Multiword expressions : a study on representation of Japanese MWEs in wordnet and other lexical databases
Lee, Hui Shan
Date of Issue2017
College of Humanities, Arts, and Social Sciences
Multiword expressions (MWEs) make up a significant portion of the lexicon and have distinctive characteristics of non-compositionality, non-substitutability, non-modifiability. They have been widely recognized as a very problematic part of natural language processing (NLP) as the current linguistic databases often do not have enough coverage on MWEs. This paper attempts to fill a gap in research by looking at how difficult it is to retrieve and process Japanese MWEs. The research presents an overview of 360 entries obtained through automatic-retrieval (AR) and manual retrieval (MR) from the corpus. These entries are then compared across seven databases; goo dictionary, imiwa? dictionary, the JDMWE, the WWWJDIC, NINJAL, wordnet, and the N-gram count corpus to test for whether thery are MWEs. The results obtained from this study suggest that the coverage of the database used, the differences in how phrases are represented in the dictionary, complications caused by the different writing systems present in Japanese, as well as the need for human judgement, are some of the main problems in determining whether a phrase is an MWE.
Final Year Project (FYP)
Nanyang Technological University