Proposal of Onomatopoeia's Senses Classification Method Using Rule based on Dictionaries

Abstract

Many Japanese onomatopoeic words have multiple meanings, which are determined by the surrounding context. Previous studies have proposed automated word sense classification using vector representations of onomatopoeia obtained from pre-trained BERT models. However, the annotation cost for creating training data is high, and it is difficult to prepare abundant training data for all onomatopoeia. In this study, we propose a rule-based word sense classification method utilizing the definitions from onomatopoeia dictionaries as a method to automate word sense classification with low annotation cost. We also define and introduce a "word sense score" into the rules, which classifies words into senses that are more characteristic and match more case parsing information, and report the accuracy of the classification.