The KOLON System: Tools for Ontological Natural Language Processing in Korean

The KOLON System: Tools for Ontological Natural Language Processing in Korean Juliano Paiva Junho, Yumi Jo, and Hyopil Shin Department of Linguistics, Seoul National University, San 56-1, Sillim-dong, Gwanak-gu, Seoul, 151-742, S. Korea jjunho@gmail.com, jmocy@snu.ac.kr, hpshin@snu.ac.kr Abstract. This paper presents and analyzes the KOLON System, created to facilitate Ko- rean Natural Language Processing (KNLP) and to improve experimental results through the KOLON Ontology. It is currently under development at our Computational Linguistics Labo- ratory, and is based on previous works, namely the Mikrokosmos Ontology and 21st Century Sejong Project. The KOLON System is also extended with software tools to simplify the han- dling and visualization of the data, as well as the creation of new programs. The mapping of words onto ontological concepts was performed automatically, with faulty information being corrected manually. In order to examine the effectiveness of using KOLON’s data, we have rerun a previous sentiment analysis (SA) experiment, changing the approach to include data from the ontology. This new experiment obtained improved results, which is a strong indica- tion that the project will be of use after its completion. Keywords: KOLON, ontology, Mikrokosmos Ontology, WordNet, FrameNet 1 Introduction In various fields related to Information Science, such as Artificial Intelligence and Natural Lan- guage Processing (NLP), ontologies, also called Knowledge Bases (KB), are formal representa- tions of knowledge and the relations among its concepts, usually composed by individuals, classes, attributes, relations, events, rules and axioms (Nirenburg, 2004). Even though researchers had been building several types of ontologies for decades and have been using them as a formal represen- tation of knowledge within certain domains, ontologies gained momentum with the creation and propagation of the World Wide Web, and, more recently, with the idea of the Semantic Web, in- troduced by Tim Berners-Lee’s homonymous article (Berners-Lee, 2000a; Nirenburg, 2004; Hirst, 2004). However, the ontological structuralization of knowledge alone is not enough to make com- plete use of information in NLP, since ontologies basically hold semantic atoms of information and interconnect them through different kinds of named relations. For a thorough analysis of lin- guistic data, we need not only the semantic data from a simple ontology – information about the lexicon, the morphology, the syntax are needed as well. With this in mind, an work-group at our Computational Linguistics Laboratory devised the KOLON Ontology (an acronym for “KOrean Lexicon mapped onto an ONtology”), based on ontological and lexical data, through the adap- tation of two large works: the Mikrokosmos Ontology and the 21st Century Sejong Project. The KOLON System is the data combined with related software to facilitate its use both for the final user, who may simply want to inspect the data, and for the programmer, who may need to create programs to utilize the data in various ways. This paper starts with a general presentation of works related to the current project, namely Mikrokosmos Ontology, 21st Century Sejong Project, WordNet and FrameNet, highlighting the Copyright 2010 by Juliano Paiva Junho, Yumi Jo, and Hyopil Shin PACLIC 24 …

Content truncated to 500 words