Korean FrameNet Expansion Based on Projection of Japanese FrameNet
Summary
Korean FrameNet previously consisted only of 4025 English FrameNet sentences rendered into Korean by trained translators. That process was expensive because translators had to keep every frame element intact, and errors required repeated revision. The resulting dataset was too small to serve as machine learning training data.
The authors propose a cheaper expansion route that exploits the structural closeness of Korean and Japanese. Both languages mark grammatical roles with postpositions and tolerate flexible word order. Annotated sentences from Japanese FrameNet are split into chunks at every start and end point of lexical units and frame elements. Translators with no frame-semantics background then translate these chunks without reordering, adding or removing boundaries. Finally, the stored annotations are automatically mapped back onto the Korean chunks.
This produced 1795 new annotated Korean sentences, and the double-checking translators reported none of them as unnatural. The project website also displays annotations through PubAnnotation and offers Korean-specific valence patterns with postpositions. Together, these changes make the resource more accessible for semantic role labeling work in Korean.
Key points
- The earlier Korean FrameNet (Park et al., 2014) held 4025 translated English FrameNet sentences, 7130 lexical units, and reused all 1019 Berkeley frame index entries unchanged.
- The new method projects Japanese FrameNet annotations by translating word chunks in their original order, so translators do not need to know frame semantics.
- Translators added 1795 Korean full-text annotated sentences and reported no cases of unnatural output from keeping the Japanese chunk order.
- Visualization uses PubAnnotation, which shows lexical units and frame-element roles as linked relations instead of relying on color tables.
- Valence patterns are encoded as frame elements together with the POS of their postpositions, and the data is publicly available at framenet.kaist.ac.kr.
Cite
@inproceedings{kim2016korean,
title = {{Korean FrameNet Expansion Based on Projection of Japanese FrameNet}},
author = {Kim, Jeong-uk and Hahm, Younggyun and Choi, Key-Sun},
booktitle = {Proceedings of COLING 2016, the 26th International Conference on Computational Linguistics: System Demonstrations},
year = {2016},
url = {https://aclanthology.org/C16-2037/}
}References
- Collin F Baker, Charles J Fillmore, and John B Lowe. 1998. The berkeley framenet project. In Proceedings of the 36th Annual Meeting of the Association for Computational Linguistics and 17th International Conference on Computational Linguistics-Volume 1, pages 86–90. Association for Computational Lin… find our card
- Charles Fillmore. 1982. Frame semantics. Linguistics in the morning calm, pages 111–137. find
- Jin-Dong Kim and Yue Wang. 2012. Pubannotation: a persistent and sharable corpus and annotation repository. In Proceedings of the 2012 Workshop on Biomedical Natural Language Processing, pages 202–205. Association for Computational Linguistics. find
- Kyoko Hirose Ohara, Seiko Fujii, Hiroaki Saito, Shun Ishizaki, Toshio Ohori, and Ryoko Suzuki. 2003. The japanese framenet project: A preliminary report. In Proceedings of pacific association for computational linguistics, pages 249–254. Citeseer. find
- Jungyeul Park, Sejin Nam, Youngsik Kim, Younggyun Hahm, Dosam Hwang, and Key-Sun Choi. 2014. Frame-semantic web: a case study for korean. In Proceedings of the 2014 International Conference on Posters & Demonstrations Track-Volume 1272, pages 257–260. CEUR-WS. org. find
- Sara Tonelli and Emanuele Pianta. 2008. Frame information transfer from english to italian. In LREC. find
This page in Markdown → https://md.research.elamur.ai/free/en/papers/kim-2016-korean-framenet-expansion-based-projection/