Metaphor recognition using automatic classification of adjectives based on statistical method

摘要

In this paper, we propose a method of metaphor recognition using automatic classification of adjectives based on statistical approach. We have defined adjective vector and calculated similarity between adjectives. An adjective vector is described by frequency of nouns which modify the adjectives. Masui et. al. have introduced two probabilistic measurements: "salience gap" and "novelty" which are based on conceptual description of metaphorical words. In the conceptual description, properties of words, which are adjectives, are classified into several categories using similarity between adjectives. We have extracted metaphorical expressions from newspaper corpus and classified into exemplify, metaphor and others by 20 human examiners. Using this test sets, we have conducted experiments of metaphor recognition and the results was almost 80 percent accuracy.