Datasets:

Languages:
French
Multilinguality:
monolingual
Size Categories:
1K<n<10K
Language Creators:
crowdsourced
found
Annotations Creators:
crowdsourced
Source Datasets:
original
ArXiv:
License:
system HF staff commited on
Commit
e57941b
1 Parent(s): 227298a

Update files from the datasets library (from 1.6.0)

Browse files

Release notes: https://github.com/huggingface/datasets/releases/tag/1.6.0

Files changed (2) hide show
  1. README.md +54 -22
  2. fquad.py +0 -1
README.md CHANGED
@@ -1,4 +1,27 @@
1
  ---
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
2
  ---
3
 
4
  # Dataset Card for "fquad"
@@ -30,9 +53,8 @@
30
  ## Dataset Description
31
 
32
  - **Homepage:** [https://fquad.illuin.tech/](https://fquad.illuin.tech/)
33
- - **Repository:** [More Information Needed](https://github.com/huggingface/datasets/blob/master/CONTRIBUTING.md#how-to-contribute-to-the-dataset-cards)
34
- - **Paper:** [More Information Needed](https://github.com/huggingface/datasets/blob/master/CONTRIBUTING.md#how-to-contribute-to-the-dataset-cards)
35
- - **Point of Contact:** [More Information Needed](https://github.com/huggingface/datasets/blob/master/CONTRIBUTING.md#how-to-contribute-to-the-dataset-cards)
36
  - **Size of downloaded dataset files:** 3.14 MB
37
  - **Size of the generated dataset:** 6.62 MB
38
  - **Total amount of disk used:** 9.76 MB
@@ -40,16 +62,19 @@
40
  ### Dataset Summary
41
 
42
  FQuAD: French Question Answering Dataset
43
- We introduce FQuAD, a native French Question Answering Dataset. FQuAD contains 25,000+ question and answer pairs.
44
- Finetuning CamemBERT on FQuAD yields a F1 score of 88% and an exact match of 77.9%.
 
 
 
45
 
46
  ### Supported Tasks
47
 
48
- [More Information Needed](https://github.com/huggingface/datasets/blob/master/CONTRIBUTING.md#how-to-contribute-to-the-dataset-cards)
49
 
50
  ### Languages
51
 
52
- [More Information Needed](https://github.com/huggingface/datasets/blob/master/CONTRIBUTING.md#how-to-contribute-to-the-dataset-cards)
53
 
54
  ## Dataset Structure
55
 
@@ -90,51 +115,57 @@ The data fields are the same among all splits.
90
 
91
  ### Data Splits Sample Size
92
 
93
- | name |train|validation|
94
- |-------|----:|---------:|
95
- |default| 4921| 768|
 
 
 
 
96
 
97
  ## Dataset Creation
98
 
99
  ### Curation Rationale
100
-
101
- [More Information Needed](https://github.com/huggingface/datasets/blob/master/CONTRIBUTING.md#how-to-contribute-to-the-dataset-cards)
102
 
103
  ### Source Data
104
 
105
- [More Information Needed](https://github.com/huggingface/datasets/blob/master/CONTRIBUTING.md#how-to-contribute-to-the-dataset-cards)
106
-
107
  ### Annotations
108
 
109
- [More Information Needed](https://github.com/huggingface/datasets/blob/master/CONTRIBUTING.md#how-to-contribute-to-the-dataset-cards)
 
 
110
 
111
  ### Personal and Sensitive Information
112
 
113
- [More Information Needed](https://github.com/huggingface/datasets/blob/master/CONTRIBUTING.md#how-to-contribute-to-the-dataset-cards)
114
 
115
  ## Considerations for Using the Data
116
 
 
 
117
  ### Social Impact of Dataset
118
 
119
- [More Information Needed](https://github.com/huggingface/datasets/blob/master/CONTRIBUTING.md#how-to-contribute-to-the-dataset-cards)
120
 
121
  ### Discussion of Biases
122
 
123
- [More Information Needed](https://github.com/huggingface/datasets/blob/master/CONTRIBUTING.md#how-to-contribute-to-the-dataset-cards)
124
 
125
  ### Other Known Limitations
126
 
127
- [More Information Needed](https://github.com/huggingface/datasets/blob/master/CONTRIBUTING.md#how-to-contribute-to-the-dataset-cards)
128
 
129
  ## Additional Information
130
 
131
  ### Dataset Curators
132
 
133
- [More Information Needed](https://github.com/huggingface/datasets/blob/master/CONTRIBUTING.md#how-to-contribute-to-the-dataset-cards)
134
 
135
  ### Licensing Information
136
 
137
- [More Information Needed](https://github.com/huggingface/datasets/blob/master/CONTRIBUTING.md#how-to-contribute-to-the-dataset-cards)
138
 
139
  ### Citation Information
140
 
@@ -159,4 +190,5 @@ archivePrefix = {arXiv},
159
 
160
  ### Contributions
161
 
162
- Thanks to [@thomwolf](https://github.com/thomwolf), [@mariamabarham](https://github.com/mariamabarham), [@patrickvonplaten](https://github.com/patrickvonplaten), [@lewtun](https://github.com/lewtun), [@albertvillanova](https://github.com/albertvillanova) for adding this dataset.
 
 
1
  ---
2
+ annotations_creators:
3
+ - crowdsourced
4
+ extended:
5
+ - original
6
+ language_creators:
7
+ - crowdsourced
8
+ - found
9
+ languages:
10
+ - fr
11
+ licenses:
12
+ - cc-by-nc-sa-3-0
13
+ multilinguality:
14
+ - monolingual
15
+ size_categories:
16
+ - 1K<n<10K
17
+ source_datasets:
18
+ - original
19
+ task_categories:
20
+ - question-answering
21
+ - text-retrieval
22
+ task_ids:
23
+ - extractive-qa
24
+ - closed-domain-qa
25
  ---
26
 
27
  # Dataset Card for "fquad"
 
53
  ## Dataset Description
54
 
55
  - **Homepage:** [https://fquad.illuin.tech/](https://fquad.illuin.tech/)
56
+ - **Paper:** [FQuAD: French Question Answering Dataset](https://arxiv.org/abs/2002.06071)
57
+ - **Point of Contact:** [https://www.illuin.tech/contact/](https://www.illuin.tech/contact/)
 
58
  - **Size of downloaded dataset files:** 3.14 MB
59
  - **Size of the generated dataset:** 6.62 MB
60
  - **Total amount of disk used:** 9.76 MB
 
62
  ### Dataset Summary
63
 
64
  FQuAD: French Question Answering Dataset
65
+ We introduce FQuAD, a native French Question Answering Dataset.
66
+
67
+ FQuAD contains 25,000+ question and answer pairs.
68
+ Finetuning CamemBERT on FQuAD yields a F1 score of 88% and an exact match of 77.9%.
69
+ Developped to provide a SQuAD equivalent in the French language. Questions are original and based on high quality Wikipedia articles.
70
 
71
  ### Supported Tasks
72
 
73
+ - `closed-domain-qa`, `text-retrieval`: This dataset is intended to be used for `closed-domain-qa`, but can also be used for information retrieval tasks.
74
 
75
  ### Languages
76
 
77
+ This dataset is exclusively in French, with context data from Wikipedia and questions from French university students (`fr`).
78
 
79
  ## Dataset Structure
80
 
 
115
 
116
  ### Data Splits Sample Size
117
 
118
+ The FQuAD dataset has 3 splits: _train_, _validation_, and _test_. The _test_ split is however not released publicly at the moment. The splits contain disjoint sets of articles. The following table contains stats about each split.
119
+
120
+ Dataset Split | Number of Articles in Split | Number of paragraphs in split | Number of questions in split
121
+ --------------|------------------------------|--------------------------|-------------------------
122
+ Train | 117 | 4921 | 20731
123
+ Validation | 768 | 51.0% | 3188
124
+ Test | 10 | 532 | 2189
125
 
126
  ## Dataset Creation
127
 
128
  ### Curation Rationale
129
+ The FQuAD dataset was created by Illuin technology. It was developped to provide a SQuAD equivalent in the French language. Questions are original and based on high quality Wikipedia articles.
 
130
 
131
  ### Source Data
132
 
133
+ The text used for the contexts are from the curated list of French High-Quality Wikipedia [articles](https://fr.wikipedia.org/wiki/Cat%C3%A9gorie:Article_de_qualit%C3%A9).
 
134
  ### Annotations
135
 
136
+ Annotations (spans and questions) are written by students of the CentraleSupélec school of engineering.
137
+ Wikipedia articles were scraped and Illuin used an internally-developped tool to help annotators ask questions and indicate the answer spans.
138
+ Annotators were given paragraph sized contexts and asked to generate 4/5 non-trivial questions about information in the context.
139
 
140
  ### Personal and Sensitive Information
141
 
142
+ No personal or sensitive information is included in this dataset. This has been manually verified by the dataset curators.
143
 
144
  ## Considerations for Using the Data
145
 
146
+ Users should consider this dataset is sampled from Wikipedia data which might not be representative of all QA use cases.
147
+
148
  ### Social Impact of Dataset
149
 
150
+ The social biases of this dataset have not yet been investigated.
151
 
152
  ### Discussion of Biases
153
 
154
+ The social biases of this dataset have not yet been investigated, though articles have been selected by their quality and objectivity.
155
 
156
  ### Other Known Limitations
157
 
158
+ The limitations of the FQuAD dataset have not yet been investigated.
159
 
160
  ## Additional Information
161
 
162
  ### Dataset Curators
163
 
164
+ Illuin Technology: [https://fquad.illuin.tech/](https://fquad.illuin.tech/)
165
 
166
  ### Licensing Information
167
 
168
+ The FQuAD dataset is licensed under the [CC BY-NC-SA 3.0](https://creativecommons.org/licenses/by-nc-sa/3.0/fr/) license.
169
 
170
  ### Citation Information
171
 
 
190
 
191
  ### Contributions
192
 
193
+ Thanks to [@thomwolf](https://github.com/thomwolf), [@mariamabarham](https://github.com/mariamabarham), [@patrickvonplaten](https://github.com/patrickvonplaten), [@lewtun](https://github.com/lewtun), [@albertvillanova](https://github.com/albertvillanova) for adding this dataset.
194
+ Thanks to [@ManuelFay](https://github.com/manuelfay) for providing information on the dataset creation process.
fquad.py CHANGED
@@ -1,6 +1,5 @@
1
  """TODO(fquad): Add a description here."""
2
 
3
- from __future__ import absolute_import, division, print_function
4
 
5
  import json
6
  import os
 
1
  """TODO(fquad): Add a description here."""
2
 
 
3
 
4
  import json
5
  import os