### Contact Email\n\n\nwebnlg-challenge@URL", "#### Has a Leaderboard?\n\n\nyes", "#### Leaderboard Link\n\n\nWebsite", "#### Leaderboard Details\n\n\nThe model outputs are evaluated against the crowdsourced references; the leaderboard reports BLEU-4, METEOR, chrF++, TER, BERTScore and BLEURT scores.", "### Languages and Intended Use", "#### Multilingual?\n\n\nyes", "#### Covered Languages\n\n\n'Russian', 'English'", "#### License\n\n\ncc-by-nc-4.0: Creative Commons Attribution Non Commercial 4.0 International", "#### Intended Use\n\n\nThe WebNLG dataset was created to promote the development (*i*) of RDF verbalisers and (*ii*) of microplanners able to handle a wide range of linguistic constructions. The dataset aims at covering knowledge in different domains (\"categories\"). The same properties and entities can appear in several categories.", "#### Primary Task\n\n\nData-to-Text", "#### Communicative Goal\n\n\nA model should verbalize all and only the provided input triples in natural language.", "### Credit", "#### Curation Organization Type(s)\n\n\n'academic'", "#### Curation Organization(s)\n\n\nUniversité de Lorraine / LORIA, France, CNRS / LORIA, France, University of Edinburgh, UK, Federal University of Minas Gerais, Brazil", "#### Dataset Creators\n\n\nThe principle curator of the dataset is Anastasia Shimorina (Université de Lorraine / LORIA, France). Throughout the WebNLG releases, several people contributed to their construction: Claire Gardent (CNRS / LORIA, France), Shashi Narayan (Google, UK), Laura Perez-Beltrachini (University of Edinburgh, UK), Elena Khasanova, and Thiago Castro Ferreira (Federal University of Minas Gerais, Brazil).", "#### Funding\n\n\nThe dataset construction was funded by the French National Research Agency (ANR).", "#### Who added the Dataset to GEM?\n\n\nSimon Mille and Sebastian Gehrmann added the dataset and wrote the data card.", "### Dataset Structure", "#### Data Fields\n\n\nSee official documentation.\n\n\n'entry': a data instance of the benchmark. Each entry has five attributes: a DBpedia category ('category'), entry ID ('eid'), shape, shape type, and triple set size ('size').\n\n\n* 'shape': a string representation of the RDF tree with nested parentheses where 'X' is a node (see Newick tree format).\n* 'shape\\_type': a type of the tree shape. We identify three types of tree shapes:\n\n\n\t+ 'chain' (the object of one triple is the subject of the other);\n\t+ 'sibling' (triples with a shared subject);\n\t+ 'mixed' (both 'chain' and 'sibling' types present).\n* 'eid': an entry ID. It is unique only within a category and a size.\n* 'category': a DBpedia category (Astronaut, City, MusicalWork, Politician, etc.).\n* 'size': the number of RDF triples in a set. Ranges from 1 to 7.\n\n\nEach 'entry' has three fields: 'originaltripleset', 'modifiedtripleset', and 'lexs'.\n\n\n'originaltripleset': a set of RDF triples as extracted from DBpedia. Each set of RDF triples is a tree. Triples have the subject-predicate-object structure.\n\n\n'modifiedtripleset': a set of RDF triples as presented to crowdworkers (for more details on modifications, see below).\n\n\nOriginal and modified triples serve different purposes: the original triples — to link data to a knowledge base (DBpedia), whereas the modified triples — to ensure consistency and homogeneity throughout the data. To train models, the modified triples should be used.\n\n\n'lexs' (shortened for lexicalisations): a natural language text verbalising the triples. Each lexicalisation has two attributes: a comment ('comment'), and a lexicalisation ID ('lid'). By default, comments have the value 'good', except rare cases when they were manually marked as 'toFix'. That was done during the corpus creation, when it was seen that a lexicalisation did not exactly match a triple set.\n\n\nRussian data has additional optional fields comparing to English:\n\n\n'': RDF triples extracted from DBpedia between English and Russian entities by means of the property 'sameAs'.\n\n\n'': RDF triples created manually for some entities to serve as pointers to translators. There are two types of them:\n\n\n* with 'sameAs' ('Spaniards | sameAs | испанцы')\n* with 'includes' ('Tomatoes, guanciale, cheese, olive oil | includes | гуанчиале'). Those were mostly created for string literals to translate some parts of them.\n\n\nLexicalisations in the Russian WebNLG have a new parameter 'lang' (values: 'en', 'ru') because original English texts were kept in the Russian version (see the example above).", "#### Example Instance\n\n\nThe XML-formatted example is here.", "#### Data Splits\n\n\n\n\nDataset in GEM\n--------------", "### Rationale for Inclusion in GEM", "#### Why is the Dataset in GEM?\n\n\nDue to the constrained generation task, this dataset can be used to evaluate very specific and narrow generation capabilities.", "#### Similar Datasets\n\n\nyes", "#### Unique Language Coverage\n\n\nyes", "#### Difference from other GEM datasets\n\n\nThe RDF-triple format is unique to WebNLG.", "#### Ability that the Dataset measures\n\n\nsurface realization", "### GEM-Specific Curation", "#### Modificatied for GEM?\n\n\nyes", "#### GEM Modifications\n\n\n'other'", "#### Modification Details\n\n\nNo changes to the main content of the dataset. The version 3.0 of the dataset is used.", "#### Additional Splits?\n\n\nyes", "#### Split Information\n\n\n23 special test sets for WebNLG were added to the GEM evaluation suite, 12 for English and 11 for Russian.\nFor both languages, we created subsets of the training and development sets of ~500 randomly selected inputs each. The inputs were sampled proportionally from each category.\n\n\nTwo types of transformations have been applied to WebNLG: (i) input scrambling (English and Russian) and (ii) numerical value replacements (English); in both cases, a subset of about 500 inputs was randomly selected. For (i), the order of the triples was randomly reassigned (each triple kept the same Subject-Property-Object internal order). For (ii), the change was performed respecting the format of the current cardinal value (e.g., alpha, integer, or floating-point) and replacing it with a new random value. The new number is lower-bounded between zero and upper bounded to be within to the highest power of 10 unit for the given value (e.g., replacing 54 would result in a random value between 0-100). Floating values maintain the degree of precision.\n\n\nFor both languages, we did identify different subsets of the test set that we could compare to each other so that we would have a better understanding of the results. There are currently 8 selections that we have made:\n\n\nSelection 1 (size): input length. This selection corresponds to the number of predicates in the input. By comparing inputs of different lengths, we can see to what extent NLG systems are able to handle different input sizes. The table below provides the relevant frequencies. Please be aware that comparing selections with fewer than 100 items may result in unreliable comparisons.\n\n\nInput length: 1, Frequency English: 369, Frequency Russian: 254\nInput length: 2, Frequency English: 349, Frequency Russian: 200\nInput length: 3, Frequency English: 350, Frequency Russian: 214\nInput length: 4, Frequency English: 305, Frequency Russian: 214\nInput length: 5, Frequency English: 213, Frequency Russian: 159\nInput length: 6, Frequency English: 114, Frequency Russian: 32\nInput length: 7, Frequency English: 79, Frequency Russian: 29\n\n\nSelection 2 (frequency): seen/unseen single predicates. This selection corresponds to the inputs with only one predicate. We compare which predicates are seen/unseen in the training data. The table below provides the relevant frequencies. Note that the comparison is only valid for English. Not for Russian, since there is only one example of unseen single predicates.\n\n\n\\_ in training: Seen, Frequency English: 297, Frequency Russian: 253\n\\_ in training: Unseen, Frequency English: 72, Frequency Russian: 1\n\n\nSelection 3 (frequency): seen/unseen combinations of predicates. This selection checks for all combinations of predicates whether that combination has been seen in the training data. For example: if the combination of predicates A and B is seen, that means that there is an input in the training data consisting of two triples, where one triple uses predicate A and the other uses predicate B. If the combination is unseen, then the converse is true. The table below provides the relevant frequencies.\n\n\n\\_ in training: unseen, Frequency English: 1295, Frequency Russian: 354\n\\_ in training: seen, Frequency English: 115, Frequency Russian: 494\n\n\nSelection 4 (frequency): seen/unseen arguments. This selection checks for all input whether or not all arg1s and arg2s in the input have been seen during the training phase. For this selection, *Seen* is the default. Only if all arg1 instances for a particular input are unseen, do we count the arg1s of the input as unseen. The same holds for arg2. So \"seen\" here really means that at least some of the arg1s or arg2s are seen in the input. The table below provides the relevant frequencies. Note that the comparison is only valid for English. Not for Russian, since there are very few examples of unseen combinations of predicates.\n\n\nArguments seen in training?: both\\_seen, Frequency English: 518, Frequency Russian: 1075\nArguments seen in training?: both\\_unseen, Frequency English: 1177, Frequency Russian: 4\nArguments seen in training?: arg1\\_unseen, Frequency English: 56, Frequency Russian: 19\nArguments seen in training?: arg2\\_unseen, Frequency English: 28, Frequency Russian: 4\n\n\nSelection 5 (shape): repeated subjects. For this selection, the subsets are based on the times a subject is repeated in the input; it only takes into account the maximum number of times a subject is repeated, that is, if in one input a subject appears 3 times and a different subject 2 times, this input will be in the \"3\\_subjects\\_same' split. Unique\\_subjects means all subjects are different.\n\n\nMax num. of repeated subjects: unique\\_subjects, Frequency English: 453, Frequency Russian: 339\nMax num. of repeated subjects: 2\\_subjects\\_same, Frequency English: 414, Frequency Russian: 316\nMax num. of repeated subjects: 3\\_subjects\\_same, Frequency English: 382, Frequency Russian: 217\nMax num. of repeated subjects: 4\\_subjects\\_same, Frequency English: 251, Frequency Russian: 143\nMax num. of repeated subjects: 5\\_subjects\\_same, Frequency English: 158, Frequency Russian: 56\nMax num. of repeated subjects: 6\\_subjects\\_same, Frequency English: 80, Frequency Russian: 19\nMax num. of repeated subjects: 7\\_subjects\\_same, Frequency English: 41, Frequency Russian: 12\n\n\nSelection 6 (shape): repeated objects. Same as for subjects above, but for objects. There are much less cases of repeated objects, so there are only two categories for this selection, unique\\_objects and some\\_objects\\_repeated; for the latter, we have up to 3 coreferring objects in English, and XXX in Russian.\n\n\nMax num. of repeated objects: unique\\_objects, Frequency English: 1654, Frequency Russian: 1099\nMax num. of repeated objects: some\\_objects\\_same, Frequency English: 125, Frequency Russian: 3\n\n\nSelection 7 (shape): repeated properties. Same as for objects above, but for properties; up to two properties can be the same in English, up to XXX in Russian.\n\n\nMax num. of repeated properties: unique\\_properties, Frequency English: 1510, Frequency Russian: 986\nMax num. of repeated properties: some\\_properties\\_same, Frequency English: 269, Frequency Russian: 116\n\n\nSelection 8 (shape): entities that appear both as subject and object. For this selection, we grouped together the inputs in which no entity is found as both subject and object, and on the other side inputs in which one or more entity/ies appear both as subject and as object. We found up to two such entities per input in English, and up to XXX in Russian.\n\n\nMax num. of objects and subjects in common: unique\\_properties, Frequency English: 1322, Frequency Russian: 642\nMax num. of objects and subjects in common: some\\_properties\\_same, Frequency English: 457, Frequency Russian: 460", "#### Split Motivation\n\n\nRobustness", "### Getting Started with the Task", "#### Pointers to Resources\n\n\nDataset construction: main dataset paper, RDF triple extraction, Russian translation\n\n\nWebNLG Challenge 2017: webpage, paper\n\n\nWebNLG Challenge 2020: webpage, paper\n\n\nEnriched version of WebNLG: repository, paper\n\n\nRelated research papers: webpage\n\n\nPrevious Results\n----------------", "### Previous Results", "#### Proposed Evaluation\n\n\nFor both languages, the participating systems are automatically evaluated in a multi-reference scenario. Each English hypothesis is compared to a maximum of 5 references, and each Russian one to a maximum of 7 references. On average, English data has 2.89 references per test instance, and Russian data has 2.52 references per instance.\n\n\nIn a human evaluation, example are uniformly sampled across size of triple sets and the following dimensions are assessed (on MTurk and Yandex.Toloka):\n\n\n1. Data Coverage: Does the text include descriptions of all predicates presented in the data?\n2. Relevance: Does the text describe only such predicates (with related subjects and objects), which are found in the data?\n3. Correctness: When describing predicates which are found in the data, does the text mention correct the objects and adequately introduces the subject for this specific predicate?\n4. Text Structure: Is the text grammatical, well-structured, written in acceptable English language?\n5. Fluency: Is it possible to say that the text progresses naturally, forms a coherent whole and it is easy to understand the text?\n\n\nFor additional information like the instructions, we refer to the original paper.", "#### Previous results available?\n\n\nyes", "#### Other Evaluation Approaches\n\n\nWe evaluated a wide range of models as part of the GEM benchmark.", "#### Relevant Previous Results\n\n\nResults can be found on the GEM website.\n\n\nBroader Social Context\n----------------------", "### Previous Work on the Social Impact of the Dataset", "#### Usage of Models based on the Data\n\n\nyes - related tasks", "#### Social Impact Observations\n\n\nWe do not foresee any negative social impact in particular from this dataset or task.\n\n\nPositive outlooks: Being able to generate good quality text from RDF data would permit, e.g., making this data more accessible to lay users, enriching existing text with information drawn from knowledge bases such as DBpedia or describing, comparing and relating entities present in these knowledge bases.", "### Impact on Under-Served Communities", "#### Addresses needs of underserved Communities?\n\n\nno", "### Discussion of Biases", "#### Any Documented Social Biases?\n\n\nyes", "#### Links and Summaries of Analysis Work\n\n\nThis dataset is created using DBpedia RDF triples which naturally exhibit biases that have been found to exist in Wikipedia such as some forms of, e.g., gender bias.\n\n\nThe choice of entities, described by RDF trees, was not controlled. As such, they may contain gender biases; for instance, all the astronauts described by RDF triples are male. Hence, in texts, pronouns *he/him/his* occur more often. Similarly, entities can be related to the Western culture more often than to other cultures.", "#### Are the Language Producers Representative of the Language?\n\n\nIn English, the dataset is limited to the language that crowdraters speak. In Russian, the language is heavily biased by the translationese of the translation system that is post-edited.\n\n\nConsiderations for Using the Data\n---------------------------------", "### PII Risks and Liability", "#### Potential PII Risk\n\n\nThere is no PII in this dataset.", "### Licenses", "#### Copyright Restrictions on the Dataset\n\n\n'non-commercial use only'", "#### Copyright Restrictions on the Language Data\n\n\n'public domain'", "### Known Technical Limitations", "#### Technical Limitations\n\n\nThe quality of the crowdsourced references is limited, in particular in terms of fluency/naturalness of the collected texts.\n\n\nRussian data was machine-translated and then post-edited by crowdworkers, so some examples may still exhibit issues related to bad translations.", "#### Unsuited Applications\n\n\nOnly a limited number of domains are covered in this dataset. As a result, it cannot be used as a general-purpose realizer." ]

[ "task_categories:text2text-generation", "task_ids:text-simplification", "annotations_creators:crowd-sourced", "language_creators:unknown", "multilinguality:unknown", "size_categories:unknown", "source_datasets:original", "language:en", "license:other", "arxiv:1910.02677", "arxiv:2005.00352", "region:us" ]

[ "1910.02677", "2005.00352" ]

[ "en" ]

[ "### Link to Main Data Card\n\n\nYou can find the main data card on the [GEM Website.", "### Dataset Summary\n\n\nWikiAuto is an English simplification dataset that we paired with ASSET and TURK, two very high-quality evaluation datasets, as test sets. The input is an English sentence taken from Wikipedia and the target a simplified sentence. ASSET and TURK contain the same test examples but have references that are simplified in different ways (splitting sentences vs. rewriting and splitting).\n\n\nYou can load the dataset via:\n\n\nThe data loader can be found here.", "#### website\n\n\nn/a", "#### paper\n\n\nWikiAuto, ASSET, TURK", "#### authors\n\n\nWikiAuto: Chao Jiang, Mounica Maddela, Wuwei Lan, Yang Zhong, Wei Xu; ASSET: Fernando Alva-Manchego, Louis Martin, Antoine Bordes, Carolina Scarton, and Benoîıt Sagot, and Lucia Specia; TURK: Wei Xu, Courtney Napoles, Ellie Pavlick, Quanze Chen, and Chris Callison-Burch\n\n\nDataset Overview\n----------------", "### Where to find the Data and its Documentation", "#### Download\n\n\nWiki-Auto repository, ASSET repository, TURKCorpus", "#### Paper\n\n\nWikiAuto, ASSET, TURK", "#### BibTex\n\n\nWikiAuto:\n\n\nASSET:\n\n\nTURK:", "#### Contact Name\n\n\nWikiAuto: Chao Jiang; ASSET: Fernando Alva-Manchego and Louis Martin; TURK: Wei Xu", "#### Contact Email\n\n\njiang.1530@URL, f.alva@URL, louismartincs@URL, URL@URL", "#### Has a Leaderboard?\n\n\nno", "### Languages and Intended Use", "#### Multilingual?\n\n\nno", "#### Covered Languages\n\n\n'English'", "#### Whose Language?\n\n\nWiki-Auto contains English text only (BCP-47: 'en'). It is presented as a translation task where Wikipedia Simple English is treated as its own idiom. For a statement of what is intended (but not always observed) to constitute Simple English on this platform, see Simple English in Wikipedia.\nBoth ASSET and TURK use crowdsourcing to change references, and their language is thus a combination of the WikiAuto data and the language of the demographic on mechanical Turk", "#### License\n\n\nother: Other license", "#### Intended Use\n\n\nWikiAuto provides a set of aligned sentences from English Wikipedia and Simple English Wikipedia as a resource to train sentence simplification systems.\n\n\nThe authors first crowd-sourced a set of manual alignments between sentences in a subset of the Simple English Wikipedia and their corresponding versions in English Wikipedia (this corresponds to the 'manual' config in this version of the dataset), then trained a neural CRF system to predict these alignments.\n\n\nThe trained alignment prediction model was then applied to the other articles in Simple English Wikipedia with an English counterpart to create a larger corpus of aligned sentences (corresponding to the 'auto' and 'auto\\_acl' configs here).\n\n\nASSET (Alva-Manchego et al., 2020) is multi-reference dataset for the evaluation of sentence simplification in English. The dataset uses the same 2,359 sentences from TurkCorpus (Xu et al., 2016) and each sentence is associated with 10 crowdsourced simplifications. Unlike previous simplification datasets, which contain a single transformation (e.g., lexical paraphrasing in TurkCorpus or sentence\nsplitting in HSplit), the simplifications in ASSET encompass a variety of rewriting transformations.\n\n\nTURKCorpus is a high quality simplification dataset where each source (not simple) sentence is associated with 8 human-written simplifications that focus on lexical paraphrasing. It is one of the two evaluation datasets for the text simplification task in GEM. It acts as the validation and test set for paraphrasing-based simplification that does not involve sentence splitting and deletion.", "#### Add. License Info\n\n\nWikiAuto: 'CC BY-NC 3.0', ASSET: 'CC BY-NC 4.0', TURK: 'GNU General Public License v3.0'", "#### Primary Task\n\n\nSimplification", "#### Communicative Goal\n\n\nThe goal is to communicate the main ideas of source sentence in a way that is easier to understand by non-native speakers of English.", "### Credit", "#### Curation Organization Type(s)\n\n\n'academic', 'industry'", "#### Curation Organization(s)\n\n\nOhio State University, University of Sheffield, Inria, Facebook AI Research, Imperial College London, University of Pennsylvania, John Hopkins University", "#### Dataset Creators\n\n\nWikiAuto: Chao Jiang, Mounica Maddela, Wuwei Lan, Yang Zhong, Wei Xu; ASSET: Fernando Alva-Manchego, Louis Martin, Antoine Bordes, Carolina Scarton, and Benoîıt Sagot, and Lucia Specia; TURK: Wei Xu, Courtney Napoles, Ellie Pavlick, Quanze Chen, and Chris Callison-Burch", "#### Funding\n\n\nWikiAuto: NSF, ODNI, IARPA, Figure Eight AI, and Criteo. ASSET: PRAIRIE Institute, ANR. TURK: NSF", "#### Who added the Dataset to GEM?\n\n\nGEM v1 had separate data cards for WikiAuto, ASSET, and TURK. They were contributed by Dhruv Kumar and Mounica Maddela. The initial data loader was written by Yacine Jernite. Sebastian Gehrmann merged and extended the data cards and migrated the loader to the v2 infrastructure.", "### Dataset Structure", "#### Data Fields\n\n\n* 'source': A source sentence from one of the datasets\n* 'target': A single simplified sentence corresponding to 'source'\n* 'references': In the case of ASSET/TURK, references is a list of strings corresponding to the different references.", "#### Reason for Structure\n\n\nThe underlying datasets have extensive secondary annotations that can be used in conjunction with the GEM version. We omit those annotations to simplify the format into one that can be used by seq2seq models.", "#### Example Instance", "#### Data Splits\n\n\nIn WikiAuto, which is used as training and validation set, the following splits are provided:\n\n\n\nASSET does not contain a training set; many models use WikiLarge (Zhang and Lapata, 2017) for training. For GEM, Wiki-Auto will be used for training the model.\n\n\nEach input sentence has 10 associated reference simplified sentences. The statistics of ASSET are given below.\n\n\n\nThe test and validation sets are the same as those of TurkCorpus. The split was random.\n\n\nThere are 19.04 tokens per reference on average (lower than 21.29 and 25.49 for TurkCorpus and HSplit, respectively). Most (17,245) of the referece sentences do not involve sentence splitting.\n\n\nTURKCorpus does not contain a training set; many models use WikiLarge (Zhang and Lapata, 2017) or Wiki-Auto (Jiang et. al 2020) for training.\n\n\nEach input sentence has 8 associated reference simplified sentences. 2,359 input sentences are randomly split into 2,000 validation and 359 test sentences.\n\n\n\nThere are 21.29 tokens per reference on average.", "#### Splitting Criteria\n\n\nIn our setup, we use WikiAuto as training/validation corpus and ASSET and TURK as test corpora. ASSET and TURK have the same inputs but differ in their reference style. Researchers can thus conduct targeted evaluations based on the strategies that a model should learn.\n\n\nDataset in GEM\n--------------", "### Rationale for Inclusion in GEM", "#### Why is the Dataset in GEM?\n\n\nWikiAuto is the largest open text simplification dataset currently available. ASSET and TURK are high quality test sets that are compatible with WikiAuto.", "#### Similar Datasets\n\n\nyes", "#### Unique Language Coverage\n\n\nno", "#### Difference from other GEM datasets\n\n\nIt's unique setup with multiple test sets makes the task interesting since it allows for evaluation of multiple generations and systems that simplify in different ways.", "#### Ability that the Dataset measures\n\n\nsimplification", "### GEM-Specific Curation", "#### Modificatied for GEM?\n\n\nyes", "#### GEM Modifications\n\n\n'other'", "#### Modification Details\n\n\nWe removed secondary annotations and focus on the simple 'input->output' format, but combine the different sub-datasets.", "#### Additional Splits?\n\n\nyes", "#### Split Information\n\n\nwe split the original test set according to syntactic complexity of the source sentences. To characterize sentence syntactic complexity, we use the 8-level developmental level (d-level) scale proposed by Covington et al. (2006) and the implementation of Lu, Xiaofei (2010).\nWe thus split the original test set into 8 subsets corresponding to the 8 d-levels assigned to source sentences. We obtain the following number of instances per level and average d-level of the dataset:", "#### Split Motivation\n\n\nThe goal was to assess performance when simplifying source sentences with different syntactic structure and complexity.", "### Getting Started with the Task", "#### Pointers to Resources\n\n\nThere are recent supervised (Martin et al., 2019, Kriz et al., 2019, Dong et al., 2019, Zhang and Lapata, 2017) and unsupervised (Martin et al., 2020, Kumar et al., 2020, Surya et al., 2019) text simplification models that can be used as baselines.", "#### Technical Terms\n\n\nThe common metric used for automatic evaluation is SARI (Xu et al., 2016).\n\n\nPrevious Results\n----------------", "### Previous Results", "#### Measured Model Abilities\n\n\nSimplification", "#### Metrics\n\n\n'Other: Other Metrics', 'BLEU'", "#### Other Metrics\n\n\nSARI: A simplification metric that considers both input and references to measure the \"goodness\" of words that are added, deleted, and kept.", "#### Proposed Evaluation\n\n\nThe original authors of WikiAuto and ASSET used human evaluation to assess the fluency, adequacy, and simplicity (details provided in the paper). For TURK, the authors measured grammaticality, meaning-preservation, and simplicity gain (details in the paper).", "#### Previous results available?\n\n\nno\n\n\nDataset Curation\n----------------", "### Original Curation", "#### Original Curation Rationale\n\n\nWiki-Auto provides a new version of the Wikipedia corpus that is larger, contains 75% less defective pairs and has more complex rewrites than the previous WIKILARGE dataset.\n\n\nASSET was created in order to improve the evaluation of sentence simplification. It uses the same input sentences as the TurkCorpus dataset from (Xu et al., 2016). The 2,359 input sentences of TurkCorpus are a sample of \"standard\" (not simple) sentences from the Parallel Wikipedia Simplification (PWKP) dataset (Zhu et al., 2010), which come from the August 22, 2009 version of Wikipedia. The sentences of TurkCorpus were chosen to be of similar length (Xu et al., 2016). No further information is provided on the sampling strategy.\n\n\nThe TurkCorpus dataset was developed in order to overcome some of the problems with sentence pairs from Standard and Simple Wikipedia: a large fraction of sentences were misaligned, or not actually simpler (Xu et al., 2016). However, TurkCorpus mainly focused on *lexical paraphrasing*, and so cannot be used to evaluate simplifications involving *compression* (deletion) or *sentence splitting*. HSplit (Sulem et al., 2018), on the other hand, can only be used to evaluate sentence splitting. The reference sentences in ASSET include a wider variety of sentence rewriting strategies, combining splitting, compression and paraphrasing. Annotators were given examples of each kind of transformation individually, as well as all three transformations used at once, but were allowed to decide which transformations to use for any given sentence.\n\n\nAn example illustrating the differences between TurkCorpus, HSplit and ASSET is given below:\n\n\n\n> \n> Original: He settled in London, devoting himself chiefly to practical teaching.\n> \n> \n> TurkCorpus: He rooted in London, devoting himself mainly to practical teaching.\n> \n> \n> HSplit: He settled in London. He devoted himself chiefly to practical teaching.\n> \n> \n> ASSET: He lived in London. He was a teacher.\n> \n> \n>", "#### Communicative Goal\n\n\nThe goal is to communicate the same information as the source sentence using simpler words and grammar.", "#### Sourced from Different Sources\n\n\nyes", "#### Source Details\n\n\nWikipedia", "### Language Data", "#### How was Language Data Obtained?\n\n\n'Found'", "#### Where was it found?\n\n\n'Single website'", "#### Language Producers\n\n\nThe dataset uses language from Wikipedia: some demographic information is provided here.", "#### Data Validation\n\n\nnot validated", "#### Was Data Filtered?\n\n\nalgorithmically", "#### Filter Criteria\n\n\nThe authors mention that they \"extracted 138,095 article pairs from the 2019/09 Wikipedia dump using an improved version of the WikiExtractor library\". The SpaCy library is used for sentence splitting.", "### Structured Annotations", "#### Additional Annotations?\n\n\ncrowd-sourced", "#### Number of Raters\n\n\n11<n<50", "#### Rater Qualifications\n\n\nWikiAuto (Figure Eight): No information provided.\n\n\nASSET (MTurk):\n\n\n* Having a HIT approval rate over 95%, and over 1000 HITs approved. No other demographic or compensation information is provided.\n* Passing a Qualification Test (appropriately simplifying sentences). Out of 100 workers, 42 passed the test.\n* Being a resident of the United States, United Kingdom or Canada.\n\n\nTURK (MTurk):\n\n\n* Reference sentences were written by workers with HIT approval rate over 95%. No other demographic or compensation information is provided.", "#### Raters per Training Example\n\n\n1", "#### Raters per Test Example\n\n\n\n> \n> 5\n> \n> \n>", "#### Annotation Service?\n\n\nyes", "#### Which Annotation Service\n\n\n'Amazon Mechanical Turk', 'Appen'", "#### Annotation Values\n\n\nWikiAuto: Sentence alignment labels were crowdsourced for 500 randomly sampled document pairs (10,123 sentence pairs total). The authors pre-selected several alignment candidates from English Wikipedia for each Simple Wikipedia sentence based on various similarity metrics, then asked the crowd-workers to annotate these pairs. Finally, they trained their alignment model on this manually annotated dataset to obtain automatically aligned sentences (138,095 document pairs, 488,332 sentence pairs).\nNo demographic annotation is provided for the crowd workers. The Figure Eight platform now part of Appen) was used for the annotation process.\n\n\nASSET: The instructions given to the annotators are available here.\n\n\nTURK: The references are crowdsourced from Amazon Mechanical Turk. The annotators were asked to provide simplifications without losing any information or splitting the input sentence. No other demographic or compensation information is provided in the TURKCorpus paper. The instructions given to the annotators are available in the paper.", "#### Any Quality Control?\n\n\nnone", "### Consent", "#### Any Consent Policy?\n\n\nyes", "#### Consent Policy Details\n\n\nBoth Figure Eight and Amazon Mechanical Turk raters forfeit the right to their data as part of their agreements.", "### Private Identifying Information (PII)", "#### Contains PII?\n\n\nno PII", "#### Justification for no PII\n\n\nSince the dataset is created from Wikipedia/Simple Wikipedia, all the information contained in the dataset is already in the public domain.", "### Maintenance", "#### Any Maintenance Plan?\n\n\nno\n\n\nBroader Social Context\n----------------------", "### Previous Work on the Social Impact of the Dataset", "#### Usage of Models based on the Data\n\n\nno", "### Impact on Under-Served Communities", "#### Addresses needs of underserved Communities?\n\n\nno", "### Discussion of Biases", "#### Any Documented Social Biases?\n\n\nyes", "#### Links and Summaries of Analysis Work\n\n\nThe dataset may contain some social biases, as the input sentences are based on Wikipedia. Studies have shown that the English Wikipedia contains both gender biases (Schmahl et al., 2020) and racial biases (Adams et al., 2019).\n\n\nConsiderations for Using the Data\n---------------------------------", "### PII Risks and Liability", "#### Potential PII Risk\n\n\nAll the data is in the public domain.", "### Licenses", "#### Copyright Restrictions on the Dataset\n\n\n'open license - commercial use allowed'", "#### Copyright Restrictions on the Language Data\n\n\n'open license - commercial use allowed'", "### Known Technical Limitations", "#### Technical Limitations\n\n\nThe dataset may contain some social biases, as the input sentences are based on Wikipedia. Studies have shown that the English Wikipedia contains both gender biases (Schmahl et al., 2020) and racial biases (Adams et al., 2019).", "#### Unsuited Applications\n\n\nSince the test datasets contains only 2,359 sentences that are derived from Wikipedia, they are limited to a small subset of topics present on Wikipedia." ]

[ "TAGS\n#task_categories-text2text-generation #task_ids-text-simplification #annotations_creators-crowd-sourced #language_creators-unknown #multilinguality-unknown #size_categories-unknown #source_datasets-original #language-English #license-other #arxiv-1910.02677 #arxiv-2005.00352 #region-us \n", "### Link to Main Data Card\n\n\nYou can find the main data card on the [GEM Website.", "### Dataset Summary\n\n\nWikiAuto is an English simplification dataset that we paired with ASSET and TURK, two very high-quality evaluation datasets, as test sets. The input is an English sentence taken from Wikipedia and the target a simplified sentence. ASSET and TURK contain the same test examples but have references that are simplified in different ways (splitting sentences vs. rewriting and splitting).\n\n\nYou can load the dataset via:\n\n\nThe data loader can be found here.", "#### website\n\n\nn/a", "#### paper\n\n\nWikiAuto, ASSET, TURK", "#### authors\n\n\nWikiAuto: Chao Jiang, Mounica Maddela, Wuwei Lan, Yang Zhong, Wei Xu; ASSET: Fernando Alva-Manchego, Louis Martin, Antoine Bordes, Carolina Scarton, and Benoîıt Sagot, and Lucia Specia; TURK: Wei Xu, Courtney Napoles, Ellie Pavlick, Quanze Chen, and Chris Callison-Burch\n\n\nDataset Overview\n----------------", "### Where to find the Data and its Documentation", "#### Download\n\n\nWiki-Auto repository, ASSET repository, TURKCorpus", "#### Paper\n\n\nWikiAuto, ASSET, TURK", "#### BibTex\n\n\nWikiAuto:\n\n\nASSET:\n\n\nTURK:", "#### Contact Name\n\n\nWikiAuto: Chao Jiang; ASSET: Fernando Alva-Manchego and Louis Martin; TURK: Wei Xu", "#### Contact Email\n\n\njiang.1530@URL, f.alva@URL, louismartincs@URL, URL@URL", "#### Has a Leaderboard?\n\n\nno", "### Languages and Intended Use", "#### Multilingual?\n\n\nno", "#### Covered Languages\n\n\n'English'", "#### Whose Language?\n\n\nWiki-Auto contains English text only (BCP-47: 'en'). It is presented as a translation task where Wikipedia Simple English is treated as its own idiom. For a statement of what is intended (but not always observed) to constitute Simple English on this platform, see Simple English in Wikipedia.\nBoth ASSET and TURK use crowdsourcing to change references, and their language is thus a combination of the WikiAuto data and the language of the demographic on mechanical Turk", "#### License\n\n\nother: Other license", "#### Intended Use\n\n\nWikiAuto provides a set of aligned sentences from English Wikipedia and Simple English Wikipedia as a resource to train sentence simplification systems.\n\n\nThe authors first crowd-sourced a set of manual alignments between sentences in a subset of the Simple English Wikipedia and their corresponding versions in English Wikipedia (this corresponds to the 'manual' config in this version of the dataset), then trained a neural CRF system to predict these alignments.\n\n\nThe trained alignment prediction model was then applied to the other articles in Simple English Wikipedia with an English counterpart to create a larger corpus of aligned sentences (corresponding to the 'auto' and 'auto\\_acl' configs here).\n\n\nASSET (Alva-Manchego et al., 2020) is multi-reference dataset for the evaluation of sentence simplification in English. The dataset uses the same 2,359 sentences from TurkCorpus (Xu et al., 2016) and each sentence is associated with 10 crowdsourced simplifications. Unlike previous simplification datasets, which contain a single transformation (e.g., lexical paraphrasing in TurkCorpus or sentence\nsplitting in HSplit), the simplifications in ASSET encompass a variety of rewriting transformations.\n\n\nTURKCorpus is a high quality simplification dataset where each source (not simple) sentence is associated with 8 human-written simplifications that focus on lexical paraphrasing. It is one of the two evaluation datasets for the text simplification task in GEM. It acts as the validation and test set for paraphrasing-based simplification that does not involve sentence splitting and deletion.", "#### Add. License Info\n\n\nWikiAuto: 'CC BY-NC 3.0', ASSET: 'CC BY-NC 4.0', TURK: 'GNU General Public License v3.0'", "#### Primary Task\n\n\nSimplification", "#### Communicative Goal\n\n\nThe goal is to communicate the main ideas of source sentence in a way that is easier to understand by non-native speakers of English.", "### Credit", "#### Curation Organization Type(s)\n\n\n'academic', 'industry'", "#### Curation Organization(s)\n\n\nOhio State University, University of Sheffield, Inria, Facebook AI Research, Imperial College London, University of Pennsylvania, John Hopkins University", "#### Dataset Creators\n\n\nWikiAuto: Chao Jiang, Mounica Maddela, Wuwei Lan, Yang Zhong, Wei Xu; ASSET: Fernando Alva-Manchego, Louis Martin, Antoine Bordes, Carolina Scarton, and Benoîıt Sagot, and Lucia Specia; TURK: Wei Xu, Courtney Napoles, Ellie Pavlick, Quanze Chen, and Chris Callison-Burch", "#### Funding\n\n\nWikiAuto: NSF, ODNI, IARPA, Figure Eight AI, and Criteo. ASSET: PRAIRIE Institute, ANR. TURK: NSF", "#### Who added the Dataset to GEM?\n\n\nGEM v1 had separate data cards for WikiAuto, ASSET, and TURK. They were contributed by Dhruv Kumar and Mounica Maddela. The initial data loader was written by Yacine Jernite. Sebastian Gehrmann merged and extended the data cards and migrated the loader to the v2 infrastructure.", "### Dataset Structure", "#### Data Fields\n\n\n* 'source': A source sentence from one of the datasets\n* 'target': A single simplified sentence corresponding to 'source'\n* 'references': In the case of ASSET/TURK, references is a list of strings corresponding to the different references.", "#### Reason for Structure\n\n\nThe underlying datasets have extensive secondary annotations that can be used in conjunction with the GEM version. We omit those annotations to simplify the format into one that can be used by seq2seq models.", "#### Example Instance", "#### Data Splits\n\n\nIn WikiAuto, which is used as training and validation set, the following splits are provided:\n\n\n\nASSET does not contain a training set; many models use WikiLarge (Zhang and Lapata, 2017) for training. For GEM, Wiki-Auto will be used for training the model.\n\n\nEach input sentence has 10 associated reference simplified sentences. The statistics of ASSET are given below.\n\n\n\nThe test and validation sets are the same as those of TurkCorpus. The split was random.\n\n\nThere are 19.04 tokens per reference on average (lower than 21.29 and 25.49 for TurkCorpus and HSplit, respectively). Most (17,245) of the referece sentences do not involve sentence splitting.\n\n\nTURKCorpus does not contain a training set; many models use WikiLarge (Zhang and Lapata, 2017) or Wiki-Auto (Jiang et. al 2020) for training.\n\n\nEach input sentence has 8 associated reference simplified sentences. 2,359 input sentences are randomly split into 2,000 validation and 359 test sentences.\n\n\n\nThere are 21.29 tokens per reference on average.", "#### Splitting Criteria\n\n\nIn our setup, we use WikiAuto as training/validation corpus and ASSET and TURK as test corpora. ASSET and TURK have the same inputs but differ in their reference style. Researchers can thus conduct targeted evaluations based on the strategies that a model should learn.\n\n\nDataset in GEM\n--------------", "### Rationale for Inclusion in GEM", "#### Why is the Dataset in GEM?\n\n\nWikiAuto is the largest open text simplification dataset currently available. ASSET and TURK are high quality test sets that are compatible with WikiAuto.", "#### Similar Datasets\n\n\nyes", "#### Unique Language Coverage\n\n\nno", "#### Difference from other GEM datasets\n\n\nIt's unique setup with multiple test sets makes the task interesting since it allows for evaluation of multiple generations and systems that simplify in different ways.", "#### Ability that the Dataset measures\n\n\nsimplification", "### GEM-Specific Curation", "#### Modificatied for GEM?\n\n\nyes", "#### GEM Modifications\n\n\n'other'", "#### Modification Details\n\n\nWe removed secondary annotations and focus on the simple 'input->output' format, but combine the different sub-datasets.", "#### Additional Splits?\n\n\nyes", "#### Split Information\n\n\nwe split the original test set according to syntactic complexity of the source sentences. To characterize sentence syntactic complexity, we use the 8-level developmental level (d-level) scale proposed by Covington et al. (2006) and the implementation of Lu, Xiaofei (2010).\nWe thus split the original test set into 8 subsets corresponding to the 8 d-levels assigned to source sentences. We obtain the following number of instances per level and average d-level of the dataset:", "#### Split Motivation\n\n\nThe goal was to assess performance when simplifying source sentences with different syntactic structure and complexity.", "### Getting Started with the Task", "#### Pointers to Resources\n\n\nThere are recent supervised (Martin et al., 2019, Kriz et al., 2019, Dong et al., 2019, Zhang and Lapata, 2017) and unsupervised (Martin et al., 2020, Kumar et al., 2020, Surya et al., 2019) text simplification models that can be used as baselines.", "#### Technical Terms\n\n\nThe common metric used for automatic evaluation is SARI (Xu et al., 2016).\n\n\nPrevious Results\n----------------", "### Previous Results", "#### Measured Model Abilities\n\n\nSimplification", "#### Metrics\n\n\n'Other: Other Metrics', 'BLEU'", "#### Other Metrics\n\n\nSARI: A simplification metric that considers both input and references to measure the \"goodness\" of words that are added, deleted, and kept.", "#### Proposed Evaluation\n\n\nThe original authors of WikiAuto and ASSET used human evaluation to assess the fluency, adequacy, and simplicity (details provided in the paper). For TURK, the authors measured grammaticality, meaning-preservation, and simplicity gain (details in the paper).", "#### Previous results available?\n\n\nno\n\n\nDataset Curation\n----------------", "### Original Curation", "#### Original Curation Rationale\n\n\nWiki-Auto provides a new version of the Wikipedia corpus that is larger, contains 75% less defective pairs and has more complex rewrites than the previous WIKILARGE dataset.\n\n\nASSET was created in order to improve the evaluation of sentence simplification. It uses the same input sentences as the TurkCorpus dataset from (Xu et al., 2016). The 2,359 input sentences of TurkCorpus are a sample of \"standard\" (not simple) sentences from the Parallel Wikipedia Simplification (PWKP) dataset (Zhu et al., 2010), which come from the August 22, 2009 version of Wikipedia. The sentences of TurkCorpus were chosen to be of similar length (Xu et al., 2016). No further information is provided on the sampling strategy.\n\n\nThe TurkCorpus dataset was developed in order to overcome some of the problems with sentence pairs from Standard and Simple Wikipedia: a large fraction of sentences were misaligned, or not actually simpler (Xu et al., 2016). However, TurkCorpus mainly focused on *lexical paraphrasing*, and so cannot be used to evaluate simplifications involving *compression* (deletion) or *sentence splitting*. HSplit (Sulem et al., 2018), on the other hand, can only be used to evaluate sentence splitting. The reference sentences in ASSET include a wider variety of sentence rewriting strategies, combining splitting, compression and paraphrasing. Annotators were given examples of each kind of transformation individually, as well as all three transformations used at once, but were allowed to decide which transformations to use for any given sentence.\n\n\nAn example illustrating the differences between TurkCorpus, HSplit and ASSET is given below:\n\n\n\n> \n> Original: He settled in London, devoting himself chiefly to practical teaching.\n> \n> \n> TurkCorpus: He rooted in London, devoting himself mainly to practical teaching.\n> \n> \n> HSplit: He settled in London. He devoted himself chiefly to practical teaching.\n> \n> \n> ASSET: He lived in London. He was a teacher.\n> \n> \n>", "#### Communicative Goal\n\n\nThe goal is to communicate the same information as the source sentence using simpler words and grammar.", "#### Sourced from Different Sources\n\n\nyes", "#### Source Details\n\n\nWikipedia", "### Language Data", "#### How was Language Data Obtained?\n\n\n'Found'", "#### Where was it found?\n\n\n'Single website'", "#### Language Producers\n\n\nThe dataset uses language from Wikipedia: some demographic information is provided here.", "#### Data Validation\n\n\nnot validated", "#### Was Data Filtered?\n\n\nalgorithmically", "#### Filter Criteria\n\n\nThe authors mention that they \"extracted 138,095 article pairs from the 2019/09 Wikipedia dump using an improved version of the WikiExtractor library\". The SpaCy library is used for sentence splitting.", "### Structured Annotations", "#### Additional Annotations?\n\n\ncrowd-sourced", "#### Number of Raters\n\n\n11<n<50", "#### Rater Qualifications\n\n\nWikiAuto (Figure Eight): No information provided.\n\n\nASSET (MTurk):\n\n\n* Having a HIT approval rate over 95%, and over 1000 HITs approved. No other demographic or compensation information is provided.\n* Passing a Qualification Test (appropriately simplifying sentences). Out of 100 workers, 42 passed the test.\n* Being a resident of the United States, United Kingdom or Canada.\n\n\nTURK (MTurk):\n\n\n* Reference sentences were written by workers with HIT approval rate over 95%. No other demographic or compensation information is provided.", "#### Raters per Training Example\n\n\n1", "#### Raters per Test Example\n\n\n\n> \n> 5\n> \n> \n>", "#### Annotation Service?\n\n\nyes", "#### Which Annotation Service\n\n\n'Amazon Mechanical Turk', 'Appen'", "#### Annotation Values\n\n\nWikiAuto: Sentence alignment labels were crowdsourced for 500 randomly sampled document pairs (10,123 sentence pairs total). The authors pre-selected several alignment candidates from English Wikipedia for each Simple Wikipedia sentence based on various similarity metrics, then asked the crowd-workers to annotate these pairs. Finally, they trained their alignment model on this manually annotated dataset to obtain automatically aligned sentences (138,095 document pairs, 488,332 sentence pairs).\nNo demographic annotation is provided for the crowd workers. The Figure Eight platform now part of Appen) was used for the annotation process.\n\n\nASSET: The instructions given to the annotators are available here.\n\n\nTURK: The references are crowdsourced from Amazon Mechanical Turk. The annotators were asked to provide simplifications without losing any information or splitting the input sentence. No other demographic or compensation information is provided in the TURKCorpus paper. The instructions given to the annotators are available in the paper.", "#### Any Quality Control?\n\n\nnone", "### Consent", "#### Any Consent Policy?\n\n\nyes", "#### Consent Policy Details\n\n\nBoth Figure Eight and Amazon Mechanical Turk raters forfeit the right to their data as part of their agreements.", "### Private Identifying Information (PII)", "#### Contains PII?\n\n\nno PII", "#### Justification for no PII\n\n\nSince the dataset is created from Wikipedia/Simple Wikipedia, all the information contained in the dataset is already in the public domain.", "### Maintenance", "#### Any Maintenance Plan?\n\n\nno\n\n\nBroader Social Context\n----------------------", "### Previous Work on the Social Impact of the Dataset", "#### Usage of Models based on the Data\n\n\nno", "### Impact on Under-Served Communities", "#### Addresses needs of underserved Communities?\n\n\nno", "### Discussion of Biases", "#### Any Documented Social Biases?\n\n\nyes", "#### Links and Summaries of Analysis Work\n\n\nThe dataset may contain some social biases, as the input sentences are based on Wikipedia. Studies have shown that the English Wikipedia contains both gender biases (Schmahl et al., 2020) and racial biases (Adams et al., 2019).\n\n\nConsiderations for Using the Data\n---------------------------------", "### PII Risks and Liability", "#### Potential PII Risk\n\n\nAll the data is in the public domain.", "### Licenses", "#### Copyright Restrictions on the Dataset\n\n\n'open license - commercial use allowed'", "#### Copyright Restrictions on the Language Data\n\n\n'open license - commercial use allowed'", "### Known Technical Limitations", "#### Technical Limitations\n\n\nThe dataset may contain some social biases, as the input sentences are based on Wikipedia. Studies have shown that the English Wikipedia contains both gender biases (Schmahl et al., 2020) and racial biases (Adams et al., 2019).", "#### Unsuited Applications\n\n\nSince the test datasets contains only 2,359 sentences that are derived from Wikipedia, they are limited to a small subset of topics present on Wikipedia." ]

[ "task_categories:summarization", "annotations_creators:automatically-created", "language_creators:unknown", "multilinguality:unknown", "size_categories:unknown", "source_datasets:original", "language:en", "license:cc-by-sa-3.0", "arxiv:1906.04687", "arxiv:1801.10198", "arxiv:2009.07032", "region:us" ]

[ "1906.04687", "1801.10198", "2009.07032" ]

[ "en" ]

[ "# Dataset Card for GEM/wiki_cat_sum", "## Dataset Description\n\n- Homepage: URL\n- Repository: URL\n- Paper: URL\n- Leaderboard: N/A\n- Point of Contact: Laura Perez-Beltrachini", "### Link to Main Data Card\n\nYou can find the main data card on the GEM Website.", "### Dataset Summary \n\nWikiCatSum is an English summarization dataset in three domains: animals, companies, and film. It provides multiple paragraphs of text paired with a summary of the paragraphs.\n\nYou can load the dataset via:\n\nThe data loader can be found here.", "#### website\nGithub", "#### paper\nArxiv", "#### authors\nLaura Perez-Beltrachini, Yang Liu, Mirella Lapata (University of Edinburgh) Peter J. Liu, Mohammad Saleh, Etienne Pot, Ben Goodrich, Ryan Sepassi, Lukasz Kaiser, Noam Shazeer (GoogleBrain)", "## Dataset Overview", "### Where to find the Data and its Documentation", "#### Webpage\n\n\n\nGithub", "#### Download\n\n\n\nWebsite", "#### Paper\n\n\n\nArxiv", "#### BibTex", "#### Contact Name\n\n\n\n\nLaura Perez-Beltrachini", "#### Contact Email\n\n\n\nlperez@URL", "#### Has a Leaderboard?\n\n\n\nno", "### Languages and Intended Use", "#### Multilingual?\n\n\n\n\nno", "#### Covered Languages\n\n\n\n\n'English'", "#### License\n\n\n\n\ncc-by-sa-3.0: Creative Commons Attribution Share Alike 3.0 Unported", "#### Intended Use\n\n\n\nResearch on multi-document abstractive summarisation.", "#### Primary Task\n\n\n\nSummarization", "#### Communicative Goal\n\n\n\n\nSummarise the most important facts of a given entity in the Film, Company, and Animal domains from a cluster of related documents.", "### Credit", "#### Curation Organization Type(s)\n\n\n\n'industry', 'academic'", "#### Curation Organization(s)\n\n\n\nGoogle Cloud Platform, University of Edinburgh", "#### Dataset Creators\n\n\n\nLaura Perez-Beltrachini, Yang Liu, Mirella Lapata (University of Edinburgh) Peter J. Liu, Mohammad Saleh, Etienne Pot, Ben Goodrich, Ryan Sepassi, Lukasz Kaiser, Noam Shazeer (GoogleBrain)", "#### Funding\n\n\n\nGoogle Cloud Platform, European Research Council", "#### Who added the Dataset to GEM?\n\n\n\nRonald Cardenas (University of Edinburgh) Laura Perez-Beltrachini (University of Edinburgh)", "### Dataset Structure", "#### Data Fields\n\n\n\n- 'id': ID of the data example \n- 'title': Is the Wikipedia article's title\n- 'paragraphs': Is the ranked list of paragraphs from the set of crawled texts\n- 'summary': Is constituted by a list of sentences together with their corresponding topic label", "#### Example Instance\n\n\n\nThis is a truncated example from the animal setting:", "#### Data Splits\n\n\n\nNb of instances in train/valid/test are 50,938/2,855/2,831", "#### Splitting Criteria\n\n\n\nThe data was split i.i.d., i.e. uniformly split into training, validation, and test datasets.", "## Dataset in GEM", "### Rationale for Inclusion in GEM", "#### Why is the Dataset in GEM?\n\n\n\nEvaluation of models' performance on noisy (document, summary) pairs and long inputs.\nEvaluate models' capabilities to generalise and mitigate biases.", "#### Similar Datasets\n\n\n\nno", "#### Unique Language Coverage\n\n\n\nno", "#### Ability that the Dataset measures\n\n\n\nCapabilities to generalise, mitigate biases, factual correctness.", "### GEM-Specific Curation", "#### Modificatied for GEM?\n\n\n\nyes", "#### GEM Modifications\n\n\n\n'annotations added'", "#### Modification Details\n\n\n\nWe provide topic labels for summary sentences.", "#### Additional Splits?\n\n\n\nno", "### Getting Started with the Task", "#### Pointers to Resources\n\n\n\n- Generating Wikipedia by Summarizing Long Sequences\n- Generating Summaries with Topic Templates and Structured Convolutional Decoders\n- Noisy Self-Knowledge Distillation for Text Summarization\n\nAnd all references in these papers.", "## Previous Results", "### Previous Results", "#### Measured Model Abilities\n\n\n\nCapabilities to generalise, mitigate biases, factual correctness.", "#### Metrics\n\n\n\n'ROUGE', 'BERT-Score', 'MoverScore', 'Other: Other Metrics'", "#### Other Metrics\n\n\n\n- Abstract/Copy\n- Factual accuracy based on the score of (Goodrich et al., 2019) and the relation extraction system of (Sorokin and Gurevych, 2017).", "#### Proposed Evaluation\n\n\n\nHuman based are Question Answering and Ranking (Content, Fluency and Repetition)", "#### Previous results available?\n\n\n\nyes", "#### Other Evaluation Approaches\n\n\n\nThose listed above.", "#### Relevant Previous Results\n\n\n\nGenerating Summaries with Topic Templates and Structured Convolutional Decoders\nURL\n\nNoisy Self-Knowledge Distillation for Text Summarization\nURL", "## Dataset Curation", "### Original Curation", "#### Original Curation Rationale\n\n\n\nThe dataset is a subset of the WikiSum (Liu et al., 2018) dataset focusing on summaries of entities in three domains (Film, Company, and Animal). It is multi-document summarisation where input-output pairs for each example entity are created as follows. The input is a set of paragraphs collected from i) documents in the Reference section of the entity's Wikipedia page plus ii) documents collected from the top ten search results after querying Google search engine with the entity name. The output summary is the Wikipedia abstract for the entity.", "#### Communicative Goal\n\n\n\nGenerate descriptive summaries with specific domains, where certain topics are discussed and generally in specific orders.", "#### Sourced from Different Sources\n\n\n\nyes", "#### Source Details\n\n\n\nWikiSum (Liu et al., 2018)", "### Language Data", "#### How was Language Data Obtained?\n\n\n\n'Other'", "#### Topics Covered\n\n\n\nThe dataset and task focuses on summaries for entities in three domains: Company, Film, and Animal.", "#### Data Validation\n\n\n\nnot validated", "#### Data Preprocessing\n\n\n\nSummary sentences are associated with a topic label. There is a topic model for each domain.", "#### Was Data Filtered?\n\n\n\nnot filtered", "### Structured Annotations", "#### Additional Annotations?\n\n\n\n\nautomatically created", "#### Annotation Service?\n\n\n\nno", "#### Annotation Values\n\n\n\nEach summary sentences was annotated with a topic label. There is a topic model for each of the three domains. This was used to guide a hierarchical decoder.", "#### Any Quality Control?\n\n\n\nvalidated by data curators", "#### Quality Control Details\n\n\n\nManual inspection of a sample of topics assigned to sentences. The number of topics was selected based on the performance of the summarisation model.", "### Consent", "#### Any Consent Policy?\n\n\n\nno", "#### Justification for Using the Data\n\n\n\nThe dataset is base on Wikipedia and referenced and retrieved documents crawled from the Web.", "### Private Identifying Information (PII)", "#### Contains PII?\n\n\n\n\nunlikely", "#### Any PII Identification?\n\n\n\nno identification", "### Maintenance", "#### Any Maintenance Plan?\n\n\n\nno", "## Broader Social Context", "### Previous Work on the Social Impact of the Dataset", "#### Usage of Models based on the Data\n\n\n\nno", "### Impact on Under-Served Communities", "#### Addresses needs of underserved Communities?\n\n\n\nno", "### Discussion of Biases", "#### Any Documented Social Biases?\n\n\n\nyes", "#### Links and Summaries of Analysis Work\n\n\n\nThis dataset is based on Wikipedia and thus biases analysis on other Wikipedia-based datasets are potentially true for WikiCatSum. For instance, see analysis for the ToTTo dataset here [1].\n\n[1] Automatic Construction of Evaluation Suites for Natural Language Generation Datasets\nURL", "## Considerations for Using the Data", "### PII Risks and Liability", "### Licenses", "#### Copyright Restrictions on the Dataset\n\n\n\n'public domain'", "#### Copyright Restrictions on the Language Data\n\n\n\n'public domain'", "### Known Technical Limitations" ]

[ "TAGS\n#task_categories-summarization #annotations_creators-automatically-created #language_creators-unknown #multilinguality-unknown #size_categories-unknown #source_datasets-original #language-English #license-cc-by-sa-3.0 #arxiv-1906.04687 #arxiv-1801.10198 #arxiv-2009.07032 #region-us \n", "# Dataset Card for GEM/wiki_cat_sum", "## Dataset Description\n\n- Homepage: URL\n- Repository: URL\n- Paper: URL\n- Leaderboard: N/A\n- Point of Contact: Laura Perez-Beltrachini", "### Link to Main Data Card\n\nYou can find the main data card on the GEM Website.", "### Dataset Summary \n\nWikiCatSum is an English summarization dataset in three domains: animals, companies, and film. It provides multiple paragraphs of text paired with a summary of the paragraphs.\n\nYou can load the dataset via:\n\nThe data loader can be found here.", "#### website\nGithub", "#### paper\nArxiv", "#### authors\nLaura Perez-Beltrachini, Yang Liu, Mirella Lapata (University of Edinburgh) Peter J. Liu, Mohammad Saleh, Etienne Pot, Ben Goodrich, Ryan Sepassi, Lukasz Kaiser, Noam Shazeer (GoogleBrain)", "## Dataset Overview", "### Where to find the Data and its Documentation", "#### Webpage\n\n\n\nGithub", "#### Download\n\n\n\nWebsite", "#### Paper\n\n\n\nArxiv", "#### BibTex", "#### Contact Name\n\n\n\n\nLaura Perez-Beltrachini", "#### Contact Email\n\n\n\nlperez@URL", "#### Has a Leaderboard?\n\n\n\nno", "### Languages and Intended Use", "#### Multilingual?\n\n\n\n\nno", "#### Covered Languages\n\n\n\n\n'English'", "#### License\n\n\n\n\ncc-by-sa-3.0: Creative Commons Attribution Share Alike 3.0 Unported", "#### Intended Use\n\n\n\nResearch on multi-document abstractive summarisation.", "#### Primary Task\n\n\n\nSummarization", "#### Communicative Goal\n\n\n\n\nSummarise the most important facts of a given entity in the Film, Company, and Animal domains from a cluster of related documents.", "### Credit", "#### Curation Organization Type(s)\n\n\n\n'industry', 'academic'", "#### Curation Organization(s)\n\n\n\nGoogle Cloud Platform, University of Edinburgh", "#### Dataset Creators\n\n\n\nLaura Perez-Beltrachini, Yang Liu, Mirella Lapata (University of Edinburgh) Peter J. Liu, Mohammad Saleh, Etienne Pot, Ben Goodrich, Ryan Sepassi, Lukasz Kaiser, Noam Shazeer (GoogleBrain)", "#### Funding\n\n\n\nGoogle Cloud Platform, European Research Council", "#### Who added the Dataset to GEM?\n\n\n\nRonald Cardenas (University of Edinburgh) Laura Perez-Beltrachini (University of Edinburgh)", "### Dataset Structure", "#### Data Fields\n\n\n\n- 'id': ID of the data example \n- 'title': Is the Wikipedia article's title\n- 'paragraphs': Is the ranked list of paragraphs from the set of crawled texts\n- 'summary': Is constituted by a list of sentences together with their corresponding topic label", "#### Example Instance\n\n\n\nThis is a truncated example from the animal setting:", "#### Data Splits\n\n\n\nNb of instances in train/valid/test are 50,938/2,855/2,831", "#### Splitting Criteria\n\n\n\nThe data was split i.i.d., i.e. uniformly split into training, validation, and test datasets.", "## Dataset in GEM", "### Rationale for Inclusion in GEM", "#### Why is the Dataset in GEM?\n\n\n\nEvaluation of models' performance on noisy (document, summary) pairs and long inputs.\nEvaluate models' capabilities to generalise and mitigate biases.", "#### Similar Datasets\n\n\n\nno", "#### Unique Language Coverage\n\n\n\nno", "#### Ability that the Dataset measures\n\n\n\nCapabilities to generalise, mitigate biases, factual correctness.", "### GEM-Specific Curation", "#### Modificatied for GEM?\n\n\n\nyes", "#### GEM Modifications\n\n\n\n'annotations added'", "#### Modification Details\n\n\n\nWe provide topic labels for summary sentences.", "#### Additional Splits?\n\n\n\nno", "### Getting Started with the Task", "#### Pointers to Resources\n\n\n\n- Generating Wikipedia by Summarizing Long Sequences\n- Generating Summaries with Topic Templates and Structured Convolutional Decoders\n- Noisy Self-Knowledge Distillation for Text Summarization\n\nAnd all references in these papers.", "## Previous Results", "### Previous Results", "#### Measured Model Abilities\n\n\n\nCapabilities to generalise, mitigate biases, factual correctness.", "#### Metrics\n\n\n\n'ROUGE', 'BERT-Score', 'MoverScore', 'Other: Other Metrics'", "#### Other Metrics\n\n\n\n- Abstract/Copy\n- Factual accuracy based on the score of (Goodrich et al., 2019) and the relation extraction system of (Sorokin and Gurevych, 2017).", "#### Proposed Evaluation\n\n\n\nHuman based are Question Answering and Ranking (Content, Fluency and Repetition)", "#### Previous results available?\n\n\n\nyes", "#### Other Evaluation Approaches\n\n\n\nThose listed above.", "#### Relevant Previous Results\n\n\n\nGenerating Summaries with Topic Templates and Structured Convolutional Decoders\nURL\n\nNoisy Self-Knowledge Distillation for Text Summarization\nURL", "## Dataset Curation", "### Original Curation", "#### Original Curation Rationale\n\n\n\nThe dataset is a subset of the WikiSum (Liu et al., 2018) dataset focusing on summaries of entities in three domains (Film, Company, and Animal). It is multi-document summarisation where input-output pairs for each example entity are created as follows. The input is a set of paragraphs collected from i) documents in the Reference section of the entity's Wikipedia page plus ii) documents collected from the top ten search results after querying Google search engine with the entity name. The output summary is the Wikipedia abstract for the entity.", "#### Communicative Goal\n\n\n\nGenerate descriptive summaries with specific domains, where certain topics are discussed and generally in specific orders.", "#### Sourced from Different Sources\n\n\n\nyes", "#### Source Details\n\n\n\nWikiSum (Liu et al., 2018)", "### Language Data", "#### How was Language Data Obtained?\n\n\n\n'Other'", "#### Topics Covered\n\n\n\nThe dataset and task focuses on summaries for entities in three domains: Company, Film, and Animal.", "#### Data Validation\n\n\n\nnot validated", "#### Data Preprocessing\n\n\n\nSummary sentences are associated with a topic label. There is a topic model for each domain.", "#### Was Data Filtered?\n\n\n\nnot filtered", "### Structured Annotations", "#### Additional Annotations?\n\n\n\n\nautomatically created", "#### Annotation Service?\n\n\n\nno", "#### Annotation Values\n\n\n\nEach summary sentences was annotated with a topic label. There is a topic model for each of the three domains. This was used to guide a hierarchical decoder.", "#### Any Quality Control?\n\n\n\nvalidated by data curators", "#### Quality Control Details\n\n\n\nManual inspection of a sample of topics assigned to sentences. The number of topics was selected based on the performance of the summarisation model.", "### Consent", "#### Any Consent Policy?\n\n\n\nno", "#### Justification for Using the Data\n\n\n\nThe dataset is base on Wikipedia and referenced and retrieved documents crawled from the Web.", "### Private Identifying Information (PII)", "#### Contains PII?\n\n\n\n\nunlikely", "#### Any PII Identification?\n\n\n\nno identification", "### Maintenance", "#### Any Maintenance Plan?\n\n\n\nno", "## Broader Social Context", "### Previous Work on the Social Impact of the Dataset", "#### Usage of Models based on the Data\n\n\n\nno", "### Impact on Under-Served Communities", "#### Addresses needs of underserved Communities?\n\n\n\nno", "### Discussion of Biases", "#### Any Documented Social Biases?\n\n\n\nyes", "#### Links and Summaries of Analysis Work\n\n\n\nThis dataset is based on Wikipedia and thus biases analysis on other Wikipedia-based datasets are potentially true for WikiCatSum. For instance, see analysis for the ToTTo dataset here [1].\n\n[1] Automatic Construction of Evaluation Suites for Natural Language Generation Datasets\nURL", "## Considerations for Using the Data", "### PII Risks and Liability", "### Licenses", "#### Copyright Restrictions on the Dataset\n\n\n\n'public domain'", "#### Copyright Restrictions on the Language Data\n\n\n\n'public domain'", "### Known Technical Limitations" ]

[ "task_categories:summarization", "annotations_creators:none", "language_creators:unknown", "multilinguality:multilingual", "size_categories:unknown", "source_datasets:original", "language:ar", "language:cs", "language:de", "language:en", "language:es", "language:fr", "language:hi", "language:id", "language:it", "language:ja", "language:ko", "language:nl", "language:pt", "language:ru", "language:th", "language:tr", "language:vi", "language:zh", "license:cc-by-nc-sa-3.0", "region:us" ]

[]

[ "ar", "cs", "de", "en", "es", "fr", "hi", "id", "it", "ja", "ko", "nl", "pt", "ru", "th", "tr", "vi", "zh" ]

[ "# Dataset Card for GEM/wiki_lingua", "## Dataset Description\n\n- Homepage: None (See Repository)\n- Repository: URL\n- Paper: URL\n- Leaderboard: N/A\n- Point of Contact: Faisal Ladhak, Esin Durmus", "### Link to Main Data Card\n\nYou can find the main data card on the GEM Website.", "### Dataset Summary \n\nPlaceholder\n\nYou can load the dataset via:\n\nThe data loader can be found here.", "#### website\nNone (See Repository)", "#### paper\nURL", "#### authors\nFaisal Ladhak (Columbia University), Esin Durmus (Stanford University), Claire Cardie (Cornell University), Kathleen McKeown (Columbia University)", "## Dataset Overview", "### Where to find the Data and its Documentation", "#### Webpage\n\n\n\nNone (See Repository)", "#### Download\n\n\n\nURL", "#### Paper\n\n\n\nURL", "#### BibTex\n\n\n\n@inproceedings{ladhak-etal-2020-wikilingua,\n title = \"{W}iki{L}ingua: A New Benchmark Dataset for Cross-Lingual Abstractive Summarization\",\n author = \"Ladhak, Faisal and\n Durmus, Esin and\n Cardie, Claire and\n McKeown, Kathleen\",\n booktitle = \"Findings of the Association for Computational Linguistics: EMNLP 2020\",\n month = nov,\n year = \"2020\",\n address = \"Online\",\n publisher = \"Association for Computational Linguistics\",\n url = \"URL\n doi = \"10.18653/v1/2020.findings-emnlp.360\",\n pages = \"4034--4048\",\n abstract = \"We introduce WikiLingua, a large-scale, multilingual dataset for the evaluation of cross-lingual abstractive summarization systems. We extract article and summary pairs in 18 languages from WikiHow, a high quality, collaborative resource of how-to guides on a diverse set of topics written by human authors. We create gold-standard article-summary alignments across languages by aligning the images that are used to describe each how-to step in an article. As a set of baselines for further studies, we evaluate the performance of existing cross-lingual abstractive summarization methods on our dataset. We further propose a method for direct cross-lingual summarization (i.e., without requiring translation at inference time) by leveraging synthetic data and Neural Machine Translation as a pre-training step. Our method significantly outperforms the baseline approaches, while being more cost efficient during inference.\",\n}", "#### Contact Name\n\n\n\n\nFaisal Ladhak, Esin Durmus", "#### Contact Email\n\n\n\nfaisal@URL, esdurmus@URL", "#### Has a Leaderboard?\n\n\n\nno", "### Languages and Intended Use", "#### Multilingual?\n\n\n\n\nyes", "#### Covered Dialects\n\n\n\nDataset does not have multiple dialects per language.", "#### Covered Languages\n\n\n\n\n'English', 'Spanish, Castilian', 'Portuguese', 'French', 'German', 'Russian', 'Italian', 'Indonesian', 'Dutch, Flemish', 'Arabic', 'Chinese', 'Vietnamese', 'Thai', 'Japanese', 'Korean', 'Hindi', 'Czech', 'Turkish'", "#### Whose Language?\n\n\n\nNo information about the user demographic is available.", "#### License\n\n\n\n\ncc-by-nc-sa-3.0: Creative Commons Attribution-NonCommercial-ShareAlike 3.0 Unported (CC BY-NC-SA 3.0)", "#### Intended Use\n\n\n\nThe dataset was intended to serve as a large-scale, high-quality benchmark dataset for cross-lingual summarization.", "#### Primary Task\n\n\n\nSummarization", "#### Communicative Goal\n\n\n\n\nProduce a high quality summary for the given input article.", "### Credit", "#### Curation Organization Type(s)\n\n\n\n'academic'", "#### Curation Organization(s)\n\n\n\nColumbia University", "#### Dataset Creators\n\n\n\nFaisal Ladhak (Columbia University), Esin Durmus (Stanford University), Claire Cardie (Cornell University), Kathleen McKeown (Columbia University)", "#### Who added the Dataset to GEM?\n\n\n\nJenny Chim (Queen Mary University of London), Faisal Ladhak (Columbia University)", "### Dataset Structure", "#### Data Fields\n\n\n\ngem_id -- The id for the data instance.\nsource_language -- The language of the source article.\ntarget_language -- The language of the target summary.\nsource -- The source document.", "#### Example Instance\n\n\n\n{\n \"gem_id\": \"wikilingua_crosslingual-train-12345\",\n \"gem_parent_id\": \"wikilingua_crosslingual-train-12345\",\n \"source_language\": \"fr\",\n \"target_language\": \"de\",\n \"source\": \"Document in fr\",\n \"target\": \"Summary in de\",\n}", "#### Data Splits\n\n\n\nThe data is split into train/dev/test. In addition to the full test set, there's also a sampled version of the test set.", "#### Splitting Criteria\n\n\n\nThe data was split to ensure the same document would appear in the same split across languages so as to ensure there's no leakage into the test set.", "## Dataset in GEM", "### Rationale for Inclusion in GEM", "#### Why is the Dataset in GEM?\n\n\n\nThis dataset provides a large-scale, high-quality resource for cross-lingual summarization in 18 languages, increasing the coverage of languages for the GEM summarization task.", "#### Similar Datasets\n\n\n\nyes", "#### Unique Language Coverage\n\n\n\nyes", "#### Difference from other GEM datasets\n\n\n\nXSum covers English news articles, and MLSum covers news articles in German and Spanish. \nIn contrast, this dataset has how-to articles in 18 languages, substantially increasing the languages covered. Moreover, it also provides a a different domain than the other two datasets.", "#### Ability that the Dataset measures\n\n\n\nThe ability to generate quality summaries across multiple languages.", "### GEM-Specific Curation", "#### Modificatied for GEM?\n\n\n\nyes", "#### GEM Modifications\n\n\n\n'other'", "#### Modification Details\n\n\n\nPrevious version had separate data loaders for each language. In this version, we've created a single monolingual data loader, which contains monolingual data in each of the 18 languages. In addition, we've also created a single cross-lingual data loader across all the language pairs in the dataset.", "#### Additional Splits?\n\n\n\nno", "### Getting Started with the Task", "## Previous Results", "### Previous Results", "#### Measured Model Abilities\n\n\n\nAbility to summarize content across different languages.", "#### Metrics\n\n\n\n'ROUGE'", "#### Proposed Evaluation\n\n\n\nROUGE is used to measure content selection by comparing word overlap with reference summaries. In addition, the authors of the dataset also used human evaluation to evaluate content selection and fluency of the systems.", "#### Previous results available?\n\n\n\nno", "## Dataset Curation", "### Original Curation", "#### Original Curation Rationale\n\n\n\nThe dataset was created in order to enable new approaches for cross-lingual and multilingual summarization, which are currently understudied as well as open up inetersting new directions for research in summarization. E.g., exploration of multi-source cross-lingual architectures, i.e. models that can summarize from multiple source languages into a target language, building models that can summarize articles from any language to any other language for a given set of languages.", "#### Communicative Goal\n\n\n\nGiven an input article, produce a high quality summary of the article in the target language.", "#### Sourced from Different Sources\n\n\n\nno", "### Language Data", "#### How was Language Data Obtained?\n\n\n\n'Found'", "#### Where was it found?\n\n\n\n'Single website'", "#### Language Producers\n\n\n\nWikiHow, which is an online resource of how-to guides (written and reviewed by human authors) is used as the data source.", "#### Topics Covered\n\n\n\nThe articles cover 19 broad categories including health, arts and entertainment, personal care and style, travel, education and communications, etc. The categories cover a broad set of genres and topics.", "#### Data Validation\n\n\n\nnot validated", "#### Was Data Filtered?\n\n\n\nnot filtered", "### Structured Annotations", "#### Additional Annotations?\n\n\n\n\nnone", "#### Annotation Service?\n\n\n\nno", "### Consent", "#### Any Consent Policy?\n\n\n\nyes", "#### Consent Policy Details\n\n\n\n(1) Text Content. All text posted by Users to the Service is sub-licensed by wikiHow to other Users under a Creative Commons license as provided herein. The Creative Commons license allows such text content be used freely for non-commercial purposes, so long as it is used and attributed to the original author as specified under the terms of the license. Allowing free republication of our articles helps wikiHow achieve its mission by providing instruction on solving the problems of everyday life to more people for free. In order to support this goal, wikiHow hereby grants each User of the Service a license to all text content that Users contribute to the Service under the terms and conditions of a Creative Commons CC BY-NC-SA 3.0 License. Please be sure to read the terms of the license carefully. You continue to own all right, title, and interest in and to your User Content, and you are free to distribute it as you wish, whether for commercial or non-commercial purposes.", "#### Other Consented Downstream Use\n\n\n\nThe data is made freely available under the Creative Commons license, therefore there are no restrictions about downstream uses as long is it's for non-commercial purposes.", "### Private Identifying Information (PII)", "#### Contains PII?\n\n\n\n\nno PII", "#### Justification for no PII\n\n\n\nOnly the article text and summaries were collected. No user information was retained in the dataset.", "### Maintenance", "#### Any Maintenance Plan?\n\n\n\nno", "## Broader Social Context", "### Previous Work on the Social Impact of the Dataset", "#### Usage of Models based on the Data\n\n\n\nyes - other datasets featuring the same task", "### Impact on Under-Served Communities", "#### Addresses needs of underserved Communities?\n\n\n\nno", "### Discussion of Biases", "#### Any Documented Social Biases?\n\n\n\nyes", "## Considerations for Using the Data", "### PII Risks and Liability", "### Licenses", "#### Copyright Restrictions on the Dataset\n\n\n\n'non-commercial use only'", "#### Copyright Restrictions on the Language Data\n\n\n\n'non-commercial use only'", "### Known Technical Limitations" ]

[ "TAGS\n#task_categories-summarization #annotations_creators-none #language_creators-unknown #multilinguality-multilingual #size_categories-unknown #source_datasets-original #language-Arabic #language-Czech #language-German #language-English #language-Spanish #language-French #language-Hindi #language-Indonesian #language-Italian #language-Japanese #language-Korean #language-Dutch #language-Portuguese #language-Russian #language-Thai #language-Turkish #language-Vietnamese #language-Chinese #license-cc-by-nc-sa-3.0 #region-us \n", "# Dataset Card for GEM/wiki_lingua", "## Dataset Description\n\n- Homepage: None (See Repository)\n- Repository: URL\n- Paper: URL\n- Leaderboard: N/A\n- Point of Contact: Faisal Ladhak, Esin Durmus", "### Link to Main Data Card\n\nYou can find the main data card on the GEM Website.", "### Dataset Summary \n\nPlaceholder\n\nYou can load the dataset via:\n\nThe data loader can be found here.", "#### website\nNone (See Repository)", "#### paper\nURL", "#### authors\nFaisal Ladhak (Columbia University), Esin Durmus (Stanford University), Claire Cardie (Cornell University), Kathleen McKeown (Columbia University)", "## Dataset Overview", "### Where to find the Data and its Documentation", "#### Webpage\n\n\n\nNone (See Repository)", "#### Download\n\n\n\nURL", "#### Paper\n\n\n\nURL", "#### BibTex\n\n\n\n@inproceedings{ladhak-etal-2020-wikilingua,\n title = \"{W}iki{L}ingua: A New Benchmark Dataset for Cross-Lingual Abstractive Summarization\",\n author = \"Ladhak, Faisal and\n Durmus, Esin and\n Cardie, Claire and\n McKeown, Kathleen\",\n booktitle = \"Findings of the Association for Computational Linguistics: EMNLP 2020\",\n month = nov,\n year = \"2020\",\n address = \"Online\",\n publisher = \"Association for Computational Linguistics\",\n url = \"URL\n doi = \"10.18653/v1/2020.findings-emnlp.360\",\n pages = \"4034--4048\",\n abstract = \"We introduce WikiLingua, a large-scale, multilingual dataset for the evaluation of cross-lingual abstractive summarization systems. We extract article and summary pairs in 18 languages from WikiHow, a high quality, collaborative resource of how-to guides on a diverse set of topics written by human authors. We create gold-standard article-summary alignments across languages by aligning the images that are used to describe each how-to step in an article. As a set of baselines for further studies, we evaluate the performance of existing cross-lingual abstractive summarization methods on our dataset. We further propose a method for direct cross-lingual summarization (i.e., without requiring translation at inference time) by leveraging synthetic data and Neural Machine Translation as a pre-training step. Our method significantly outperforms the baseline approaches, while being more cost efficient during inference.\",\n}", "#### Contact Name\n\n\n\n\nFaisal Ladhak, Esin Durmus", "#### Contact Email\n\n\n\nfaisal@URL, esdurmus@URL", "#### Has a Leaderboard?\n\n\n\nno", "### Languages and Intended Use", "#### Multilingual?\n\n\n\n\nyes", "#### Covered Dialects\n\n\n\nDataset does not have multiple dialects per language.", "#### Covered Languages\n\n\n\n\n'English', 'Spanish, Castilian', 'Portuguese', 'French', 'German', 'Russian', 'Italian', 'Indonesian', 'Dutch, Flemish', 'Arabic', 'Chinese', 'Vietnamese', 'Thai', 'Japanese', 'Korean', 'Hindi', 'Czech', 'Turkish'", "#### Whose Language?\n\n\n\nNo information about the user demographic is available.", "#### License\n\n\n\n\ncc-by-nc-sa-3.0: Creative Commons Attribution-NonCommercial-ShareAlike 3.0 Unported (CC BY-NC-SA 3.0)", "#### Intended Use\n\n\n\nThe dataset was intended to serve as a large-scale, high-quality benchmark dataset for cross-lingual summarization.", "#### Primary Task\n\n\n\nSummarization", "#### Communicative Goal\n\n\n\n\nProduce a high quality summary for the given input article.", "### Credit", "#### Curation Organization Type(s)\n\n\n\n'academic'", "#### Curation Organization(s)\n\n\n\nColumbia University", "#### Dataset Creators\n\n\n\nFaisal Ladhak (Columbia University), Esin Durmus (Stanford University), Claire Cardie (Cornell University), Kathleen McKeown (Columbia University)", "#### Who added the Dataset to GEM?\n\n\n\nJenny Chim (Queen Mary University of London), Faisal Ladhak (Columbia University)", "### Dataset Structure", "#### Data Fields\n\n\n\ngem_id -- The id for the data instance.\nsource_language -- The language of the source article.\ntarget_language -- The language of the target summary.\nsource -- The source document.", "#### Example Instance\n\n\n\n{\n \"gem_id\": \"wikilingua_crosslingual-train-12345\",\n \"gem_parent_id\": \"wikilingua_crosslingual-train-12345\",\n \"source_language\": \"fr\",\n \"target_language\": \"de\",\n \"source\": \"Document in fr\",\n \"target\": \"Summary in de\",\n}", "#### Data Splits\n\n\n\nThe data is split into train/dev/test. In addition to the full test set, there's also a sampled version of the test set.", "#### Splitting Criteria\n\n\n\nThe data was split to ensure the same document would appear in the same split across languages so as to ensure there's no leakage into the test set.", "## Dataset in GEM", "### Rationale for Inclusion in GEM", "#### Why is the Dataset in GEM?\n\n\n\nThis dataset provides a large-scale, high-quality resource for cross-lingual summarization in 18 languages, increasing the coverage of languages for the GEM summarization task.", "#### Similar Datasets\n\n\n\nyes", "#### Unique Language Coverage\n\n\n\nyes", "#### Difference from other GEM datasets\n\n\n\nXSum covers English news articles, and MLSum covers news articles in German and Spanish. \nIn contrast, this dataset has how-to articles in 18 languages, substantially increasing the languages covered. Moreover, it also provides a a different domain than the other two datasets.", "#### Ability that the Dataset measures\n\n\n\nThe ability to generate quality summaries across multiple languages.", "### GEM-Specific Curation", "#### Modificatied for GEM?\n\n\n\nyes", "#### GEM Modifications\n\n\n\n'other'", "#### Modification Details\n\n\n\nPrevious version had separate data loaders for each language. In this version, we've created a single monolingual data loader, which contains monolingual data in each of the 18 languages. In addition, we've also created a single cross-lingual data loader across all the language pairs in the dataset.", "#### Additional Splits?\n\n\n\nno", "### Getting Started with the Task", "## Previous Results", "### Previous Results", "#### Measured Model Abilities\n\n\n\nAbility to summarize content across different languages.", "#### Metrics\n\n\n\n'ROUGE'", "#### Proposed Evaluation\n\n\n\nROUGE is used to measure content selection by comparing word overlap with reference summaries. In addition, the authors of the dataset also used human evaluation to evaluate content selection and fluency of the systems.", "#### Previous results available?\n\n\n\nno", "## Dataset Curation", "### Original Curation", "#### Original Curation Rationale\n\n\n\nThe dataset was created in order to enable new approaches for cross-lingual and multilingual summarization, which are currently understudied as well as open up inetersting new directions for research in summarization. E.g., exploration of multi-source cross-lingual architectures, i.e. models that can summarize from multiple source languages into a target language, building models that can summarize articles from any language to any other language for a given set of languages.", "#### Communicative Goal\n\n\n\nGiven an input article, produce a high quality summary of the article in the target language.", "#### Sourced from Different Sources\n\n\n\nno", "### Language Data", "#### How was Language Data Obtained?\n\n\n\n'Found'", "#### Where was it found?\n\n\n\n'Single website'", "#### Language Producers\n\n\n\nWikiHow, which is an online resource of how-to guides (written and reviewed by human authors) is used as the data source.", "#### Topics Covered\n\n\n\nThe articles cover 19 broad categories including health, arts and entertainment, personal care and style, travel, education and communications, etc. The categories cover a broad set of genres and topics.", "#### Data Validation\n\n\n\nnot validated", "#### Was Data Filtered?\n\n\n\nnot filtered", "### Structured Annotations", "#### Additional Annotations?\n\n\n\n\nnone", "#### Annotation Service?\n\n\n\nno", "### Consent", "#### Any Consent Policy?\n\n\n\nyes", "#### Consent Policy Details\n\n\n\n(1) Text Content. All text posted by Users to the Service is sub-licensed by wikiHow to other Users under a Creative Commons license as provided herein. The Creative Commons license allows such text content be used freely for non-commercial purposes, so long as it is used and attributed to the original author as specified under the terms of the license. Allowing free republication of our articles helps wikiHow achieve its mission by providing instruction on solving the problems of everyday life to more people for free. In order to support this goal, wikiHow hereby grants each User of the Service a license to all text content that Users contribute to the Service under the terms and conditions of a Creative Commons CC BY-NC-SA 3.0 License. Please be sure to read the terms of the license carefully. You continue to own all right, title, and interest in and to your User Content, and you are free to distribute it as you wish, whether for commercial or non-commercial purposes.", "#### Other Consented Downstream Use\n\n\n\nThe data is made freely available under the Creative Commons license, therefore there are no restrictions about downstream uses as long is it's for non-commercial purposes.", "### Private Identifying Information (PII)", "#### Contains PII?\n\n\n\n\nno PII", "#### Justification for no PII\n\n\n\nOnly the article text and summaries were collected. No user information was retained in the dataset.", "### Maintenance", "#### Any Maintenance Plan?\n\n\n\nno", "## Broader Social Context", "### Previous Work on the Social Impact of the Dataset", "#### Usage of Models based on the Data\n\n\n\nyes - other datasets featuring the same task", "### Impact on Under-Served Communities", "#### Addresses needs of underserved Communities?\n\n\n\nno", "### Discussion of Biases", "#### Any Documented Social Biases?\n\n\n\nyes", "## Considerations for Using the Data", "### PII Risks and Liability", "### Licenses", "#### Copyright Restrictions on the Dataset\n\n\n\n'non-commercial use only'", "#### Copyright Restrictions on the Language Data\n\n\n\n'non-commercial use only'", "### Known Technical Limitations" ]

[ "task_categories:summarization", "annotations_creators:none", "language_creators:unknown", "multilinguality:unknown", "size_categories:unknown", "source_datasets:original", "language:und", "license:cc-by-nc-sa-4.0", "arxiv:1607.01759", "region:us" ]

[ "1607.01759" ]

[ "und" ]

[ "### Link to Main Data Card\n\n\nYou can find the main data card on the GEM Website.", "### Dataset Summary\n\n\nXLSum is a highly multilingual summarization dataset supporting 44 language. The data stems from BBC news articles.\n\n\nYou can load the dataset via:\n\n\nThe data loader can be found here.", "#### website\n\n\nGithub", "#### paper\n\n\nACL Anthology\n\n\nDataset Overview\n----------------", "### Where to find the Data and its Documentation", "#### Webpage\n\n\nGithub", "#### Download\n\n\nHuggingface", "#### Paper\n\n\nACL Anthology", "#### BibTex", "#### Contact Name\n\n\nTahmid Hasan", "#### Contact Email\n\n\ntahmidhasan@URL", "#### Has a Leaderboard?\n\n\nyes", "#### Leaderboard Link\n\n\nExplainaboard", "#### Leaderboard Details\n\n\nThe leaderboard ranks models based on ROUGE scores (R1/R2/RL) of the generated summaries.", "### Languages and Intended Use", "#### Multilingual?\n\n\nyes", "#### Covered Languages\n\n\n'Amharic', 'Arabic', 'Azerbaijani', 'Bengali, Bangla', 'Burmese', 'Chinese (family)', 'English', 'French', 'Gujarati', 'Hausa', 'Hindi', 'Igbo', 'Indonesian', 'Japanese', 'Rundi', 'Korean', 'Kirghiz, Kyrgyz', 'Marathi', 'Nepali (individual language)', 'Oromo', 'Pushto, Pashto', 'Persian', 'Ghanaian Pidgin English', 'Portuguese', 'Panjabi, Punjabi', 'Russian', 'Scottish Gaelic, Gaelic', 'Serbian', 'Romano-Serbian', 'Sinhala, Sinhalese', 'Somali', 'Spanish, Castilian', 'Swahili (individual language), Kiswahili', 'Tamil', 'Telugu', 'Thai', 'Tigrinya', 'Turkish', 'Ukrainian', 'Urdu', 'Uzbek', 'Vietnamese', 'Welsh', 'Yoruba'", "#### License\n\n\ncc-by-nc-sa-4.0: Creative Commons Attribution Non Commercial Share Alike 4.0 International", "#### Intended Use\n\n\nAbstractive summarization has centered around the English language, as most large abstractive summarization datasets are available in English only. Though there have been some recent efforts for curating multilingual abstractive summarization datasets, they are limited in terms of the number of languages covered, the number of training samples, or both. To this end, XL-Sum presents a large-scale abstractive summarization dataset of 1.35 million news articles from 45 languages crawled from the British Broadcasting Corporation website. It is intended to be used for both multilingual and per-language summarization tasks.", "#### Primary Task\n\n\nSummarization", "#### Communicative Goal\n\n\nSummarize news-like text in one of 45 languages.", "### Credit", "#### Curation Organization Type(s)\n\n\n'academic'", "#### Curation Organization(s)\n\n\nBangladesh University of Engineering and Technology", "#### Who added the Dataset to GEM?\n\n\nTahmid Hasan (Bangladesh University of Engineering and Technology), Abhik Bhattacharjee (Bangladesh University of Engineering and Technology)", "### Dataset Structure", "#### Data Fields\n\n\n* 'gem\\_id': A string representing the article ID.\n* 'url': A string representing the article URL.\n* 'title': A string containing the article title.\n* 'summary': A string containing the article summary.\n* 'text' : A string containing the article text.", "#### Example Instance", "#### Data Splits\n\n\nThe splits in the dataset are specified by the language names, which are as follows:\n\n\n* 'amharic'\n* 'arabic'\n* 'azerbaijani'\n* 'bengali'\n* 'burmese'\n* 'chinese\\_simplified'\n* 'chinese\\_traditional'\n* 'english'\n* 'french'\n* 'gujarati'\n* 'hausa'\n* 'hindi'\n* 'igbo'\n* 'indonesian'\n* 'japanese'\n* 'kirundi'\n* 'korean'\n* 'kyrgyz'\n* 'marathi'\n* 'nepali'\n* 'oromo'\n* 'pashto'\n* 'persian'\n* 'pidgin'\n* 'portuguese'\n* 'punjabi'\n* 'russian'\n* 'scottish\\_gaelic'\n* 'serbian\\_cyrillic'\n* 'serbian\\_latin'\n* 'sinhala'\n* 'somali'\n* 'spanish'\n* 'swahili'\n* 'tamil'\n* 'telugu'\n* 'thai'\n* 'tigrinya'\n* 'turkish'\n* 'ukrainian'\n* 'urdu'\n* 'uzbek'\n* 'vietnamese'\n* 'welsh'\n* 'yoruba'", "#### Splitting Criteria\n\n\nWe used a 80%-10%-10% split for all languages with a few exceptions. 'English' was split 93%-3.5%-3.5% for the evaluation set size to resemble that of 'CNN/DM' and 'XSum'; 'Scottish Gaelic', 'Kyrgyz' and 'Sinhala' had relatively fewer samples, their evaluation sets were increased to 500 samples for more reliable evaluation. Same articles were used for evaluation in the two variants of Chinese and Serbian to prevent data leakage in multilingual training. Individual dataset download links with train-dev-test example counts are given below:\n\n\n\n'\\*' A lot of articles in BBC Sinhala and BBC Ukrainian were written in English and Russian respectively. They were identified using Fasttext and moved accordingly.\n'' West African Pidgin English\n\n\nDataset in GEM\n--------------", "### Rationale for Inclusion in GEM", "#### Why is the Dataset in GEM?\n\n\nTraditional abstractive text summarization has been centered around English and other high-resource languages. XL-Sum provides a large collection of high-quality article-summary pairs for 45 languages where the languages range from high-resource to extremely low-resource. This enables the research community to explore the summarization capabilities of different models for multiple languages and languages in isolation. We believe the addition of XL-Sum to GEM makes the domain of abstractive text summarization more diversified and inclusive to the research community. We hope our efforts in this work will encourage the community to push the boundaries of abstractive text summarization beyond the English language, especially for low and mid-resource languages, bringing technological advances to communities of these languages that have been traditionally under-served.", "#### Similar Datasets\n\n\nyes", "#### Unique Language Coverage\n\n\nyes", "#### Difference from other GEM datasets\n\n\nThe summaries are highly concise and abstractive.", "#### Ability that the Dataset measures\n\n\nConciseness, abstractiveness, and overall summarization capability.", "### GEM-Specific Curation", "#### Modificatied for GEM?\n\n\nno", "#### Additional Splits?\n\n\nno", "### Getting Started with the Task\n\n\nPrevious Results\n----------------", "### Previous Results", "#### Measured Model Abilities\n\n\nConciseness, abstractiveness, and overall summarization capability.", "#### Metrics\n\n\n'ROUGE'", "#### Proposed Evaluation\n\n\nROUGE is the de facto evaluation metric used for text summarization. However, it was designed specifically for evaluating English texts. Due to the nature of the metric, scores are heavily dependent on text tokenization / stemming / unnecessary character removal, etc. Some modifications to the original ROUGE evaluation were done such as punctuation only removal, language specific tokenization/stemming to enable reliable comparison of source and target summaries across different scripts.", "#### Previous results available?\n\n\nno\n\n\nDataset Curation\n----------------", "### Original Curation", "#### Original Curation Rationale\n\n\nState-of-the-art text summarization models are heavily data-driven, i.e., a large number of article-summary pairs are required to train them effectively. As a result, abstractive summarization has centered around the English language, as most large abstractive summarization datasets are available in English only. Though there have been some recent efforts for curating multilingual abstractive summarization datasets, they are limited in terms of the number of languages covered, the number of training samples, or both. To this end, we curate XL-Sum, a large-scale abstractive summarization dataset of 1.35 million news articles from 45 languages crawled from the British Broadcasting Corporation website.", "#### Communicative Goal\n\n\nIntroduce new languages in the english-centric domain of abstractive text summarization and enable both multilingual and per-language summarization.", "#### Sourced from Different Sources\n\n\nyes", "#### Source Details\n\n\nBritish Broadcasting Corporation (BBC) news websites.", "### Language Data", "#### How was Language Data Obtained?\n\n\n'Found'", "#### Where was it found?\n\n\n'Multiple websites'", "#### Language Producers\n\n\nThe language content was written by professional news editors hired by BBC.", "#### Topics Covered\n\n\nNews", "#### Data Validation\n\n\nnot validated", "#### Data Preprocessing\n\n\nWe used 'NFKC' normalization on all text instances.", "#### Was Data Filtered?\n\n\nalgorithmically", "#### Filter Criteria\n\n\nWe designed a crawler to recursively crawl pages starting from the homepage by visiting different article links present in each page visited. We were able to take advantage of the fact that all BBC sites have somewhat similar structures, and were able to scrape articles from all sites. We discarded pages with no textual contents (mostly pages consisting of multimedia contents) before further processing. We designed a number of heuristics to make the extraction effective by carefully examining the HTML structures of the crawled pages:\n\n\n1. The desired summary must be present within the beginning two paragraphs of an article.\n2. The summary paragraph must have some portion of texts in bold format.\n3. The summary paragraph may contain some hyperlinks that may not be bold. The proportion of bold texts and hyperlinked texts to the total length of the paragraph in consideration must be at least 95%.\n4. All texts except the summary and the headline must be included in the input text (including image captions).\n5. The input text must be at least twice as large as the summary.", "### Structured Annotations", "#### Additional Annotations?\n\n\nnone", "#### Annotation Service?\n\n\nno", "### Consent", "#### Any Consent Policy?\n\n\nyes", "#### Consent Policy Details\n\n\nBBC's policy specifies that the text content within its websites can be used for non-commercial research only.", "### Private Identifying Information (PII)", "#### Contains PII?\n\n\nlikely", "#### Categories of PII\n\n\n'generic PII'", "#### Any PII Identification?\n\n\nno identification", "### Maintenance", "#### Any Maintenance Plan?\n\n\nno\n\n\nBroader Social Context\n----------------------", "### Previous Work on the Social Impact of the Dataset", "#### Usage of Models based on the Data\n\n\nno", "### Impact on Under-Served Communities", "#### Addresses needs of underserved Communities?\n\n\nyes", "#### Details on how Dataset Addresses the Needs\n\n\nThis dataset introduces summarization corpus for many languages where there weren't any datasets like this curated before.", "### Discussion of Biases", "#### Any Documented Social Biases?\n\n\nno", "#### Are the Language Producers Representative of the Language?\n\n\nYes\n\n\nConsiderations for Using the Data\n---------------------------------", "### PII Risks and Liability", "### Licenses", "#### Copyright Restrictions on the Dataset\n\n\n'research use only', 'non-commercial use only'", "#### Copyright Restrictions on the Language Data\n\n\n'research use only', 'non-commercial use only'", "### Known Technical Limitations", "#### Technical Limitations\n\n\nHuman evaluation showed most languages had a high percentage of good summaries in the upper nineties, almost none of the summaries contained any conflicting information, while about one-third on average had information that was not directly inferrable from the source article. Since generally multiple articles are written regarding an important event, there could be an overlap between the training and evaluation data in terms on content.", "#### Unsuited Applications\n\n\nThe dataset is limited to news domain only. Hence it wouldn't be advisable to use a model trained on this dataset for summarizing texts from a different domain i.e. literature, scientific text etc. Another pitfall could be hallucinations in the model generated summary.", "#### Discouraged Use Cases\n\n\nROUGE evaluates the quality of the summary as a whole by considering up to 4-gram overlaps. Therefore, in an article about India if the word \"India\" in the generated summary gets replaced by \"Pakistan\" due to model hallucination, the overall score wouldn't be reduced significantly, but the entire meaning could get changed." ]

[ "TAGS\n#task_categories-summarization #annotations_creators-none #language_creators-unknown #multilinguality-unknown #size_categories-unknown #source_datasets-original #language-Undetermined #license-cc-by-nc-sa-4.0 #arxiv-1607.01759 #region-us \n", "### Link to Main Data Card\n\n\nYou can find the main data card on the GEM Website.", "### Dataset Summary\n\n\nXLSum is a highly multilingual summarization dataset supporting 44 language. The data stems from BBC news articles.\n\n\nYou can load the dataset via:\n\n\nThe data loader can be found here.", "#### website\n\n\nGithub", "#### paper\n\n\nACL Anthology\n\n\nDataset Overview\n----------------", "### Where to find the Data and its Documentation", "#### Webpage\n\n\nGithub", "#### Download\n\n\nHuggingface", "#### Paper\n\n\nACL Anthology", "#### BibTex", "#### Contact Name\n\n\nTahmid Hasan", "#### Contact Email\n\n\ntahmidhasan@URL", "#### Has a Leaderboard?\n\n\nyes", "#### Leaderboard Link\n\n\nExplainaboard", "#### Leaderboard Details\n\n\nThe leaderboard ranks models based on ROUGE scores (R1/R2/RL) of the generated summaries.", "### Languages and Intended Use", "#### Multilingual?\n\n\nyes", "#### Covered Languages\n\n\n'Amharic', 'Arabic', 'Azerbaijani', 'Bengali, Bangla', 'Burmese', 'Chinese (family)', 'English', 'French', 'Gujarati', 'Hausa', 'Hindi', 'Igbo', 'Indonesian', 'Japanese', 'Rundi', 'Korean', 'Kirghiz, Kyrgyz', 'Marathi', 'Nepali (individual language)', 'Oromo', 'Pushto, Pashto', 'Persian', 'Ghanaian Pidgin English', 'Portuguese', 'Panjabi, Punjabi', 'Russian', 'Scottish Gaelic, Gaelic', 'Serbian', 'Romano-Serbian', 'Sinhala, Sinhalese', 'Somali', 'Spanish, Castilian', 'Swahili (individual language), Kiswahili', 'Tamil', 'Telugu', 'Thai', 'Tigrinya', 'Turkish', 'Ukrainian', 'Urdu', 'Uzbek', 'Vietnamese', 'Welsh', 'Yoruba'", "#### License\n\n\ncc-by-nc-sa-4.0: Creative Commons Attribution Non Commercial Share Alike 4.0 International", "#### Intended Use\n\n\nAbstractive summarization has centered around the English language, as most large abstractive summarization datasets are available in English only. Though there have been some recent efforts for curating multilingual abstractive summarization datasets, they are limited in terms of the number of languages covered, the number of training samples, or both. To this end, XL-Sum presents a large-scale abstractive summarization dataset of 1.35 million news articles from 45 languages crawled from the British Broadcasting Corporation website. It is intended to be used for both multilingual and per-language summarization tasks.", "#### Primary Task\n\n\nSummarization", "#### Communicative Goal\n\n\nSummarize news-like text in one of 45 languages.", "### Credit", "#### Curation Organization Type(s)\n\n\n'academic'", "#### Curation Organization(s)\n\n\nBangladesh University of Engineering and Technology", "#### Who added the Dataset to GEM?\n\n\nTahmid Hasan (Bangladesh University of Engineering and Technology), Abhik Bhattacharjee (Bangladesh University of Engineering and Technology)", "### Dataset Structure", "#### Data Fields\n\n\n* 'gem\\_id': A string representing the article ID.\n* 'url': A string representing the article URL.\n* 'title': A string containing the article title.\n* 'summary': A string containing the article summary.\n* 'text' : A string containing the article text.", "#### Example Instance", "#### Data Splits\n\n\nThe splits in the dataset are specified by the language names, which are as follows:\n\n\n* 'amharic'\n* 'arabic'\n* 'azerbaijani'\n* 'bengali'\n* 'burmese'\n* 'chinese\\_simplified'\n* 'chinese\\_traditional'\n* 'english'\n* 'french'\n* 'gujarati'\n* 'hausa'\n* 'hindi'\n* 'igbo'\n* 'indonesian'\n* 'japanese'\n* 'kirundi'\n* 'korean'\n* 'kyrgyz'\n* 'marathi'\n* 'nepali'\n* 'oromo'\n* 'pashto'\n* 'persian'\n* 'pidgin'\n* 'portuguese'\n* 'punjabi'\n* 'russian'\n* 'scottish\\_gaelic'\n* 'serbian\\_cyrillic'\n* 'serbian\\_latin'\n* 'sinhala'\n* 'somali'\n* 'spanish'\n* 'swahili'\n* 'tamil'\n* 'telugu'\n* 'thai'\n* 'tigrinya'\n* 'turkish'\n* 'ukrainian'\n* 'urdu'\n* 'uzbek'\n* 'vietnamese'\n* 'welsh'\n* 'yoruba'", "#### Splitting Criteria\n\n\nWe used a 80%-10%-10% split for all languages with a few exceptions. 'English' was split 93%-3.5%-3.5% for the evaluation set size to resemble that of 'CNN/DM' and 'XSum'; 'Scottish Gaelic', 'Kyrgyz' and 'Sinhala' had relatively fewer samples, their evaluation sets were increased to 500 samples for more reliable evaluation. Same articles were used for evaluation in the two variants of Chinese and Serbian to prevent data leakage in multilingual training. Individual dataset download links with train-dev-test example counts are given below:\n\n\n\n'\\*' A lot of articles in BBC Sinhala and BBC Ukrainian were written in English and Russian respectively. They were identified using Fasttext and moved accordingly.\n'' West African Pidgin English\n\n\nDataset in GEM\n--------------", "### Rationale for Inclusion in GEM", "#### Why is the Dataset in GEM?\n\n\nTraditional abstractive text summarization has been centered around English and other high-resource languages. XL-Sum provides a large collection of high-quality article-summary pairs for 45 languages where the languages range from high-resource to extremely low-resource. This enables the research community to explore the summarization capabilities of different models for multiple languages and languages in isolation. We believe the addition of XL-Sum to GEM makes the domain of abstractive text summarization more diversified and inclusive to the research community. We hope our efforts in this work will encourage the community to push the boundaries of abstractive text summarization beyond the English language, especially for low and mid-resource languages, bringing technological advances to communities of these languages that have been traditionally under-served.", "#### Similar Datasets\n\n\nyes", "#### Unique Language Coverage\n\n\nyes", "#### Difference from other GEM datasets\n\n\nThe summaries are highly concise and abstractive.", "#### Ability that the Dataset measures\n\n\nConciseness, abstractiveness, and overall summarization capability.", "### GEM-Specific Curation", "#### Modificatied for GEM?\n\n\nno", "#### Additional Splits?\n\n\nno", "### Getting Started with the Task\n\n\nPrevious Results\n----------------", "### Previous Results", "#### Measured Model Abilities\n\n\nConciseness, abstractiveness, and overall summarization capability.", "#### Metrics\n\n\n'ROUGE'", "#### Proposed Evaluation\n\n\nROUGE is the de facto evaluation metric used for text summarization. However, it was designed specifically for evaluating English texts. Due to the nature of the metric, scores are heavily dependent on text tokenization / stemming / unnecessary character removal, etc. Some modifications to the original ROUGE evaluation were done such as punctuation only removal, language specific tokenization/stemming to enable reliable comparison of source and target summaries across different scripts.", "#### Previous results available?\n\n\nno\n\n\nDataset Curation\n----------------", "### Original Curation", "#### Original Curation Rationale\n\n\nState-of-the-art text summarization models are heavily data-driven, i.e., a large number of article-summary pairs are required to train them effectively. As a result, abstractive summarization has centered around the English language, as most large abstractive summarization datasets are available in English only. Though there have been some recent efforts for curating multilingual abstractive summarization datasets, they are limited in terms of the number of languages covered, the number of training samples, or both. To this end, we curate XL-Sum, a large-scale abstractive summarization dataset of 1.35 million news articles from 45 languages crawled from the British Broadcasting Corporation website.", "#### Communicative Goal\n\n\nIntroduce new languages in the english-centric domain of abstractive text summarization and enable both multilingual and per-language summarization.", "#### Sourced from Different Sources\n\n\nyes", "#### Source Details\n\n\nBritish Broadcasting Corporation (BBC) news websites.", "### Language Data", "#### How was Language Data Obtained?\n\n\n'Found'", "#### Where was it found?\n\n\n'Multiple websites'", "#### Language Producers\n\n\nThe language content was written by professional news editors hired by BBC.", "#### Topics Covered\n\n\nNews", "#### Data Validation\n\n\nnot validated", "#### Data Preprocessing\n\n\nWe used 'NFKC' normalization on all text instances.", "#### Was Data Filtered?\n\n\nalgorithmically", "#### Filter Criteria\n\n\nWe designed a crawler to recursively crawl pages starting from the homepage by visiting different article links present in each page visited. We were able to take advantage of the fact that all BBC sites have somewhat similar structures, and were able to scrape articles from all sites. We discarded pages with no textual contents (mostly pages consisting of multimedia contents) before further processing. We designed a number of heuristics to make the extraction effective by carefully examining the HTML structures of the crawled pages:\n\n\n1. The desired summary must be present within the beginning two paragraphs of an article.\n2. The summary paragraph must have some portion of texts in bold format.\n3. The summary paragraph may contain some hyperlinks that may not be bold. The proportion of bold texts and hyperlinked texts to the total length of the paragraph in consideration must be at least 95%.\n4. All texts except the summary and the headline must be included in the input text (including image captions).\n5. The input text must be at least twice as large as the summary.", "### Structured Annotations", "#### Additional Annotations?\n\n\nnone", "#### Annotation Service?\n\n\nno", "### Consent", "#### Any Consent Policy?\n\n\nyes", "#### Consent Policy Details\n\n\nBBC's policy specifies that the text content within its websites can be used for non-commercial research only.", "### Private Identifying Information (PII)", "#### Contains PII?\n\n\nlikely", "#### Categories of PII\n\n\n'generic PII'", "#### Any PII Identification?\n\n\nno identification", "### Maintenance", "#### Any Maintenance Plan?\n\n\nno\n\n\nBroader Social Context\n----------------------", "### Previous Work on the Social Impact of the Dataset", "#### Usage of Models based on the Data\n\n\nno", "### Impact on Under-Served Communities", "#### Addresses needs of underserved Communities?\n\n\nyes", "#### Details on how Dataset Addresses the Needs\n\n\nThis dataset introduces summarization corpus for many languages where there weren't any datasets like this curated before.", "### Discussion of Biases", "#### Any Documented Social Biases?\n\n\nno", "#### Are the Language Producers Representative of the Language?\n\n\nYes\n\n\nConsiderations for Using the Data\n---------------------------------", "### PII Risks and Liability", "### Licenses", "#### Copyright Restrictions on the Dataset\n\n\n'research use only', 'non-commercial use only'", "#### Copyright Restrictions on the Language Data\n\n\n'research use only', 'non-commercial use only'", "### Known Technical Limitations", "#### Technical Limitations\n\n\nHuman evaluation showed most languages had a high percentage of good summaries in the upper nineties, almost none of the summaries contained any conflicting information, while about one-third on average had information that was not directly inferrable from the source article. Since generally multiple articles are written regarding an important event, there could be an overlap between the training and evaluation data in terms on content.", "#### Unsuited Applications\n\n\nThe dataset is limited to news domain only. Hence it wouldn't be advisable to use a model trained on this dataset for summarizing texts from a different domain i.e. literature, scientific text etc. Another pitfall could be hallucinations in the model generated summary.", "#### Discouraged Use Cases\n\n\nROUGE evaluates the quality of the summary as a whole by considering up to 4-gram overlaps. Therefore, in an article about India if the word \"India\" in the generated summary gets replaced by \"Pakistan\" due to model hallucination, the overall score wouldn't be reduced significantly, but the entire meaning could get changed." ]

[ "task_categories:summarization", "annotations_creators:none", "language_creators:unknown", "multilinguality:unknown", "size_categories:unknown", "source_datasets:original", "language:en", "license:cc-by-sa-4.0", "region:us" ]

[]

[ "en" ]

[ "### Link to Main Data Card\n\n\nYou can find the main data card on the GEM Website.", "### Dataset Summary\n\n\nXSum is an English news summarization dataset where the task is to predict the first sentence of an article from the rest of it.\n\n\nYou can load the dataset via:\n\n\nThe data loader can be found here.", "#### website\n\n\nn/a", "#### paper\n\n\nACL Anthology", "#### authors\n\n\nShashi Narayan, Shay B. Cohen, Mirella Lapata (all affiliated with University of Edinburgh at the time of dataset creation)\n\n\nDataset Overview\n----------------", "### Where to find the Data and its Documentation", "#### Download\n\n\nGithub", "#### Paper\n\n\nACL Anthology", "#### BibTex", "#### Contact Name\n\n\nShashi Narayan", "#### Contact Email\n\n\nshashinarayan@URL", "#### Has a Leaderboard?\n\n\nno", "### Languages and Intended Use", "#### Multilingual?\n\n\nno", "#### Covered Dialects\n\n\nSince the source of the dataset are BBC articles, the language is in British English of the variation written by journalists.", "#### Covered Languages\n\n\n'English'", "#### Whose Language?\n\n\nProfessional journalists", "#### License\n\n\ncc-by-sa-4.0: Creative Commons Attribution Share Alike 4.0 International", "#### Intended Use\n\n\nThe dataset is for the task of abstractive summarization in its extreme form, its about summarizing a document in a single sentence. The idea is to create a short, one-sentence news summary answering the question \"What is the article about?\".", "#### Primary Task\n\n\nSummarization", "#### Communicative Goal\n\n\nGiven a news article, produce a single sentence summary of the content of the article.", "### Credit", "#### Curation Organization Type(s)\n\n\n'academic'", "#### Curation Organization(s)\n\n\nUniversity of Edinburgh", "#### Dataset Creators\n\n\nShashi Narayan, Shay B. Cohen, Mirella Lapata (all affiliated with University of Edinburgh at the time of dataset creation)", "#### Funding\n\n\nEuropean Research Council (Lapata; award number 681760), the European Union under the Horizon 2020 SUMMA project (Narayan, Cohen; grant agreement 688139), and Huawei Technologies (Cohen).", "#### Who added the Dataset to GEM?\n\n\nThe original data card was written by Laura Perez-Beltrachini and the data loader by Yacine Jernite. Sebastian Gehrmann migrated the data card to the new format and extended it. The v2 data loader was migrated by Abinaya Mahendiran", "### Dataset Structure", "#### Data Fields\n\n\n* 'Document': Input news article.\n* 'Summary': One sentence summary of the article.\n* 'Id': BBC ID of the article.", "#### Reason for Structure\n\n\nThe Document/Summary format is standard for summarization datasets.", "#### How were labels chosen?\n\n\nThe labels are the first sentence of the source article.", "#### Example Instance", "#### Data Splits", "#### Splitting Criteria\n\n\nThe identifiers in the URLs were used to randomly split the dataset into training (90%, 204,045), validation (5%, 11,332), and test (5%, 11,334) sets.\n\n\nDataset Curation\n----------------", "### Original Curation", "#### Original Curation Rationale\n\n\nComparable datasets are often very extractive which is not a strategy that works for one-sentence summaries. The dataset curators thus created this dataset as a way to evaluate truly abstractive models", "#### Communicative Goal\n\n\nSame as the communicative goal in GEM: A model should summarize a news article in a single sentence", "#### Sourced from Different Sources\n\n\nno", "### Language Data", "#### How was Language Data Obtained?\n\n\n'Found'", "#### Where was it found?\n\n\n'Single website'", "#### Language Producers\n\n\nThe data was collected from articles between 2010 and 2017. No other information", "#### Topics Covered\n\n\nThe collected articles included the following topics: News, Politics, Sports, Weather, Business, Technology, Science, Health, Family, Education, Entertainment and Arts\n\n\nThe dataset curators also used LDA to gain insight into this question and found that the following were the top keywords associated with each topic:\n\n\n* T1: charge, court, murder, police, arrest, guilty, sentence, boy, bail, space, crown, trial\n* T2: church, abuse, bishop, child, catholic, gay, pope, school, christian, priest, cardinal\n* T3: council, people, government, local, housing, home, house, property, city, plan, authority\n* T4: clinton, party, trump, climate, poll, vote, plaid, election, debate, change, candidate, campaign\n* T5: country, growth, report, business, export, fall, bank, security, economy, rise, global, inflation\n* T6: hospital, patient, trust, nhs, people, care, health, service, staff, report, review, system, child", "#### Data Validation\n\n\nnot validated", "#### Data Preprocessing\n\n\nThe text was extracted from the HTML of the webpage. No further processing was done.", "#### Was Data Filtered?\n\n\nnot filtered", "### Structured Annotations", "#### Additional Annotations?\n\n\nnone", "#### Annotation Service?\n\n\nno", "### Consent", "#### Any Consent Policy?\n\n\nno", "#### Justification for Using the Data\n\n\nThe copyright license of the data allows reusing it for this purpose.", "### Private Identifying Information (PII)", "#### Contains PII?\n\n\nyes/very likely", "#### Categories of PII\n\n\n'generic PII'", "#### Any PII Identification?\n\n\nno identification", "### Maintenance", "#### Any Maintenance Plan?\n\n\nno\n\n\nBroader Social Context\n----------------------", "### Previous Work on the Social Impact of the Dataset", "#### Usage of Models based on the Data\n\n\nno", "### Impact on Under-Served Communities", "#### Addresses needs of underserved Communities?\n\n\nno", "### Discussion of Biases", "#### Any Documented Social Biases?\n\n\nunsure", "#### Are the Language Producers Representative of the Language?\n\n\nThe language and content of the data is focused on news and language in the UK and as such not representative of the speakers world-wide. Existing selection biases of the BBC exist in this dataset." ]

[ "TAGS\n#task_categories-summarization #annotations_creators-none #language_creators-unknown #multilinguality-unknown #size_categories-unknown #source_datasets-original #language-English #license-cc-by-sa-4.0 #region-us \n", "### Link to Main Data Card\n\n\nYou can find the main data card on the GEM Website.", "### Dataset Summary\n\n\nXSum is an English news summarization dataset where the task is to predict the first sentence of an article from the rest of it.\n\n\nYou can load the dataset via:\n\n\nThe data loader can be found here.", "#### website\n\n\nn/a", "#### paper\n\n\nACL Anthology", "#### authors\n\n\nShashi Narayan, Shay B. Cohen, Mirella Lapata (all affiliated with University of Edinburgh at the time of dataset creation)\n\n\nDataset Overview\n----------------", "### Where to find the Data and its Documentation", "#### Download\n\n\nGithub", "#### Paper\n\n\nACL Anthology", "#### BibTex", "#### Contact Name\n\n\nShashi Narayan", "#### Contact Email\n\n\nshashinarayan@URL", "#### Has a Leaderboard?\n\n\nno", "### Languages and Intended Use", "#### Multilingual?\n\n\nno", "#### Covered Dialects\n\n\nSince the source of the dataset are BBC articles, the language is in British English of the variation written by journalists.", "#### Covered Languages\n\n\n'English'", "#### Whose Language?\n\n\nProfessional journalists", "#### License\n\n\ncc-by-sa-4.0: Creative Commons Attribution Share Alike 4.0 International", "#### Intended Use\n\n\nThe dataset is for the task of abstractive summarization in its extreme form, its about summarizing a document in a single sentence. The idea is to create a short, one-sentence news summary answering the question \"What is the article about?\".", "#### Primary Task\n\n\nSummarization", "#### Communicative Goal\n\n\nGiven a news article, produce a single sentence summary of the content of the article.", "### Credit", "#### Curation Organization Type(s)\n\n\n'academic'", "#### Curation Organization(s)\n\n\nUniversity of Edinburgh", "#### Dataset Creators\n\n\nShashi Narayan, Shay B. Cohen, Mirella Lapata (all affiliated with University of Edinburgh at the time of dataset creation)", "#### Funding\n\n\nEuropean Research Council (Lapata; award number 681760), the European Union under the Horizon 2020 SUMMA project (Narayan, Cohen; grant agreement 688139), and Huawei Technologies (Cohen).", "#### Who added the Dataset to GEM?\n\n\nThe original data card was written by Laura Perez-Beltrachini and the data loader by Yacine Jernite. Sebastian Gehrmann migrated the data card to the new format and extended it. The v2 data loader was migrated by Abinaya Mahendiran", "### Dataset Structure", "#### Data Fields\n\n\n* 'Document': Input news article.\n* 'Summary': One sentence summary of the article.\n* 'Id': BBC ID of the article.", "#### Reason for Structure\n\n\nThe Document/Summary format is standard for summarization datasets.", "#### How were labels chosen?\n\n\nThe labels are the first sentence of the source article.", "#### Example Instance", "#### Data Splits", "#### Splitting Criteria\n\n\nThe identifiers in the URLs were used to randomly split the dataset into training (90%, 204,045), validation (5%, 11,332), and test (5%, 11,334) sets.\n\n\nDataset Curation\n----------------", "### Original Curation", "#### Original Curation Rationale\n\n\nComparable datasets are often very extractive which is not a strategy that works for one-sentence summaries. The dataset curators thus created this dataset as a way to evaluate truly abstractive models", "#### Communicative Goal\n\n\nSame as the communicative goal in GEM: A model should summarize a news article in a single sentence", "#### Sourced from Different Sources\n\n\nno", "### Language Data", "#### How was Language Data Obtained?\n\n\n'Found'", "#### Where was it found?\n\n\n'Single website'", "#### Language Producers\n\n\nThe data was collected from articles between 2010 and 2017. No other information", "#### Topics Covered\n\n\nThe collected articles included the following topics: News, Politics, Sports, Weather, Business, Technology, Science, Health, Family, Education, Entertainment and Arts\n\n\nThe dataset curators also used LDA to gain insight into this question and found that the following were the top keywords associated with each topic:\n\n\n* T1: charge, court, murder, police, arrest, guilty, sentence, boy, bail, space, crown, trial\n* T2: church, abuse, bishop, child, catholic, gay, pope, school, christian, priest, cardinal\n* T3: council, people, government, local, housing, home, house, property, city, plan, authority\n* T4: clinton, party, trump, climate, poll, vote, plaid, election, debate, change, candidate, campaign\n* T5: country, growth, report, business, export, fall, bank, security, economy, rise, global, inflation\n* T6: hospital, patient, trust, nhs, people, care, health, service, staff, report, review, system, child", "#### Data Validation\n\n\nnot validated", "#### Data Preprocessing\n\n\nThe text was extracted from the HTML of the webpage. No further processing was done.", "#### Was Data Filtered?\n\n\nnot filtered", "### Structured Annotations", "#### Additional Annotations?\n\n\nnone", "#### Annotation Service?\n\n\nno", "### Consent", "#### Any Consent Policy?\n\n\nno", "#### Justification for Using the Data\n\n\nThe copyright license of the data allows reusing it for this purpose.", "### Private Identifying Information (PII)", "#### Contains PII?\n\n\nyes/very likely", "#### Categories of PII\n\n\n'generic PII'", "#### Any PII Identification?\n\n\nno identification", "### Maintenance", "#### Any Maintenance Plan?\n\n\nno\n\n\nBroader Social Context\n----------------------", "### Previous Work on the Social Impact of the Dataset", "#### Usage of Models based on the Data\n\n\nno", "### Impact on Under-Served Communities", "#### Addresses needs of underserved Communities?\n\n\nno", "### Discussion of Biases", "#### Any Documented Social Biases?\n\n\nunsure", "#### Are the Language Producers Representative of the Language?\n\n\nThe language and content of the data is focused on news and language in the UK and as such not representative of the speakers world-wide. Existing selection biases of the BBC exist in this dataset." ]

[ "benchmark:gem", "evaluation", "benchmark", "region:us" ]

[]

[]

[ "# GEM Submission\n\nSubmission name: Hugging Face test URL 36bf2a59" ]

[ "TAGS\n#benchmark-gem #evaluation #benchmark #region-us \n", "# GEM Submission\n\nSubmission name: Hugging Face test URL 36bf2a59" ]

[ "benchmark:gem", "evaluation", "benchmark", "region:us" ]

[]

[]

[ "# GEM Submission\n\nSubmission name: This is a test" ]

[ "TAGS\n#benchmark-gem #evaluation #benchmark #region-us \n", "# GEM Submission\n\nSubmission name: This is a test" ]

[ "benchmark:gem", "evaluation", "benchmark", "region:us" ]

[]

[]

[ "# GEM Submission\n\nSubmission name: This is a test" ]

[ "TAGS\n#benchmark-gem #evaluation #benchmark #region-us \n", "# GEM Submission\n\nSubmission name: This is a test" ]

[ "task_categories:text-retrieval", "task_categories:text-classification", "task_ids:semantic-similarity-classification", "size_categories:10K<n<100K", "language:sv", "license:mit", "question-pairing", "semantic-search", "region:us" ]

[]

[ "sv" ]

[ "# Dataset Card for \"quora_swe\"\n\nThe dataset quora_swe is a subset of the automatically translated (MNT) Swedish Semantic Textual Similarity dataset: quora-deduplicates ." ]

[ "TAGS\n#task_categories-text-retrieval #task_categories-text-classification #task_ids-semantic-similarity-classification #size_categories-10K<n<100K #language-Swedish #license-mit #question-pairing #semantic-search #region-us \n", "# Dataset Card for \"quora_swe\"\n\nThe dataset quora_swe is a subset of the automatically translated (MNT) Swedish Semantic Textual Similarity dataset: quora-deduplicates ." ]

[ "region:us" ]

[]

[]

[]

[ "TAGS\n#region-us \n" ]

[ "region:us" ]

[]

[]

[ "### Dataset Summary\n\n\nThe GonzaloA / Fake\\_News\\_TFG Dataset repository is an English-language dataset containing just over 45k unique news articles. This articles are classified as true (1) or false (0). The current version is the first of the study the Fake News identification using Transformers models.", "### Supported Tasks and Leaderboards", "### Languages\n\n\nThe dataset is code for English as generally spoken in the United States is en-US\n\n\nDataset Structure\n-----------------\n\n\nThe structure of this dataSet is composed by 40587 fields about News. This fields are composed by three types of fields; title of the news, the text or content of the news, and finally, the value of the news, who said if the new are fake (0) or true (1).", "### Data Instances\n\n\nFor each instance, there is a string for the title, a string for the article and a label to mark if it's true or false. See the CNN / Daily Mail dataset viewer to explore more examples.", "### Data Fields\n\n\n* 'id': an integer value to count the rows in the dataset\n* 'title': a string that summarize the article\n* 'text': a string that contains the article\n* 'label': a boolean that mark the article true or false", "### Data Splits\n\n\nThe GonzaloA/FakeNews dataset has 3 splits: train, evaluation, and test. Below are the statistics for the version 1.0 of the dataset:\n\n\n\nDataset Creation\n----------------\n\n\nThis dataset was created with python, using pandas library as the main processing data. Also, this dataset are the mix of other datasets which are the same scope, the Fake News. All of the process is available in this repository: URL", "### Source Data\n\n\nThe source data is a mix of multiple fake\\_news datasets in Kaggle, a platform for train your skills and learnings about Artificial Intelligence. The main datasets who are based this dataset are:", "#### Initial Data Collection and Normalization\n\n\nVersion 1.0.0 aimed to support supervised neural methodologies for deep learning and study the new Transformers models in the Natural Language Processing with News of the United States.", "### Annotations", "#### Annotation process", "#### Who are the annotators?", "### Personal and Sensitive Information\n\n\nConsiderations for Using the Data\n---------------------------------\n\n\nThis Dataset is compose for 3 types: Training phase, for training your model of NLP, validation phase, because we need to validate if the training was successful or our model has overfitting, and train phase, who count the probability and the mistakes in the model fine-tuning.", "### Social Impact of Dataset", "### Discussion of Biases", "### Other Known Limitations\n\n\nAdditional Information\n----------------------", "### Dataset Curators", "### Licensing Information", "### Contributions\n\n\nThanks to @github-username for adding this dataset." ]

[ "TAGS\n#region-us \n", "### Dataset Summary\n\n\nThe GonzaloA / Fake\\_News\\_TFG Dataset repository is an English-language dataset containing just over 45k unique news articles. This articles are classified as true (1) or false (0). The current version is the first of the study the Fake News identification using Transformers models.", "### Supported Tasks and Leaderboards", "### Languages\n\n\nThe dataset is code for English as generally spoken in the United States is en-US\n\n\nDataset Structure\n-----------------\n\n\nThe structure of this dataSet is composed by 40587 fields about News. This fields are composed by three types of fields; title of the news, the text or content of the news, and finally, the value of the news, who said if the new are fake (0) or true (1).", "### Data Instances\n\n\nFor each instance, there is a string for the title, a string for the article and a label to mark if it's true or false. See the CNN / Daily Mail dataset viewer to explore more examples.", "### Data Fields\n\n\n* 'id': an integer value to count the rows in the dataset\n* 'title': a string that summarize the article\n* 'text': a string that contains the article\n* 'label': a boolean that mark the article true or false", "### Data Splits\n\n\nThe GonzaloA/FakeNews dataset has 3 splits: train, evaluation, and test. Below are the statistics for the version 1.0 of the dataset:\n\n\n\nDataset Creation\n----------------\n\n\nThis dataset was created with python, using pandas library as the main processing data. Also, this dataset are the mix of other datasets which are the same scope, the Fake News. All of the process is available in this repository: URL", "### Source Data\n\n\nThe source data is a mix of multiple fake\\_news datasets in Kaggle, a platform for train your skills and learnings about Artificial Intelligence. The main datasets who are based this dataset are:", "#### Initial Data Collection and Normalization\n\n\nVersion 1.0.0 aimed to support supervised neural methodologies for deep learning and study the new Transformers models in the Natural Language Processing with News of the United States.", "### Annotations", "#### Annotation process", "#### Who are the annotators?", "### Personal and Sensitive Information\n\n\nConsiderations for Using the Data\n---------------------------------\n\n\nThis Dataset is compose for 3 types: Training phase, for training your model of NLP, validation phase, because we need to validate if the training was successful or our model has overfitting, and train phase, who count the probability and the mistakes in the model fine-tuning.", "### Social Impact of Dataset", "### Discussion of Biases", "### Other Known Limitations\n\n\nAdditional Information\n----------------------", "### Dataset Curators", "### Licensing Information", "### Contributions\n\n\nThanks to @github-username for adding this dataset." ]

[ "language:en", "license:cc-by-4.0", "region:us" ]

[]

[ "en" ]

[]

[ "TAGS\n#language-English #license-cc-by-4.0 #region-us \n" ]

[ "language:en", "license:cc-by-4.0", "region:us" ]

[]

[ "en" ]

[]

[ "TAGS\n#language-English #license-cc-by-4.0 #region-us \n" ]

[ "language:en", "license:cc-by-4.0", "region:us" ]

[]

[ "en" ]

[]

[ "TAGS\n#language-English #license-cc-by-4.0 #region-us \n" ]

[ "task_categories:translation", "annotations_creators:machine-generated", "annotations_creators:expert-generated", "language_creators:found", "multilinguality:translation", "size_categories:1K<n<10K", "source_datasets:original", "language:en", "language:it", "license:other", "region:us" ]

[]

[ "en", "it" ]

[ "### Dataset Summary\n\n\nThis dataset contains a sample of sentences taken from the FLORES-101 dataset that were either translated from scratch or post-edited from an existing automatic translation by three human translators. Translation were performed for the English-Italian language pair, and translators' behavioral data (keystrokes, pauses, editing times) were collected using the PET platform.\n\n\nThis dataset is made available for final projects of the 2022 edition of the Natural Language Processing course at the Information Science Master's Degree at the University of Groningen, taught by Arianna Bisazza and Gabriele Sarti with the assistance of Anjali Nair.\n\n\nDisclaimer: *This repository is provided without direct data access due to currently unpublished results.* *For this reason, it is strictly forbidden to share or publish all the data associated to this repository*. *Students will be provided with a compressed folder containing the data upon choosing a project based on this dataset. To load the dataset using Datasets, download and unzip the provided folder and pass it to the* 'load\\_dataset' *method as:* 'datasets.load\\_dataset('GroNLP/ik-nlp-22\\_pestyle', 'full', data\\_dir='path/to/unzipped/folder')'", "### Languages\n\n\nThe language data of is in English (BCP-47 'en') and Italian (BCP-47 'it')\n\n\nDataset Structure\n-----------------", "### Data Instances\n\n\nThe dataset contains four configurations: 'full', 'test\\_mask\\_subject', 'test\\_mask\\_modality', 'test\\_mask\\_time'. 'full' contains the main 'train' split in which all fields are available. The other three, 'test\\_mask\\_subject', 'test\\_mask\\_modality', 'test\\_mask\\_time', contain a 'test' split each with different fields removed to avoid information leaking during evaluation. See more details in the Data Splits section.", "### Data Fields\n\n\nThe following fields are contained in the training set:", "### Data Splits", "#### Train Split\n\n\nThe 'train' split contains a total of 1170 triplets (or pairs, when translation from scratch is performed) annotated with behavioral data produced during the translation. The following is an example of the subject 't3' post-editing a machine translation produced by system 2 (tasktype 'pe2') taken from the 'train' split. The field 'aligned\\_edit' is showed over three lines to provide a visual understanding of its contents.\n\n\nThe text is provided as-is, without further preprocessing or tokenization.", "#### Test splits\n\n\nThe three 'test' splits (one per configuration) contain the same 120 entries each, following the same structure as 'train'. Each test split omit some of the fields to prevent leakage of information:\n\n\n* In 'test\\_mask\\_subject' the 'subject\\_id' is absent, for the main task of post-editor stylometry.\n* In 'test\\_mask\\_modality' the following fields are absent for the modality prediction extra task: 'modality', 'mt\\_text', 'n\\_insert', 'n\\_delete', 'n\\_substitute', 'n\\_shift', 'ter', 'bleu', 'chrf', 'aligned\\_edit'.\n* In 'test\\_mask\\_time' the following fields are absent for the time and pause prediction extra task: 'edit\\_time', 'n\\_pause\\_geq\\_300', 'len\\_pause\\_geq\\_300', 'n\\_pause\\_geq\\_1000', and 'len\\_pause\\_geq\\_1000'.", "### Dataset Creation\n\n\nThe dataset was parsed from PET XML files into CSV format using a script adapted from the one by Antonio Toral found at the following link: URL\n\n\nAdditional Information\n----------------------", "### Dataset Curators\n\n\nFor problems related to this Datasets version, please contact us at ik-nlp-course@URL.", "### Licensing Information\n\n\nIt is forbidden to share or publish the data associated with this Dataset version.\n\n\nNo citation information is provided for this dataset." ]

[ "TAGS\n#task_categories-translation #annotations_creators-machine-generated #annotations_creators-expert-generated #language_creators-found #multilinguality-translation #size_categories-1K<n<10K #source_datasets-original #language-English #language-Italian #license-other #region-us \n", "### Dataset Summary\n\n\nThis dataset contains a sample of sentences taken from the FLORES-101 dataset that were either translated from scratch or post-edited from an existing automatic translation by three human translators. Translation were performed for the English-Italian language pair, and translators' behavioral data (keystrokes, pauses, editing times) were collected using the PET platform.\n\n\nThis dataset is made available for final projects of the 2022 edition of the Natural Language Processing course at the Information Science Master's Degree at the University of Groningen, taught by Arianna Bisazza and Gabriele Sarti with the assistance of Anjali Nair.\n\n\nDisclaimer: *This repository is provided without direct data access due to currently unpublished results.* *For this reason, it is strictly forbidden to share or publish all the data associated to this repository*. *Students will be provided with a compressed folder containing the data upon choosing a project based on this dataset. To load the dataset using Datasets, download and unzip the provided folder and pass it to the* 'load\\_dataset' *method as:* 'datasets.load\\_dataset('GroNLP/ik-nlp-22\\_pestyle', 'full', data\\_dir='path/to/unzipped/folder')'", "### Languages\n\n\nThe language data of is in English (BCP-47 'en') and Italian (BCP-47 'it')\n\n\nDataset Structure\n-----------------", "### Data Instances\n\n\nThe dataset contains four configurations: 'full', 'test\\_mask\\_subject', 'test\\_mask\\_modality', 'test\\_mask\\_time'. 'full' contains the main 'train' split in which all fields are available. The other three, 'test\\_mask\\_subject', 'test\\_mask\\_modality', 'test\\_mask\\_time', contain a 'test' split each with different fields removed to avoid information leaking during evaluation. See more details in the Data Splits section.", "### Data Fields\n\n\nThe following fields are contained in the training set:", "### Data Splits", "#### Train Split\n\n\nThe 'train' split contains a total of 1170 triplets (or pairs, when translation from scratch is performed) annotated with behavioral data produced during the translation. The following is an example of the subject 't3' post-editing a machine translation produced by system 2 (tasktype 'pe2') taken from the 'train' split. The field 'aligned\\_edit' is showed over three lines to provide a visual understanding of its contents.\n\n\nThe text is provided as-is, without further preprocessing or tokenization.", "#### Test splits\n\n\nThe three 'test' splits (one per configuration) contain the same 120 entries each, following the same structure as 'train'. Each test split omit some of the fields to prevent leakage of information:\n\n\n* In 'test\\_mask\\_subject' the 'subject\\_id' is absent, for the main task of post-editor stylometry.\n* In 'test\\_mask\\_modality' the following fields are absent for the modality prediction extra task: 'modality', 'mt\\_text', 'n\\_insert', 'n\\_delete', 'n\\_substitute', 'n\\_shift', 'ter', 'bleu', 'chrf', 'aligned\\_edit'.\n* In 'test\\_mask\\_time' the following fields are absent for the time and pause prediction extra task: 'edit\\_time', 'n\\_pause\\_geq\\_300', 'len\\_pause\\_geq\\_300', 'n\\_pause\\_geq\\_1000', and 'len\\_pause\\_geq\\_1000'.", "### Dataset Creation\n\n\nThe dataset was parsed from PET XML files into CSV format using a script adapted from the one by Antonio Toral found at the following link: URL\n\n\nAdditional Information\n----------------------", "### Dataset Curators\n\n\nFor problems related to this Datasets version, please contact us at ik-nlp-course@URL.", "### Licensing Information\n\n\nIt is forbidden to share or publish the data associated with this Dataset version.\n\n\nNo citation information is provided for this dataset." ]

[ "task_categories:question-answering", "task_categories:summarization", "task_categories:text-retrieval", "annotations_creators:expert-generated", "language_creators:expert-generated", "multilinguality:monolingual", "size_categories:1K<n<10K", "source_datasets:original", "language:en", "license:unknown", "question-generation", "region:us" ]

[]

[ "en" ]

[ "### Dataset Summary\n\n\nThis dataset contains chapters extracted from the Speech and Language Processing book (3ed draft of January 2022) by Jurafsky and Martin via a semi-automatic procedure (see below for additional details). Moreover, a small set of conceptual questions associated with each chapter is provided alongside possible answers.\n\n\nOnly the content of chapters 2 to 11 of the book draft are provided, since these are the ones relevant to the contents of the 2022 edition of the Natural Language Processing course at the Information Science Master's Degree (IK) at the University of Groningen, taught by Arianna Bisazza with the assistance of Gabriele Sarti.\n\n\n*The Speech and Language Processing book was made freely available by the authors Dan Jurafsky and James H. Martin on the Stanford University website. The present dataset was created for educational purposes, and is based on the draft of the 3rd edition of the book accessed on December 29th, 2021. All rights of the present contents are attributed to the original authors.*", "### Projects\n\n\nSee the course page for a description of possible research directions.", "### Languages\n\n\nThe language data of Speech and Language Processing is in English (BCP-47 'en')\n\n\nDataset Structure\n-----------------", "### Data Instances\n\n\nThe dataset contains two configurations: 'paragraphs' (default), containing the full set of parsed paragraphs associated to the respective chapter and sections, and 'questions', containing a small subset of example questions matched with the relevant paragraph, and with the answer span extracted.", "#### Paragraphs Configuration\n\n\nThe 'paragraphs' configuration contains all the paragraphs of the selected book chapters, each associated with the respective chapter, section and subsection. An example from the 'train' split of the 'paragraphs' config is provided below. The example belongs to section 2.3 but not to a subsection, so the 'n\\_subsection' and 'subsection' fields are empty strings.\n\n\nThe text is provided as-is, without further preprocessing or tokenization.", "#### Questions Configuration\n\n\nThe 'questions' configuration contains a small subset of questions, the top retrieved paragraph relevant to the question and the answer spans. An example from the 'test' split of the 'questions' config is provided below.", "### Data Splits", "### Dataset Creation\n\n\nThe contents of the Speech and Language Processing book PDF were extracted using the PDF to S2ORC JSON Converter by AllenAI. The texts extracted by the converter were then manually cleaned to remove end-of-chapter exercises and other irrelevant content (e.g. tables, TikZ figures, etc.). Some issues in the parsed content were preserved in the final version to maintain a naturalistic setting for the associated projects, promoting the use of data filtering heuristics for students.\n\n\nThe question-answer pairs were created manually by Gabriele Sarti.\n\n\nAdditional Information\n----------------------", "### Dataset Curators\n\n\nFor problems on this Datasets version, please contact us at ik-nlp-course@URL.", "### Licensing Information\n\n\nPlease refer to the authors' websites for licensing information.\n\n\nPlease cite the authors if you use these corpora in your work:" ]

[ "TAGS\n#task_categories-question-answering #task_categories-summarization #task_categories-text-retrieval #annotations_creators-expert-generated #language_creators-expert-generated #multilinguality-monolingual #size_categories-1K<n<10K #source_datasets-original #language-English #license-unknown #question-generation #region-us \n", "### Dataset Summary\n\n\nThis dataset contains chapters extracted from the Speech and Language Processing book (3ed draft of January 2022) by Jurafsky and Martin via a semi-automatic procedure (see below for additional details). Moreover, a small set of conceptual questions associated with each chapter is provided alongside possible answers.\n\n\nOnly the content of chapters 2 to 11 of the book draft are provided, since these are the ones relevant to the contents of the 2022 edition of the Natural Language Processing course at the Information Science Master's Degree (IK) at the University of Groningen, taught by Arianna Bisazza with the assistance of Gabriele Sarti.\n\n\n*The Speech and Language Processing book was made freely available by the authors Dan Jurafsky and James H. Martin on the Stanford University website. The present dataset was created for educational purposes, and is based on the draft of the 3rd edition of the book accessed on December 29th, 2021. All rights of the present contents are attributed to the original authors.*", "### Projects\n\n\nSee the course page for a description of possible research directions.", "### Languages\n\n\nThe language data of Speech and Language Processing is in English (BCP-47 'en')\n\n\nDataset Structure\n-----------------", "### Data Instances\n\n\nThe dataset contains two configurations: 'paragraphs' (default), containing the full set of parsed paragraphs associated to the respective chapter and sections, and 'questions', containing a small subset of example questions matched with the relevant paragraph, and with the answer span extracted.", "#### Paragraphs Configuration\n\n\nThe 'paragraphs' configuration contains all the paragraphs of the selected book chapters, each associated with the respective chapter, section and subsection. An example from the 'train' split of the 'paragraphs' config is provided below. The example belongs to section 2.3 but not to a subsection, so the 'n\\_subsection' and 'subsection' fields are empty strings.\n\n\nThe text is provided as-is, without further preprocessing or tokenization.", "#### Questions Configuration\n\n\nThe 'questions' configuration contains a small subset of questions, the top retrieved paragraph relevant to the question and the answer spans. An example from the 'test' split of the 'questions' config is provided below.", "### Data Splits", "### Dataset Creation\n\n\nThe contents of the Speech and Language Processing book PDF were extracted using the PDF to S2ORC JSON Converter by AllenAI. The texts extracted by the converter were then manually cleaned to remove end-of-chapter exercises and other irrelevant content (e.g. tables, TikZ figures, etc.). Some issues in the parsed content were preserved in the final version to maintain a naturalistic setting for the associated projects, promoting the use of data filtering heuristics for students.\n\n\nThe question-answer pairs were created manually by Gabriele Sarti.\n\n\nAdditional Information\n----------------------", "### Dataset Curators\n\n\nFor problems on this Datasets version, please contact us at ik-nlp-course@URL.", "### Licensing Information\n\n\nPlease refer to the authors' websites for licensing information.\n\n\nPlease cite the authors if you use these corpora in your work:" ]

[ "task_categories:text-classification", "task_ids:natural-language-inference", "annotations_creators:expert-generated", "language_creators:expert-generated", "language_creators:machine-generated", "multilinguality:translation", "size_categories:unknown", "source_datasets:extended|esnli", "language:en", "language:nl", "license:apache-2.0", "quality-estimation", "region:us" ]

[]

[ "en", "nl" ]

[ "### Dataset Summary\n\n\nThis dataset contains the full e-SNLI dataset, automatically translated to Dutch using the Helsinki-NLP/opus-mt-en-nl neural machine translation model. The translation of each field has been anotated with two quality estimation scores using the referenceless version of the COMET metric by Unbabel.\n\n\nThe intended usage of this corpus is restricted to the scope of final project for the 2022 edition of the Natural Language Processing course at the Information Science Master's Degree (IK) at the University of Groningen, taught by Arianna Bisazza and Gabriele Sarti, with the assistance of Anjali Nair.\n\n\n*The e-SNLI corpus was made freely available by the authors on Github. The present dataset was created for educational purposes, and is based on the original e-SNLI dataset by Camburu et al..All rights of the present contents are attributed to the original authors.*", "### Languages\n\n\nThe language data of this corpus is in English (BCP-47 'en') and Dutch (BCP-47 'nl').\n\n\nDataset Structure\n-----------------", "### Data Instances\n\n\nThe dataset contains a single condiguration by default, named 'plain\\_text', with the three original splits 'train', 'validation' and 'test'. Every split contains the following fields:\n\n\n\nExplanation 2 and 3 and related quality estimation scores are only present in the 'validation' and 'test' splits.", "### Data Splits\n\n\n\nFor your analyses, use the amount of data that is the most reasonable for your computational setup. The more, the better.", "### Data Example\n\n\nThe following is an example of entry 2000 taken from the 'test' split:", "### Dataset Creation\n\n\nThe dataset was created through the following steps:\n\n\n* Translating every field of the original e-SNLI corpus to Dutch using the Helsinki-NLP/opus-mt-en-nl neural machine translation model.\n* Annotating the quality estimation of the translations with two referenceless versions of the COMET metric by Unbabel.\n\n\nAdditional Information\n----------------------", "### Dataset Curators\n\n\nFor problems on this Datasets version, please contact us at ik-nlp-course@URL.", "### Licensing Information\n\n\nThe dataset is licensed under the Apache 2.0 License.\n\n\nPlease cite the authors if you use these corpora in your work:" ]

[ "TAGS\n#task_categories-text-classification #task_ids-natural-language-inference #annotations_creators-expert-generated #language_creators-expert-generated #language_creators-machine-generated #multilinguality-translation #size_categories-unknown #source_datasets-extended|esnli #language-English #language-Dutch #license-apache-2.0 #quality-estimation #region-us \n", "### Dataset Summary\n\n\nThis dataset contains the full e-SNLI dataset, automatically translated to Dutch using the Helsinki-NLP/opus-mt-en-nl neural machine translation model. The translation of each field has been anotated with two quality estimation scores using the referenceless version of the COMET metric by Unbabel.\n\n\nThe intended usage of this corpus is restricted to the scope of final project for the 2022 edition of the Natural Language Processing course at the Information Science Master's Degree (IK) at the University of Groningen, taught by Arianna Bisazza and Gabriele Sarti, with the assistance of Anjali Nair.\n\n\n*The e-SNLI corpus was made freely available by the authors on Github. The present dataset was created for educational purposes, and is based on the original e-SNLI dataset by Camburu et al..All rights of the present contents are attributed to the original authors.*", "### Languages\n\n\nThe language data of this corpus is in English (BCP-47 'en') and Dutch (BCP-47 'nl').\n\n\nDataset Structure\n-----------------", "### Data Instances\n\n\nThe dataset contains a single condiguration by default, named 'plain\\_text', with the three original splits 'train', 'validation' and 'test'. Every split contains the following fields:\n\n\n\nExplanation 2 and 3 and related quality estimation scores are only present in the 'validation' and 'test' splits.", "### Data Splits\n\n\n\nFor your analyses, use the amount of data that is the most reasonable for your computational setup. The more, the better.", "### Data Example\n\n\nThe following is an example of entry 2000 taken from the 'test' split:", "### Dataset Creation\n\n\nThe dataset was created through the following steps:\n\n\n* Translating every field of the original e-SNLI corpus to Dutch using the Helsinki-NLP/opus-mt-en-nl neural machine translation model.\n* Annotating the quality estimation of the translations with two referenceless versions of the COMET metric by Unbabel.\n\n\nAdditional Information\n----------------------", "### Dataset Curators\n\n\nFor problems on this Datasets version, please contact us at ik-nlp-course@URL.", "### Licensing Information\n\n\nThe dataset is licensed under the Apache 2.0 License.\n\n\nPlease cite the authors if you use these corpora in your work:" ]

[ "arxiv:2101.08382", "region:us" ]

[ "2101.08382" ]

[]

[]

[ "TAGS\n#arxiv-2101.08382 #region-us \n" ]

[ "task_categories:fill-mask", "task_categories:summarization", "task_categories:text-classification", "task_categories:token-classification", "task_ids:masked-language-modeling", "task_ids:multi-class-classification", "task_ids:topic-classification", "task_ids:named-entity-recognition", "language:en", "license:cc-by-sa-4.0", "patents", "arxiv:2207.04043", "region:us" ]

[ "2207.04043" ]

[ "en" ]

[ "# Dataset Card for The Harvard USPTO Patent Dataset (HUPD)\n\n!HUPD-Diagram", "## Dataset Description\n\n- Homepage: URL\n- Repository: HUPD GitHub repository\n- Paper: HUPD arXiv Submission\n- Point of Contact: Mirac Suzgun", "### Dataset Summary\n\nThe Harvard USPTO Dataset (HUPD) is a large-scale, well-structured, and multi-purpose corpus of English-language utility patent applications filed to the United States Patent and Trademark Office (USPTO) between January 2004 and December 2018.", "### Experiments and Tasks Considered in the Paper\n\n- Patent Acceptance Prediction: Given a section of a patent application (in particular, the abstract, claims, or description), predict whether the application will be accepted by the USPTO.\n- Automated Subject (IPC/CPC) Classification: Predict the primary IPC or CPC code of a patent application given (some subset of) the text of the application.\n- Language Modeling: Masked/autoregressive language modeling on the claims and description sections of patent applications.\n- Abstractive Summarization: Given the claims or claims section of a patent application, generate the abstract.", "### Languages\n\nThe dataset contains English text only.", "### Domain\nPatents (intellectual property).", "### Dataset Curators\nThe dataset was created by Mirac Suzgun, Luke Melas-Kyriazi, Suproteem K. Sarkar, Scott Duke Kominers, and Stuart M. Shieber.", "## Dataset Structure\n\nEach patent application is defined by a distinct JSON file, named after its application number, and includes information about \nthe application and publication numbers, \ntitle, \ndecision status, \nfiling and publication dates, \nprimary and secondary classification codes, \ninventor(s), \nexaminer,\nattorney,\nabstract, \nclaims, \nbackground, \nsummary, and \nfull description of the proposed invention, among other fields. There are also supplementary variables, such as the small-entity indicator (which denotes whether the applicant is considered to be a small entity by the USPTO) and the foreign-filing indicator (which denotes whether the application was originally filed in a foreign country).\n\nIn total, there are 34 data fields for each application. A full list of data fields used in the dataset is listed in the next section.", "### Data Instances\n\nEach patent application in our patent dataset is defined by a distinct JSON file (e.g., ''URL''), named after its unique application number. The format of the JSON files is as follows:", "## Usage", "### Loading the Dataset", "#### Sample (January 2016 Subset) \nThe following command can be used to load the 'sample' version of the dataset, which contains all the patent applications that were filed to the USPTO during the month of January in 2016. This small subset of the dataset can be used for debugging and exploration purposes.", "#### Full Dataset\n\nIf you would like to use the full version of the dataset, please make sure that change the 'name' field from 'sample' to 'all', specify the training and validation start and end dates carefully, and set 'force_extract' to be 'True' (so that you would only untar the files that you are interested in and not squander your disk storage space). In the following example, for instance, we set the training set year range to be [2011, 2016] (inclusive) and the validation set year range to be 2017.", "### Google Colab Notebook\nYou can also use the following Google Colab notebooks to explore HUPD. \n- ![Open In Colab](URL HUPD Examples: Loading the Dataset\n- ![Open In Colab](URL HUPD Examples: Loading HUPD By Using HuggingFace's Libraries\n- ![Open In Colab](URL HUPD Examples: Using the HUPD DistilRoBERTa Model\n- ![Open In Colab](URL HUPD Examples: Using the HUPD T5-Small Summarization Model", "## Dataset Creation", "### Source Data\n\nHUPD synthesizes multiple data sources from the USPTO: While the full patent application texts were obtained from the USPTO Bulk Data Storage System (Patent Application Data/XML Versions 4.0, 4.1, 4.2, 4.3, 4.4 ICE, as well as Version 1.5) as XML files, the bibliographic filing metadata were obtained from the USPTO Patent Examination Research Dataset (in February, 2021).", "### Annotations\n\nBeyond our patent decision label, for which construction details are provided in the paper, the dataset does not contain any human-written or computer-generated annotations beyond those produced by patent applicants or the USPTO.", "### Data Shift\n\nA major feature of HUPD is its structure, which allows it to demonstrate the evolution of concepts over time. As we illustrate in the paper, the criteria for patent acceptance evolve over time at different rates, depending on category. We believe this is an important feature of the dataset, not only because of the social scientific questions it raises, but also because it facilitates research on models that can accommodate concept shift in a real-world setting.", "### Personal and Sensitive Information\n\nThe dataset contains information about the inventor(s) and examiner of each patent application. These details are, however, already in the public domain and available on the USPTO's Patent Application Information Retrieval (PAIR) system, as well as on Google Patents and PatentsView.", "### Social Impact of the Dataset\n\nThe authors of the dataset hope that HUPD will have a positive social impact on the ML/NLP and Econ/IP communities. They discuss these considerations in more detail in the paper.", "### Impact on Underserved Communities and Discussion of Biases\nThe dataset contains patent applications in English, a language with heavy attention from the NLP community. However, innovation is spread across many languages, cultures, and communities that are not reflected in this dataset. HUPD is thus not representative of all kinds of innovation. Furthermore, patent applications require a fixed cost to draft and file and are not accessible to everyone. One goal of this dataset is to spur research that reduces the cost of drafting applications, potentially allowing for more people to seek intellectual property protection for their innovations.", "### Discussion of Biases\nSection 4 of the HUPD paper provides an examination of the dataset for potential biases. It shows, among other things, that female inventors are notably underrepresented in the U.S. patenting system, that small and micro entities (e.g., independent inventors, small companies, non-profit organizations) are less likely to have positive outcomes in patent obtaining than large entities (e.g., companies with more than 500 employees), and that patent filing and acceptance rates are not uniformly distributed across the US. Our empirical findings suggest that any study focusing on the acceptance prediction task, especially if it is using the inventor information or the small-entity indicator as part of the input, should be aware of the the potential biases present in the dataset and interpret their results carefully in light of those biases.\n\n- Please refer to Section 4 and Section D for an in-depth discussion of potential biases embedded in the dataset.", "### Licensing Information\n\nHUPD is released under the CreativeCommons Attribution-NonCommercial-ShareAlike 4.0 International." ]

[ "TAGS\n#task_categories-fill-mask #task_categories-summarization #task_categories-text-classification #task_categories-token-classification #task_ids-masked-language-modeling #task_ids-multi-class-classification #task_ids-topic-classification #task_ids-named-entity-recognition #language-English #license-cc-by-sa-4.0 #patents #arxiv-2207.04043 #region-us \n", "# Dataset Card for The Harvard USPTO Patent Dataset (HUPD)\n\n!HUPD-Diagram", "## Dataset Description\n\n- Homepage: URL\n- Repository: HUPD GitHub repository\n- Paper: HUPD arXiv Submission\n- Point of Contact: Mirac Suzgun", "### Dataset Summary\n\nThe Harvard USPTO Dataset (HUPD) is a large-scale, well-structured, and multi-purpose corpus of English-language utility patent applications filed to the United States Patent and Trademark Office (USPTO) between January 2004 and December 2018.", "### Experiments and Tasks Considered in the Paper\n\n- Patent Acceptance Prediction: Given a section of a patent application (in particular, the abstract, claims, or description), predict whether the application will be accepted by the USPTO.\n- Automated Subject (IPC/CPC) Classification: Predict the primary IPC or CPC code of a patent application given (some subset of) the text of the application.\n- Language Modeling: Masked/autoregressive language modeling on the claims and description sections of patent applications.\n- Abstractive Summarization: Given the claims or claims section of a patent application, generate the abstract.", "### Languages\n\nThe dataset contains English text only.", "### Domain\nPatents (intellectual property).", "### Dataset Curators\nThe dataset was created by Mirac Suzgun, Luke Melas-Kyriazi, Suproteem K. Sarkar, Scott Duke Kominers, and Stuart M. Shieber.", "## Dataset Structure\n\nEach patent application is defined by a distinct JSON file, named after its application number, and includes information about \nthe application and publication numbers, \ntitle, \ndecision status, \nfiling and publication dates, \nprimary and secondary classification codes, \ninventor(s), \nexaminer,\nattorney,\nabstract, \nclaims, \nbackground, \nsummary, and \nfull description of the proposed invention, among other fields. There are also supplementary variables, such as the small-entity indicator (which denotes whether the applicant is considered to be a small entity by the USPTO) and the foreign-filing indicator (which denotes whether the application was originally filed in a foreign country).\n\nIn total, there are 34 data fields for each application. A full list of data fields used in the dataset is listed in the next section.", "### Data Instances\n\nEach patent application in our patent dataset is defined by a distinct JSON file (e.g., ''URL''), named after its unique application number. The format of the JSON files is as follows:", "## Usage", "### Loading the Dataset", "#### Sample (January 2016 Subset) \nThe following command can be used to load the 'sample' version of the dataset, which contains all the patent applications that were filed to the USPTO during the month of January in 2016. This small subset of the dataset can be used for debugging and exploration purposes.", "#### Full Dataset\n\nIf you would like to use the full version of the dataset, please make sure that change the 'name' field from 'sample' to 'all', specify the training and validation start and end dates carefully, and set 'force_extract' to be 'True' (so that you would only untar the files that you are interested in and not squander your disk storage space). In the following example, for instance, we set the training set year range to be [2011, 2016] (inclusive) and the validation set year range to be 2017.", "### Google Colab Notebook\nYou can also use the following Google Colab notebooks to explore HUPD. \n- ![Open In Colab](URL HUPD Examples: Loading the Dataset\n- ![Open In Colab](URL HUPD Examples: Loading HUPD By Using HuggingFace's Libraries\n- ![Open In Colab](URL HUPD Examples: Using the HUPD DistilRoBERTa Model\n- ![Open In Colab](URL HUPD Examples: Using the HUPD T5-Small Summarization Model", "## Dataset Creation", "### Source Data\n\nHUPD synthesizes multiple data sources from the USPTO: While the full patent application texts were obtained from the USPTO Bulk Data Storage System (Patent Application Data/XML Versions 4.0, 4.1, 4.2, 4.3, 4.4 ICE, as well as Version 1.5) as XML files, the bibliographic filing metadata were obtained from the USPTO Patent Examination Research Dataset (in February, 2021).", "### Annotations\n\nBeyond our patent decision label, for which construction details are provided in the paper, the dataset does not contain any human-written or computer-generated annotations beyond those produced by patent applicants or the USPTO.", "### Data Shift\n\nA major feature of HUPD is its structure, which allows it to demonstrate the evolution of concepts over time. As we illustrate in the paper, the criteria for patent acceptance evolve over time at different rates, depending on category. We believe this is an important feature of the dataset, not only because of the social scientific questions it raises, but also because it facilitates research on models that can accommodate concept shift in a real-world setting.", "### Personal and Sensitive Information\n\nThe dataset contains information about the inventor(s) and examiner of each patent application. These details are, however, already in the public domain and available on the USPTO's Patent Application Information Retrieval (PAIR) system, as well as on Google Patents and PatentsView.", "### Social Impact of the Dataset\n\nThe authors of the dataset hope that HUPD will have a positive social impact on the ML/NLP and Econ/IP communities. They discuss these considerations in more detail in the paper.", "### Impact on Underserved Communities and Discussion of Biases\nThe dataset contains patent applications in English, a language with heavy attention from the NLP community. However, innovation is spread across many languages, cultures, and communities that are not reflected in this dataset. HUPD is thus not representative of all kinds of innovation. Furthermore, patent applications require a fixed cost to draft and file and are not accessible to everyone. One goal of this dataset is to spur research that reduces the cost of drafting applications, potentially allowing for more people to seek intellectual property protection for their innovations.", "### Discussion of Biases\nSection 4 of the HUPD paper provides an examination of the dataset for potential biases. It shows, among other things, that female inventors are notably underrepresented in the U.S. patenting system, that small and micro entities (e.g., independent inventors, small companies, non-profit organizations) are less likely to have positive outcomes in patent obtaining than large entities (e.g., companies with more than 500 employees), and that patent filing and acceptance rates are not uniformly distributed across the US. Our empirical findings suggest that any study focusing on the acceptance prediction task, especially if it is using the inventor information or the small-entity indicator as part of the input, should be aware of the the potential biases present in the dataset and interpret their results carefully in light of those biases.\n\n- Please refer to Section 4 and Section D for an in-depth discussion of potential biases embedded in the dataset.", "### Licensing Information\n\nHUPD is released under the CreativeCommons Attribution-NonCommercial-ShareAlike 4.0 International." ]

[ "annotations_creators:crowdsourced", "language_creators:crowdsourced", "multilinguality:multilingual", "language:hi", "language:mr", "language:or", "language:ta", "language:te", "language:gu", "robust-speech-event", "region:us" ]

[]

[ "hi", "mr", "or", "ta", "te", "gu" ]

[]

[ "TAGS\n#annotations_creators-crowdsourced #language_creators-crowdsourced #multilinguality-multilingual #language-Hindi #language-Marathi #language-Oriya (macrolanguage) #language-Tamil #language-Telugu #language-Gujarati #robust-speech-event #region-us \n" ]

[ "task_categories:question-answering", "task_categories:summarization", "task_categories:text-retrieval", "task_categories:text-generation", "task_ids:abstractive-qa", "task_ids:document-retrieval", "task_ids:extractive-qa", "annotations_creators:expert-generated", "language_creators:crowdsourced", "multilinguality:monolingual", "size_categories:100K<n<1M", "source_datasets:original", "language:en", "license:mit", "conditional-text-generation", "arxiv:2011.07251", "region:us" ]

[ "2011.07251" ]

[ "en" ]

[ "# DebateSum\nCorresponding code repo for the upcoming paper at ARGMIN 2020: \"DebateSum: A large-scale argument mining and summarization dataset\"\n\nArxiv pre-print available here: URL\n\nCheck out the presentation date and time here: URL\n\nFull paper as presented by the ACL is here: URL\n\nVideo of presentation at COLING 2020: URL\n\nThe dataset is distributed as csv files. \n\nA search engine over DebateSum (as well as some additional evidence not included in DebateSum) is available as URL. It's very good quality and allows for the evidence to be viewed in the format that debaters use.", "# Data \n\nDebateSum consists of 187328 debate documents, arguements (also can be thought of as abstractive summaries, or queries), word-level extractive summaries, citations, and associated metadata organized by topic-year. This data is ready for analysis by NLP systems.", "## Download\nAll data is accesable in a parsed format organized by topic year here\n\nAddtionally, the trained word-vectors for debate2vec are also found in that folder.", "## Regenerating it yourself \n\nThis is useful as the debaters who produce the evidence release their work every year. Soon enough I will update to include the 2020-2021 topic. \n\n*Step 1: Download all open evidence files from Open Evidence and unzip them into a directory. The links are as follows:*\n\n* 2019 - Resolved: The United States federal government should substantially reduce Direct Commercial Sales and/or Foreign Military Sales of arms from the United States.\n* 2018 - Resolved: The United States federal government should substantially reduce its restrictions on legal immigration to the United States.\n* 2017 - Resolved: The United States federal government should substantially increase its funding and/or regulation of elementary and/or secondary education in the United States.\n* 2016 - Resolved: The United States federal government should substantially increase its economic and/or diplomatic engagement with the People’s Republic of China.\n* 2015 - Resolved: The United States federal government should substantially curtail its domestic surveil-lance.\n* 2014 - Resolved: The United States federal government should substantially increase its non-military exploration and/or development of the Earth’s oceans.\n* 2013 - Resolved: The United States federal government should substantially increase its economic en-gagement toward Cuba, Mexico or Venezuela.\n\n\n*Step 2: Convert all evidence from docx files to html5 files using pandoc with this command:*\n\n\n*Step 3: install the dependencies for make_debate_dataset.py.*\n\n\n\n*Step 4: Modify the folder and file locations as needed for your system, and run make_debate_dataset.py*", "# Credits \nHuge thanks to Arvind Balaji for making URL and being second author on this paper!" ]

[ "TAGS\n#task_categories-question-answering #task_categories-summarization #task_categories-text-retrieval #task_categories-text-generation #task_ids-abstractive-qa #task_ids-document-retrieval #task_ids-extractive-qa #annotations_creators-expert-generated #language_creators-crowdsourced #multilinguality-monolingual #size_categories-100K<n<1M #source_datasets-original #language-English #license-mit #conditional-text-generation #arxiv-2011.07251 #region-us \n", "# DebateSum\nCorresponding code repo for the upcoming paper at ARGMIN 2020: \"DebateSum: A large-scale argument mining and summarization dataset\"\n\nArxiv pre-print available here: URL\n\nCheck out the presentation date and time here: URL\n\nFull paper as presented by the ACL is here: URL\n\nVideo of presentation at COLING 2020: URL\n\nThe dataset is distributed as csv files. \n\nA search engine over DebateSum (as well as some additional evidence not included in DebateSum) is available as URL. It's very good quality and allows for the evidence to be viewed in the format that debaters use.", "# Data \n\nDebateSum consists of 187328 debate documents, arguements (also can be thought of as abstractive summaries, or queries), word-level extractive summaries, citations, and associated metadata organized by topic-year. This data is ready for analysis by NLP systems.", "## Download\nAll data is accesable in a parsed format organized by topic year here\n\nAddtionally, the trained word-vectors for debate2vec are also found in that folder.", "## Regenerating it yourself \n\nThis is useful as the debaters who produce the evidence release their work every year. Soon enough I will update to include the 2020-2021 topic. \n\n*Step 1: Download all open evidence files from Open Evidence and unzip them into a directory. The links are as follows:*\n\n* 2019 - Resolved: The United States federal government should substantially reduce Direct Commercial Sales and/or Foreign Military Sales of arms from the United States.\n* 2018 - Resolved: The United States federal government should substantially reduce its restrictions on legal immigration to the United States.\n* 2017 - Resolved: The United States federal government should substantially increase its funding and/or regulation of elementary and/or secondary education in the United States.\n* 2016 - Resolved: The United States federal government should substantially increase its economic and/or diplomatic engagement with the People’s Republic of China.\n* 2015 - Resolved: The United States federal government should substantially curtail its domestic surveil-lance.\n* 2014 - Resolved: The United States federal government should substantially increase its non-military exploration and/or development of the Earth’s oceans.\n* 2013 - Resolved: The United States federal government should substantially increase its economic en-gagement toward Cuba, Mexico or Venezuela.\n\n\n*Step 2: Convert all evidence from docx files to html5 files using pandoc with this command:*\n\n\n*Step 3: install the dependencies for make_debate_dataset.py.*\n\n\n\n*Step 4: Modify the folder and file locations as needed for your system, and run make_debate_dataset.py*", "# Credits \nHuge thanks to Arvind Balaji for making URL and being second author on this paper!" ]

[ "annotations_creators:no-annotation", "language_creators:crowdsourced", "multilinguality:translation", "size_categories:unknown", "source_datasets:original", "language:af", "language:ar", "language:az", "language:be", "language:bg", "language:bn", "language:br", "language:bs", "language:ca", "language:ch", "language:cs", "language:cv", "language:cy", "language:da", "language:de", "language:el", "language:en", "language:eo", "language:es", "language:et", "language:eu", "language:fa", "language:fi", "language:fo", "language:fr", "language:fy", "language:ga", "language:gd", "language:gl", "language:gn", "language:he", "language:hi", "language:hr", "language:hu", "language:hy", "language:ia", "language:id", "language:ie", "language:io", "language:is", "language:it", "language:ja", "language:jv", "language:ka", "language:kk", "language:km", "language:ko", "language:ku", "language:kw", "language:la", "language:lb", "language:lt", "language:lv", "language:mi", "language:mk", "language:ml", "language:mn", "language:mr", "language:ms", "language:mt", "language:my", "language:nb", "language:nl", "language:nn", "language:no", "language:oc", "language:pl", "language:pt", "language:qu", "language:rn", "language:ro", "language:ru", "language:sh", "language:sl", "language:sq", "language:sr", "language:sv", "language:sw", "language:ta", "language:te", "language:th", "language:tk", "language:tl", "language:tr", "language:tt", "language:ug", "language:uk", "language:ur", "language:uz", "language:vi", "language:vo", "language:yi", "language:zh", "license:cc-by-2.0", "region:us" ]

[]

[ "af", "ar", "az", "be", "bg", "bn", "br", "bs", "ca", "ch", "cs", "cv", "cy", "da", "de", "el", "en", "eo", "es", "et", "eu", "fa", "fi", "fo", "fr", "fy", "ga", "gd", "gl", "gn", "he", "hi", "hr", "hu", "hy", "ia", "id", "ie", "io", "is", "it", "ja", "jv", "ka", "kk", "km", "ko", "ku", "kw", "la", "lb", "lt", "lv", "mi", "mk", "ml", "mn", "mr", "ms", "mt", "my", "nb", "nl", "nn", "no", "oc", "pl", "pt", "qu", "rn", "ro", "ru", "sh", "sl", "sq", "sr", "sv", "sw", "ta", "te", "th", "tk", "tl", "tr", "tt", "ug", "uk", "ur", "uz", "vi", "vo", "yi", "zh" ]

[ "# Dataset Card for [Dataset Name]", "## Table of Contents\n- Table of Contents\n- Dataset Description\n - Dataset Summary\n - Supported Tasks and Leaderboards\n - Languages\n- Dataset Structure\n - Data Instances\n - Data Fields\n - Data Splits\n- Dataset Creation\n - Curation Rationale\n - Source Data\n - Annotations\n - Personal and Sensitive Information\n- Considerations for Using the Data\n - Social Impact of Dataset\n - Discussion of Biases\n - Other Known Limitations\n- Additional Information\n - Dataset Curators\n - Licensing Information\n - Citation Information\n - Contributions", "## Dataset Description\n\n- Homepage: URL\n- Repository: URL\n- Paper: The Tatoeba Translation Challenge – Realistic Data Sets for Low Resource and Multilingual MT\n- Leaderboard:\n- Point of Contact: Jörg Tiedemann", "### Dataset Summary\n\nThe Tatoeba Translation Challenge is a multilingual data set of machine translation benchmarks derived from user-contributed translations collected by URL and provided as parallel corpus from OPUS. This dataset includes test and development data sorted by language pair. It includes test sets for hundreds of language pairs and is continuously updated. Please, check the version number tag to refer to the release that your are using.", "### Supported Tasks and Leaderboards\n\nThe translation task is described in detail in the Tatoeba-Challenge repository and covers various sub-tasks with different data coverage and resources. Training data is also available from the same repository and results are published and collected as well. Models are also released for public use and are also partially available from the huggingface model hub.", "### Languages\n\nThe data set covers hundreds of languages and language pairs and are organized by ISO-639-3 languages. The current release covers the following language: Afrikaans, Arabic, Azerbaijani, Belarusian, Bulgarian, Bengali, Breton, Bosnian, Catalan, Chamorro, Czech, Chuvash, Welsh, Danish, German, Modern Greek, English, Esperanto, Spanish, Estonian, Basque, Persian, Finnish, Faroese, French, Western Frisian, Irish, Scottish Gaelic, Galician, Guarani, Hebrew, Hindi, Croatian, Hungarian, Armenian, Interlingua, Indonesian, Interlingue, Ido, Icelandic, Italian, Japanese, Javanese, Georgian, Kazakh, Khmer, Korean, Kurdish, Cornish, Latin, Luxembourgish, Lithuanian, Latvian, Maori, Macedonian, Malayalam, Mongolian, Marathi, Malay, Maltese, Burmese, Norwegian Bokmål, Dutch, Norwegian Nynorsk, Norwegian, Occitan, Polish, Portuguese, Quechua, Rundi, Romanian, Russian, Serbo-Croatian, Slovenian, Albanian, Serbian, Swedish, Swahili, Tamil, Telugu, Thai, Turkmen, Tagalog, Turkish, Tatar, Uighur, Ukrainian, Urdu, Uzbek, Vietnamese, Volapük, Yiddish, Chinese", "## Dataset Structure", "### Data Instances\n\nData instances are given as translation units in TAB-separated files with four columns: source and target language ISO-639-3 codes, source language text and target language text. Note that we do not imply a translation direction and consider the data set to be symmetric and to be used as a test set in both directions. Language-pair-specific subsets are only provided under the label of one direction using sorted ISO-639-3 language IDs.\n\nSome subsets contain several sub-languages or language variants. They may refer to macro-languages such as Serbo-Croatian languages that are covered by the ISO code 'hbs'. Language variants may also include different writing systems and in that case the ISO15924 script codes are attached to the language codes. Here are a few examples from the English to Serbo-Croation test set including examples for Bosnian, Croatian and Serbian in Cyrillic and in Latin scripts:\n\n\n\nThere are also data sets with sentence pairs in the same language. In most cases, those are variants with minor spelling differences but they also include rephrased sentences. Here are a few examples from the English test set:", "### Data Splits\n\nTest and development data sets are disjoint with respect to sentence pairs but may include overlaps in individual source or target language sentences. Development data should not be used in training directly. The goal of the data splits is to create test sets of reasonable size with a large language coverage. Test sets include at most 10,000 instances. Development data do not exist for all language pairs.\n\nTo be comparable with other results, models should use the training data distributed from the Tatoeba MT Challenge Repository including monolingual data sets also listed there.", "## Dataset Creation", "### Curation Rationale\n\n\nThe Tatoeba MT data set will be updated continuously and the data preparation procedures are also public and released on github. High language coverage is the main goal of the project and data sets are prepared to be consistent and systematic with standardized language labels and distribution formats.", "### Source Data", "#### Initial Data Collection and Normalization\n\nThe Tatoeba data sets are collected from user-contributed translations submitted to URL and compiled into a multi-parallel corpus in OPUS. The test and development sets are incrementally updated with new releases of the Tatoeba data collection at OPUS. New releases extend the existing data sets. Test sets should not overlap with any of the released development data sets.", "#### Who are the source language producers?\n\nThe data sets come from URL, which provides a large database of sentences and their translations into a wide varity of languages. Its content is constantly growing as a result of voluntary contributions of thousands of users.\nThe original project was founded by Trang Ho in 2006, hosted on Sourceforge under the codename of multilangdict.", "### Annotations", "#### Annotation process\n\nSentences are translated by volunteers and the Tatoeba database also provides additional metadata about each record including user ratings etc. However, the metadata is currently not used in any way for the compilation of the MT benchmark. Language skills of contributors naturally vary quite a bit and not all translations are done by native speakers of the target language. More information about the contributions can be found at URL.", "#### Who are the annotators?", "### Personal and Sensitive Information\n\nFor information about handling personal and sensitive information we refer to the original provider of the data. This data set has not been processed in any way to detect or remove potentially sensitve or personal information.", "## Considerations for Using the Data", "### Social Impact of Dataset\n\nThe language coverage is high and with that it represents a highly valuable resource for machine translation development especially for lesser resourced languages and language pairs. The constantly growing database also represents a dynamic resource and its value wil grow further.", "### Discussion of Biases\n\nThe original source lives from its contributors and there interest and background will to certain subjective and cultural biases. Language coverage and translation quality is also biased by the skills of the contributors.", "### Other Known Limitations\n\nThe sentences are typically quite short and, therefore, rather easy to translate. For high-resource languages, this leads to results that will be less useful than more challenging benchmarks. For lesser resource language pairs, the limited complexity of the examples is actually a good thing to measure progress even in very challenging setups.", "## Additional Information", "### Dataset Curators\n\nThe data set is curated by the University of Helsinki and its language technology research group. Data and tools used for creating and using the resource are open source and will be maintained as part of the OPUS ecosystem for parallel data and machine translation research.", "### Licensing Information\n\nThe data sets are distributed under the same licence agreement as the original Tatoeba database using a\nCC-BY 2.0 license. More information about the terms of use of the original data sets is listed here.\n\n\n\n\nIf you use the data sets then, please, cite the following paper: The Tatoeba Translation Challenge – Realistic Data Sets for Low Resource and Multilingual MT", "### Contributions\n\nThanks to @jorgtied and @Helsinki-NLP for adding this dataset.\nThanks also to CSC Finland for providing computational resources and storage space for the work on OPUS and other MT projects." ]

[ "TAGS\n#annotations_creators-no-annotation #language_creators-crowdsourced #multilinguality-translation #size_categories-unknown #source_datasets-original #language-Afrikaans #language-Arabic #language-Azerbaijani #language-Belarusian #language-Bulgarian #language-Bengali #language-Breton #language-Bosnian #language-Catalan #language-Chamorro #language-Czech #language-Chuvash #language-Welsh #language-Danish #language-German #language-Modern Greek (1453-) #language-English #language-Esperanto #language-Spanish #language-Estonian #language-Basque #language-Persian #language-Finnish #language-Faroese #language-French #language-Western Frisian #language-Irish #language-Scottish Gaelic #language-Galician #language-Guarani #language-Hebrew #language-Hindi #language-Croatian #language-Hungarian #language-Armenian #language-Interlingua (International Auxiliary Language Association) #language-Indonesian #language-Interlingue #language-Ido #language-Icelandic #language-Italian #language-Japanese #language-Javanese #language-Georgian #language-Kazakh #language-Khmer #language-Korean #language-Kurdish #language-Cornish #language-Latin #language-Luxembourgish #language-Lithuanian #language-Latvian #language-Maori #language-Macedonian #language-Malayalam #language-Mongolian #language-Marathi #language-Malay (macrolanguage) #language-Maltese #language-Burmese #language-Norwegian Bokmål #language-Dutch #language-Norwegian Nynorsk #language-Norwegian #language-Occitan (post 1500) #language-Polish #language-Portuguese #language-Quechua #language-Rundi #language-Romanian #language-Russian #language-Serbo-Croatian #language-Slovenian #language-Albanian #language-Serbian #language-Swedish #language-Swahili (macrolanguage) #language-Tamil #language-Telugu #language-Thai #language-Turkmen #language-Tagalog #language-Turkish #language-Tatar #language-Uighur #language-Ukrainian #language-Urdu #language-Uzbek #language-Vietnamese #language-Volapük #language-Yiddish #language-Chinese #license-cc-by-2.0 #region-us \n", "# Dataset Card for [Dataset Name]", "## Table of Contents\n- Table of Contents\n- Dataset Description\n - Dataset Summary\n - Supported Tasks and Leaderboards\n - Languages\n- Dataset Structure\n - Data Instances\n - Data Fields\n - Data Splits\n- Dataset Creation\n - Curation Rationale\n - Source Data\n - Annotations\n - Personal and Sensitive Information\n- Considerations for Using the Data\n - Social Impact of Dataset\n - Discussion of Biases\n - Other Known Limitations\n- Additional Information\n - Dataset Curators\n - Licensing Information\n - Citation Information\n - Contributions", "## Dataset Description\n\n- Homepage: URL\n- Repository: URL\n- Paper: The Tatoeba Translation Challenge – Realistic Data Sets for Low Resource and Multilingual MT\n- Leaderboard:\n- Point of Contact: Jörg Tiedemann", "### Dataset Summary\n\nThe Tatoeba Translation Challenge is a multilingual data set of machine translation benchmarks derived from user-contributed translations collected by URL and provided as parallel corpus from OPUS. This dataset includes test and development data sorted by language pair. It includes test sets for hundreds of language pairs and is continuously updated. Please, check the version number tag to refer to the release that your are using.", "### Supported Tasks and Leaderboards\n\nThe translation task is described in detail in the Tatoeba-Challenge repository and covers various sub-tasks with different data coverage and resources. Training data is also available from the same repository and results are published and collected as well. Models are also released for public use and are also partially available from the huggingface model hub.", "### Languages\n\nThe data set covers hundreds of languages and language pairs and are organized by ISO-639-3 languages. The current release covers the following language: Afrikaans, Arabic, Azerbaijani, Belarusian, Bulgarian, Bengali, Breton, Bosnian, Catalan, Chamorro, Czech, Chuvash, Welsh, Danish, German, Modern Greek, English, Esperanto, Spanish, Estonian, Basque, Persian, Finnish, Faroese, French, Western Frisian, Irish, Scottish Gaelic, Galician, Guarani, Hebrew, Hindi, Croatian, Hungarian, Armenian, Interlingua, Indonesian, Interlingue, Ido, Icelandic, Italian, Japanese, Javanese, Georgian, Kazakh, Khmer, Korean, Kurdish, Cornish, Latin, Luxembourgish, Lithuanian, Latvian, Maori, Macedonian, Malayalam, Mongolian, Marathi, Malay, Maltese, Burmese, Norwegian Bokmål, Dutch, Norwegian Nynorsk, Norwegian, Occitan, Polish, Portuguese, Quechua, Rundi, Romanian, Russian, Serbo-Croatian, Slovenian, Albanian, Serbian, Swedish, Swahili, Tamil, Telugu, Thai, Turkmen, Tagalog, Turkish, Tatar, Uighur, Ukrainian, Urdu, Uzbek, Vietnamese, Volapük, Yiddish, Chinese", "## Dataset Structure", "### Data Instances\n\nData instances are given as translation units in TAB-separated files with four columns: source and target language ISO-639-3 codes, source language text and target language text. Note that we do not imply a translation direction and consider the data set to be symmetric and to be used as a test set in both directions. Language-pair-specific subsets are only provided under the label of one direction using sorted ISO-639-3 language IDs.\n\nSome subsets contain several sub-languages or language variants. They may refer to macro-languages such as Serbo-Croatian languages that are covered by the ISO code 'hbs'. Language variants may also include different writing systems and in that case the ISO15924 script codes are attached to the language codes. Here are a few examples from the English to Serbo-Croation test set including examples for Bosnian, Croatian and Serbian in Cyrillic and in Latin scripts:\n\n\n\nThere are also data sets with sentence pairs in the same language. In most cases, those are variants with minor spelling differences but they also include rephrased sentences. Here are a few examples from the English test set:", "### Data Splits\n\nTest and development data sets are disjoint with respect to sentence pairs but may include overlaps in individual source or target language sentences. Development data should not be used in training directly. The goal of the data splits is to create test sets of reasonable size with a large language coverage. Test sets include at most 10,000 instances. Development data do not exist for all language pairs.\n\nTo be comparable with other results, models should use the training data distributed from the Tatoeba MT Challenge Repository including monolingual data sets also listed there.", "## Dataset Creation", "### Curation Rationale\n\n\nThe Tatoeba MT data set will be updated continuously and the data preparation procedures are also public and released on github. High language coverage is the main goal of the project and data sets are prepared to be consistent and systematic with standardized language labels and distribution formats.", "### Source Data", "#### Initial Data Collection and Normalization\n\nThe Tatoeba data sets are collected from user-contributed translations submitted to URL and compiled into a multi-parallel corpus in OPUS. The test and development sets are incrementally updated with new releases of the Tatoeba data collection at OPUS. New releases extend the existing data sets. Test sets should not overlap with any of the released development data sets.", "#### Who are the source language producers?\n\nThe data sets come from URL, which provides a large database of sentences and their translations into a wide varity of languages. Its content is constantly growing as a result of voluntary contributions of thousands of users.\nThe original project was founded by Trang Ho in 2006, hosted on Sourceforge under the codename of multilangdict.", "### Annotations", "#### Annotation process\n\nSentences are translated by volunteers and the Tatoeba database also provides additional metadata about each record including user ratings etc. However, the metadata is currently not used in any way for the compilation of the MT benchmark. Language skills of contributors naturally vary quite a bit and not all translations are done by native speakers of the target language. More information about the contributions can be found at URL.", "#### Who are the annotators?", "### Personal and Sensitive Information\n\nFor information about handling personal and sensitive information we refer to the original provider of the data. This data set has not been processed in any way to detect or remove potentially sensitve or personal information.", "## Considerations for Using the Data", "### Social Impact of Dataset\n\nThe language coverage is high and with that it represents a highly valuable resource for machine translation development especially for lesser resourced languages and language pairs. The constantly growing database also represents a dynamic resource and its value wil grow further.", "### Discussion of Biases\n\nThe original source lives from its contributors and there interest and background will to certain subjective and cultural biases. Language coverage and translation quality is also biased by the skills of the contributors.", "### Other Known Limitations\n\nThe sentences are typically quite short and, therefore, rather easy to translate. For high-resource languages, this leads to results that will be less useful than more challenging benchmarks. For lesser resource language pairs, the limited complexity of the examples is actually a good thing to measure progress even in very challenging setups.", "## Additional Information", "### Dataset Curators\n\nThe data set is curated by the University of Helsinki and its language technology research group. Data and tools used for creating and using the resource are open source and will be maintained as part of the OPUS ecosystem for parallel data and machine translation research.", "### Licensing Information\n\nThe data sets are distributed under the same licence agreement as the original Tatoeba database using a\nCC-BY 2.0 license. More information about the terms of use of the original data sets is listed here.\n\n\n\n\nIf you use the data sets then, please, cite the following paper: The Tatoeba Translation Challenge – Realistic Data Sets for Low Resource and Multilingual MT", "### Contributions\n\nThanks to @jorgtied and @Helsinki-NLP for adding this dataset.\nThanks also to CSC Finland for providing computational resources and storage space for the work on OPUS and other MT projects." ]

[ "region:us" ]

[]

[]

[]

[ "TAGS\n#region-us \n" ]

[ "region:us" ]

[]

[]

[]

[ "TAGS\n#region-us \n" ]

[ "region:us" ]

[]

[]

[]

[ "TAGS\n#region-us \n" ]

[ "region:us" ]

[]

[]

[ "# Dataset Card for [Dataset Name]", "## Table of Contents\n- Table of Contents\n- Dataset Description\n - Dataset Summary\n - Supported Tasks and Leaderboards\n - Languages\n- Dataset Structure\n - Data Instances\n - Data Fields\n - Data Splits\n- Dataset Creation\n - Curation Rationale\n - Source Data\n - Annotations\n - Personal and Sensitive Information\n- Considerations for Using the Data\n - Social Impact of Dataset\n - Discussion of Biases\n - Other Known Limitations\n- Additional Information\n - Dataset Curators\n - Licensing Information\n - Citation Information\n - Contributions", "## Dataset Description\n\n- Homepage:\n- Repository:\n- Paper:\n- Leaderboard:\n- Point of Contact:", "### Dataset Summary", "### Supported Tasks and Leaderboards", "### Languages", "## Dataset Structure", "### Data Instances", "### Data Fields", "### Data Splits", "## Dataset Creation", "### Curation Rationale", "### Source Data", "#### Initial Data Collection and Normalization", "#### Who are the source language producers?", "### Annotations", "#### Annotation process", "#### Who are the annotators?", "### Personal and Sensitive Information", "## Considerations for Using the Data", "### Social Impact of Dataset", "### Discussion of Biases", "### Other Known Limitations", "## Additional Information", "### Dataset Curators", "### Licensing Information", "### Contributions\n\nThanks to @github-username for adding this dataset." ]

[ "TAGS\n#region-us \n", "# Dataset Card for [Dataset Name]", "## Table of Contents\n- Table of Contents\n- Dataset Description\n - Dataset Summary\n - Supported Tasks and Leaderboards\n - Languages\n- Dataset Structure\n - Data Instances\n - Data Fields\n - Data Splits\n- Dataset Creation\n - Curation Rationale\n - Source Data\n - Annotations\n - Personal and Sensitive Information\n- Considerations for Using the Data\n - Social Impact of Dataset\n - Discussion of Biases\n - Other Known Limitations\n- Additional Information\n - Dataset Curators\n - Licensing Information\n - Citation Information\n - Contributions", "## Dataset Description\n\n- Homepage:\n- Repository:\n- Paper:\n- Leaderboard:\n- Point of Contact:", "### Dataset Summary", "### Supported Tasks and Leaderboards", "### Languages", "## Dataset Structure", "### Data Instances", "### Data Fields", "### Data Splits", "## Dataset Creation", "### Curation Rationale", "### Source Data", "#### Initial Data Collection and Normalization", "#### Who are the source language producers?", "### Annotations", "#### Annotation process", "#### Who are the annotators?", "### Personal and Sensitive Information", "## Considerations for Using the Data", "### Social Impact of Dataset", "### Discussion of Biases", "### Other Known Limitations", "## Additional Information", "### Dataset Curators", "### Licensing Information", "### Contributions\n\nThanks to @github-username for adding this dataset." ]

[ "region:us" ]

[]

[]

[]

[ "TAGS\n#region-us \n" ]

[ "task_categories:summarization", "annotations_creators:expert-generated", "annotations_creators:found", "language_creators:expert-generated", "language_creators:found", "multilinguality:monolingual", "size_categories:10K<n<100K", "source_datasets:original", "language:ru", "license:unknown", "arxiv:2006.11063", "region:us" ]

[ "2006.11063" ]

[ "ru" ]

[ "### Dataset Summary\n\n\nDataset for automatic summarization of Russian news. News and their summaries are from the Gazeta website. Summaries were parsed as the content of an HTML tag with “description” property. Additional selection of good summaries was performed. There are two versions of this dataset.", "### Supported Tasks and Leaderboards\n\n\nLeaderboard on Papers With Code: text-summarization-on-gazeta.\n\n\nPlease use the original evaluation script with the same parameters. Example:", "### Languages\n\n\nThe dataset is in Russian.", "### Usage\n\n\nLoading version 1.0:\n\n\nLoading version 2.0:", "### Other datasets\n\n\nOther Russian summarization datasets:\n\n\n* Russian part of XL-Sum, parsed from URL 77803 samples\n* Russian part of MLSUM, parsed from URL, 27063 samples\n\n\nDataset Structure\n-----------------", "### Data Instances\n\n\nFor each instance, there is a string for the article, a string for the summary, and a string for the url. Additionally, a string for the title and a date are provided.\n\n\nSome dataset statistics are below:", "### Data Splits\n\n\n\nDataset Creation\n----------------", "### Curation Rationale\n\n\nWhen the first version of the dataset was collected, there were no other datasets for Russian text summarization. Even now, it is one of the few datasets for this task.", "### Source Data", "#### Initial Data Collection and Normalization\n\n\n* The source of data is the Gazeta website.\n* Parsing scripts are here.\n* Cleaning and normalization Colab notebook is here", "#### Who are the source language producers?\n\n\nTexts and summaries were written by journalists at Gazeta.", "### Annotations", "#### Annotation process\n\n\n[N/A]", "#### Who are the annotators?\n\n\n[N/A]", "### Personal and Sensitive Information\n\n\nThe dataset is not anonymized, so individuals' names can be found in the dataset. Information about the original author is not included in the dataset.\n\n\nConsiderations for Using the Data\n---------------------------------", "### Social Impact of Dataset", "### Discussion of Biases\n\n\nIt is a dataset from a single source. Thus it has a constrained text style and event perspective.", "### Other Known Limitations\n\n\nAdditional Information\n----------------------", "### Dataset Curators\n\n\nThe data was collected by Ilya Gusev.", "### Licensing Information\n\n\nLegal basis for distribution of the dataset: URL paragraph 2.1.2. All rights belong to \"URL\". Usage of this dataset is possible only for personal purposes on a non-commercial basis.", "### Contributions\n\n\n[N/A]" ]

[ "TAGS\n#task_categories-summarization #annotations_creators-expert-generated #annotations_creators-found #language_creators-expert-generated #language_creators-found #multilinguality-monolingual #size_categories-10K<n<100K #source_datasets-original #language-Russian #license-unknown #arxiv-2006.11063 #region-us \n", "### Dataset Summary\n\n\nDataset for automatic summarization of Russian news. News and their summaries are from the Gazeta website. Summaries were parsed as the content of an HTML tag with “description” property. Additional selection of good summaries was performed. There are two versions of this dataset.", "### Supported Tasks and Leaderboards\n\n\nLeaderboard on Papers With Code: text-summarization-on-gazeta.\n\n\nPlease use the original evaluation script with the same parameters. Example:", "### Languages\n\n\nThe dataset is in Russian.", "### Usage\n\n\nLoading version 1.0:\n\n\nLoading version 2.0:", "### Other datasets\n\n\nOther Russian summarization datasets:\n\n\n* Russian part of XL-Sum, parsed from URL 77803 samples\n* Russian part of MLSUM, parsed from URL, 27063 samples\n\n\nDataset Structure\n-----------------", "### Data Instances\n\n\nFor each instance, there is a string for the article, a string for the summary, and a string for the url. Additionally, a string for the title and a date are provided.\n\n\nSome dataset statistics are below:", "### Data Splits\n\n\n\nDataset Creation\n----------------", "### Curation Rationale\n\n\nWhen the first version of the dataset was collected, there were no other datasets for Russian text summarization. Even now, it is one of the few datasets for this task.", "### Source Data", "#### Initial Data Collection and Normalization\n\n\n* The source of data is the Gazeta website.\n* Parsing scripts are here.\n* Cleaning and normalization Colab notebook is here", "#### Who are the source language producers?\n\n\nTexts and summaries were written by journalists at Gazeta.", "### Annotations", "#### Annotation process\n\n\n[N/A]", "#### Who are the annotators?\n\n\n[N/A]", "### Personal and Sensitive Information\n\n\nThe dataset is not anonymized, so individuals' names can be found in the dataset. Information about the original author is not included in the dataset.\n\n\nConsiderations for Using the Data\n---------------------------------", "### Social Impact of Dataset", "### Discussion of Biases\n\n\nIt is a dataset from a single source. Thus it has a constrained text style and event perspective.", "### Other Known Limitations\n\n\nAdditional Information\n----------------------", "### Dataset Curators\n\n\nThe data was collected by Ilya Gusev.", "### Licensing Information\n\n\nLegal basis for distribution of the dataset: URL paragraph 2.1.2. All rights belong to \"URL\". Usage of this dataset is possible only for personal purposes on a non-commercial basis.", "### Contributions\n\n\n[N/A]" ]

[ "task_categories:text-classification", "task_ids:multi-class-classification", "annotations_creators:crowdsourced", "language_creators:found", "multilinguality:multilingual", "size_categories:10K<n<100K", "source_datasets:original", "language:ru", "language:en", "license:cc0-1.0", "causal-reasoning", "arxiv:2108.12626", "region:us" ]

[ "2108.12626" ]

[ "ru", "en" ]

[ "### Dataset Summary\n\n\nA dataset for detecting implicit causal relations between pairs of news headlines. The dataset includes over 5000 headline pairs from English news and over 9000 headline pairs from Russian news labeled through crowdsourcing. The pairs vary from totally unrelated or belonging to the same general topic to the ones including causation and refutation relations.", "### Usage\n\n\nLoading Russian Simple task:", "### Supported Tasks and Leaderboards", "### Languages\n\n\nThis dataset consists of two parts, Russian and English.\n\n\nDataset Structure\n-----------------", "### Data Instances\n\n\nThere is an URL, a title, and a timestamp for each of the two headlines in every data instance. A label is presented in three fields. 'Result' field is a textual label, 'label' field is a numeric label, and the 'agreement' field shows the majority vote agreement between annotators. Additional information includes instance ID and the presence of the link between two articles.", "### Data Splits\n\n\n\nDataset Creation\n----------------", "### Curation Rationale", "### Source Data", "#### Initial Data Collection and Normalization", "#### Who are the source language producers?", "### Annotations", "#### Annotation process\n\n\nEvery candidate pair was annotated with Yandex Toloka, a crowdsourcing platform. The task was to determine a relationship between two headlines, A and B. There were seven possible options: titles are almost the same, A causes B, B causes A, A refutes B, B refutes A, A linked with B in another way, A is not linked to B. An annotation guideline was in Russian for Russian news and in English for English news.\n\n\nGuidelines:\n\n\n* Russian: link\n* English: link\n\n\nTen workers annotated every pair. The total annotation budget was 870$, with the estimated hourly wage paid to participants of 45 cents. Annotation management was semi-automatic. Scripts are available in the Github repository.", "#### Who are the annotators?\n\n\nYandex Toloka workers were the annotators, 457 workers for the Russian part, 180 workers for the English part.", "### Personal and Sensitive Information\n\n\nThe dataset is not anonymized, so individuals' names can be found in the dataset. Information about the original author is not included in the dataset. No information about annotators is included except a platform worker ID.\n\n\nConsiderations for Using the Data\n---------------------------------", "### Social Impact of Dataset\n\n\nWe do not see any direct malicious applications of our work. The data probably do not contain offensive content, as news agencies usually do not produce it, and a keyword search returned nothing. However, there are news documents in the dataset on several sensitive topics.", "### Discussion of Biases", "### Other Known Limitations\n\n\nAdditional Information\n----------------------", "### Dataset Curators\n\n\nThe data was collected by Ilya Gusev.", "### Licensing Information", "### Contributions\n\n\n[N/A]" ]

[ "TAGS\n#task_categories-text-classification #task_ids-multi-class-classification #annotations_creators-crowdsourced #language_creators-found #multilinguality-multilingual #size_categories-10K<n<100K #source_datasets-original #language-Russian #language-English #license-cc0-1.0 #causal-reasoning #arxiv-2108.12626 #region-us \n", "### Dataset Summary\n\n\nA dataset for detecting implicit causal relations between pairs of news headlines. The dataset includes over 5000 headline pairs from English news and over 9000 headline pairs from Russian news labeled through crowdsourcing. The pairs vary from totally unrelated or belonging to the same general topic to the ones including causation and refutation relations.", "### Usage\n\n\nLoading Russian Simple task:", "### Supported Tasks and Leaderboards", "### Languages\n\n\nThis dataset consists of two parts, Russian and English.\n\n\nDataset Structure\n-----------------", "### Data Instances\n\n\nThere is an URL, a title, and a timestamp for each of the two headlines in every data instance. A label is presented in three fields. 'Result' field is a textual label, 'label' field is a numeric label, and the 'agreement' field shows the majority vote agreement between annotators. Additional information includes instance ID and the presence of the link between two articles.", "### Data Splits\n\n\n\nDataset Creation\n----------------", "### Curation Rationale", "### Source Data", "#### Initial Data Collection and Normalization", "#### Who are the source language producers?", "### Annotations", "#### Annotation process\n\n\nEvery candidate pair was annotated with Yandex Toloka, a crowdsourcing platform. The task was to determine a relationship between two headlines, A and B. There were seven possible options: titles are almost the same, A causes B, B causes A, A refutes B, B refutes A, A linked with B in another way, A is not linked to B. An annotation guideline was in Russian for Russian news and in English for English news.\n\n\nGuidelines:\n\n\n* Russian: link\n* English: link\n\n\nTen workers annotated every pair. The total annotation budget was 870$, with the estimated hourly wage paid to participants of 45 cents. Annotation management was semi-automatic. Scripts are available in the Github repository.", "#### Who are the annotators?\n\n\nYandex Toloka workers were the annotators, 457 workers for the Russian part, 180 workers for the English part.", "### Personal and Sensitive Information\n\n\nThe dataset is not anonymized, so individuals' names can be found in the dataset. Information about the original author is not included in the dataset. No information about annotators is included except a platform worker ID.\n\n\nConsiderations for Using the Data\n---------------------------------", "### Social Impact of Dataset\n\n\nWe do not see any direct malicious applications of our work. The data probably do not contain offensive content, as news agencies usually do not produce it, and a keyword search returned nothing. However, there are news documents in the dataset on several sensitive topics.", "### Discussion of Biases", "### Other Known Limitations\n\n\nAdditional Information\n----------------------", "### Dataset Curators\n\n\nThe data was collected by Ilya Gusev.", "### Licensing Information", "### Contributions\n\n\n[N/A]" ]

[ "region:us" ]

[]

[]

[ "### Languages\n\n\nEnglish\n\n\nLoad Dataset\n------------\n\n\nYou can read in WEC-Eng files as follows (using the huggingface\\_hub library):\n\n\nDataset Structure\n-----------------", "### Data Splits\n\n\n* Final version of the English CD event coreference dataset \n\n\t+ Train - Train\\_Event\\_gold\\_mentions.json\n\t+ Dev - Dev\\_Event\\_gold\\_mentions\\_validated.json\n\t+ Test - Test\\_Event\\_gold\\_mentions\\_validated.json\n\n\n\n* The non (within clusters) controlled version of the dataset (lexical diversity) \n\n\t+ All (experimental) - All\\_Event\\_gold\\_mentions\\_unfiltered.json", "### Data Instances", "### Data Fields\n\n\n\nLicense\n-------\n\n\nWe provide the following data sets under a <a href=\"URL Commons Attribution-ShareAlike 3.0 Unported License. It is based on content extracted from Wikipedia that is licensed under the Creative Commons Attribution-ShareAlike 3.0 Unported License\n\n\nContact\n-------\n\n\nIf you have any questions please create a Github issue at URL" ]

[ "TAGS\n#region-us \n", "### Languages\n\n\nEnglish\n\n\nLoad Dataset\n------------\n\n\nYou can read in WEC-Eng files as follows (using the huggingface\\_hub library):\n\n\nDataset Structure\n-----------------", "### Data Splits\n\n\n* Final version of the English CD event coreference dataset \n\n\t+ Train - Train\\_Event\\_gold\\_mentions.json\n\t+ Dev - Dev\\_Event\\_gold\\_mentions\\_validated.json\n\t+ Test - Test\\_Event\\_gold\\_mentions\\_validated.json\n\n\n\n* The non (within clusters) controlled version of the dataset (lexical diversity) \n\n\t+ All (experimental) - All\\_Event\\_gold\\_mentions\\_unfiltered.json", "### Data Instances", "### Data Fields\n\n\n\nLicense\n-------\n\n\nWe provide the following data sets under a <a href=\"URL Commons Attribution-ShareAlike 3.0 Unported License. It is based on content extracted from Wikipedia that is licensed under the Creative Commons Attribution-ShareAlike 3.0 Unported License\n\n\nContact\n-------\n\n\nIf you have any questions please create a Github issue at URL" ]

[ "region:us" ]

[]

[]

[]

[ "TAGS\n#region-us \n" ]

[ "task_categories:token-classification", "language:fr", "region:us" ]

[]

[ "fr" ]

[ "# Dataset Card for \"wikiner_fr\"\n\n\nDataset Description:\n\n- Homepage: URL\n- Repository: \n- Paper: URL\n- Leaderboard:\n- Point of Contact:" ]

[ "TAGS\n#task_categories-token-classification #language-French #region-us \n", "# Dataset Card for \"wikiner_fr\"\n\n\nDataset Description:\n\n- Homepage: URL\n- Repository: \n- Paper: URL\n- Leaderboard:\n- Point of Contact:" ]

[ "task_categories:text-classification", "region:us" ]

[]

[]

[ "### Languages\n\n\nThe BCP-47 code for the dataset's language is unk.\n\n\nDataset Structure\n-----------------", "### Data Instances\n\n\nA sample from this dataset looks as follows:", "### Dataset Fields\n\n\nThe dataset has the following fields (also called \"features\"):", "### Dataset Splits\n\n\nThis dataset is split into a train and validation split. The split sizes are as follow:" ]

[ "TAGS\n#task_categories-text-classification #region-us \n", "### Languages\n\n\nThe BCP-47 code for the dataset's language is unk.\n\n\nDataset Structure\n-----------------", "### Data Instances\n\n\nA sample from this dataset looks as follows:", "### Dataset Fields\n\n\nThe dataset has the following fields (also called \"features\"):", "### Dataset Splits\n\n\nThis dataset is split into a train and validation split. The split sizes are as follow:" ]

[ "task_categories:text-classification", "task_ids:natural-language-inference", "task_ids:semantic-similarity-classification", "task_ids:sentiment-classification", "task_ids:text-scoring", "annotations_creators:other", "language_creators:other", "multilinguality:translation", "size_categories:unknown", "source_datasets:extended|glue", "source_datasets:extended|super_glue", "language:sv", "language:da", "language:nb", "license:cc-by-4.0", "qa-nli", "paraphrase-identification", "region:us" ]

[]

[ "sv", "da", "nb" ]

[ "# Dataset Card for OverLim", "## Dataset Description\n\n- Homepage:\n- Repository:\n- Paper:\n- Leaderboard:\n- Point of Contact:", "### Dataset Summary\n\nThe _OverLim_ dataset contains some of the GLUE and SuperGLUE tasks automatically\ntranslated to Swedish, Danish, and Norwegian (bokmål), using the OpusMT models\nfor MarianMT.\n\nThe translation quality was not manually checked and may thus be faulty.\nResults on these datasets should thus be interpreted carefully.\n\nIf you want to have an easy script to train and evaluate your models have a look here", "### Supported Tasks and Leaderboards\n\nThe data contains the following tasks from GLUE and SuperGLUE:\n\n- GLUE\n - 'mnli'\n - 'mrpc'\n - 'qnli'\n - 'qqp'\n - 'rte'\n - 'sst'\n - 'stsb'\n - 'wnli'\n- SuperGLUE\n - 'boolq'\n - 'cb'\n - 'copa'\n - 'rte'", "### Languages\n\n- Swedish\n- Danish\n- Norwegian (bokmål)", "## Dataset Structure", "### Data Instances\n\nEvery task has their own set of features, but all share an 'idx' and 'label'.\n\n- GLUE\n - 'mnli'\n - 'premise', 'hypothesis'\n - 'mrpc'\n - 'text_a', 'text_b'\n - 'qnli'\n - 'premise', 'hypothesis'\n - 'qqp'\n - 'text_a', 'text_b'\n - 'sst'\n - 'text'\n - 'stsb'\n - 'text_a', 'text_b'\n - 'wnli'\n - 'premise', 'hypothesis'\n- SuperGLUE\n - 'boolq'\n - 'question', 'passage'\n - 'cb'\n - 'premise', 'hypothesis'\n - 'copa'\n - 'premise', 'choice1', 'choice2', 'question'\n - 'rte'\n - 'premise', 'hypothesis'", "### Data Splits\n\nIn order to have test-split, we repurpose the original validation-split as\ntest-split, and split the training-split into a new training- and\nvalidation-split, with an 80-20 distribution.", "## Dataset Creation\n\nFor more information about the individual tasks see (URL) and (URL).", "### Curation Rationale\n\nTraining non-English models is easy, but there is a lack of evaluation datasets to compare their actual performance.", "### Source Data", "#### Initial Data Collection and Normalization", "#### Who are the source language producers?", "### Annotations", "#### Annotation process", "#### Who are the annotators?", "### Personal and Sensitive Information", "## Considerations for Using the Data", "### Social Impact of Dataset", "### Discussion of Biases", "### Other Known Limitations", "## Additional Information", "### Dataset Curators", "### Licensing Information", "### Contributions\n\nThanks to @kb-labb for adding this dataset." ]

[ "TAGS\n#task_categories-text-classification #task_ids-natural-language-inference #task_ids-semantic-similarity-classification #task_ids-sentiment-classification #task_ids-text-scoring #annotations_creators-other #language_creators-other #multilinguality-translation #size_categories-unknown #source_datasets-extended|glue #source_datasets-extended|super_glue #language-Swedish #language-Danish #language-Norwegian Bokmål #license-cc-by-4.0 #qa-nli #paraphrase-identification #region-us \n", "# Dataset Card for OverLim", "## Dataset Description\n\n- Homepage:\n- Repository:\n- Paper:\n- Leaderboard:\n- Point of Contact:", "### Dataset Summary\n\nThe _OverLim_ dataset contains some of the GLUE and SuperGLUE tasks automatically\ntranslated to Swedish, Danish, and Norwegian (bokmål), using the OpusMT models\nfor MarianMT.\n\nThe translation quality was not manually checked and may thus be faulty.\nResults on these datasets should thus be interpreted carefully.\n\nIf you want to have an easy script to train and evaluate your models have a look here", "### Supported Tasks and Leaderboards\n\nThe data contains the following tasks from GLUE and SuperGLUE:\n\n- GLUE\n - 'mnli'\n - 'mrpc'\n - 'qnli'\n - 'qqp'\n - 'rte'\n - 'sst'\n - 'stsb'\n - 'wnli'\n- SuperGLUE\n - 'boolq'\n - 'cb'\n - 'copa'\n - 'rte'", "### Languages\n\n- Swedish\n- Danish\n- Norwegian (bokmål)", "## Dataset Structure", "### Data Instances\n\nEvery task has their own set of features, but all share an 'idx' and 'label'.\n\n- GLUE\n - 'mnli'\n - 'premise', 'hypothesis'\n - 'mrpc'\n - 'text_a', 'text_b'\n - 'qnli'\n - 'premise', 'hypothesis'\n - 'qqp'\n - 'text_a', 'text_b'\n - 'sst'\n - 'text'\n - 'stsb'\n - 'text_a', 'text_b'\n - 'wnli'\n - 'premise', 'hypothesis'\n- SuperGLUE\n - 'boolq'\n - 'question', 'passage'\n - 'cb'\n - 'premise', 'hypothesis'\n - 'copa'\n - 'premise', 'choice1', 'choice2', 'question'\n - 'rte'\n - 'premise', 'hypothesis'", "### Data Splits\n\nIn order to have test-split, we repurpose the original validation-split as\ntest-split, and split the training-split into a new training- and\nvalidation-split, with an 80-20 distribution.", "## Dataset Creation\n\nFor more information about the individual tasks see (URL) and (URL).", "### Curation Rationale\n\nTraining non-English models is easy, but there is a lack of evaluation datasets to compare their actual performance.", "### Source Data", "#### Initial Data Collection and Normalization", "#### Who are the source language producers?", "### Annotations", "#### Annotation process", "#### Who are the annotators?", "### Personal and Sensitive Information", "## Considerations for Using the Data", "### Social Impact of Dataset", "### Discussion of Biases", "### Other Known Limitations", "## Additional Information", "### Dataset Curators", "### Licensing Information", "### Contributions\n\nThanks to @kb-labb for adding this dataset." ]

[ "task_categories:other", "task_ids:named-entity-recognition", "task_ids:part-of-speech", "annotations_creators:expert-generated", "language_creators:other", "multilinguality:monolingual", "size_categories:unknown", "source_datasets:original", "language:sv", "license:cc-by-4.0", "structure-prediction", "region:us" ]

[]

[ "sv" ]

[ "### Dataset Summary\n\n\nThe dataset is a conversion of the venerable SUC 3.0 dataset into the\nhuggingface ecosystem.\nThe original dataset does not contain an official train-dev-test split, which is\nintroduced here; the tag distribution for the NER tags between the three splits\nis mostly the same.\n\n\nThe dataset has three different types of tagsets: manually annotated POS,\nmanually annotated NER, and automatically annotated NER.\nFor the automatically annotated NER tags, only sentences were chosen, where the\nautomatic and manual annotations would match (with their respective categories).\n\n\nAdditionally we provide remixes of the same data with some or all sentences\nbeing lowercased.", "### Supported Tasks and Leaderboards\n\n\n* Part-of-Speech tagging\n* Named-Entity-Recognition", "### Languages\n\n\nSwedish\n\n\nDataset Structure\n-----------------", "### Data Remixes\n\n\n* 'original\\_tags' contain the manual NER annotations\n\t+ 'lower' the whole dataset uncased\n\t+ 'lower\\_mix' some of the dataset uncased\n\t+ 'lower\\_both' every instance both cased and uncased\n* 'simple\\_tags' contain the automatic NER annotations\n\t+ 'lower' the whole dataset uncased\n\t+ 'lower\\_mix' some of the dataset uncased\n\t+ 'lower\\_both' every instance both cased and uncased", "### Data Instances\n\n\nFor each instance, there is an 'id', with an optional '\\_lower' suffix to mark\nthat it has been modified, a 'tokens' list of strings containing tokens, a\n'pos\\_tags' list of strings containing POS-tags, and a 'ner\\_tags' list of strings\ncontaining NER-tags.", "### Data Fields\n\n\n* 'id': a string containing the sentence-id\n* 'tokens': a list of strings containing the sentence's tokens\n* 'pos\\_tags': a list of strings containing the tokens' POS annotations\n* 'ner\\_tags': a list of strings containing the tokens' NER annotations", "### Data Splits\n\n\nDataset Split: train, Size Percentage of Total Dataset Size: 64%, Number of Instances for the Original Tags: 46,026\nDataset Split: dev, Size Percentage of Total Dataset Size: 16%, Number of Instances for the Original Tags: 11,506\nDataset Split: test, Size Percentage of Total Dataset Size: 20%, Number of Instances for the Original Tags: 14,383\n\n\nThe 'simple\\_tags' remix has fewer instances due to the requirement to match\ntags.\n\n\nDataset Creation\n----------------\n\n\nSee the original webpage\n\n\nAdditional Information\n----------------------", "### Dataset Curators\n\n\nSpråkbanken", "### Licensing Information\n\n\nCC BY 4.0 (attribution)\n\n\nSUC 2.0 manual", "### Contributions\n\n\nThanks to @robinqrtz for adding this dataset." ]

[ "TAGS\n#task_categories-other #task_ids-named-entity-recognition #task_ids-part-of-speech #annotations_creators-expert-generated #language_creators-other #multilinguality-monolingual #size_categories-unknown #source_datasets-original #language-Swedish #license-cc-by-4.0 #structure-prediction #region-us \n", "### Dataset Summary\n\n\nThe dataset is a conversion of the venerable SUC 3.0 dataset into the\nhuggingface ecosystem.\nThe original dataset does not contain an official train-dev-test split, which is\nintroduced here; the tag distribution for the NER tags between the three splits\nis mostly the same.\n\n\nThe dataset has three different types of tagsets: manually annotated POS,\nmanually annotated NER, and automatically annotated NER.\nFor the automatically annotated NER tags, only sentences were chosen, where the\nautomatic and manual annotations would match (with their respective categories).\n\n\nAdditionally we provide remixes of the same data with some or all sentences\nbeing lowercased.", "### Supported Tasks and Leaderboards\n\n\n* Part-of-Speech tagging\n* Named-Entity-Recognition", "### Languages\n\n\nSwedish\n\n\nDataset Structure\n-----------------", "### Data Remixes\n\n\n* 'original\\_tags' contain the manual NER annotations\n\t+ 'lower' the whole dataset uncased\n\t+ 'lower\\_mix' some of the dataset uncased\n\t+ 'lower\\_both' every instance both cased and uncased\n* 'simple\\_tags' contain the automatic NER annotations\n\t+ 'lower' the whole dataset uncased\n\t+ 'lower\\_mix' some of the dataset uncased\n\t+ 'lower\\_both' every instance both cased and uncased", "### Data Instances\n\n\nFor each instance, there is an 'id', with an optional '\\_lower' suffix to mark\nthat it has been modified, a 'tokens' list of strings containing tokens, a\n'pos\\_tags' list of strings containing POS-tags, and a 'ner\\_tags' list of strings\ncontaining NER-tags.", "### Data Fields\n\n\n* 'id': a string containing the sentence-id\n* 'tokens': a list of strings containing the sentence's tokens\n* 'pos\\_tags': a list of strings containing the tokens' POS annotations\n* 'ner\\_tags': a list of strings containing the tokens' NER annotations", "### Data Splits\n\n\nDataset Split: train, Size Percentage of Total Dataset Size: 64%, Number of Instances for the Original Tags: 46,026\nDataset Split: dev, Size Percentage of Total Dataset Size: 16%, Number of Instances for the Original Tags: 11,506\nDataset Split: test, Size Percentage of Total Dataset Size: 20%, Number of Instances for the Original Tags: 14,383\n\n\nThe 'simple\\_tags' remix has fewer instances due to the requirement to match\ntags.\n\n\nDataset Creation\n----------------\n\n\nSee the original webpage\n\n\nAdditional Information\n----------------------", "### Dataset Curators\n\n\nSpråkbanken", "### Licensing Information\n\n\nCC BY 4.0 (attribution)\n\n\nSUC 2.0 manual", "### Contributions\n\n\nThanks to @robinqrtz for adding this dataset." ]

[ "region:us" ]

[]

[]

[ "# Korean Language Understanding Evaluation (KLUE)" ]

[ "TAGS\n#region-us \n", "# Korean Language Understanding Evaluation (KLUE)" ]

[ "region:us" ]

[]

[]

[ "# KorQuAD" ]

[ "TAGS\n#region-us \n", "# KorQuAD" ]

[ "region:us" ]

[]

[]

[ "# National Institute of Korean Language(NIKL) Corpus" ]

[ "TAGS\n#region-us \n", "# National Institute of Korean Language(NIKL) Corpus" ]

[ "region:us" ]

[]

[]

[]

[ "TAGS\n#region-us \n" ]

[ "task_categories:automatic-speech-recognition", "language:sv", "license:cc0-1.0", "region:us" ]

[]

[ "sv" ]

[ "# NST Swedish ASR Database (16 kHz) – reorganized\n\nThis database was created by Nordic Language Technology for the development of automatic speech recognition and dictation in Swedish. In this updated version, the organization of the data have been altered to improve the usefulness of the database.\n\nIn the original version of the material, the files were organized in a specific folder structure where the folder names were meaningful. However, the file names were not meaningful, and there were also cases of files with identical names in different folders. This proved to be impractical, since users had to keep the original folder structure in order to use the data. The files have been renamed, such that the file names are unique and meaningful regardless of the folder structure. The original metadata files were in spl format. These have been converted to JSON format. The converted metadata files are also anonymized and the text encoding has been converted from ANSI to UTF-8.\n\nSee the documentation file for a full description of the data and the changes made to the database.\n\nThe data is originally hosted on the National Library of Norway website.\nURL\n\nHosting on Hugging Face datasets for convenience. \n\nLicence\nCC0 1.0 Universal (CC0 1.0)\nPublic Domain Dedication" ]

[ "TAGS\n#task_categories-automatic-speech-recognition #language-Swedish #license-cc0-1.0 #region-us \n", "# NST Swedish ASR Database (16 kHz) – reorganized\n\nThis database was created by Nordic Language Technology for the development of automatic speech recognition and dictation in Swedish. In this updated version, the organization of the data have been altered to improve the usefulness of the database.\n\nIn the original version of the material, the files were organized in a specific folder structure where the folder names were meaningful. However, the file names were not meaningful, and there were also cases of files with identical names in different folders. This proved to be impractical, since users had to keep the original folder structure in order to use the data. The files have been renamed, such that the file names are unique and meaningful regardless of the folder structure. The original metadata files were in spl format. These have been converted to JSON format. The converted metadata files are also anonymized and the text encoding has been converted from ANSI to UTF-8.\n\nSee the documentation file for a full description of the data and the changes made to the database.\n\nThe data is originally hosted on the National Library of Norway website.\nURL\n\nHosting on Hugging Face datasets for convenience. \n\nLicence\nCC0 1.0 Universal (CC0 1.0)\nPublic Domain Dedication" ]

[ "region:us" ]

[]

[]

[ "# Speechdat\nSpeechdat dataset", "## Loading the dataset\nYou need to download the dataset from another folder. We assume the wav files are located inside a /wav folder inside speechdat." ]

[ "TAGS\n#region-us \n", "# Speechdat\nSpeechdat dataset", "## Loading the dataset\nYou need to download the dataset from another folder. We assume the wav files are located inside a /wav folder inside speechdat." ]

[ "task_categories:automatic-speech-recognition", "language:sv", "region:us" ]

[]

[ "sv" ]

[ "### SAMPLED FILES\n\n\nThe .smp files contain the speech signal. The identity\nof the speaker is coded by the two digits after 'fp20' in the file\nname. The smp file format was developed by TMH. Recording information\nis stored in a header as a 1024 byte text string. The speech signal in\nthe Waxholm corpus is quantised into 16 bits, 2 bytes/sample and the\nbyte order is big-endian (most significant byte first). The sampling\nfrequency is 16 kHz. Here is an example of a file header:", "### LABEL FILES\n\n\nNormally, each sample file has a label file. This has been\nproduced in four steps. The first step was to manually enter the\northographic text by listening. From this text a sequence of phonemes\nwere produced by a rule-based text-to-phoneme module. The endpoint\ntime positions of the phonemes were computed by an automatic alignment\nprogram, followed by manual correction. Some of the speech files have\nno label file, due to different problems in this process. These files\nshould not be used for training or testing.\n\n\nThe labels are stored in .mix files. Below is an example of the\nbeginning of a mix file.\n\n\nThe orthographic text representation is after the label 'TEXT:' CT is\nthe frame length in number of sample points. (Always = 1 in Waxholm\nmix files) Each line starting with 'FR' contains up to three labels at\nthe phonetic, phonemic and word levels. FR is immediately followed by\nthe frame number of the start of the segment. Since CT = 1, FR is the\nsample index in the file. If a frame duration is = 0, the label has\nbeen judged as a non-pronounced segment and deleted by the manual\nlabeller, although it was generated by the text-to-phoneme or the\nautomatic alignment modules. Column 3 in an FR line is the phonetic\nlabel. Initial '#' indicates word initial position. '$' indicates\nother positions. The optional label '>pm' precedes the phonemic label,\nwhich has been generated by the text-to-phoneme rules. Often, the\nphonemic and the phonetic labels are identical. The optional '>w' is\nfollowed by the identity of the word beginning at this frame. The\nphoneme symbol inventory is mainly STA, used by the KTH/TMH RULSYS\nsystem. It is specified in the included file 'sampa\\_latex\\_se.pdf'.\n\n\nSome extra labels at the phonetic level have been defined.\nThe most common ones are:\n\n\n\nThe label 'Labels:' before the FR lines is a text string assembled\nfrom the FR labels\n\n\nThe mix files in this archive correspond to those with the name\nextension .URL in the original corpus. Besides a few other\ncorrections, the main difference is that burst segments after\nretroflex stops were not labelled as retroflex in the original .mix\nfiles ( d, t after 2D and 2T have been changed to 2d and 2t).\n\n\nREFERENCES\n----------\n\n\nBertenstam, J., Blomberg, M., Carlson, R., Elenius, K., Granström,\nB., Gustafson, J., Hunnicutt, S., Högberg, J., Lindell, R., Neovius,\nL., Nord, L., de Serpa-Leitao, A., and Ström, N.,(1995). \"Spoken\ndialogue data collected in the WAXHOLM project\" STL-QPSR 1/1995,\nKTH/TMH, Stockholm.\n\n\nBertenstam, J., Blomberg, M., Carlson, R.,\nElenius, K., Granström, B., Gustafson, J., Hunnicutt, S., Högberg, J.,\nLindell, R., Neovius, L., de Serpa-Leitao, A., Nord, L., & Ström,\nN. (1995). The Waxholm application data-base. In Pardo, J.M. (Ed.),\nProceedings Eurospeech 1995 (pp. 833-836). Madrid.\n\n\nComments and error reports are welcome. These should be sent to:\nMats Blomberg [matsb@URL](mailto:matsb@URL) or Kjell Elenius [kjell@URL](mailto:kjell@URL)" ]

[ "TAGS\n#task_categories-automatic-speech-recognition #language-Swedish #region-us \n", "### SAMPLED FILES\n\n\nThe .smp files contain the speech signal. The identity\nof the speaker is coded by the two digits after 'fp20' in the file\nname. The smp file format was developed by TMH. Recording information\nis stored in a header as a 1024 byte text string. The speech signal in\nthe Waxholm corpus is quantised into 16 bits, 2 bytes/sample and the\nbyte order is big-endian (most significant byte first). The sampling\nfrequency is 16 kHz. Here is an example of a file header:", "### LABEL FILES\n\n\nNormally, each sample file has a label file. This has been\nproduced in four steps. The first step was to manually enter the\northographic text by listening. From this text a sequence of phonemes\nwere produced by a rule-based text-to-phoneme module. The endpoint\ntime positions of the phonemes were computed by an automatic alignment\nprogram, followed by manual correction. Some of the speech files have\nno label file, due to different problems in this process. These files\nshould not be used for training or testing.\n\n\nThe labels are stored in .mix files. Below is an example of the\nbeginning of a mix file.\n\n\nThe orthographic text representation is after the label 'TEXT:' CT is\nthe frame length in number of sample points. (Always = 1 in Waxholm\nmix files) Each line starting with 'FR' contains up to three labels at\nthe phonetic, phonemic and word levels. FR is immediately followed by\nthe frame number of the start of the segment. Since CT = 1, FR is the\nsample index in the file. If a frame duration is = 0, the label has\nbeen judged as a non-pronounced segment and deleted by the manual\nlabeller, although it was generated by the text-to-phoneme or the\nautomatic alignment modules. Column 3 in an FR line is the phonetic\nlabel. Initial '#' indicates word initial position. '$' indicates\nother positions. The optional label '>pm' precedes the phonemic label,\nwhich has been generated by the text-to-phoneme rules. Often, the\nphonemic and the phonetic labels are identical. The optional '>w' is\nfollowed by the identity of the word beginning at this frame. The\nphoneme symbol inventory is mainly STA, used by the KTH/TMH RULSYS\nsystem. It is specified in the included file 'sampa\\_latex\\_se.pdf'.\n\n\nSome extra labels at the phonetic level have been defined.\nThe most common ones are:\n\n\n\nThe label 'Labels:' before the FR lines is a text string assembled\nfrom the FR labels\n\n\nThe mix files in this archive correspond to those with the name\nextension .URL in the original corpus. Besides a few other\ncorrections, the main difference is that burst segments after\nretroflex stops were not labelled as retroflex in the original .mix\nfiles ( d, t after 2D and 2T have been changed to 2d and 2t).\n\n\nREFERENCES\n----------\n\n\nBertenstam, J., Blomberg, M., Carlson, R., Elenius, K., Granström,\nB., Gustafson, J., Hunnicutt, S., Högberg, J., Lindell, R., Neovius,\nL., Nord, L., de Serpa-Leitao, A., and Ström, N.,(1995). \"Spoken\ndialogue data collected in the WAXHOLM project\" STL-QPSR 1/1995,\nKTH/TMH, Stockholm.\n\n\nBertenstam, J., Blomberg, M., Carlson, R.,\nElenius, K., Granström, B., Gustafson, J., Hunnicutt, S., Högberg, J.,\nLindell, R., Neovius, L., de Serpa-Leitao, A., Nord, L., & Ström,\nN. (1995). The Waxholm application data-base. In Pardo, J.M. (Ed.),\nProceedings Eurospeech 1995 (pp. 833-836). Madrid.\n\n\nComments and error reports are welcome. These should be sent to:\nMats Blomberg [matsb@URL](mailto:matsb@URL) or Kjell Elenius [kjell@URL](mailto:kjell@URL)" ]

[ "multilinguality:monolingual", "language:hy", "arxiv:2009.12615", "region:us" ]

[ "2009.12615" ]

[ "hy" ]

[ "### Dataset Summary\n\n\nThe sentences in the dataset are taken from Hetq and Panarmenian news articles. To generate paraphrase for the sentences, we used back translation from Armenian to English. We repeated the step twice, after which the generated paraphrases were manually reviewed. Invalid sentences were filtered out, while the rest were labelled as either paraphrase, near paraphrase or non-paraphrase. Test examples were reviewed by 3 different annotators. In addition, to increase the number of non-paraphrase pairs, we padded the dataset with automatically generated negative examples, including pairs of consecutive sentences and random pairs.\n\n\nDataset Structure\n-----------------\n\n\nEach row consists of 2 sentences and their label. This sentences were labelled as either paraphrase, near paraphrase or non-paraphrase (with 1, 0, -1 labels respectively). The sentences are divided into train and test sets.", "### Dataset Evaluation\n\n\nWe finetuned Multilingual BERT on several training sets, including the proposed ARPA dataset, and evaluated their performance on our test set. During training and\nevaluation, near paraphrase and non-paraphrase pairs were combined into one class. The results are provided below:", "#### Additional Information\n\n\nThe model trained on ARPA is available for use, and can be downloaded using this link.\n\n\nFor more details about the models and dataset construction, refer to the paper." ]

[ "TAGS\n#multilinguality-monolingual #language-Armenian #arxiv-2009.12615 #region-us \n", "### Dataset Summary\n\n\nThe sentences in the dataset are taken from Hetq and Panarmenian news articles. To generate paraphrase for the sentences, we used back translation from Armenian to English. We repeated the step twice, after which the generated paraphrases were manually reviewed. Invalid sentences were filtered out, while the rest were labelled as either paraphrase, near paraphrase or non-paraphrase. Test examples were reviewed by 3 different annotators. In addition, to increase the number of non-paraphrase pairs, we padded the dataset with automatically generated negative examples, including pairs of consecutive sentences and random pairs.\n\n\nDataset Structure\n-----------------\n\n\nEach row consists of 2 sentences and their label. This sentences were labelled as either paraphrase, near paraphrase or non-paraphrase (with 1, 0, -1 labels respectively). The sentences are divided into train and test sets.", "### Dataset Evaluation\n\n\nWe finetuned Multilingual BERT on several training sets, including the proposed ARPA dataset, and evaluated their performance on our test set. During training and\nevaluation, near paraphrase and non-paraphrase pairs were combined into one class. The results are provided below:", "#### Additional Information\n\n\nThe model trained on ARPA is available for use, and can be downloaded using this link.\n\n\nFor more details about the models and dataset construction, refer to the paper." ]

[ "task_categories:text-classification", "multilinguality:monolingual", "language:hy", "license:apache-2.0", "region:us" ]

[]

[ "hy" ]

[ "## Table of Contents\n- Table of Contents\n- News Texts Dataset", "## News Texts Dataset\n\nWe release a dataset of over 12000 news articles from URL, categorized into 7 classes: sport, politics, weather, economy, accidents, art, society. The articles are split into train (2242k tokens) and test sets (425k tokens).\n\nFor more details, refer to the paper." ]

[ "TAGS\n#task_categories-text-classification #multilinguality-monolingual #language-Armenian #license-apache-2.0 #region-us \n", "## Table of Contents\n- Table of Contents\n- News Texts Dataset", "## News Texts Dataset\n\nWe release a dataset of over 12000 news articles from URL, categorized into 7 classes: sport, politics, weather, economy, accidents, art, society. The articles are split into train (2242k tokens) and test sets (425k tokens).\n\nFor more details, refer to the paper." ]

[ "task_ids:named-entity-recognition", "multilinguality:monolingual", "language:hy", "license:apache-2.0", "region:us" ]

[]

[ "hy" ]

[]

[ "TAGS\n#task_ids-named-entity-recognition #multilinguality-monolingual #language-Armenian #license-apache-2.0 #region-us \n" ]

[ "region:us" ]

[]

[]

[]

[ "TAGS\n#region-us \n" ]

[ "task_categories:text-classification", "task_ids:sentiment-classification", "annotations_creators:expert-generated", "language_creators:expert-generated", "language:bn", "region:us" ]

[]

[ "bn" ]

[ "# Dataset Card for \"SentNoB\"", "### Dataset Summary\n\nSocial Media User Comments' Sentiment Analysis Dataset. Each user comments are labeled with either positive (1), negative (2), or neutral (0)." ]

[ "TAGS\n#task_categories-text-classification #task_ids-sentiment-classification #annotations_creators-expert-generated #language_creators-expert-generated #language-Bengali #region-us \n", "# Dataset Card for \"SentNoB\"", "### Dataset Summary\n\nSocial Media User Comments' Sentiment Analysis Dataset. Each user comments are labeled with either positive (1), negative (2), or neutral (0)." ]

[ "task_categories:object-detection", "size_categories:1K<n<10K", "source_datasets:original", "other-object-detection", "region:us" ]

[]

[]

[ "# Plastic in river\nThis dataset is an export of the annotated assets from the Kili's Community Challenge - Plastic in River dataset.\n\nThe Hugging Face dataset will be updated every day during the challenge with the latest annotations." ]

[ "TAGS\n#task_categories-object-detection #size_categories-1K<n<10K #source_datasets-original #other-object-detection #region-us \n", "# Plastic in river\nThis dataset is an export of the annotated assets from the Kili's Community Challenge - Plastic in River dataset.\n\nThe Hugging Face dataset will be updated every day during the challenge with the latest annotations." ]

[ "region:us" ]

[]

[]

[ "# Gender classification from Clinical Trial Public Data" ]

[ "TAGS\n#region-us \n", "# Gender classification from Clinical Trial Public Data" ]

[ "region:us" ]

[]

[]

[]

[ "TAGS\n#region-us \n" ]

[ "task_categories:text-generation", "annotations_creators:expert-generated", "language_creators:expert-generated", "multilinguality:monolingual", "size_categories:1K<n<10K", "source_datasets:original", "language:en", "license:mit", "region:us" ]

[]

[ "en" ]

[ "# Dataset Card for [Dataset Name]", "## Table of Contents\n- Table of Contents\n- Dataset Description\n - Dataset Summary\n - Supported Tasks and Leaderboards\n - Languages\n- Dataset Structure\n - Data Instances\n - Data Fields\n - Data Splits\n- Dataset Creation\n - Curation Rationale\n - Source Data\n - Annotations\n - Personal and Sensitive Information\n- Considerations for Using the Data\n - Social Impact of Dataset\n - Discussion of Biases\n - Other Known Limitations\n- Additional Information\n - Dataset Curators\n - Licensing Information\n - Citation Information\n - Contributions", "## Dataset Description\n\n- Homepage:\n- Repository:\n- Paper:\n- Leaderboard:\n- Point of Contact:", "### Dataset Summary", "### Supported Tasks and Leaderboards", "### Languages", "## Dataset Structure", "### Data Instances", "### Data Fields", "### Data Splits", "## Dataset Creation", "### Curation Rationale", "### Source Data", "#### Initial Data Collection and Normalization", "#### Who are the source language producers?", "### Annotations", "#### Annotation process", "#### Who are the annotators?", "### Personal and Sensitive Information", "## Considerations for Using the Data", "### Social Impact of Dataset", "### Discussion of Biases", "### Other Known Limitations", "## Additional Information", "### Dataset Curators", "### Licensing Information", "### Contributions\n\nThanks to @github-username for adding this dataset." ]

[ "TAGS\n#task_categories-text-generation #annotations_creators-expert-generated #language_creators-expert-generated #multilinguality-monolingual #size_categories-1K<n<10K #source_datasets-original #language-English #license-mit #region-us \n", "# Dataset Card for [Dataset Name]", "## Table of Contents\n- Table of Contents\n- Dataset Description\n - Dataset Summary\n - Supported Tasks and Leaderboards\n - Languages\n- Dataset Structure\n - Data Instances\n - Data Fields\n - Data Splits\n- Dataset Creation\n - Curation Rationale\n - Source Data\n - Annotations\n - Personal and Sensitive Information\n- Considerations for Using the Data\n - Social Impact of Dataset\n - Discussion of Biases\n - Other Known Limitations\n- Additional Information\n - Dataset Curators\n - Licensing Information\n - Citation Information\n - Contributions", "## Dataset Description\n\n- Homepage:\n- Repository:\n- Paper:\n- Leaderboard:\n- Point of Contact:", "### Dataset Summary", "### Supported Tasks and Leaderboards", "### Languages", "## Dataset Structure", "### Data Instances", "### Data Fields", "### Data Splits", "## Dataset Creation", "### Curation Rationale", "### Source Data", "#### Initial Data Collection and Normalization", "#### Who are the source language producers?", "### Annotations", "#### Annotation process", "#### Who are the annotators?", "### Personal and Sensitive Information", "## Considerations for Using the Data", "### Social Impact of Dataset", "### Discussion of Biases", "### Other Known Limitations", "## Additional Information", "### Dataset Curators", "### Licensing Information", "### Contributions\n\nThanks to @github-username for adding this dataset." ]

[ "region:us" ]

[]

[]

[ "# Langame/waiting-messages\nGenerated using OpenAI GPT-3 davinci-codex based on random initial samples written by a human.\n\n️ The dataset has not been de-duplicated, so there may be duplicates. ️" ]

[ "TAGS\n#region-us \n", "# Langame/waiting-messages\nGenerated using OpenAI GPT-3 davinci-codex based on random initial samples written by a human.\n\n️ The dataset has not been de-duplicated, so there may be duplicates. ️" ]

[ "task_categories:summarization", "task_categories:text-generation", "annotations_creators:no-annotation", "language_creators:found", "multilinguality:monolingual", "size_categories:100K<n<1M", "source_datasets:cc-news", "language:es", "license:mit", "conditional-text-generation", "region:us" ]

[]

[ "es" ]

[ "### Dataset Summary\n\n\nCC-NEWS-ES-titles is a Spanish-language dataset for news titles generation. The text and titles comes from 2019 and 2020 CC-NEWS data (which is part of Common Crawl).\n\n\nIt contains 402.310 pairs of news title and body, splitted in :\n\n\n* Train: 370.125\n* Eval: 16.092\n* Test: 16.092", "### Supported Tasks and Leaderboards\n\n\n* 'text-classification', 'sentiment-classification': The dataset can be used to train a model for news title generation which can be considered a subset of abstractive summarization.", "### Languages\n\n\nThe text is in Spanish. The BCP-47 code for Spanish is es.\n\n\nDataset Structure\n-----------------", "### Data Instances\n\n\nEach data instance contains the following features: *text* and *output\\_text*.\n\n\n* *text* is the body of the news.\n* *output\\_text* is the title of the news.\n\n\nAn example from the CC-NEWS-ES-titles train set looks like the following:", "### Data Fields\n\n\n* 'text': a string containing the body of the news.\n* 'output\\_text': a string containing the title of the news.", "### Data Splits\n\n\nThe CC-NEWS-ES-titles dataset has 3 splits: *train*, *validation*, and *test*. The splits contain disjoint sets of news.\n\n\n\nDataset Creation\n----------------", "### Curation Rationale\n\n\n[N/A]", "### Source Data", "#### Initial Data Collection and Normalization\n\n\nTODO", "#### Who are the source language producers?\n\n\nCommon Crawl: URL", "### Annotations\n\n\nThe dataset does not contain any additional annotations.", "#### Annotation process\n\n\n[N/A]", "#### Who are the annotators?\n\n\n[N/A]", "### Personal and Sensitive Information\n\n\n[N/A]\n\n\nConsiderations for Using the Data\n---------------------------------", "### Social Impact of Dataset\n\n\nAbstractive summarization is a complex task and Spanish is a underrepresented language in the NLP domain. As a consequence, adding a Spanish resource may help others to improve their research and educational activities.", "### Discussion of Biases\n\n\n[N/A]", "### Other Known Limitations\n\n\n[N/A]\n\n\nAdditional Information\n----------------------", "### Dataset Curators\n\n\nThis dataset is maintained by Leonardo Ignacio Córdoba and was built with the help of María Gaska.", "### Licensing Information\n\n\n[N/A]\n\n\nTODO", "### Contributions\n\n\n[N/A]" ]

[ "TAGS\n#task_categories-summarization #task_categories-text-generation #annotations_creators-no-annotation #language_creators-found #multilinguality-monolingual #size_categories-100K<n<1M #source_datasets-cc-news #language-Spanish #license-mit #conditional-text-generation #region-us \n", "### Dataset Summary\n\n\nCC-NEWS-ES-titles is a Spanish-language dataset for news titles generation. The text and titles comes from 2019 and 2020 CC-NEWS data (which is part of Common Crawl).\n\n\nIt contains 402.310 pairs of news title and body, splitted in :\n\n\n* Train: 370.125\n* Eval: 16.092\n* Test: 16.092", "### Supported Tasks and Leaderboards\n\n\n* 'text-classification', 'sentiment-classification': The dataset can be used to train a model for news title generation which can be considered a subset of abstractive summarization.", "### Languages\n\n\nThe text is in Spanish. The BCP-47 code for Spanish is es.\n\n\nDataset Structure\n-----------------", "### Data Instances\n\n\nEach data instance contains the following features: *text* and *output\\_text*.\n\n\n* *text* is the body of the news.\n* *output\\_text* is the title of the news.\n\n\nAn example from the CC-NEWS-ES-titles train set looks like the following:", "### Data Fields\n\n\n* 'text': a string containing the body of the news.\n* 'output\\_text': a string containing the title of the news.", "### Data Splits\n\n\nThe CC-NEWS-ES-titles dataset has 3 splits: *train*, *validation*, and *test*. The splits contain disjoint sets of news.\n\n\n\nDataset Creation\n----------------", "### Curation Rationale\n\n\n[N/A]", "### Source Data", "#### Initial Data Collection and Normalization\n\n\nTODO", "#### Who are the source language producers?\n\n\nCommon Crawl: URL", "### Annotations\n\n\nThe dataset does not contain any additional annotations.", "#### Annotation process\n\n\n[N/A]", "#### Who are the annotators?\n\n\n[N/A]", "### Personal and Sensitive Information\n\n\n[N/A]\n\n\nConsiderations for Using the Data\n---------------------------------", "### Social Impact of Dataset\n\n\nAbstractive summarization is a complex task and Spanish is a underrepresented language in the NLP domain. As a consequence, adding a Spanish resource may help others to improve their research and educational activities.", "### Discussion of Biases\n\n\n[N/A]", "### Other Known Limitations\n\n\n[N/A]\n\n\nAdditional Information\n----------------------", "### Dataset Curators\n\n\nThis dataset is maintained by Leonardo Ignacio Córdoba and was built with the help of María Gaska.", "### Licensing Information\n\n\n[N/A]\n\n\nTODO", "### Contributions\n\n\n[N/A]" ]

[ "task_categories:summarization", "task_categories:text-generation", "annotations_creators:no-annotation", "language_creators:found", "multilinguality:monolingual", "size_categories:n<1K", "size_categories:1K<n<10K", "size_categories:10K<n<100K", "size_categories:100K<n<1M", "size_categories:1M<n<10M", "source_datasets:cc-news", "language:es", "license:mit", "conditional-text-generation", "region:us" ]

[]

[ "es" ]

[ "### Dataset Summary\n\n\nCC-NEWS-ES is a Spanish-language dataset of news. The corpus was generated by extracting the Spanish articles from CC-NEWS (news index of Common Crawl) of 2019. For doing that FastText model was used for language prediction.\n\n\nIt contains a total of 7,473,286 texts and 1,812,009,283 words distributed as follows:", "### Supported Tasks and Leaderboards\n\n\nTODO\n----", "### Languages\n\n\nThe text is in Spanish. The BCP-47 code for Spanish is es.\n\n\nDataset Structure\n-----------------", "### Data Instances\n\n\nEach data instance contains the following features: ...\n\n\n* country: top level domain, usually refers to a country (except in the case of .com).\n* text: body of the news\n* id: internal id\n\n\nAn example from CC-NEWS-ES looks like the following:", "### Data Fields\n\n\n* ...\n* ...", "### Data Splits\n\n\n...\n\n\nDataset Creation\n----------------", "### Curation Rationale\n\n\n[N/A]", "### Source Data", "#### Initial Data Collection and Normalization\n\n\nTODO", "#### Who are the source language producers?\n\n\nCommon Crawl: URL", "### Annotations\n\n\nThe dataset does not contain any additional annotations.", "#### Annotation process\n\n\n[N/A]", "#### Who are the annotators?\n\n\n[N/A]", "### Personal and Sensitive Information\n\n\n[N/A]\n\n\nConsiderations for Using the Data\n---------------------------------", "### Social Impact of Dataset\n\n\n...", "### Discussion of Biases\n\n\n[N/A]", "### Other Known Limitations\n\n\n[N/A]\n\n\nAdditional Information\n----------------------", "### Dataset Curators\n\n\nThis dataset is maintained by Leonardo Ignacio Córdoba and was built with the help of María Gaska.", "### Licensing Information\n\n\n[N/A]\n\n\nTODO", "### Contributions\n\n\n[N/A]" ]

[ "TAGS\n#task_categories-summarization #task_categories-text-generation #annotations_creators-no-annotation #language_creators-found #multilinguality-monolingual #size_categories-n<1K #size_categories-1K<n<10K #size_categories-10K<n<100K #size_categories-100K<n<1M #size_categories-1M<n<10M #source_datasets-cc-news #language-Spanish #license-mit #conditional-text-generation #region-us \n", "### Dataset Summary\n\n\nCC-NEWS-ES is a Spanish-language dataset of news. The corpus was generated by extracting the Spanish articles from CC-NEWS (news index of Common Crawl) of 2019. For doing that FastText model was used for language prediction.\n\n\nIt contains a total of 7,473,286 texts and 1,812,009,283 words distributed as follows:", "### Supported Tasks and Leaderboards\n\n\nTODO\n----", "### Languages\n\n\nThe text is in Spanish. The BCP-47 code for Spanish is es.\n\n\nDataset Structure\n-----------------", "### Data Instances\n\n\nEach data instance contains the following features: ...\n\n\n* country: top level domain, usually refers to a country (except in the case of .com).\n* text: body of the news\n* id: internal id\n\n\nAn example from CC-NEWS-ES looks like the following:", "### Data Fields\n\n\n* ...\n* ...", "### Data Splits\n\n\n...\n\n\nDataset Creation\n----------------", "### Curation Rationale\n\n\n[N/A]", "### Source Data", "#### Initial Data Collection and Normalization\n\n\nTODO", "#### Who are the source language producers?\n\n\nCommon Crawl: URL", "### Annotations\n\n\nThe dataset does not contain any additional annotations.", "#### Annotation process\n\n\n[N/A]", "#### Who are the annotators?\n\n\n[N/A]", "### Personal and Sensitive Information\n\n\n[N/A]\n\n\nConsiderations for Using the Data\n---------------------------------", "### Social Impact of Dataset\n\n\n...", "### Discussion of Biases\n\n\n[N/A]", "### Other Known Limitations\n\n\n[N/A]\n\n\nAdditional Information\n----------------------", "### Dataset Curators\n\n\nThis dataset is maintained by Leonardo Ignacio Córdoba and was built with the help of María Gaska.", "### Licensing Information\n\n\n[N/A]\n\n\nTODO", "### Contributions\n\n\n[N/A]" ]

[ "region:us" ]

[]

[]

[]

[ "TAGS\n#region-us \n" ]