id
stringlengths 7
12
| sentence1
stringlengths 6
1.27k
| sentence2
stringlengths 6
926
| label
stringclasses 4
values |
|---|---|---|---|
train_16900 | (2015) propose a sparse coding method to generate summaries that not only cover key content in news but also focuses highlighted by readers' comments. | they do not consider semantic information. | contrasting |
train_16901 | But they do not tackle sequential context information among sentences and treat them as separate instances. | we deal with sequential context information within each document and the relationships among documents. | contrasting |
train_16902 | In fact, one can also try to use a CNN to get t d . | our experiments suggest RNN performs better than CNN. | contrasting |
train_16903 | Cao and Clark (2019) factor the generation process leveraging syntactic information to improve the performance. | they linearize both AMR and constituency graphs, which implies that important parts of the graphs cannot well be represented (e.g., coreference). | contrasting |
train_16904 | On the same dataset, we have competitive results to Damonte and Cohen (2019). | we do not rely on preprocessing anonymisation not to lose semantic signals. | contrasting |
train_16905 | By penalizing such inconsistencies, the model enables the generation of more consistent outputs. | perfect consistency between attention weights occasionally disturbs the model to generate proper outputs. | contrasting |
train_16906 | It would, however, take the expertise of at least a teacher of English and also a computer engineer to achieve it; the former would have to think of what is typical in the given topic and then to make a set of rules; the latter then would have to turn them into computer-readable forms. | the neural retrieval-based method only requires a teacher of English to annotate a given corpus with feedback comments, without examining what is typical, which is much more effective and efficient. | contrasting |
train_16907 | On the one hand, it would require a successful parsing to recognize the sources of the errors. | it would require successful error detection to parse them correctly. | contrasting |
train_16908 | Compared with rule-based approaches, neural models (Yuan et al., 2017) can generate more fluent and grammatical questions. | question generation is a one-to-many sequence generation problem, i.e., several aspects can be asked given a sentence, which confuses the model during train and prevents concrete automatic evaluation. | contrasting |
train_16909 | From this correlation analysis against human judgments, we observe that, as expected, the Language Model metric captures readability better than ROUGE, while falling short on relevance. | the results obtained using the proposed QA-based metrics indicate their potential benefits especially under the unsupervised setting, with QA conf and QA f score capturing readability and relevance better than all the others reported metrics, including ROUGE. | contrasting |
train_16910 | To elaborate further, we notice that applying the learned coefficients for 1 to the results obtained by models reinforced on QA learned and QA equally , see Table 2, we obtain very similar scores (namely, 136.43 for QA equally and 136.4 for QA learned ). | the qualitative analysis reported in Tables 3 and 4 shows that while they perform sim-ilarly in terms of relevance, a significantly lower score for readability is obtained using QA equally . | contrasting |
train_16911 | Nema et al., 2018) using well-designed data encoder and attention mechanisms. | as demonstrated in Wiseman et al. | contrasting |
train_16912 | Improving the parser and deriving a more semantically-aware set of compression rules can help achieving better grammaticality and readability. | we note that such errors are largely orthogonal to the core of our approach; a more refined set of compression options could be dropped into our system and used without changing our fundamental model. | contrasting |
train_16913 | Without the manual deduplication mechanism, our model matches the ground truth around 80% of the time. | a low accuracy here may not actually cause a low final ROUGE score, as many compression choices only affect the final ROUGE score by a small amount. | contrasting |
train_16914 | A natural solution to the data-scarcity issue is to resort to massive data from other domains. | directly leveraging abundant data from other domains is problematic due to the discrepancies in data distribution on different domains. | contrasting |
train_16915 | This enables the model to learn generic style information from both domains. | explicitly learning precise stylized information within each domain is crucial to generate domain-specific styles. | contrasting |
train_16916 | Existing approaches focus on encoding the passage, the answer and the relationship between them using complex functions and then generate the question in one single pass. | by carefully analysing the generated questions, we observe that these approaches tend to miss one or more of the important aspects of the question. | contrasting |
train_16917 | Over 68.6%, 66.7% and 64.2% of the generated questions from RefNet were respectively more fluent, complete and answerable when compared to the EAD model. | there are some cases where EAD does better than RefNet. | contrasting |
train_16918 | It guarantees the maximum volume size of selected points with minimum number of points (Figure 2 (c)). | it does not reduce a redundancy between the points over the convex-hull, and usually choose larger number of sentences than k. Marcu (1999) shows an interesting study regarding an importance of sentences: given a document, if one deletes the least central sentence from the source text, then at some point the similarity with the reference text rapidly drops at sudden called the waterfall phenomena. | contrasting |
train_16919 | In most evaluations, ROUGE scores are linear to SO ratios as expected. | vO has high variance across algorithms and aspects. | contrasting |
train_16920 | One thing to note is that XSum and AMI have less new words in their target summaries. | paper datasets (i.e., PeerRead and PubMed) include a lot, indicating that abstract text in academic paper is indeed "abstract". | contrasting |
train_16921 | LexRank is highly biased toward the position aspect. | mmR is extremely biased to the importance aspect on XSum and Reddit. | contrasting |
train_16922 | (2018) investigate how to evaluate semi-supervised training algorithms in a realistic way; they differ from us in that they focus exclusively on semi-supervised learning (SSL) algorithms, and do not consider NLP explicitly. | in line with our conclusion, they report that recent practices for evaluating SSL techniques do not address the question of the algorithms' real-word applicability in a satisfying way. | contrasting |
train_16923 | As we will see in later sections, this is one of the main sources of search errors. | in many cases, the model score found by beam search is a reasonable approximation to the global best model score. | contrasting |
train_16924 | ELMo and BERT improve naive baselines by a large margin, indicating that a notable amount of commonsense knowledge has been acquired via pre-training. | even BERT still falls far behind human performance, indicating the need of further research. | contrasting |
train_16925 | Standard accuracy metrics indicate that modern reading comprehension systems have achieved strong performance in many question answering datasets. | the extent these systems truly understand language remains unknown, and existing systems are not good at distinguishing distractor sentences, which look related but do not actually answer the question. | contrasting |
train_16926 | Question answering tasks are widely used for training and testing machine comprehension and reasoning (Rajpurkar et al., 2016;Joshi et al., 2017). | high performance in standard automatic metrics has been achieved with only superficial understanding, as models exploit simple correlations in the data that happen to be predictive on most test examples. | contrasting |
train_16927 | Intuitively, the model is expected to choose the answer span after fully considering the entire question and paragraph. | traditional QA models suffered the overstability problem, and tended to be fooled by distractor answers, such as the one containing an unrelated human name. | contrasting |
train_16928 | and then search the target entity America from KGs as the answer. | as many KGs are constructed automatically and face serious incompleteness problems (Bordes et al., 2013), it is often hard to directly get target entities for queries. | contrasting |
train_16929 | These demonstrate that LM-based methods perform very well on the associative sentences, as expected. | their performance drops significantly on the non-associative subset, when information related to the candidates themselves does not give away the answer. | contrasting |
train_16930 | It shows that Pun-GAN can generate more vivid pun sentences compared with the previous best model CLM+JD. | there still exists a big gap between generated puns and human-written puns. | contrasting |
train_16931 | Sentences that describe varying levels of respect for a demographic tend to contain more adjectives that are strongly indicative of the overall sentiment. | sentences describing occupations are usually more neutrally worded, though some occupations are socially perceived to be more positive or negative than others. | contrasting |
train_16932 | expresses conflict sentiment towards ambience aspect. | most of existing studies ignore conflict opinions, for the reason that they are sparse in the datasets (Tang et al., 2016b;He et al., 2018). | contrasting |
train_16933 | Recently, researches have explored the graph neural network (GNN) techniques on text classification, since GNN does well in handling complex structures and preserving global information. | previous methods based on GNN are mainly faced with the practical problems of fixed corpus level graph structure which do not support online testing and high memory consumption. | contrasting |
train_16934 | cific points in the sequence when computing its output. | in this case, the attention attends the wrong context, as there are many words have no correlation or do not correspond to actual words. | contrasting |
train_16935 | Such approaches may achieve high-quality extraction and labeling. | they rely on extracted PDF source markup (not always available, e.g. | contrasting |
train_16936 | Aletras and Stevenson (2013) devised a new method by mapping the topic words into a semantic space and then computing the pairwise distributional similarity (DS) of words in that space. | the semantic space is still built on PMI or NPMI. | contrasting |
train_16937 | In recent years, Variational Autoencoder (VAE) has been proved more effective and efficient to approximating deep, complex and underestimated variance in integrals (Kingma and Welling, 2013;He et al., 2017). | the VAE-based topic models focus on the construction of deep neural networks to approximate the § The two authors contributed equally to this work. | contrasting |
train_16938 | When performing cross-language information retrieval (CLIR) for lower-resourced languages, a common approach is to retrieve over the output of machine translation (MT). | there is no established guidance on how to optimize the resulting MT-IR system. | contrasting |
train_16939 | The BM25 model was evaluated against both the Europarl and Wikipedia collections. | to avoid the performance degradation caused by crosscollection evaluation (Cohen et al., 2018), we only evaluate the Wikipedia-trained neural model on the Wikipedia evaluation collection. | contrasting |
train_16940 | We conclude that none of the n-gram precision components of BLEU or variations on it provide consistently better correlations with IR performance. | given a specific collection and model, it is likely one of the alternatives we explored here or other metrics (e.g. | contrasting |
train_16941 | (2017) built on this work, exploring several strategies for rotating embeddings to obtain more semantically meaningful dimensions. | to our knowledge, orthogonal transformations themselves have not been used to represent word relationships; our work is novel in this respect. | contrasting |
train_16942 | (2017a) convert WSD task to a sequence labeling task, thus building a unified model for all polysemous words. | neither of them can totally beat the best word expert supervised methods. | contrasting |
train_16943 | So far, we have obtained the adjective-noun pairs and value polarity relations for nouns. | it is still unclear whether the polarity is positive or negative. | contrasting |
train_16944 | While the meanings of defining words are important in dictionary definitions, it is crucial to capture the lexical semantic relations between defined words and defining words. | thus far, the utilization of such relations has not been explored for definition modeling. | contrasting |
train_16945 | Word analogy test results show that word representations trained with pre-trained Chinese n-grams perform better than those trained without (SISG(cjhr)), supporting our claim that our approach is able to transfer relevant knowledge from the Chinese language for detecting analogical relationships. | for the word similarity test, word vectors trained without Chinese embeddings perform better, suggesting that there are some trade-offs. | contrasting |
train_16946 | Previous works addressing this challenge mainly focused on word-level aspects such as word embeddings. | in many cases, languages share common subwords, especially for closely related languages, but also for languages that are seemingly irrelevant. | contrasting |
train_16947 | One popular approach is to estimate the relation between noisy and clean, gold-standard labels and use this noise model to improve the training procedure. | most of these approaches only assume a dependency between the labels and do not take the features into account when modeling the label noise. | contrasting |
train_16948 | Increasing the number of clusters introduces smaller clusters for which it is difficult to estimate the noise matrix, due to the limited training resources. | decreasing the number of clusters can generalize too much, resulting in loss of information on the noise distribution. | contrasting |
train_16949 | GPA was recently used to jointly transform multiple languages into a shared vector space (Kementchedjhieva et al., 2018). | gPA assumes that a multi-way word correspondence is available, which is often not the case. | contrasting |
train_16950 | Note that MGPA needs a multi-way dictionary constructed from the bilingual dictionaries. | mPPA uses directly the raw data (the bilingual dictionaries). | contrasting |
train_16951 | Adversarial training is a popular method to ensure the transferred sentences have the desired target styles. | previous works often suffer from content leaking problem. | contrasting |
train_16952 | (2018) also made use of a conditional discriminator for multiple style transfer. | a few works including, Li et al. | contrasting |
train_16953 | The optimized network is inferred by choosing the edges with maximum weights in softmax. | dARTS is a "local" model because the softmax-based relaxation is imposed on each bundle of edges between two nodes. | contrasting |
train_16954 | Bilinear models such as DistMult and ComplEx are effective methods for knowledge graph (KG) completion. | they require large batch sizes, which becomes a performance bottleneck when training on large scale datasets due to memory constraints. | contrasting |
train_16955 | This technique is possible for standard benchmarks but not for large KGs, and we report results in Appendix D for all datasets small enough to allow for full contrastive training. | our main experiments use NLL of sampled softmax since our focus is on scalability. | contrasting |
train_16956 | As shown in Table 6, while AE achieves the best reconstruction when the noise is small (k = 1), its reconstruction deteriorates dramatically when k > 1, which suggests AE fails to learn a smooth latent space. | our method outperforms all the baselines by a large margin when k > 1. | contrasting |
train_16957 | BIOBERT is trained on PubMed abstracts and PMC full text articles, and CLIN-ICALBERT is trained on clinical text from the MIMIC-III database . | sCIBERT is trained on the full text of 1.14M biomedical and computer science papers from the semantic scholar corpus (Ammar et al., 2018). | contrasting |
train_16958 | We find that combining the sparse global gradient with the dense local gradient improves convergence. | adding local information means that nodes' parameters will diverge over time. | contrasting |
train_16959 | mMiniBERT Effectiveness The multilingual baseline mMeta-LSTM does not do well on lowresource languages. | mMiniBERT performs well and outperforms the state-of-the-art Meta-LSTM on the POS tagging task and on four out of size languages of the Morphology task. | contrasting |
train_16960 | Deep learning has achieved great success in the SLU field (Mesnil et al., 2015;Liu and Lane, 2016;Zhao et al., 2019). | it is notorious for requiring large labelled data, which limits the scalability of SLU models. | contrasting |
train_16961 | We observe no significant trend of favoring one branching direction over the other. | after training with the language modeling objective, PaLM-U shows a clear right-skewness more than it should: it produces much more right-branching structures than the gold annotation. | contrasting |
train_16962 | When we manually change the sentence pattern into "List the most common hometown of teachers", the parser gives the correct keyword. | the characterbased model is less sensitive to question sentences, which is likely because characters are less sparse compared with words. | contrasting |
train_16963 | In their work, they compare EigenSent with various sentence embedding models, including a different implementation of the Discrete Cosine Transform (DCT*). | to our implementation described in section 2.2, DCT* is applied at the word level along the word embedding dimension. | contrasting |
train_16964 | These models encode and contextualize sentences in two consecutive steps. | we propose an input representation which allows the Transformer layers in BERT to directly leverage contextualized representations of all words in all sentences, while still utilizing the pretrained weights from BERT. | contrasting |
train_16965 | (2017) is that they use the ROUGE scores to label the top (bottom) 20 sentences as positive (negative), and the rest are neutral. | we found it better to train our model to directly predict the ROUGE scores, and the loss function we used is Mean Square Error. | contrasting |
train_16966 | We observe that before finetuning, the attention patterns on [SEP] tokens and periods is almost identical between sentences. | after finetuning, the model attends to sentences differently, likely based on their different role in the sentence that requires different contextual information. | contrasting |
train_16967 | achieving 65.6% success rate between agents that never interacted with each other. | when the inter-group interaction occurred only half as frequently as the intra-group interaction, the agents from the two groups can play together with a much lower 52.4% success rate. | contrasting |
train_16968 | To the best of our knowledge, ours is the first work which builds end-to-end models on a large-scale dataset for topic-focused summarization. | generating Wikipedia to our work focusing on content selection for topic-focused summaries, there have been previous work interested in generating Wikipedia articles. | contrasting |
train_16969 | Another series of works focus on template-based methods such as (Oya et al., 2014). | template-based methods are too rigid for our patternized summary generation task. | contrasting |
train_16970 | On one hand, the sections in the prototype summary that are not highly related to the prototype document are the universal patternized words and should be emphasized when generating the new summary. | the sections in the prototype document that are highly related to the prototype summary are useful facts that can guide the process of extracting facts from input document. | contrasting |
train_16971 | With the IB objective (eq 3), there is no benefit to keeping any information from Z, which strictly makes the first term worse (more mutual information between source and summary) and does not affect the second (Z is unrelated to Y ). | 1 In IB, this is a strict statistical relationship. | contrasting |
train_16972 | Thus, these supervised models do not generalize well to other kinds of sentence summarization or domains. | our method is applicable to any domain for which examples of the inputs to be summarized are available in context. | contrasting |
train_16973 | We also tried applying more complex transitions to − → x i like diagonal mapping (Trouillon et al., 2016), but did not observe improvements. | another option is to estimate this leads to poor alignment and performance drop because y t is not explicitly grounded on x i 3 . | contrasting |
train_16974 | Neural attention models (Bahdanau et al., 2015) with the seq2seq architecture (Sutskever et al., 2014) have achieved impressive results in text summarization tasks. | the attention vector comes from a weighted sum of source information and does not model the source-target alignment in a probabilistic sense. | contrasting |
train_16975 | Pointer generators are slightly better as it is trained to directly copy keyword from the source. | once it starts to enter the generation mode ("of british" in example 2 and "has been arrested" in example 3), the generation also loses control. | contrasting |
train_16976 | For this reason, our decomposition alone may not be very beneficial if coupled with standard attention. | our structured-attention model consistently performs much better than both baselines. | contrasting |
train_16977 | This kind of re-categorization has been shown to have considerable effects on the performance Guo and Lu, 2018). | one issue is that the precise set of re-categorization rules differs among different models, making it difficult to distinguish the performance improvement from model optimization or carefully designed rules. | contrasting |
train_16978 | In contrast, our method shows a strong capacity in capturing the main idea "the solution is about some patterns and a balance". | on the ordinary Smatch met- ric, their graph obtains a higher score (68% vs. 66%), which indicates that the ordinary Smatch is not a proper metric for evaluating the quality of capturing core semantics. | contrasting |
train_16979 | 6 The reason is that the random order potentially produces a larger set of training pairs since each random order strategy can be considered as a different training pair. | the deterministic order stabilizes the maximum likelihood estimate training. | contrasting |
train_16980 | One prominent approach for data collection has been to automatically generate pseudo-language paired with logical forms, and paraphrase the pseudo-language to natural language through crowdsourcing (Wang et al., 2015). | this data collection procedure often leads to low performance on real data, due to a mismatch between the true distribution of examples and the distribution induced by the data collection procedure. | contrasting |
train_16981 | Knowledge Graphs (KGs) such as Freebase and DBpedia have shown their strong power in many natural language processing tasks including question answering and dialog generation (Zhou et al., 2018). | these KGs are far from complete. | contrasting |
train_16982 | MultiR (Hoffmann et al., 2011) and MIMLRE (Surdeanu et al., 2012) introduce multi-instance learning where the instances mentioning the same entity pair are processed at a bag level. | these methods rely heavily on handcrafted features. | contrasting |
train_16983 | Fortunately, the automatically constructed lexicon contains rich word boundaries information and word semantic information. | integrating lexical knowledge in Chinese NER tasks still faces challenges when it comes to self-matched lexical words as well as the nearest contextual lexical words. | contrasting |
train_16984 | Since without word boundaries information, it is intuitive to use character information 1 The code is available at https://github.com/ DianboWork/Graph4CNER only for Chinese NER (He and Wang, 2008;Liu et al., 2010;Li et al., 2014), although such methods could result in the disregard of word information. | word information is very useful in Chinese NER, because word boundaries are usually the same as named entity boundaries. | contrasting |
train_16985 | We observe significant improvements on distantly supervised datasets (i.e., KBP and NYT), with a up to 19% relative F1 improvement (Bi-GRU from 37.77% to 45.01% on KBP). | on the human-annotated corpus, the performance gain can be hardly noticed. | contrasting |
train_16986 | Many existing relation extraction (RE) models make decisions globally using integer linear programming (ILP). | it is nontrivial to make use of integer linear programming as a blackbox solver for RE. | contrasting |
train_16987 | Currently, research efforts have derived useful discrete features from dependency structures (Sasano and Kurohashi, 2008;Cucchiarelli and Velardi, 2001;Ling and Weld, 2012) or structural constraints (Jie et al., 2017) to help the NER task. | how to make good use of the rich relational information as well as complex long-distance interactions among words as conveyed by the complete dependency structures for improved NER remains a research question to be answered. | contrasting |
train_16988 | Empirically, we also found that those correctly retrieved entities of the DGLSTM-CRF (compared against the baseline) mostly correlate with the following dependency relations: "nn", "nsubj", "nummod". | dGLSTM-CRF achieves lower precisions against BiLSTM-CRF, which indicates that the dGLSTM-CRF model makes more false-positive predictions. | contrasting |
train_16989 | Note that, the bilingual data needed for this approach is coarsely taken from the same domain. | the texts need not be aligned beyond this coarse level. | contrasting |
train_16990 | We also experimented with the following neural network architectures for the classification: LSTM-RNN (Hochreiter and Schmidhuber, 1997), HAN (Yang et al., 2016), QRNN (Bradbury et al., 2017), and VDCNN (Conneau et al., 2017). | these models did not achieve any substantial performance gain to justify their additional complexity. | contrasting |
train_16991 | However, they are limited to short-text classification tasks. | our model effectively uses contextual information and combines recurrent and projection operations to achieve efficiency and enable learning more powerful neural networks that generalize well and can solve more complex language classification tasks. | contrasting |
train_16992 | Previous methods have proposed to overcome this by relying on character-level embeddings and other neural models like character-CNNs. | these methods are often complex and slow to compute for long text (e.g., convolution kernels on devices without significant computational capacity) and still require explicitly storing character or sub-word sequences. | contrasting |
train_16993 | In such challenging scenarios, recent studies have used meta-learning to simulate the few-shot task, in which new queries are compared to a small support set at the samplewise level. | this sample-wise comparison may be severely disturbed by the various expressions in the same class. | contrasting |
train_16994 | Such non-parametric models only need to learn the representation of the samples and the metric measure. | instances in the same class are interlinked and have their uniform fraction and their specific fractions. | contrasting |
train_16995 | We have a large labeled training set with a set of classes C train . | after training, our ultimate goal is to produce classifiers on the testing set with a disjoint set of new classes C test , for which only a small labeled support set will be available. | contrasting |
train_16996 | Style of a text is a very general notion that is hard to define in rigorous terms (Xu, 2017). | the style of a text can be characterized quantitatively (Hughes et al., 2012); stylized texts could be generated if a system is trained on a dataset of stylistically similar texts (Potash et al., 2015); and authorstyle could be learned end-to-end (Tikhonov and Yamshchikov, 2018b,c;. | contrasting |
train_16997 | Indeed, the output that copies input gives maximal BLEU yet clearly fails in terms of the style transfer. | a wholly rephrased sentence could provide a low BLEU between input and output but high accuracy. | contrasting |
train_16998 | Learning-based approaches alone, or combined with lexicon-based methods, usually produce state-of-the-art (SOTA) performance (Mudinas et al., 2012;Zhang et al., 2018), and thus are widely used nowadays. | learning-based methods usually demand a large amount of annotated data to train models, which has become one of the performance bottlenecks. | contrasting |
train_16999 | BERT uses a cross-encoder: Two sentences are passed to the transformer network and the target value is predicted. | this setup is unsuitable for various pair regression tasks due to too many possible combinations. | contrasting |
Subsets and Splits
No community queries yet
The top public SQL queries from the community will appear here once available.