Datasets:
Tasks:
Question Answering
Modalities:
Text
Formats:
json
Sub-tasks:
extractive-qa
Languages:
Catalan
Size:
1K - 10K
ArXiv:
License:
Upload dataset
Browse files- README.md +236 -0
- dev.json +0 -0
- test.json +0 -0
- train.json +0 -0
- vilaquad.py +197 -0
README.md
ADDED
@@ -0,0 +1,236 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
---
|
2 |
+
|
3 |
+
languages:
|
4 |
+
|
5 |
+
- ca
|
6 |
+
|
7 |
+
---
|
8 |
+
|
9 |
+
# VilaQuAD, An extractive QA dataset for catalan, from Vilaweb newswire text
|
10 |
+
|
11 |
+
## BibTeX citation
|
12 |
+
|
13 |
+
If you use any of these resources (datasets or models) in your work, please cite our latest paper:
|
14 |
+
|
15 |
+
```bibtex
|
16 |
+
|
17 |
+
@inproceedings{armengol-estape-etal-2021-multilingual,
|
18 |
+
|
19 |
+
title = "Are Multilingual Models the Best Choice for Moderately Under-resourced Languages? {A} Comprehensive Assessment for {C}atalan",
|
20 |
+
|
21 |
+
author = "Armengol-Estap{\'e}, Jordi and
|
22 |
+
|
23 |
+
Carrino, Casimiro Pio and
|
24 |
+
|
25 |
+
Rodriguez-Penagos, Carlos and
|
26 |
+
|
27 |
+
de Gibert Bonet, Ona and
|
28 |
+
|
29 |
+
Armentano-Oller, Carme and
|
30 |
+
|
31 |
+
Gonzalez-Agirre, Aitor and
|
32 |
+
|
33 |
+
Melero, Maite and
|
34 |
+
|
35 |
+
Villegas, Marta",
|
36 |
+
|
37 |
+
booktitle = "Findings of the Association for Computational Linguistics: ACL-IJCNLP 2021",
|
38 |
+
|
39 |
+
month = aug,
|
40 |
+
|
41 |
+
year = "2021",
|
42 |
+
|
43 |
+
address = "Online",
|
44 |
+
|
45 |
+
publisher = "Association for Computational Linguistics",
|
46 |
+
|
47 |
+
url = "https://aclanthology.org/2021.findings-acl.437",
|
48 |
+
|
49 |
+
doi = "10.18653/v1/2021.findings-acl.437",
|
50 |
+
|
51 |
+
pages = "4933--4946",
|
52 |
+
|
53 |
+
}
|
54 |
+
|
55 |
+
```
|
56 |
+
|
57 |
+
## Digital Object Identifier (DOI) and access to dataset files
|
58 |
+
|
59 |
+
https://doi.org/10.5281/zenodo.4562337
|
60 |
+
|
61 |
+
|
62 |
+
## Introduction
|
63 |
+
|
64 |
+
This dataset contains 2095 of Catalan language news articles along with 1 to 5 questions referring to each fragment (or context).
|
65 |
+
VilaQuad articles are extracted from the daily Vilaweb (www.vilaweb.cat) and used under CC-by-nc-sa-nd (https://creativecommons.org/licenses/by-nc-nd/3.0/deed.ca) licence.
|
66 |
+
This dataset can be used to build extractive-QA and Language Models.
|
67 |
+
|
68 |
+
### Supported Tasks and Leaderboards
|
69 |
+
|
70 |
+
Extractive-QA, Language Model
|
71 |
+
|
72 |
+
### Languages
|
73 |
+
|
74 |
+
CA- Catalan
|
75 |
+
|
76 |
+
### Directory structure
|
77 |
+
|
78 |
+
* README.md
|
79 |
+
* dev.json
|
80 |
+
* test.json
|
81 |
+
* train.json
|
82 |
+
* vilaquad.py
|
83 |
+
|
84 |
+
## Dataset Structure
|
85 |
+
|
86 |
+
### Data Instances
|
87 |
+
|
88 |
+
Three json files
|
89 |
+
|
90 |
+
### Data Fields
|
91 |
+
|
92 |
+
Follows ((Rajpurkar, Pranav et al., 2016) for squad v1 datasets. (see below for full reference)
|
93 |
+
|
94 |
+
### Example:
|
95 |
+
<pre>
|
96 |
+
{
|
97 |
+
"data": [
|
98 |
+
{
|
99 |
+
"title": "Com celebrar el Cap d'Any 2020? Deu propostes per a acomiadar-se del 2019",
|
100 |
+
"paragraphs": [
|
101 |
+
{
|
102 |
+
"context": "Hi ha moltes propostes per a acomiadar-se d'aquest 2019. Els uns es queden a casa, els altres volen anar lluny o sortir al teatre. També s'organitzen festes o festivals a l'engròs, fins i tot hi ha propostes diürnes. Tot és possible per Cap d'Any. Encara no sabeu com celebrar l'entrada el 2020? Us oferim una llista amb deu propostes variades arreu dels Països Catalans: Festivern El Festivern enguany celebra quinze anys.",
|
103 |
+
"qas": [
|
104 |
+
|
105 |
+
{
|
106 |
+
"answers": [
|
107 |
+
{
|
108 |
+
"text": "festes o festivals",
|
109 |
+
"answer_start": 150
|
110 |
+
}
|
111 |
+
],
|
112 |
+
"id": "P_23_C_23_Q2",
|
113 |
+
"question": "Què s'organitza a l'engròs per acomiadar el 2019?"
|
114 |
+
},
|
115 |
+
...
|
116 |
+
]
|
117 |
+
}
|
118 |
+
]
|
119 |
+
},
|
120 |
+
...
|
121 |
+
]
|
122 |
+
}
|
123 |
+
|
124 |
+
</pre>
|
125 |
+
|
126 |
+
### Data Splits
|
127 |
+
|
128 |
+
train.json: 1295 contexts, 3882 questions
|
129 |
+
dev.json: 400 contexts, 1200 questions
|
130 |
+
test.json: 400 contexts, 1200 questions
|
131 |
+
|
132 |
+
## Content analysis
|
133 |
+
|
134 |
+
### Number of articles, paragraphs and questions
|
135 |
+
|
136 |
+
* Number of contexts: 2095
|
137 |
+
* Number of questions: 6282
|
138 |
+
* Questions/context: 2.99
|
139 |
+
* Number of sentences in contexts: 11901
|
140 |
+
* Sentences/context: 5.6
|
141 |
+
|
142 |
+
### Number of tokens
|
143 |
+
|
144 |
+
* tokens in context: 422477
|
145 |
+
* tokens/context 201.66
|
146 |
+
* tokens in questons: 65849
|
147 |
+
* tokens/questions: 10.48
|
148 |
+
* tokens in answers: 27716
|
149 |
+
* tokens/answers: 4.41
|
150 |
+
|
151 |
+
### Question type
|
152 |
+
|
153 |
+
| Question | Count | % |
|
154 |
+
|--------|-----|------|
|
155 |
+
| què | 1698 | 27.03 % |
|
156 |
+
| qui | 1161 | 18.48 % |
|
157 |
+
| com | 574 | 9.14 % |
|
158 |
+
| quan | 468 | 7.45 % |
|
159 |
+
| on | 559 | 8.9 % |
|
160 |
+
| quant | 601 | 9.57 % |
|
161 |
+
| quin | 1301 | 20.87 % |
|
162 |
+
| no question mark | 0 | 0.0 % |
|
163 |
+
|
164 |
+
|
165 |
+
### Question-answer relationships
|
166 |
+
|
167 |
+
From 100 randomly selected samples:
|
168 |
+
|
169 |
+
* Lexical variation: 32.0%
|
170 |
+
* World knowledge: 16.0%
|
171 |
+
* Syntactic variation: 22.0%
|
172 |
+
* Multiple sentence: 16.0%
|
173 |
+
|
174 |
+
## Dataset Creation
|
175 |
+
|
176 |
+
### Methodology
|
177 |
+
From a the online edition of the catalan newspaper Vilaweb (https://www.vilaweb.cat), 2095 articles were randomnly selected. These headlines were also used to create a Textual Entailment dataset. For the extractive QA dataset, creation of between 1 and 5 questions for each news context was commissioned, following an adaptation of the guidelines from SQUAD 1.0 (Rajpurkar, Pranav et al. “SQuAD: 100, 000+ Questions for Machine Comprehension of Text.” EMNLP (2016)), http://arxiv.org/abs/1606.05250. In total, 6282 pairs of a question and an extracted fragment that contains the answer were created.
|
178 |
+
|
179 |
+
### Curation Rationale
|
180 |
+
|
181 |
+
For compatibility with similar datasets in other languages, we followed as close as possible existing curation guidelines. We also created another QA dataset with wikipedia to ensure thematic and stylistic variety.
|
182 |
+
|
183 |
+
### Source Data
|
184 |
+
|
185 |
+
- https://www.vilaweb.cat/
|
186 |
+
|
187 |
+
#### Initial Data Collection and Normalization
|
188 |
+
|
189 |
+
The source data are scraped articles from archives of Catalan newspaper website Vilaweb (https://www.vilaweb.cat).
|
190 |
+
|
191 |
+
#### Who are the source language producers?
|
192 |
+
|
193 |
+
[More Information Needed]
|
194 |
+
|
195 |
+
### Annotations
|
196 |
+
|
197 |
+
#### Annotation process
|
198 |
+
|
199 |
+
We comissioned the creation of 1 to 5 questions for each context, following an adaptation of the guidelines from SQUAD 1.0 (Rajpurkar, Pranav et al. “SQuAD: 100, 000+ Questions for Machine Comprehension of Text.” EMNLP (2016)), http://arxiv.org/abs/1606.05250.
|
200 |
+
|
201 |
+
#### Who are the annotators?
|
202 |
+
|
203 |
+
Annotation was commissioned to an specialized company that hired a team of native language speakers.
|
204 |
+
|
205 |
+
### Dataset Curators
|
206 |
+
|
207 |
+
Carlos Rodríguez and Carme Armentano, from BSC-CNS
|
208 |
+
|
209 |
+
### Personal and Sensitive Information
|
210 |
+
|
211 |
+
No personal or sensitive information included.
|
212 |
+
|
213 |
+
## Considerations for Using the Data
|
214 |
+
|
215 |
+
### Social Impact of Dataset
|
216 |
+
|
217 |
+
[More Information Needed]
|
218 |
+
|
219 |
+
### Discussion of Biases
|
220 |
+
|
221 |
+
[More Information Needed]
|
222 |
+
|
223 |
+
### Other Known Limitations
|
224 |
+
|
225 |
+
[More Information Needed]
|
226 |
+
|
227 |
+
|
228 |
+
## Contact
|
229 |
+
|
230 |
+
Carlos Rodríguez-Penagos ([email protected]) and Carme Armentano-Oller ([email protected])
|
231 |
+
|
232 |
+
|
233 |
+
## License
|
234 |
+
|
235 |
+
<a rel="license" href="https://creativecommons.org/licenses/by-sa/4.0/"><img alt="Attribution-ShareAlike 4.0 International License" style="border-width:0" src="https://i.creativecommons.org/l/by/4.0/88x31.png" /></a><br />This work is licensed under a <a rel="license" href="https://creativecommons.org/licenses/by-sa/4.0/">Attribution-ShareAlike 4.0 International License</a>.
|
236 |
+
|
dev.json
ADDED
The diff for this file is too large to render.
See raw diff
|
|
test.json
ADDED
The diff for this file is too large to render.
See raw diff
|
|
train.json
ADDED
The diff for this file is too large to render.
See raw diff
|
|
vilaquad.py
ADDED
@@ -0,0 +1,197 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
# Loading script for the VilaQuAD dataset.
|
2 |
+
|
3 |
+
import json
|
4 |
+
|
5 |
+
import datasets
|
6 |
+
|
7 |
+
logger = datasets.logging.get_logger(__name__)
|
8 |
+
|
9 |
+
_CITATION = """
|
10 |
+
|
11 |
+
Rodriguez-Penagos, Carlos Gerardo, & Armentano-Oller, Carme. (2021).
|
12 |
+
|
13 |
+
VilaQuAD: an extractive QA dataset for catalan, from Vilaweb newswire text
|
14 |
+
|
15 |
+
[Data set]. Zenodo. https://doi.org/10.5281/zenodo.4562337
|
16 |
+
|
17 |
+
"""
|
18 |
+
|
19 |
+
_DESCRIPTION = """
|
20 |
+
|
21 |
+
This dataset contains 2095 of Catalan language news articles along with 1 to 5 questions referring to each fragment (or context).
|
22 |
+
|
23 |
+
VilaQuad articles are extracted from the daily Vilaweb (www.vilaweb.cat) and used under CC-by-nc-sa-nd (https://creativecommons.org/licenses/by-nc-nd/3.0/deed.ca) licence.
|
24 |
+
|
25 |
+
This dataset can be used to build extractive-QA and Language Models.
|
26 |
+
|
27 |
+
Funded by the Generalitat de Catalunya, Departament de Polítiques Digitals i Administració Pública (AINA),
|
28 |
+
|
29 |
+
MT4ALL and Plan de Impulso de las Tecnologías del Lenguaje (Plan TL).
|
30 |
+
|
31 |
+
"""
|
32 |
+
|
33 |
+
_HOMEPAGE = """https://doi.org/10.5281/zenodo.4562337"""
|
34 |
+
|
35 |
+
_URL = "https://huggingface.co/datasets/BSC-TeMU/vilaquad/resolve/main/"
|
36 |
+
_URL = "/home/carme/Desktop/tasques/QA/HF/vilaquad/"
|
37 |
+
|
38 |
+
_TRAINING_FILE = "train.json"
|
39 |
+
|
40 |
+
_DEV_FILE = "dev.json"
|
41 |
+
|
42 |
+
_TEST_FILE = "test.json"
|
43 |
+
|
44 |
+
class VilaQuADConfig(datasets.BuilderConfig):
|
45 |
+
|
46 |
+
""" Builder config for the VilaQuAD dataset """
|
47 |
+
|
48 |
+
def __init__(self, **kwargs):
|
49 |
+
|
50 |
+
"""BuilderConfig for VilaQuAD.
|
51 |
+
|
52 |
+
Args:
|
53 |
+
|
54 |
+
**kwargs: keyword arguments forwarded to super.
|
55 |
+
|
56 |
+
"""
|
57 |
+
|
58 |
+
super(VilaQuADConfig, self).__init__(**kwargs)
|
59 |
+
|
60 |
+
class VilaQuAD(datasets.GeneratorBasedBuilder):
|
61 |
+
|
62 |
+
"""VilaQuAD Dataset."""
|
63 |
+
|
64 |
+
BUILDER_CONFIGS = [
|
65 |
+
|
66 |
+
VilaQuADConfig(
|
67 |
+
|
68 |
+
name="VilaQuAD",
|
69 |
+
|
70 |
+
version=datasets.Version("1.0.1"),
|
71 |
+
|
72 |
+
description="VilaQuAD dataset",
|
73 |
+
|
74 |
+
),
|
75 |
+
|
76 |
+
]
|
77 |
+
|
78 |
+
def _info(self):
|
79 |
+
|
80 |
+
return datasets.DatasetInfo(
|
81 |
+
|
82 |
+
description=_DESCRIPTION,
|
83 |
+
|
84 |
+
features=datasets.Features(
|
85 |
+
|
86 |
+
{
|
87 |
+
|
88 |
+
"id": datasets.Value("string"),
|
89 |
+
|
90 |
+
"title": datasets.Value("string"),
|
91 |
+
|
92 |
+
"context": datasets.Value("string"),
|
93 |
+
|
94 |
+
"question": datasets.Value("string"),
|
95 |
+
|
96 |
+
"answers": [
|
97 |
+
|
98 |
+
{
|
99 |
+
|
100 |
+
"text": datasets.Value("string"),
|
101 |
+
|
102 |
+
"answer_start": datasets.Value("int32"),
|
103 |
+
|
104 |
+
}
|
105 |
+
|
106 |
+
],
|
107 |
+
|
108 |
+
}
|
109 |
+
|
110 |
+
),
|
111 |
+
|
112 |
+
# No default supervised_keys (as we have to pass both question
|
113 |
+
|
114 |
+
# and context as input).
|
115 |
+
|
116 |
+
supervised_keys=None,
|
117 |
+
|
118 |
+
homepage=_HOMEPAGE,
|
119 |
+
|
120 |
+
citation=_CITATION,
|
121 |
+
|
122 |
+
)
|
123 |
+
|
124 |
+
def _split_generators(self, dl_manager):
|
125 |
+
|
126 |
+
"""Returns SplitGenerators."""
|
127 |
+
|
128 |
+
urls_to_download = {
|
129 |
+
|
130 |
+
"train": f"{_URL}{_TRAINING_FILE}",
|
131 |
+
|
132 |
+
"dev": f"{_URL}{_DEV_FILE}",
|
133 |
+
|
134 |
+
"test": f"{_URL}{_TEST_FILE}",
|
135 |
+
|
136 |
+
}
|
137 |
+
|
138 |
+
downloaded_files = dl_manager.download_and_extract(urls_to_download)
|
139 |
+
|
140 |
+
return [
|
141 |
+
|
142 |
+
datasets.SplitGenerator(name=datasets.Split.TRAIN, gen_kwargs={"filepath": downloaded_files["train"]}),
|
143 |
+
|
144 |
+
datasets.SplitGenerator(name=datasets.Split.VALIDATION, gen_kwargs={"filepath": downloaded_files["dev"]}),
|
145 |
+
|
146 |
+
datasets.SplitGenerator(name=datasets.Split.TEST, gen_kwargs={"filepath": downloaded_files["test"]}),
|
147 |
+
|
148 |
+
]
|
149 |
+
|
150 |
+
def _generate_examples(self, filepath):
|
151 |
+
|
152 |
+
"""This function returns the examples in the raw (text) form."""
|
153 |
+
|
154 |
+
logger.info("generating examples from = %s", filepath)
|
155 |
+
|
156 |
+
with open(filepath, encoding="utf-8") as f:
|
157 |
+
|
158 |
+
vilaquad = json.load(f, encoding="utf-8")
|
159 |
+
|
160 |
+
for article in vilaquad["data"]:
|
161 |
+
|
162 |
+
title = article.get("title", "").strip()
|
163 |
+
|
164 |
+
for paragraph in article["paragraphs"]:
|
165 |
+
|
166 |
+
context = paragraph["context"].strip()
|
167 |
+
|
168 |
+
for qa in paragraph["qas"]:
|
169 |
+
|
170 |
+
question = qa["question"].strip()
|
171 |
+
|
172 |
+
id_ = qa["id"]
|
173 |
+
|
174 |
+
#answer_starts = [answer["answer_start"] for answer in qa["answers"]]
|
175 |
+
|
176 |
+
#answers = [answer["text"].strip() for answer in qa["answers"]]
|
177 |
+
|
178 |
+
|
179 |
+
# Features currently used are "context", "question", and "answers".
|
180 |
+
|
181 |
+
# Others are extracted here for the ease of future expansions.
|
182 |
+
text = qa["answers"][0]["text"]
|
183 |
+
answer_start = qa["answers"][0]["answer_start"]
|
184 |
+
|
185 |
+
yield id_, {
|
186 |
+
|
187 |
+
"title": title,
|
188 |
+
|
189 |
+
"context": context,
|
190 |
+
|
191 |
+
"question": question,
|
192 |
+
|
193 |
+
"id": id_,
|
194 |
+
|
195 |
+
"answers": [{"text": text, "answer_start": answer_start}]
|
196 |
+
|
197 |
+
}
|