A Similarity Measure for the Topic of Legal Texts using a Domain-Specific Lexical Network

Abstract

This paper proposes a similarity measure for extremely short texts in legal domain, especially aiming at the task of retrieving related laws given a question text of the bar examination of Japan. Through this task we discuss the effectiveness of utilizing structured knowledge in text processing in the legal domain. We construct a lexical network providing both linguistic and domain-specific relations between legal terminologies, and evaluate the utilities and limitations for the application to this retrieval task.