Statistical Corpus Analysis for Kt{treasure : Korea Telecom Train Ticket Reservation Aid System Based upon Speech Recognition

This paper describes statistical analysis results of the corpus for KT{TREASURE (Korea Telecom Train ticket REservation Aid System based Upon speech REcognition). As the beginning of this development, two sets of speech corpus were collected. One was based on human-human(H-H) dialogues and the other was based on human-computer(H-C) dialogues. Wizard of Oz(WOZ) experiment was carried out to collect speech corpus based on H-C spoken dialogue. Linguistic analysis results show that people respond diierently when they are talking to a computer compared to when talking to a human. Since the basic unit of grammar in Korean is a morpheme, Korean-language model based on a morpheme was designed in addition to a word-based language model. We also deened the subword unit which lies between word and morpheme, then constructed a subword-based language model. Language-model analysis results reveal that a morpheme-based language model gives 50% reduction in perplexity(PP) over a word-based one. It also shows that a morpheme-based language model is least aaected by vocabulary reduction.