BaseNP Supersense Tagging for Japanese Texts

要旨

This paper describes baseNP supersense tagging for Japanese texts. The task extracts base noun phrases (baseNPs) from raw texts in Japanese, and labels their baseNPs with supersenses. This task has a number of applications including predicate argument structure analysis and question answering. While the definition of baseNP in English is relatively clear, its definition in Japanese has not yet been clear. In this paper, we defined Japanese baseNP analogous to English and defined Japanese supersenses using a broad-coverage Japanese thesaurus, Nihongo Goi Taikei (comprehensive outline of Japanese vocabulary). We then adopted a sequential tagging algorithm for the task, namely the averaged perceptron with HMM, and achieve high performance compared to a baseline.

主題