Detecting Metonymic Expressions with Associative Information from Words

In natural language processing, metonymic expressions need to be detected and interpreted because sentences including metonymic expressions have different meanings from literal ones. Previous studies focused mainly on interpreting metonymic expressions, but not on detecting ones in sentences. The purposes of this study are to propose an associative approach for detecting them and to evaluate it. By using associative concept dictionaries and Japanese WordNet, we constructed an automatic system that can detect metonymic expressions. We evaluated the system by comparing it with a baseline system that also detects such expressions. As a result, our system showed higher rates of recall (80.0%), precision (72.0%), and F-measure (75.8%) than those of a baseline system.