© 1991 by Association for Literary & Linguistic Computing
Articles |
Classification of Conjunctive Postpositions in Spoken Language Towards Speech Recognition*
ATR Interpreting Telephony Research Laboratories Kyoto, Japan
ATR Interpreting Telephony Research Laboratories, Sanpeidani, Inuidani, Seika-cho, Sorakugun, Kyoto 619-02, Japan
This paper proposes a classification of postpositions for spoken Japanese speech recognition based on empirical data Applying syntactic rules to speech recognition is drawing a great deal of attention because of its potential for raising recognition accuracy. At the same time, this application reveals the lack of a detailed study of syntactic phenomena, which results in producing many unacceptable sentences. In contrast to traditional intuitive classification, we examine the behaviour of postpositions quantitatively. First, we retrieve our empirical dialogue database. We will investigate the syntactic varieties in the sentence ending phrase, where the speaker's attitude toward his utterance is best expressed, especially focusing on the conjunctive postpositions. Secondly, we investigate how they behave in speech recognition by applying loosely constrained syntactic rules. Based on the retrieval results and speech recognition results, we refine the rules. Finally, we establish the validity of their improvement by both speech recognition rate and acceptance rate.
* This is the extended version of "Classification of S-Postpositions in Spoken Language toward Speech Recognition" (Hosaka et al, 1991) presented at ACH/ALLC-91
Currently he is working at NHK science and technical research laboratories