Muennighoff
commited on
Commit
·
f65642d
1
Parent(s):
c9b4dc1
Add data
Browse files- data/xwinograd.tsv +0 -0
- xwinograd.py +147 -0
data/xwinograd.tsv
ADDED
The diff for this file is too large to render.
See raw diff
|
|
xwinograd.py
ADDED
@@ -0,0 +1,147 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
# coding=utf-8
|
2 |
+
|
3 |
+
# Lint as: python3
|
4 |
+
"""XWinograd"""
|
5 |
+
|
6 |
+
|
7 |
+
import json
|
8 |
+
|
9 |
+
import pandas as pd
|
10 |
+
|
11 |
+
import datasets
|
12 |
+
|
13 |
+
|
14 |
+
logger = datasets.logging.get_logger(__name__)
|
15 |
+
|
16 |
+
|
17 |
+
_CITATION = """\
|
18 |
+
@misc{tikhonov2021heads,
|
19 |
+
title={It's All in the Heads: Using Attention Heads as a Baseline for Cross-Lingual Transfer in Commonsense Reasoning},
|
20 |
+
author={Alexey Tikhonov and Max Ryabinin},
|
21 |
+
year={2021},
|
22 |
+
eprint={2106.12066},
|
23 |
+
archivePrefix={arXiv},
|
24 |
+
primaryClass={cs.CL}
|
25 |
+
}
|
26 |
+
"""
|
27 |
+
|
28 |
+
_DESCRIPTION = """\
|
29 |
+
A multilingual collection of Winograd Schemas in six languages \
|
30 |
+
that can be used for evaluation of cross-lingual commonsense reasoning capabilities.
|
31 |
+
"""
|
32 |
+
|
33 |
+
|
34 |
+
#_URL = "https://github.com/yandex-research/crosslingual_winograd/blob/main/dataset.tsv"
|
35 |
+
_URL = "https://huggingface.co/datasets/muennighoff/xwinograd/resolve/main/data/mt/dataset.tsv"
|
36 |
+
|
37 |
+
import json
|
38 |
+
import random
|
39 |
+
|
40 |
+
def winogrande_format(row):
|
41 |
+
array = row["pronoun"]
|
42 |
+
position_idx = json.loads(array)[1][0]
|
43 |
+
# Turn unicode into proper chinese characters
|
44 |
+
sent = str(u"{}".format(row["sent"]))
|
45 |
+
start_idx = 0
|
46 |
+
for i, tok in enumerate(json.loads(row["toks"])):
|
47 |
+
tok = str(u"{}".format(tok))
|
48 |
+
cur_start_idx = sent.find(tok)
|
49 |
+
if i == position_idx:
|
50 |
+
break
|
51 |
+
sent = sent[cur_start_idx + len(tok):]
|
52 |
+
start_idx += cur_start_idx + len(tok)
|
53 |
+
# +1 to give room for an optional space
|
54 |
+
row["sentence"] = row["sent"][:start_idx] + row["sent"][start_idx:start_idx+len(tok)+1].replace(tok, "_") + row["sent"][start_idx+len(tok)+1:]
|
55 |
+
|
56 |
+
sol = json.loads(row["solution"])
|
57 |
+
|
58 |
+
cor_answer_idx = random.choice([1, 2])
|
59 |
+
incor_answer_idx = 2 if cor_answer_idx == 1 else 1
|
60 |
+
|
61 |
+
cor_answer = str(u"{}".format(sol[0][0])) if sol[0][-1] == True else str(u"{}".format(sol[1][0]))
|
62 |
+
incor_answer = str(u"{}".format(sol[0][0])) if sol[0][-1] == False else str(u"{}".format(sol[1][0]))
|
63 |
+
|
64 |
+
row[f"option{cor_answer_idx}"] = cor_answer
|
65 |
+
row[f"option{incor_answer_idx}"] = incor_answer
|
66 |
+
row["answer"] = cor_answer_idx
|
67 |
+
return row
|
68 |
+
|
69 |
+
|
70 |
+
class XWinograd(datasets.GeneratorBasedBuilder):
|
71 |
+
"""XWinograd"""
|
72 |
+
|
73 |
+
VERSION = datasets.Version("1.0.0")
|
74 |
+
BUILDER_CONFIGS = [
|
75 |
+
datasets.BuilderConfig(
|
76 |
+
name="en",
|
77 |
+
version=VERSION,
|
78 |
+
description="X",
|
79 |
+
),
|
80 |
+
datasets.BuilderConfig(
|
81 |
+
name="fr",
|
82 |
+
version=VERSION,
|
83 |
+
description="X",
|
84 |
+
),
|
85 |
+
datasets.BuilderConfig(
|
86 |
+
name="jp",
|
87 |
+
version=VERSION,
|
88 |
+
description="X",
|
89 |
+
),
|
90 |
+
datasets.BuilderConfig(
|
91 |
+
name="pt",
|
92 |
+
version=VERSION,
|
93 |
+
description="X",
|
94 |
+
),
|
95 |
+
datasets.BuilderConfig(
|
96 |
+
name="ru",
|
97 |
+
version=VERSION,
|
98 |
+
description="X",
|
99 |
+
),
|
100 |
+
datasets.BuilderConfig(
|
101 |
+
name="zh",
|
102 |
+
version=VERSION,
|
103 |
+
description="X",
|
104 |
+
),
|
105 |
+
]
|
106 |
+
DEFAULT_CONFIG_NAME = "en"
|
107 |
+
|
108 |
+
def _info(self):
|
109 |
+
return datasets.DatasetInfo(
|
110 |
+
description=_DESCRIPTION,
|
111 |
+
features=datasets.Features(
|
112 |
+
{
|
113 |
+
"sentence": datasets.Value("string"),
|
114 |
+
"option1": datasets.Value("string"),
|
115 |
+
"option2": datasets.Value("string"),
|
116 |
+
"answer": datasets.Value("string")
|
117 |
+
}
|
118 |
+
),
|
119 |
+
supervised_keys=None,
|
120 |
+
citation=_CITATION,
|
121 |
+
)
|
122 |
+
|
123 |
+
def _split_generators(self, dl_manager):
|
124 |
+
downloaded_files = dl_manager.download_and_extract(_URL)
|
125 |
+
return [
|
126 |
+
datasets.SplitGenerator(name=datasets.Split.VALIDATION, gen_kwargs={"filepath": downloaded_files}),
|
127 |
+
]
|
128 |
+
|
129 |
+
def _generate_examples(self, filepath):
|
130 |
+
"""This function returns the examples in the raw (text) form."""
|
131 |
+
logger.info("generating examples from = %s", filepath)
|
132 |
+
ds = pd.read_csv(
|
133 |
+
filepath, sep='\t', header=None,
|
134 |
+
names=["lang", "type", "original", "sent", "toks", "pronoun", "solution"]
|
135 |
+
)
|
136 |
+
if self.config.name:
|
137 |
+
ds = ds[ds["lang"] == self.config.name]
|
138 |
+
ds = ds.apply(winogrande_format, axis=1)
|
139 |
+
|
140 |
+
for idx, row in ds.iterrows():
|
141 |
+
yield idx, {
|
142 |
+
"sentence": row["sentence"],
|
143 |
+
"option1": row["option1"],
|
144 |
+
"option2": row["option2"],
|
145 |
+
"answer": row["answer"],
|
146 |
+
}
|
147 |
+
|