Datasets:
Tasks:
Text Classification
Modalities:
Text
Formats:
parquet
Sub-tasks:
natural-language-inference
Languages:
English
Size:
100K - 1M
License:
Create defeasible-nli.py
Browse files- defeasible-nli.py +60 -0
defeasible-nli.py
ADDED
@@ -0,0 +1,60 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
import datasets
|
2 |
+
import json
|
3 |
+
import os
|
4 |
+
|
5 |
+
citation='''
|
6 |
+
@inproceedings{rudinger-etal-2020-thinking,
|
7 |
+
title = "Thinking Like a Skeptic: Defeasible Inference in Natural Language",
|
8 |
+
author = "Rudinger, Rachel and
|
9 |
+
Shwartz, Vered and
|
10 |
+
Hwang, Jena D. and
|
11 |
+
Bhagavatula, Chandra and
|
12 |
+
Forbes, Maxwell and
|
13 |
+
Le Bras, Ronan and
|
14 |
+
Smith, Noah A. and
|
15 |
+
Choi, Yejin",
|
16 |
+
booktitle = "Findings of the Association for Computational Linguistics: EMNLP 2020",
|
17 |
+
month = nov,
|
18 |
+
year = "2020",
|
19 |
+
address = "Online",
|
20 |
+
publisher = "Association for Computational Linguistics",
|
21 |
+
url = "https://www.aclweb.org/anthology/2020.findings-emnlp.418",
|
22 |
+
doi = "10.18653/v1/2020.findings-emnlp.418",
|
23 |
+
pages = "4661--4675"
|
24 |
+
}
|
25 |
+
'''
|
26 |
+
|
27 |
+
class DefeasibleNLIConfig(datasets.BuilderConfig):
|
28 |
+
citation=citation
|
29 |
+
|
30 |
+
configs = ['atomic','snli','social']
|
31 |
+
splits=['train', 'test', 'dev']
|
32 |
+
_URLs = {(f,s):f"https://huggingface.co/datasets/metaeval/defeasible-nli/resolve/main/{f}_{s}.jsonl" for f in configs for s in splits}
|
33 |
+
|
34 |
+
class DefeasibleNLI(datasets.GeneratorBasedBuilder):
|
35 |
+
|
36 |
+
BUILDER_CONFIGS = [
|
37 |
+
DefeasibleNLIConfig(
|
38 |
+
name=n,
|
39 |
+
data_dir=n
|
40 |
+
) for n in configs
|
41 |
+
]
|
42 |
+
|
43 |
+
def _split_generators(self, dl_manager: datasets.DownloadManager):
|
44 |
+
path = lambda split: dl_manager.download(_URLs[self.config.name,split])
|
45 |
+
return [ datasets.SplitGenerator(name=name, gen_kwargs={'path':path(split),'split':split})
|
46 |
+
for name,split in zip([datasets.Split.TRAIN,datasets.Split.VALIDATION,datasets.Split.TEST],
|
47 |
+
['train','dev','test'])]
|
48 |
+
|
49 |
+
def _info(self):
|
50 |
+
return datasets.DatasetInfo()
|
51 |
+
|
52 |
+
def _generate_examples(self,path,split):
|
53 |
+
"""Yields examples."""
|
54 |
+
with open(path, "r", encoding="utf-8") as f:
|
55 |
+
for id_, line in enumerate(f):
|
56 |
+
line_dict = json.loads(line)
|
57 |
+
if not line_dict['UpdateTypeImpossible']:
|
58 |
+
fields = ["Premise","Hypothesis","Update","UpdateType"]#,"UpdateTypeImpossible","UpdateTypeImpossibleReason"]
|
59 |
+
line_dict = {k:v for k,v in line_dict.items() if k in fields}
|
60 |
+
yield id_, line_dict
|