KoCHET: A Korean Cultural Heritage Corpus for Entity-related Tasks

Proceedings of the 29th International Conference on Computational Linguistics, pages 3496–3505 October 12–17, 2022. 3496 KOCHET : a Korean Cultural Heritage corpus for Entity-related Tasks Gyeongmin Kim∗, Jinsung Kim∗, Junyoung Son∗, Heuiseok Lim† Korea University, Korea {totoro4007, jin62304, s0ny, limhseok}@korea.ac.kr Abstract As digitized traditional cultural heritage docu- ments have rapidly increased, resulting in an increased need for preservation and manage- ment, practical recognition of entities and typ- ification of their classes has become essen- tial. To achieve this, we propose KOCHET - a Korean cultural heritage corpus for the typical entity-related tasks, i.e., named entity recognition (NER), relation extraction (RE), and entity typing (ET). Advised by cultural heritage experts based on the data construc- tion guidelines of government-affiliated orga- nizations, KOCHET consists of respectively 112,362, 38,765, 113,198 examples for NER, RE, and ET tasks, covering all entity types re- lated to Korean cultural heritage. Moreover, unlike the existing public corpora, modified redistribution can be allowed both domes- tic and foreign researchers. Our experimen- tal results make the practical usability of KO- CHET more valuable in terms of cultural her- itage. We also provide practical insights of KOCHET in terms of statistical and linguis- tic analysis. Our corpus is freely available at https://github.com/Gyeongmin47/KoCHET. 1 Introduction Recently there has been an increasing interest in the preservation of national historical artifacts and tra- ditional cultural heritage, and also grows up the im- portance of effective management of them through digitization and archival. As the amount of digi- tized information materials increases rapidly, in- formation extraction (IE) tasks in natural language processing (NLP), such as named entity recognition (NER), relation extraction (RE), and entity typing (ET), have become an essential and fundamental step in the field of historical document analysis. Despite the necessity of a well-refined entity- centric corpus specialized in domestic cultural her- ∗These authors have equally contributed to this work †Corresponding author itage, unfortunately, there no exists any cultural her- itage domain-specialized corpus in Korean. More- over, conventional entity-related systems deal only with a coarse set of entity types such as person, lo- cation, and organization which is significantly lim- ited in terms of application (Kim et al., 2020). This absence of cultural heritage domain-specialized corpus and narrow coverage of entity types hin- ders the effective digitization of domestic historical documents because training the model with general corpus for entity-related tasks cannot afford to learn enough significant entity types such as pagodas, historical sites and intangible heritage, and their relations. Furthermore, not in the cultural heritage domain, the existing entity-related datasets super- vised by the public institutions have a complicated procedure for data acquisition, and they are also re- stricted from modification and redistribution. These cumbersome procedures and restrictions have been stumbling blocks for researchers against the rapid increase in digitized cultural heritage materials over the past few decades. To address these difficulties against the conser- vation of Korean cultural heritage, we introduce a new dataset collection called KOCHET - Korean Cultural Heritage corpus for Entity-related Tasks, a high-quality Korean cultural heritage …

Content truncated to 500 words