대용량음성인식을 위한 parser 연구

Abstract

In this paper, we construct Parser for sentence analysis in order to improve performance of Large-vocabulary continuous speech recognition system. We use HPSG which is suitable for Korean of its partially-tree word order characteristics, as a grammar for Parsing. We propose a combination of sentence pattern and characteristic of predicate in structural for Korean sentence. As recognition engine for Morpheme analysis, we use H2MA developed by ROSP laboratory and FSL laboratory in Hanyang University. As a result, we obtain the recognition rate 91.82% about short sentences including double subjects and synthesize verb sentences.