Update README.md
Browse files
README.md
CHANGED
@@ -1,4 +1,8 @@
|
|
1 |
---
|
|
|
|
|
|
|
|
|
2 |
dataset_info:
|
3 |
- config_name: pair
|
4 |
features:
|
@@ -25,7 +29,12 @@ dataset_info:
|
|
25 |
- name: hypothesis
|
26 |
dtype: string
|
27 |
- name: label
|
28 |
-
dtype:
|
|
|
|
|
|
|
|
|
|
|
29 |
splits:
|
30 |
- name: train
|
31 |
num_bytes: 138755142
|
@@ -115,4 +124,61 @@ configs:
|
|
115 |
path: triplet/dev-*
|
116 |
- split: test
|
117 |
path: triplet/test-*
|
|
|
|
|
|
|
|
|
|
|
|
|
118 |
---
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
---
|
2 |
+
language:
|
3 |
+
- en
|
4 |
+
multilinguality:
|
5 |
+
- monolingual
|
6 |
dataset_info:
|
7 |
- config_name: pair
|
8 |
features:
|
|
|
29 |
- name: hypothesis
|
30 |
dtype: string
|
31 |
- name: label
|
32 |
+
dtype:
|
33 |
+
class_label:
|
34 |
+
names:
|
35 |
+
'0': entailment
|
36 |
+
'1': neutral
|
37 |
+
'2': contradiction
|
38 |
splits:
|
39 |
- name: train
|
40 |
num_bytes: 138755142
|
|
|
124 |
path: triplet/dev-*
|
125 |
- split: test
|
126 |
path: triplet/test-*
|
127 |
+
task_categories:
|
128 |
+
- feature-extraction
|
129 |
+
- sentence-similarity
|
130 |
+
pretty_name: AllNLI
|
131 |
+
size_categories:
|
132 |
+
- 1M<n<10M
|
133 |
---
|
134 |
+
|
135 |
+
# Dataset Card for AllNLI
|
136 |
+
|
137 |
+
This dataset is a concatenation of the [SNLI](https://huggingface.co/datasets/stanfordnlp/snli) and [MultiNLI](https://huggingface.co/datasets/nyu-mll/multi_nli) datasets.
|
138 |
+
Despite originally being intended for Natural Language Inference (NLI), this dataset can be used for training/finetuning an embedding model for semantic textual similarity.
|
139 |
+
|
140 |
+
## Dataset Subsets
|
141 |
+
|
142 |
+
### `pair-class` subset
|
143 |
+
|
144 |
+
* Columns: "premise", "hypothesis", "label"
|
145 |
+
* Column types: `str`, `str`, `class` with {"0": "entailment", "1": "neutral", "2", "contradiction"}
|
146 |
+
* Examples:
|
147 |
+
```python
|
148 |
+
{'premise': 'A person on a horse jumps over a broken down airplane.', 'hypothesis': 'A person is training his horse for a competition.', 'label': 1}
|
149 |
+
```
|
150 |
+
* Collection strategy: Reading the premise, hypothesis and integer label from SNLI & MultiNLI datasets.
|
151 |
+
* Deduplified: Yes
|
152 |
+
|
153 |
+
### `pair-score` subset
|
154 |
+
|
155 |
+
* Columns: "sentence_1", "sentence_2", "label"
|
156 |
+
* Column types: `str`, `str`, `float`
|
157 |
+
* Examples:
|
158 |
+
```python
|
159 |
+
{'premise': 'A person on a horse jumps over a broken down airplane.', 'hypothesis': 'A person is training his horse for a competition.', 'label': 1.0}
|
160 |
+
```
|
161 |
+
* Collection strategy: Taking the `pair-class` subset and remapping "entailment", "neutral" and "contradiction" to 1.0, 0.5 and 0.0, respectively.
|
162 |
+
* Deduplified: Yes
|
163 |
+
|
164 |
+
### `pair` subset
|
165 |
+
|
166 |
+
* Columns: "anchor", "positive"
|
167 |
+
* Column types: `str`, `str`
|
168 |
+
* Examples:
|
169 |
+
```python
|
170 |
+
{'anchor': 'A person on a horse jumps over a broken down airplane.', 'positive': 'A person is training his horse for a competition.'}
|
171 |
+
```
|
172 |
+
* Collection strategy: Reading the SNLI & MultiNLI datasets and considering the "premise" as the "anchor" and the "hypothesis" as the "positive" if the label is "entailment". The reverse ("entailment" as "anchor" and "premise" as "positive") is not included.
|
173 |
+
* Deduplified: Yes
|
174 |
+
|
175 |
+
### `triplet` subset
|
176 |
+
|
177 |
+
* Columns: "anchor", "positive", "negative"
|
178 |
+
* Column types: `str`, `str`, `str`
|
179 |
+
* Examples:
|
180 |
+
```python
|
181 |
+
{'anchor': 'A person on a horse jumps over a broken down airplane.', 'positive': 'A person is outdoors, on a horse.', 'negative': 'A person is at a diner, ordering an omelette.'}
|
182 |
+
```
|
183 |
+
* Collection strategy: Reading the SNLI & MultiNLI datasets, for each "premise" making a list of entailing and contradictory sentences using the dataset labels. Then, considering all possible triplets out of these entailing and contradictory lists. The reverse ("entailment" as "anchor" and "premise" as "positive") is also included.
|
184 |
+
* Deduplified: Yes
|