This article was accepted into the corpus but its outbound wikilinks were never NER-processed — typical at the deepest BFS hop or when the run's entity cap was reached. No expansion funnel to show.
| Sejong Corpus | |
|---|---|
| Name | Sejong Corpus |
| Country | South Korea |
| Language | Korean |
| Created | 1990s |
| Developers | National Institute of Korean Language; Academy of Korean Studies; Yonsei University; Korea University; KAIST |
| Size | multi-genre |
| Format | annotated text |
Sejong Corpus The Sejong Corpus is a major annotated Korean language resource developed to support computational linguistics, natural language processing, and language technology in South Korea, designed and maintained by national and academic institutions. It serves researchers working with corpora from multiple domains and has influenced projects at organizations such as Naver Corporation, Samsung Electronics, LG Electronics, Microsoft Research Asia, and Google Research. The corpus interfaces with standards and projects tied to institutions like the Korean Language Society, Ministry of Culture, Sports and Tourism (South Korea), Asian Language Resources, ACL and regional research groups at Seoul National University and Sogang University.
The project originated to create a balanced, machine-readable dataset for Korean morphology, syntax, and semantics with annotations compatible with tools from Stanford University, University of Pennsylvania, MIT, Carnegie Mellon University, University of Cambridge, and University of Oxford. It includes written text genres such as newswire reported by outlets like Yonhap News Agency and The Korea Herald, literary text referencing works by authors in the Korean Literature Translation Institute, and technical documents similar to material produced at Korea Institute of Science and Technology and Electronics and Telecommunications Research Institute. The resource aligns with international efforts exemplified by projects at Linguistic Data Consortium and ELRA and has been cited in conferences including ACL (annual meeting), COLING, EMNLP, ICPR, and LREC.
Initial planning occurred in the 1990s through collaborations among National Institute of Korean Language, Academy of Korean Studies, Yonsei University, and industry partners such as POSCO and KT Corporation. Funding and oversight involved agencies like Korea Institute of Science and Technology Information and initiatives coordinated with the Ministry of Culture, Sports and Tourism (South Korea). Technical leadership drew on methods from teams at KAIST and algorithms influenced by work at Max Planck Institute for Psycholinguistics, University of Tokyo, and Chinese Academy of Sciences. Over successive phases, annotation schemes were refined following best practices documented by ISO committees and influenced by treebanks like those from Penn Treebank and corpora developed at Brown University and University of Pennsylvania.
The corpus contains a mix of genres: journalism (newswire), literature, bureaucracy-style text similar to publications from Blue House (South Korea), transcripts of broadcasts from Korean Broadcasting System and Munhwa Broadcasting Corporation, and domain-specific texts comparable to outputs of Samsung Medical Center and KHIDI. Annotations include morphological tags (part-of-speech), phrase structure trees, dependency relations, named entities, and semantic role labels, using conventions comparable to frameworks from Universal Dependencies and methodologies in datasets from OntoNotes and FrameNet. The annotation pipeline incorporated manual adjudication by linguists affiliated with Sejong Center for the Performing Arts and computational verification tools developed by research groups at Hanyang University, Pohang University of Science and Technology, and Chonnam National University.
Researchers and companies have used the corpus for tasks such as part-of-speech tagging, syntactic parsing, machine translation, information extraction, named entity recognition, and language modeling, informing products at Naver Corporation, Kakao Corporation, Samsung SDS, and academic systems in projects at KAIST and Seoul National University. It has supported shared tasks in workshops organized by ACL and LREC, powered baseline models compared against neural architectures from teams at Facebook AI Research and Google DeepMind, and enabled evaluation of speech recognition pipelines akin to those at Nuance Communications and IBM Watson. Educational use includes curricula at Yonsei University, Korea University, and training data for competitions run by Kaggle and regional hackathons hosted by Seoul Metropolitan Government.
Distribution and licensing were coordinated by the National Institute of Korean Language with agreements negotiated among universities and publishers similar to arrangements seen with Linguistic Data Consortium and ELRA. Access tiers permit academic research and commercial licensing under terms reflecting intellectual property policies comparable to those of Creative Commons variants and institutional repositories used by DANBI and national archives. Data release schedules and updates have been announced at venues such as International Conference on Computational Linguistics and workshops at Sejong University and Korean Studies Association.
The corpus has been a benchmark in Korean NLP, enabling evaluation of parsing accuracy, tagset consistency, and cross-genre generalization in studies presented at ACL (annual meeting), EMNLP, COLING, and Korean Society of Computational Linguistics conferences. It influenced subsequent resources and treebanks developed at KAIST Treebank projects and informed multilingual initiatives linked to Universal Dependencies conversions and datasets curated by Linguistic Data Consortium. The resource’s availability improved baseline performance in commercial systems by organizations like Naver Corporation and academic labs at Seoul National University, helping close the gap between Korean and high-resource languages in tasks that mirror evaluations at GLUE and SuperGLUE.
Category:Korean language corpora