Automatic Wordnet Mapping: from CoreNet to Princeton WordNet

Automatic Wordnet Mapping: from CoreNet to Princeton WordNet Jiseong Kim, Younggyun Hahm, Sunggoo Kwon, Key-Sun Choi Semantic Web Research Center, School of Computing, KAIST 291 Daehak-ro, Yuseong-gu, Daejeon, Korea {jiseong, hahmyg, fanafa, kschoi}@kaist.ac.kr Abstract CoreNet is a lexico-semantic network of 73,100 Korean word senses, which are categorized under 2,937 semantic categories organized in a taxonomy. Recently, to foster the more widespread use of CoreNet, there was an attempt to map the semantic categories of CoreNet into synsets of Princeton WordNet by lexical relations such as synonymy, hyponymy, and hypernymy relations. One of the limitations of the existing mapping is that it is only focused on mapping the semantic categories, but not on mapping the word senses, which are the majority part (96%) of CoreNet. To boost bridging the gap between CoreNet and WordNet, we introduce the automatic mapping approach to link the word senses of CoreNet into WordNet synsets. The evaluation shows that our approach successfully maps previously unmapped 38,028 word senses into WordNet synsets with the precision of 91.2% (±1.14 with 99% confidence). Keywords: wordnet, automatic mapping, automatic construction 1. Introduction CoreNet (Choi et al., 2004) is a semantic hierarchy of Ko- rean word senses, which has been built by KAIST since 1994 based on CoreNet concept hierarchy originated from NTT Goi-Taikei (Ikehara et al., 1997) concept hierarchy. The CoreNet hierarchy comprises mainly two parts, non- terminal part and terminal part. The non-terminal part of the hierarchy comprises 2,937 semantic categories, called CoreNet concept, as non-terminal nodes, which are orga- nized by a taxonomic relation, while the terminal part of the hierarchy comprises 73,100 Korean word senses as termi- nal nodes, which are separated from each other (i.e., there is no link between the word senses) and there is only a link be- tween word sense and its semantic category with unknown lexical relations such as is-a and part-of; the unknown re- lation means there may be is-a or part-of relation between word sense and its semantic category, but a label of the re- lation is not revealed. An example of the CoreNet hierarchy is shown in Figure 1. Figure 1: An example of the hierarchy of CoreNet To extend CoreNet into other languages and to promote its broader utilization for diverse NLP application, Kang et al. (2010) made an attempt to map the CoreNet hierarchy into the Princeton WordNet hierarchy. The scope of the map- ping encompassed all of 2,937 semantic categories, which were successfully mapped to WordNet synsets with syn- onymy, hypernymy, and hyponymy relations. Although Kang et al. (2010) mapped the almost all of the semantic categories of CoreNet, word senses of CoreNet is not in the scope of the mapping, and still remains out of mapping; this leads to the fact that, from the perspective of NLP application, WordNet operations such as path similar- ity hard to be applied to the word senses, which are the ma- jority part (96%) of CoreNet, because there is no mapping for the word senses, and, moreover, lexical relations …

Content truncated to 500 words