# K-UniMorph: Korean Universal Morphology and its Feature Schema

**Authors:** Eunkyul Leah Jo, Kyuwon Kim, Xihan Wu, KyungTae Lim, Jungyeul Park, Chulwoo Park  
**arXiv:** [2305.06335](https://arxiv.org/abs/2305.06335)  
**Date:** 2023-05-10  
**Summary page:** https://www.summarizepaper.com/en/paper/2305.06335v3/

## Key points

- Authors propose a new Universal Morphology dataset for the Korean language
- Introduce the K-UniMorph dataset to address underrepresentation of Korean in morphological paradigms
- Adopt a morphological feature schema from previous works by Sylak-Glassman et al. (2015) and Sylak-Glassman (2016)
- Extract inflected verb forms from the Sejong morphologically analyzed corpus
- Focus on annotating morphological data for verbs, separating postpositions from substantive elements
- Detailed explanations on how to extract inflected verbal forms and grammatical criteria
- Morphological schema includes four types of verbal endings: sentence final ending (ef), non-final ending (ep), conjunctive ending (ec), and modifier ending (etm)
- Discuss two grammatical categories: evidentiality and interrogativity, reflected in sentence final endings denoting declarative or interrogative forms
- Conclude by discussing future perspectives on Korean morphological paradigms and the dataset created

## Summary

In this paper, the authors propose a new Universal Morphology dataset for the Korean language. To address the underrepresentation of Korean in the field of morphological paradigms compared to other languages, they introduce the K-UniMorph dataset which aims to preserve its distinct characteristics. The authors adopt a morphological feature schema from previous works by Sylak-Glassman et al. (2015) and Sylak-Glassman (2016) for the Korean language. They extract inflected verb forms from the Sejong morphologically analyzed corpus - one of the largest annotated corpora for Korean consisting of over 0.6 million sentences and 9.5 million words - and investigate its correctness during data creation. Additionally, they generate morphological schemata based on their findings and focus specifically on annotating morphological data for verbs (V), separating postpositions from substantive elements like noun phrases. The paper provides detailed explanations on how to extract inflected verbal forms and outlines each grammatical criterion in detail for these forms. It also presents a morphological schema for Korean UniMorph that incorporates features from Sylak-Glassman et al. (2015) and Sylak-Glassman (2016). This schema includes four types of verbal endings: sentence final ending (ef), non-final ending (ep), conjunctive ending (ec), and modifier ending (etm). Furthermore, it discusses two grammatical categories: evidentiality and interrogativity which reflect source of information conveyed in a proposition or indicate whether a statement or question is being expressed respectively with different sentence final endings denoting declarative or interrogative forms. The authors conclude by discussing future perspectives on Korean morphological paradigms and the dataset they have created. Overall, this paper presents a comprehensive and detailed approach to creating a Universal Morphology dataset for the Korean language offering valuable insights into its morphological features as well as potential directions for further research in this area.