Skip to content

README.md

Latest commit

 

History

History
69 lines (52 loc) · 3.87 KB

README.md

File metadata and controls

69 lines (52 loc) · 3.87 KB

Summary

UD_Korean-KSL is a dependency treebank of second-language (L2) Korean.

Introduction

The annotated sentences are sourced from three datasets: (1) the Kyung Hee dataset, with sentence IDs starting with KH and annotated with classroom proficiency levels (A1–C2); (2) the KoLLA dataset, with sentence IDs starting with KL and grouped as fb (foreign beginners), fi (foreign intermediates), and hb (heritage beginners); (3) argumentative essays, with sentence IDs starting with ARG and grouped by first language: CHN (Chinese), CZH (Czech), ENG (English), and JPN (Japanese).

Acknowledgements

We acknowledge the original data contributors: (1) the Kyung Hee dataset (credit to Jungyeul Park and Jung Hee Lee; note that its sentences are not used for further annotation); (2) the KoLLA dataset (credit to Markus Dickinson, Ross Israel, and Sun-Hee Lee); and (3) the argumentative essays (credit to Boo Kyung Jung and Jeong Eun Shin).

References

Please refer to the supplementary repo for more information on recent papers, annotation guidelines, and fine-tuned models for L2-Korean.

Changelog

  • 2026-04-24

    • Added 1,694 new sentences; Includes: 14,076 train, 1,480 dev, 1,813 test — total 17,369 sentences.
    • Fixed validation issues:
      • L3 Warning pron-det-without-prontype: Added missing PronType features to pronominal and determiner forms by assigning PronType=Prs to PRON (e.g., 나, 저, 우리), and PronType=Dem or PronType=Int to DET (e.g., 그, 이, 무엇), and updated the Korean language-specific validator settings to permit these PronType × UPOS combinations.
      • L3 Syntax obl-should-be-nmod: Corrected case-marked noun phrases attached to nominal heads by changing the dependency label from obl to nmod.

  • 2025-10-17

    • Includes: 12,721 train, 1,311 dev, 1,643 test — total 15,675 sentences.
    • Fixed validation issues:
      • L3 Syntax too-many-objects: Reanalyzed light-verb constructions (e.g., 이야기/계획/연습/추천/선택/치료/생각/극복/시작/경험 하다) by changing the noun dependency relation from obj to compound:lvc.
      • L3 Warning obl-should-be-nmod: Corrected case-marked noun phrases functioning as nominal modifiers by changing the label from obl to nmod.
      • Length mismatches: Resolved tokenization inconsistencies between surface forms and corresponding morphological analyses.
      • Comment=Case/Spelling/SpacingError (10th column): Temporarily retained learner error annotations in the MISC column; these entries will be reviewed and standardized in the next annotation phase.

  • 2025-07-03 (UD-KSL v1.3)

    • Added 2,998 new sentences
    • Introduced a semi-automated framework that identifies morphosyntactic constructions from XPOS sequences and aligns them with corresponding UPOS categories (db here)

  • 2025-02-25 (UD-KSL v1.2)

    • Added 5,447 new sentences
    • Introduced the Typo=Yes feature to indicate erroneous or unexpected word forms
    • Revised annotation guidelines to better align with the UD framework

  • 2024-11-11 v2.15

    • Initial release in Universal Dependencies
    • Minor update on the README

  • 2024-10-16

    • Updated the source data information

  • 2024-05-15 v2.14

    • Release in Universal Dependencies (dev-repo)
=== Machine-readable metadata (DO NOT REMOVE!) ================================
Data available since: UD v2.15
License: CC BY-SA 4.0
Includes text: yes
Parallel: no
Genre: learner-essays
Lemmas: manual native
UPOS: converted with corrections
XPOS: manual native
Features: converted from manual
Relations: manual native
Contributors: Sung, Hakyung; Shin, Gyu-Ho
Contributing: here
Contact: hksung001@gmail.com; Shin, ghshin@uic.edu
===============================================================================