hpprc commited on
Commit
01fa168
β€’
1 Parent(s): e15e78a

:memo: Update README.md

Browse files
Files changed (1) hide show
  1. README.md +115 -0
README.md CHANGED
@@ -21,6 +21,14 @@ license: cc-by-sa-4.0
21
  - [Dataset Description](#dataset-description)
22
  - [Dataset Summary](#dataset-summary)
23
  - [Languages](#languages)
 
 
 
 
 
 
 
 
24
  - [Annotations](#annotations)
25
  - [Additional Information](#additional-information)
26
  - [Licensing Information](#licensing-information)
@@ -41,6 +49,113 @@ The JaNLI (Japanese Adversarial NLI) dataset, inspired by the English HANS datas
41
 
42
  The language data in JaNLI is in Japanese (BCP-47 [ja-JP](https://www.rfc-editor.org/info/bcp47)).
43
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
44
  ### Annotations
45
 
46
  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.
 
21
  - [Dataset Description](#dataset-description)
22
  - [Dataset Summary](#dataset-summary)
23
  - [Languages](#languages)
24
+ - [Dataset Structure](#dataset-structure)
25
+ - [Data Instances](#data-instances)
26
+ - [base](#base)
27
+ - [original](#original)
28
+ - [Data Fields](#data-fields)
29
+ - [base](#base-1)
30
+ - [original](#original-1)
31
+ - [Data Splits](#data-splits)
32
  - [Annotations](#annotations)
33
  - [Additional Information](#additional-information)
34
  - [Licensing Information](#licensing-information)
 
49
 
50
  The language data in JaNLI is in Japanese (BCP-47 [ja-JP](https://www.rfc-editor.org/info/bcp47)).
51
 
52
+
53
+
54
+ ## Dataset Structure
55
+
56
+
57
+ ### Data Instances
58
+ When loading a specific configuration, users has to append a version dependent suffix:
59
+
60
+ ```python
61
+ import datasets as ds
62
+
63
+ dataset: ds.DatasetDict = ds.load_dataset("hpprc/janli")
64
+ print(dataset)
65
+ # DatasetDict({
66
+ # train: Dataset({
67
+ # features: ['id', 'premise', 'hypothesis', 'label', 'heuristics', 'number_of_NPs', 'semtag'],
68
+ # num_rows: 13680
69
+ # })
70
+ # test: Dataset({
71
+ # features: ['id', 'premise', 'hypothesis', 'label', 'heuristics', 'number_of_NPs', 'semtag'],
72
+ # num_rows: 720
73
+ # })
74
+ # })
75
+
76
+ dataset: ds.DatasetDict = ds.load_dataset("hpprc/janli", name="original")
77
+ print(dataset)
78
+ # DatasetDict({
79
+ # train: Dataset({
80
+ # features: ['id', 'sentence_A_Ja', 'sentence_B_Ja', 'entailment_label_Ja', 'heuristics', 'number_of_NPs', 'semtag'],
81
+ # num_rows: 13680
82
+ # })
83
+ # test: Dataset({
84
+ # features: ['id', 'sentence_A_Ja', 'sentence_B_Ja', 'entailment_label_Ja', 'heuristics', 'number_of_NPs', 'semtag'],
85
+ # num_rows: 720
86
+ # })
87
+ # })
88
+ ```
89
+
90
+
91
+ #### base
92
+
93
+ An example of looks as follows:
94
+
95
+ ```json
96
+ {
97
+ 'id': 12,
98
+ 'premise': 'θ‹₯θ€…γŒγƒ•γƒƒγƒˆγƒœγƒΌγƒ«ιΈζ‰‹γ‚’θ¦‹γ¦γ„γ‚‹',
99
+ 'hypothesis': 'γƒ•γƒƒγƒˆγƒœγƒΌγƒ«ιΈζ‰‹γ‚’θ‹₯θ€…γŒθ¦‹γ¦γ„γ‚‹',
100
+ 'label': 0,
101
+ 'heuristics': 'overlap-full',
102
+ 'number_of_NPs': 2,
103
+ 'semtag': 'scrambling'
104
+ }
105
+ ```
106
+
107
+ #### original
108
+
109
+ An example of looks as follows:
110
+
111
+ ```json
112
+ {
113
+ 'id': 12,
114
+ 'sentence_A_Ja': 'θ‹₯θ€…γŒγƒ•γƒƒγƒˆγƒœγƒΌγƒ«ιΈζ‰‹γ‚’θ¦‹γ¦γ„γ‚‹',
115
+ 'sentence_B_Ja': 'γƒ•γƒƒγƒˆγƒœγƒΌγƒ«ιΈζ‰‹γ‚’θ‹₯θ€…γŒθ¦‹γ¦γ„γ‚‹',
116
+ 'entailment_label_Ja': 0,
117
+ 'heuristics': 'overlap-full',
118
+ 'number_of_NPs': 2,
119
+ 'semtag': 'scrambling'
120
+ }
121
+ ```
122
+
123
+ ### Data Fields
124
+
125
+ #### base
126
+
127
+ A version adopting the column names of a typical NLI dataset.
128
+
129
+ - `id`: The number of the sentence pair.
130
+ - `premise`: The premise (sentence_A_Ja).
131
+ - `hypothesis`: The hypothesis (sentence_B_Ja).
132
+ - `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).
133
+ - `heuristics`: The heuristics (structural pattern) tag. The tags are: subsequence, constituent, full-overlap, order-subset, and mixed-subset.
134
+ - `number_of_NPs`: The number of noun phrase in a sentence.
135
+ - `semtag`: The linguistic phenomena tag.
136
+
137
+ #### original
138
+
139
+ The original version retaining the unaltered column names.
140
+
141
+ - `id`: The number of the sentence pair.
142
+ - `sentence_A_Ja`: The premise.
143
+ - `sentence_B_Ja`: The hypothesis.
144
+ - `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
145
+ - `heuristics`: The heuristics (structural pattern) tag. The tags are: subsequence, constituent, full-overlap, order-subset, and mixed-subset.
146
+ - `number_of_NPs`: The number of noun phrase in a sentence.
147
+ - `semtag`: The linguistic phenomena tag.
148
+
149
+
150
+ ### Data Splits
151
+
152
+ | name | train | validation | test |
153
+ | -------- | -----: | ---------: | ---: |
154
+ | base | 13,680 | | 720 |
155
+ | original | 13,680 | | 720 |
156
+
157
+
158
+
159
  ### Annotations
160
 
161
  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.