CLIcK / README.md
taeminlee's picture
init
8a95bfb
metadata
task_categories:
  - multiple-choice
language:
  - ko
tags:
  - Culture
  - Language
size_categories:
  - 1K<n<10K
configs:
  - config_name: KL_Grammar
    data_files:
      - path:
          - Dataset/Language/Grammar/Grammar_CSAT.json
          - Dataset/Language/Grammar/Grammar_TOPIK.json
          - Dataset/Language/Grammar/Grammar_Kedu.json
        split: test
  - config_name: KL_Textual
    data_files:
      - path:
          - Dataset/Language/Textual/Textual_TOPIK.json
          - Dataset/Language/Textual/Textual_CSAT.json
        split: test
  - config_name: KL_Functional
    data_files:
      - path:
          - Dataset/Language/Functional/Functional_Kedu.json
          - Dataset/Language/Functional/Functional_PSE.json
          - Dataset/Language/Functional/Functional_CSAT.json
        split: test
  - config_name: KC_Law
    data_files:
      - path:
          - Dataset/Culture/Korean Law/Law_KIIP.json
          - Dataset/Culture/Korean Law/Law_PSAT.json
        split: test
  - config_name: KC_Popular
    data_files:
      - path:
          - Dataset/Culture/Korean Popular/Popular_Kedu.json
          - Dataset/Culture/Korean Popular/Popular_KIIP.json
        split: test
  - config_name: KC_Politics
    data_files:
      - path:
          - Dataset/Culture/Korean Politics/Politics_Kedu.json
          - Dataset/Culture/Korean Politics/Politics_KIIP.json
        split: test
  - config_name: KC_Geography
    data_files:
      - path:
          - Dataset/Culture/Korean Geography/Geography_KIIP.json
          - Dataset/Culture/Korean Geography/Geography_Kedu.json
          - Dataset/Culture/Korean Geography/Geography_CSAT.json
        split: test
  - config_name: KC_Economy
    data_files:
      - path:
          - Dataset/Culture/Korean Economy/Economy_KIIP.json
          - Dataset/Culture/Korean Economy/Economy_Kedu.json
        split: test
  - config_name: KC_History
    data_files:
      - path:
          - Dataset/Culture/Korean History/History_Kedu.json
          - Dataset/Culture/Korean History/History_PSE.json
          - Dataset/Culture/Korean History/History_KHB.json
        split: test
  - config_name: KC_Society
    data_files:
      - path:
          - Dataset/Culture/Korean Society/Society_Kedu.json
          - Dataset/Culture/Korean Society/Society_KIIP.json
        split: test
  - config_name: KC_Tradition
    data_files:
      - path:
          - Dataset/Culture/Korean Tradition/Tradition_Kedu.json
          - Dataset/Culture/Korean Tradition/Tradition_KIIP.json
        split: test

This dataset is the same as https://huggingface.co/datasets/EunsuKim/CLIcK. This dataset has been subdivided for simplified viewing and evaluation.

CLIcK πŸ‡°πŸ‡·πŸ§ 

Evaluation of Cultural and Linguistic Intelligence in Korean

Dataset Paper

Introduction πŸŽ‰

CLIcK (Cultural and Linguistic Intelligence in Korean) is a comprehensive dataset designed to evaluate cultural and linguistic intelligence in the context of Korean language models. In an era where diverse language models are continually emerging, there is a pressing need for robust evaluation datasets, especially for non-English languages like Korean. CLIcK fills this gap by providing a rich, well-categorized dataset focusing on both cultural and linguistic aspects, enabling a nuanced assessment of Korean language models.

News πŸ“°

  • [LREC-COLING] Our paper introducing CLIcK has been accepted to LREC-COLING 2024!πŸŽ‰

Dataset Description πŸ“Š

The CLIcK benchmark comprises two broad categories: Culture and Language, which are further divided into 11 fine-grained subcategories.

Categories πŸ“‚

  • Language πŸ—£οΈ

    • Textual Knowledge
    • Grammatical Knowledge
    • Functional Knowledge
  • Culture 🌍

    • Korean Society
    • Korean Tradition
    • Korean Politics
    • Korean Economy
    • Korean Law
    • Korean History
    • Korean Geography
    • Korean Popular Culture (K-Pop)

Construction πŸ—οΈ

CLIcK was developed using two human-centric approaches:

  1. Reclassification of official and well-designed exam data into our defined categories.
  2. Generation of questions using ChatGPT, based on official educational materials from the Korean Ministry of Justice, followed by our own validation process.

Structure πŸ›οΈ

The dataset is organized as follows, with each subcategory containing relevant JSON files:

πŸ“¦CLIcK
 └─ Dataset
    β”œβ”€ Culture
    β”‚  β”œβ”€ [Each cultural subcategory with associated JSON files]
    └─ Language
       β”œβ”€ [Each language subcategory with associated JSON files]

Exam Code Descriptions πŸ“œ

  • KIIP: Korea Immigration & Integration Program (Website)
  • CSAT: College Scholastic Ability Test for Korean (Website)
  • Kedu: Test of Teaching Korean as a Foreign Language exams (Website)
  • PSE: Public Service Exam for 9th grade
  • TOPIK: Test of Proficiency in Korean (Website)
  • KHB: Korean History Exam Basic (Website)
  • PSAT: Public Service Aptitude Test in Korea

Results

Models Average Accuracy (Korean Culture) Average Accuracy (Korean Language)
Polyglot-Ko 1.3B 32.71% 22.88%
Polyglot-Ko 3.8B 32.90% 22.38%
Polyglot-Ko 5.8B 33.14% 23.27%
Polyglot-Ko 12.8B 33.40% 22.24%
KULLM 5.8B 33.79% 23.50%
KULLM 12.8B 33.51% 23.78%
KoAlpaca 5.8B 32.33% 23.87%
KoAlpaca 12.8B 33.80% 22.42%
LLaMA-Ko 7B 33.26% 25.69%
LLaMA 7B 35.44% 27.17%
LLaMA 13B 36.22% 26.71%
GPT-3.5 49.30% 42.32%
Claude2 51.72% 45.39%

Dataset Link πŸ”—

The CLIcK dataset is available on the Hugging Face Hub: CLIcK Dataset

Citation πŸ“

If you use CLIcK in your research, please cite our paper:

@misc{kim2024click,
      title={CLIcK: A Benchmark Dataset of Cultural and Linguistic Intelligence in Korean}, 
      author={Eunsu Kim and Juyoung Suk and Philhoon Oh and Haneul Yoo and James Thorne and Alice Oh},
      year={2024},
      eprint={2403.06412},
      archivePrefix={arXiv},
      primaryClass={cs.CL}
}

Contact πŸ“§

For any questions or inquiries, please contact [email protected].