Unlocking Korean Verbs: A User-Friendly Exploration into the Verb Lexicon

A web interface and Python library (pySejongFrame) for the subcategorization frames of the Sejong verb dictionary

NAACL 2025 Demo

Seohyun Song1,*, Eunkyul Leah Jo2,*, Yige Chen3,*, Jeen-Pyo Hong4,*, Kyuwon Kim5, Jin Wee6, Miyoung Kang6, KyungTae Lim7,†, Jungyeul Park2,†, Chulwoo Park8

1Seoultech, South Korea 2The University of British Columbia, Canada 3The Chinese University of Hong Kong, Hong Kong 442dot Inc., South Korea 5Seoul National University, South Korea 6National Institute of Korean Language, South Korea 7KAIST, South Korea 8Anyang University, South Korea

*Equal contribution   †Corresponding authors

Paper (ACL Anthology) PDF arXiv
Screenshot of the word details page for 확립하다 (hwaglibhada, 'establish'): the headword, its variant 확립을 하다, structure N.하 and origin 確立, then Sense 1 with semantic class 종료행위 and translation 'establish, settle, fix', Frame 1 'X=N0-이 Y=N1-을 V' with X as agent (AGT) and Y as theme (THM), and three Korean example sentences.
Figure 3. Interface screenshot of a word details page for 확립하다 hwaglibhada (‘establish’), showing morphological, semantic, and syntactic frame information (frame X=N0-이 Y=N1-을 V with X = AGENT and Y = THEME) together with sentence examples, all read from the Sejong dictionary XML entry.

Abstract

The Sejong dictionary dataset offers a valuable resource, providing extensive coverage of morphology, syntax, and semantic representation. This dataset can be utilized to explore linguistic information in greater depth. The labeled linguistic structures within this dataset form the basis for uncovering relationships between words and phrases and their associations with target verbs. This paper introduces a user-friendly web interface designed for the collection and consolidation of verb-related information, with a particular focus on subcategorization frames. Additionally, it outlines our efforts in mapping this information by aligning subcategorization frames with corresponding illustrative sentence examples. Furthermore, we provide a Python library that would simplify syntactic parsing and semantic role labeling. These tools are intended to assist individuals interested in harnessing the Sejong dictionary dataset to develop applications for Korean language processing.

Keywords Korean verb lexicon Sejong dictionary Subcategorization frames Valency Semantic role labeling (SRL) Korean PropBank Korean FrameNet pySejongFrame Korean NLP

System

Frame projection view for the sentence 많은 사람들이 사회의 질서를 확립하려는 노력을 기울여 왔다: entry, sense, frame X=N0-이 Y=N1-을 V, selectional restrictions (X agent: human or human group; Y theme: factual proposition, norm, custom, institution and others), and a brat visualization labeling 많은 사람들이 as AGT, 사회의 질서를 as THM and 확립하려는 as TARGET.
Figure 4. Subcategorization frame of the verb with semantic role information for its arguments, and chunked argument boundaries in the example sentence ‘Many people have made efforts to establish order in society’, visualized with brat: [많은 사람들이]AGENT [사회의 질서를]THEME [확립하려는]TARGET.

The Sejong dictionary, built in the 21st Century Sejong Project, stores each Korean verb as an XML file with morphology, senses (semantic class and English translation), subcategorization frames, selectional restrictions with semantic roles such as AGENT (AGT) and THEME (THM), and example sentences. The paper turns this resource into two tools:

  1. Web interface. Verbs are listed in Korean alphabetical order and can be searched by verb, frame, argument, morphological inflection, and semantic group. Each word page renders the XML tags (orthography, morphology, senses, frames, selectional restrictions, examples) for users without programming experience.
  2. Frame-to-sentence projection. Argument boundaries in each example sentence are found by heuristic chunking at postpositions, which in Korean usually mark the end of arguments and modifiers, with dependency structures as a fallback. The postposition of each chunk is compared with the postpositions in the frame (e.g. -이, -을) to decide whether it is an argument, and the frame's semantic roles are assigned to it.
  3. Python library (pySejongFrame). A corpus reader for the Sejong Frame Dictionary with direct loading, lazy loading (indexing postponed until the data is queried), and NLTK integration, plus queries for entries (e.g. entries("수정하다.vv")), senses, semantic classes, translations, frames, and arguments.

Resource Overview

  • Coverage. The Sejong dictionary describes 15,181 verbs with 1.812 subcategorization frames per verb on average, far more verbs than Korean PropBank (2,749 verbs, 1.408 frames) or the NIKL SRL lexicon (3,660 verbs, 1.593 frames).
  • Contribution. A corpus of Sejong dictionary example sentences annotated with argument–predicate boundaries, plus a web interface and a Python library for accessing the dictionary.
  • Distribution. The National Institute of Korean Language, which holds the Sejong Project copyright, granted permission to distribute these derivative materials and will host the web system; the dictionary files for the library must still be obtained from the institute.
  • Uses. Lexicalized parsing, semantic role labeling, and FrameNet parsing for Korean, and a planned Korean VerbNet that harmonizes Sejong, PropBank, and FrameNet frames.
Table 1. Basic statistics of Korean verb lexicons: number of verbs and average number of frames per verb.
SejongK-PropBankNIKL
# of verbs15,1812,7493,660
avg. # of frames1.8121.4081.593

BibTeX

@inproceedings{song-etal-2025-unlocking,
    title = "Unlocking {K}orean Verbs: A User-Friendly Exploration into the Verb Lexicon",
    author = "Song, Seohyun  and
      Jo, Eunkyul Leah  and
      Chen, Yige  and
      Hong, Jeen-Pyo  and
      Kim, Kyuwon  and
      Wee, Jin  and
      Miyoung, Kang  and
      Lim, KyungTae  and
      Park, Jungyeul  and
      Park, Chulwoo",
    editor = "Dziri, Nouha  and
      Ren, Sean (Xiang)  and
      Diao, Shizhe",
    booktitle = "Proceedings of the 2025 Conference of the Nations of the {A}mericas Chapter of the {A}ssociation for {C}omputational {L}inguistics: Human Language Technologies (System Demonstrations)",
    month = apr,
    year = "2025",
    address = "Albuquerque, New Mexico",
    publisher = "Association for Computational Linguistics",
    url = "https://aclanthology.org/2025.naacl-demo.24/",
    doi = "10.18653/v1/2025.naacl-demo.24",
    pages = "287--294",
    ISBN = "979-8-89176-191-9"
}
More projects from KAIST MLP Lab →