Datasets:
Size:
10K - 100K
License:
:memo: Update documents
Browse files
README.md
CHANGED
@@ -1,3 +1,215 @@
|
|
1 |
---
|
2 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
3 |
---
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
---
|
2 |
+
annotations_creators:
|
3 |
+
- expert-generated
|
4 |
+
language:
|
5 |
+
- ja
|
6 |
+
- en
|
7 |
+
language_creators:
|
8 |
+
- expert-generated
|
9 |
+
license:
|
10 |
+
- cc-by-sa-4.0
|
11 |
+
multilinguality:
|
12 |
+
- translation
|
13 |
+
pretty_name: JSICK
|
14 |
+
size_categories:
|
15 |
+
- 1K<n<10K
|
16 |
+
source_datasets:
|
17 |
+
- extended|sick
|
18 |
+
tags:
|
19 |
+
- semantic-textual-similarity
|
20 |
+
- sts
|
21 |
+
task_categories:
|
22 |
+
- sentence-similarity
|
23 |
+
- text-classification
|
24 |
+
task_ids:
|
25 |
+
- natural-language-inference
|
26 |
+
- semantic-similarity-scoring
|
27 |
---
|
28 |
+
|
29 |
+
# Dataset Card for JaNLI
|
30 |
+
|
31 |
+
## Table of Contents
|
32 |
+
- [Dataset Card for JaNLI](#dataset-card-for-janli)
|
33 |
+
- [Table of Contents](#table-of-contents)
|
34 |
+
- [Dataset Description](#dataset-description)
|
35 |
+
- [Dataset Summary](#dataset-summary)
|
36 |
+
- [Languages](#languages)
|
37 |
+
- [Dataset Structure](#dataset-structure)
|
38 |
+
- [Data Instances](#data-instances)
|
39 |
+
- [base](#base)
|
40 |
+
- [original](#original)
|
41 |
+
- [Data Fields](#data-fields)
|
42 |
+
- [base](#base-1)
|
43 |
+
- [original](#original-1)
|
44 |
+
- [Data Splits](#data-splits)
|
45 |
+
- [Annotations](#annotations)
|
46 |
+
- [Additional Information](#additional-information)
|
47 |
+
- [Licensing Information](#licensing-information)
|
48 |
+
- [Citation Information](#citation-information)
|
49 |
+
- [Contributions](#contributions)
|
50 |
+
|
51 |
+
## Dataset Description
|
52 |
+
|
53 |
+
- **Homepage:** https://github.com/verypluming/JSICK
|
54 |
+
- **Repository:** https://github.com/verypluming/JSICK
|
55 |
+
- **Paper:** https://direct.mit.edu/tacl/article/doi/10.1162/tacl_a_00518/113850/Compositional-Evaluation-on-Japanese-Textual
|
56 |
+
- **Paper:** https://www.jstage.jst.go.jp/article/pjsai/JSAI2021/0/JSAI2021_4J3GS6f02/_pdf/-char/ja
|
57 |
+
|
58 |
+
### Dataset Summary
|
59 |
+
|
60 |
+
From official [GitHub](https://github.com/verypluming/JSICK):
|
61 |
+
|
62 |
+
Japanese Sentences Involving Compositional Knowledge (JSICK) Dataset.
|
63 |
+
JSICK is the Japanese NLI and STS dataset by manually translating the English dataset [SICK (Marelli et al., 2014)](https://aclanthology.org/L14-1314/) into Japanese.
|
64 |
+
We hope that our dataset will be useful in research for realizing more advanced models that are capable of appropriately performing multilingual compositional inference.
|
65 |
+
|
66 |
+
### Languages
|
67 |
+
|
68 |
+
The language data in JSICK is in Japanese and English.
|
69 |
+
|
70 |
+
|
71 |
+
## Dataset Structure
|
72 |
+
|
73 |
+
|
74 |
+
### Data Instances
|
75 |
+
When loading a specific configuration, users has to append a version dependent suffix:
|
76 |
+
|
77 |
+
```python
|
78 |
+
import datasets as ds
|
79 |
+
|
80 |
+
dataset: ds.DatasetDict = ds.load_dataset("hpprc/jsick")
|
81 |
+
print(dataset)
|
82 |
+
# DatasetDict({
|
83 |
+
# train: Dataset({
|
84 |
+
# features: ['id', 'premise', 'hypothesis', 'label', 'heuristics', 'number_of_NPs', 'semtag'],
|
85 |
+
# num_rows: 13680
|
86 |
+
# })
|
87 |
+
# test: Dataset({
|
88 |
+
# features: ['id', 'premise', 'hypothesis', 'label', 'heuristics', 'number_of_NPs', 'semtag'],
|
89 |
+
# num_rows: 720
|
90 |
+
# })
|
91 |
+
# })
|
92 |
+
|
93 |
+
dataset: ds.DatasetDict = ds.load_dataset("hpprc/jsick", name="original")
|
94 |
+
print(dataset)
|
95 |
+
# DatasetDict({
|
96 |
+
# train: Dataset({
|
97 |
+
# features: ['id', 'sentence_A_Ja', 'sentence_B_Ja', 'entailment_label_Ja', 'heuristics', 'number_of_NPs', 'semtag'],
|
98 |
+
# num_rows: 13680
|
99 |
+
# })
|
100 |
+
# test: Dataset({
|
101 |
+
# features: ['id', 'sentence_A_Ja', 'sentence_B_Ja', 'entailment_label_Ja', 'heuristics', 'number_of_NPs', 'semtag'],
|
102 |
+
# num_rows: 720
|
103 |
+
# })
|
104 |
+
# })
|
105 |
+
```
|
106 |
+
|
107 |
+
|
108 |
+
#### base
|
109 |
+
|
110 |
+
An example of looks as follows:
|
111 |
+
|
112 |
+
```json
|
113 |
+
{
|
114 |
+
'id': 12,
|
115 |
+
'premise': 'θ₯θ
γγγγγγΌγ«ιΈζγθ¦γ¦γγ',
|
116 |
+
'hypothesis': 'γγγγγΌγ«ιΈζγθ₯θ
γθ¦γ¦γγ',
|
117 |
+
'label': 0,
|
118 |
+
'heuristics': 'overlap-full',
|
119 |
+
'number_of_NPs': 2,
|
120 |
+
'semtag': 'scrambling'
|
121 |
+
}
|
122 |
+
```
|
123 |
+
|
124 |
+
#### original
|
125 |
+
|
126 |
+
An example of looks as follows:
|
127 |
+
|
128 |
+
```json
|
129 |
+
{
|
130 |
+
'id': 12,
|
131 |
+
'sentence_A_Ja': 'θ₯θ
γγγγγγΌγ«ιΈζγθ¦γ¦γγ',
|
132 |
+
'sentence_B_Ja': 'γγγγγΌγ«ιΈζγθ₯θ
γθ¦γ¦γγ',
|
133 |
+
'entailment_label_Ja': 0,
|
134 |
+
'heuristics': 'overlap-full',
|
135 |
+
'number_of_NPs': 2,
|
136 |
+
'semtag': 'scrambling'
|
137 |
+
}
|
138 |
+
```
|
139 |
+
|
140 |
+
### Data Fields
|
141 |
+
|
142 |
+
#### base
|
143 |
+
|
144 |
+
A version adopting the column names of a typical NLI dataset.
|
145 |
+
|
146 |
+
- `id`: The number of the sentence pair.
|
147 |
+
- `premise`: The premise (sentence_A_Ja).
|
148 |
+
- `hypothesis`: The hypothesis (sentence_B_Ja).
|
149 |
+
- `label`: The correct label for this sentence pair (either `entailment` or `non-entailment`); in the setting described in the paper, non-entailment = neutral + contradiction (entailment_label_Ja).
|
150 |
+
- `heuristics`: The heuristics (structural pattern) tag. The tags are: subsequence, constituent, full-overlap, order-subset, and mixed-subset.
|
151 |
+
- `number_of_NPs`: The number of noun phrase in a sentence.
|
152 |
+
- `semtag`: The linguistic phenomena tag.
|
153 |
+
|
154 |
+
#### original
|
155 |
+
|
156 |
+
The original version retaining the unaltered column names.
|
157 |
+
|
158 |
+
- `id`: The number of the sentence pair.
|
159 |
+
- `sentence_A_Ja`: The premise.
|
160 |
+
- `sentence_B_Ja`: The hypothesis.
|
161 |
+
- `entailment_label_Ja`: The correct label for this sentence pair (either `entailment` or `non-entailment`); in the setting described in the paper, non-entailment = neutral + contradiction
|
162 |
+
- `heuristics`: The heuristics (structural pattern) tag. The tags are: subsequence, constituent, full-overlap, order-subset, and mixed-subset.
|
163 |
+
- `number_of_NPs`: The number of noun phrase in a sentence.
|
164 |
+
- `semtag`: The linguistic phenomena tag.
|
165 |
+
|
166 |
+
|
167 |
+
### Data Splits
|
168 |
+
|
169 |
+
| name | train | validation | test |
|
170 |
+
| -------- | -----: | ---------: | ---: |
|
171 |
+
| base | 13,680 | | 720 |
|
172 |
+
| original | 13,680 | | 720 |
|
173 |
+
|
174 |
+
|
175 |
+
|
176 |
+
### Annotations
|
177 |
+
|
178 |
+
The annotation process for this Japanese NLI dataset involves tagging each pair (P, H) of a premise and hypothesis with a label for structural pattern and linguistic phenomenon.
|
179 |
+
The structural relationship between premise and hypothesis sentences is classified into five patterns, with each pattern associated with a type of heuristic that can lead to incorrect predictions of the entailment relation.
|
180 |
+
Additionally, 11 categories of Japanese linguistic phenomena and constructions are focused on for generating the five patterns of adversarial inferences.
|
181 |
+
|
182 |
+
For each linguistic phenomenon, a template for the premise sentence P is fixed, and multiple templates for hypothesis sentences H are created.
|
183 |
+
In total, 144 templates for (P, H) pairs are produced.
|
184 |
+
Each pair of premise and hypothesis sentences is tagged with an entailment label (entailment or non-entailment), a structural pattern, and a linguistic phenomenon label.
|
185 |
+
|
186 |
+
The JaNLI dataset is generated by instantiating each template 100 times, resulting in a total of 14,400 examples.
|
187 |
+
The same number of entailment and non-entailment examples are generated for each phenomenon.
|
188 |
+
The structural patterns are annotated with the templates for each linguistic phenomenon, and the ratio of entailment and non-entailment examples is not necessarily 1:1 for each pattern.
|
189 |
+
The dataset uses a total of 158 words (nouns and verbs), which occur more than 20 times in the JSICK and JSNLI datasets.
|
190 |
+
|
191 |
+
|
192 |
+
## Additional Information
|
193 |
+
|
194 |
+
- [verypluming/JaNLI](https://github.com/verypluming/JaNLI)
|
195 |
+
- [Hitomi Yanaka, Koji Mineshima, Assessing the Generalization Capacity of Pre-trained Language Models through Japanese Adversarial Natural Language Inference, Proceedings of the 2021 EMNLP Workshop BlackboxNLP: Analyzing and Interpreting Neural Networks for NLP (BlackboxNLP2021), 2021.](https://aclanthology.org/2021.blackboxnlp-1.26/)
|
196 |
+
|
197 |
+
### Licensing Information
|
198 |
+
|
199 |
+
CC BY-SA 4.0
|
200 |
+
|
201 |
+
### Citation Information
|
202 |
+
|
203 |
+
```bibtex
|
204 |
+
@InProceedings{yanaka-EtAl:2021:blackbox,
|
205 |
+
author = {Yanaka, Hitomi and Mineshima, Koji},
|
206 |
+
title = {Assessing the Generalization Capacity of Pre-trained Language Models through Japanese Adversarial Natural Language Inference},
|
207 |
+
booktitle = {Proceedings of the 2021 EMNLP Workshop BlackboxNLP: Analyzing and Interpreting Neural Networks for NLP (BlackboxNLP2021)},
|
208 |
+
url = {https://aclanthology.org/2021.blackboxnlp-1.26/},
|
209 |
+
year = {2021},
|
210 |
+
}
|
211 |
+
```
|
212 |
+
|
213 |
+
### Contributions
|
214 |
+
|
215 |
+
Thanks to [Hitomi Yanaka](https://hitomiyanaka.mystrikingly.com/) and Koji Mineshima for creating this dataset.
|
jsick.py
CHANGED
@@ -17,7 +17,10 @@ _CITATION = """\
|
|
17 |
"""
|
18 |
|
19 |
_DESCRIPTION = """\
|
20 |
-
|
|
|
|
|
|
|
21 |
"""
|
22 |
|
23 |
_HOMEPAGE = "https://github.com/verypluming/JSICK"
|
@@ -51,7 +54,7 @@ class JSICKDataset(ds.GeneratorBasedBuilder):
|
|
51 |
description="fuga",
|
52 |
),
|
53 |
ds.BuilderConfig(
|
54 |
-
name="
|
55 |
version=VERSION,
|
56 |
description="fuga",
|
57 |
),
|
|
|
17 |
"""
|
18 |
|
19 |
_DESCRIPTION = """\
|
20 |
+
Japanese Sentences Involving Compositional Knowledge (JSICK) Dataset.
|
21 |
+
JSICK is the Japanese NLI and STS dataset by manually translating the English dataset SICK (Marelli et al., 2014) into Japanese.
|
22 |
+
We hope that our dataset will be useful in research for realizing more advanced models that are capable of appropriately performing multilingual compositional inference.
|
23 |
+
(from official website)
|
24 |
"""
|
25 |
|
26 |
_HOMEPAGE = "https://github.com/verypluming/JSICK"
|
|
|
54 |
description="fuga",
|
55 |
),
|
56 |
ds.BuilderConfig(
|
57 |
+
name="stress-original",
|
58 |
version=VERSION,
|
59 |
description="fuga",
|
60 |
),
|