pretty_name: KoBEST
annotations_creators:
- expert-generated
language_creators:
- expert-generated
language:
- ko
license:
- cc-by-sa-4.0
multilinguality:
- monolingual
size_categories:
- 10K<n<100K
source_datasets:
- original
configs:
- config_name: boolq
data_files: boolq/*
- config_name: copa
data_files: copa/*
- config_name: hellaswag
data_files: hellaswag/*
- config_name: sentineg
data_files: sentineg/*
- config_name: wic
data_files: wic/*
Dataset Card for KoBEST
Table of Contents
- Table of Contents
- Dataset Description
- Dataset Structure
- Dataset Creation
- Considerations for Using the Data
- Additional Information
Dataset Description
- Repository: https://github.com/SKT-LSL/KoBEST_datarepo
- Paper:
- Point of Contact: https://github.com/SKT-LSL/KoBEST_datarepo/issues
Dataset Summary
KoBEST is a Korean benchmark suite consists of 5 natural language understanding tasks that requires advanced knowledge in Korean.
Supported Tasks and Leaderboards
Boolean Question Answering, Choice of Plausible Alternatives, Words-in-Context, HellaSwag, Sentiment Negation Recognition
Languages
ko-KR
Dataset Structure
Data Instances
KB-BoolQ
An example of a data point looks as follows.
{'paragraph': '๋์ ๋ฆฌํ(Dua Lipa, 1995๋
8์ 22์ผ ~ )๋ ์๊ธ๋๋์ ์ฑ์ด์ก๋ผ์ดํฐ, ๋ชจ๋ธ์ด๋ค. BBC ์ฌ์ด๋ ์ค๋ธ 2016 ๋ช
๋จ์ ๋
ธ๋ฏธ๋๋์๋ค. ์ฑ๊ธ "Be the One"๊ฐ ์๊ตญ ์ฑ๊ธ ์ฐจํธ 9์๊น์ง ์ค๋ฅด๋ ๋ฑ ์ฑ๊ณผ๋ฅผ ๋ณด์ฌ์ฃผ์๋ค.',
'question': '๋์ ๋ฆฌํ๋ ์๊ตญ์ธ์ธ๊ฐ?',
'label': 1}
KB-COPA
An example of a data point looks as follows.
{'premise': '๋ฌผ์ ์ค๋ ๋์๋ค.',
'question': '๊ฒฐ๊ณผ',
'alternative_1': '๋ฌผ์ ์์ด ๋์ด๋ฌ๋ค.',
'alternative_2': '๋ฌผ์ ์์ด ์ค์ด๋ค์๋ค.',
'label': 1}
KB-WiC
An example of a data point looks as follows.
{'word': '์๋ถ',
'context_1': 'ํ ์์ [์๋ถ]์ด ํ๋ถํ์ฌ ๋๋ฌด๊ฐ ์ ์๋๋ค. ',
'context_2': 'ํ์๋ ๋ชจ์ฒด๋ก๋ถํฐ [์๋ถ]๊ณผ ์ฐ์๋ฅผ ๊ณต๊ธ๋ฐ๊ฒ ๋๋ค.',
'label': 1}
KB-HellaSwag
An example of a data point looks as follows.
{'context': '๋ชจ์๋ฅผ ์ด ํฌ์๊ฐ ํ์์๊ฒ ์จ ํ์ ๋คํด ๊ณต์ ๋์ง๋ค. ๊ณต์ด ํ์์๊ฒ ๋น ๋ฅธ ์๋๋ก ๋ค๊ฐ์จ๋ค. ํ์๊ฐ ๊ณต์ ๋ฐฐํธ๋ก ์น๋ค. ๋ฐฐํธ์์ ๊นก ์๋ฆฌ๊ฐ ๋๋ค. ๊ณต์ด ํ๋ ์๋ก ๋ ์๊ฐ๋ค.',
'ending_1': '์ธ์ผ์๊ฐ ๋จ์ด์ง๋ ๊ณต์ ๊ธ๋ฌ๋ธ๋ก ์ก๋๋ค.',
'ending_2': '์ธ์ผ์๊ฐ ๊ณต์ด ๋จ์ด์ง ์์น์ ์๋ฆฌ๋ฅผ ์ก๋๋ค.',
'ending_3': '์ฌํ์ด ์์์ ์ธ์น๋ค.',
'ending_4': '์ธ์ผ์๊ฐ ๊ณต์ ๋ฐ๋ผ ๋ฐ๊ธฐ ์์ํ๋ค.',
'label': 3}
KB-SentiNeg
An example of a data point looks as follows.
{'sentence': 'ํ๋ฐฐ์ฌ ์ ๋ง ๋ง์์ ๋ฌ',
'label': 1}
Data Fields
KB-BoolQ
paragraph
: astring
featurequestion
: astring
featurelabel
: a classification label, with possible valuesFalse
(0) andTrue
(1)
KB-COPA
premise
: astring
featurequestion
: astring
featurealternative_1
: astring
featurealternative_2
: astring
featurelabel
: an answer candidate label, with possible valuesalternative_1
(0) andalternative_2
(1)
KB-WiC
target_word
: astring
featurecontext_1
: astring
featurecontext_2
: astring
featurelabel
: a classification label, with possible valuesFalse
(0) andTrue
(1)
KB-HellaSwag
target_word
: astring
featurecontext_1
: astring
featurecontext_2
: astring
featurelabel
: a classification label, with possible valuesFalse
(0) andTrue
(1)
KB-SentiNeg
sentence
: astring
featurelabel
: a classification label, with possible valuesNegative
(0) andPositive
(1)
Data Splits
KB-BoolQ
- train: 3,665
- dev: 700
- test: 1,404
KB-COPA
- train: 3,076
- dev: 1,000
- test: 1,000
KB-WiC
- train: 3,318
- dev: 1,260
- test: 1,260
KB-HellaSwag
- train: 3,665
- dev: 700
- test: 1,404
KB-SentiNeg
- train: 3,649
- dev: 400
- test: 397
- test_originated: 397 (Corresponding training data where the test set is originated from.)
Dataset Creation
Curation Rationale
[More Information Needed]
Source Data
Initial Data Collection and Normalization
[More Information Needed]
Who are the source language producers?
[More Information Needed]
Annotations
Annotation process
[More Information Needed]
Who are the annotators?
[More Information Needed]
Personal and Sensitive Information
[More Information Needed]
Considerations for Using the Data
Social Impact of Dataset
[More Information Needed]
Discussion of Biases
[More Information Needed]
Other Known Limitations
[More Information Needed]
Additional Information
Dataset Curators
[More Information Needed]
Licensing Information
@misc{https://doi.org/10.48550/arxiv.2204.04541,
doi = {10.48550/ARXIV.2204.04541},
url = {https://arxiv.org/abs/2204.04541},
author = {Kim, Dohyeong and Jang, Myeongjun and Kwon, Deuk Sin and Davis, Eric},
title = {KOBEST: Korean Balanced Evaluation of Significant Tasks},
publisher = {arXiv},
year = {2022},
}
[More Information Needed]
Contributions
Thanks to @MJ-Jang for adding this dataset.