Datasets:
skt
/

Modalities:
Text
Formats:
json
Languages:
Korean
ArXiv:
Libraries:
Datasets
pandas
License:
kobest_v1 / README.md
korca's picture
add citation info
8f15a82
|
raw
history blame
6.1 kB
metadata
pretty_name: KoBEST
annotations_creators:
  - expert-generated
language_creators:
  - expert-generated
languages:
  - ko
licenses:
  - cc-by-sa-4-0
multilinguality:
  - monolingual
size_categories:
  - 10K<n<100K
source_datasets:
  - original

Dataset Card for KoBEST

Table of Contents

Dataset Description

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': '๋‘์•„ ๋ฆฌํŒŒ(Dua Lipa, 1995๋…„ 8์›” 22์ผ ~ )๋Š” ์ž‰๊ธ€๋žœ๋“œ์˜ ์‹ฑ์–ด์†ก๋ผ์ดํ„ฐ, ๋ชจ๋ธ์ด๋‹ค. BBC ์‚ฌ์šด๋“œ ์˜ค๋ธŒ 2016 ๋ช…๋‹จ์— ๋…ธ๋ฏธ๋‹›๋˜์—ˆ๋‹ค. ์‹ฑ๊ธ€ "Be the One"๊ฐ€ ์˜๊ตญ ์‹ฑ๊ธ€ ์ฐจํŠธ 9์œ„๊นŒ์ง€ ์˜ค๋ฅด๋Š” ๋“ฑ ์„ฑ๊ณผ๋ฅผ ๋ณด์—ฌ์ฃผ์—ˆ๋‹ค.',
 'question': '๋‘์•„ ๋ฆฌํŒŒ๋Š” ์˜๊ตญ์ธ์ธ๊ฐ€?',
 'label': 1}

KB-COPA

An example of a data point looks as follows.

{'premise': '๋ฌผ์„ ์˜ค๋ž˜ ๋“์˜€๋‹ค.',
 'question': '๊ฒฐ๊ณผ',
 'alternative_1': '๋ฌผ์˜ ์–‘์ด ๋Š˜์–ด๋‚ฌ๋‹ค.',
 'alternative_2': '๋ฌผ์˜ ์–‘์ด ์ค„์–ด๋“ค์—ˆ๋‹ค.',
 'label': 1}

KB-WiC

An example of a data point looks as follows.

{'word': '์–‘๋ถ„',
 'context_1': 'ํ† ์–‘์— [์–‘๋ถ„]์ด ํ’๋ถ€ํ•˜์—ฌ ๋‚˜๋ฌด๊ฐ€ ์ž˜ ์ž๋ž€๋‹ค.	',
 'context_2': 'ํƒœ์•„๋Š” ๋ชจ์ฒด๋กœ๋ถ€ํ„ฐ [์–‘๋ถ„]๊ณผ ์‚ฐ์†Œ๋ฅผ ๊ณต๊ธ‰๋ฐ›๊ฒŒ ๋œ๋‹ค.',
 'label': 1}

KB-HellaSwag

An example of a data point looks as follows.

{'context': '๋ชจ์ž๋ฅผ ์“ด ํˆฌ์ˆ˜๊ฐ€ ํƒ€์ž์—๊ฒŒ ์˜จ ํž˜์„ ๋‹คํ•ด ๊ณต์„ ๋˜์ง„๋‹ค. ๊ณต์ด ํƒ€์ž์—๊ฒŒ ๋น ๋ฅธ ์†๋„๋กœ ๋‹ค๊ฐ€์˜จ๋‹ค. ํƒ€์ž๊ฐ€ ๊ณต์„ ๋ฐฐํŠธ๋กœ ์นœ๋‹ค. ๋ฐฐํŠธ์—์„œ ๊นก ์†Œ๋ฆฌ๊ฐ€ ๋‚œ๋‹ค. ๊ณต์ด ํ•˜๋Š˜ ์œ„๋กœ ๋‚ ์•„๊ฐ„๋‹ค.',
 'ending_1': '์™ธ์•ผ์ˆ˜๊ฐ€ ๋–จ์–ด์ง€๋Š” ๊ณต์„ ๊ธ€๋Ÿฌ๋ธŒ๋กœ ์žก๋Š”๋‹ค.',
 'ending_2': '์™ธ์•ผ์ˆ˜๊ฐ€ ๊ณต์ด ๋–จ์–ด์งˆ ์œ„์น˜์— ์ž๋ฆฌ๋ฅผ ์žก๋Š”๋‹ค.',
 'ending_3': '์‹ฌํŒ์ด ์•„์›ƒ์„ ์™ธ์นœ๋‹ค.',
 'ending_4': '์™ธ์•ผ์ˆ˜๊ฐ€ ๊ณต์„ ๋”ฐ๋ผ ๋›ฐ๊ธฐ ์‹œ์ž‘ํ•œ๋‹ค.',
 'label': 3}

KB-SentiNeg

An example of a data point looks as follows.

{'sentence': 'ํƒ๋ฐฐ์‚ฌ ์ •๋ง ๋งˆ์Œ์— ๋“ฌ',
 'label': 1}

Data Fields

KB-BoolQ

  • paragraph: a string feature
  • question: a string feature
  • label: a classification label, with possible values False(0) and True(1)

KB-COPA

  • premise: a string feature
  • question: a string feature
  • alternative_1: a string feature
  • alternative_2: a string feature
  • label: an answer candidate label, with possible values alternative_1(0) and alternative_2(1)

KB-WiC

  • target_word: a string feature
  • context_1: a string feature
  • context_2: a string feature
  • label: a classification label, with possible values False(0) and True(1)

KB-HellaSwag

  • target_word: a string feature
  • context_1: a string feature
  • context_2: a string feature
  • label: a classification label, with possible values False(0) and True(1)

KB-SentiNeg

  • sentence: a string feature
  • label: a classification label, with possible values Negative(0) and Positive(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.