Summary

The importance of this investigation is three-folds：firstly，utterance patterns are extracted directly from authentic spoken English data which reflects the real time conversational exchange as speech recognition is aiming to interpret authentic discourse；secondly，this investigation produces some new utterance patterns due to its capability in revealing authentic interface between syntax and semantics which have not been described by other studies；and thirdly，no only verbs，as studied by Levin and Jones，but also noun，adjective，and adverbs are investigated as authors believe that there also relationships existed between these semantic categories and their discoursal patterns.

Implication of this study is obviously as it provides a base for language modelling on top of an acoustic model.That is，when a word is identified，not only its utterance patterns can be determined，but also its neighbour word's syntactic status can be confirmed and the recognition system may use these patterns and categories to retrieve or filter the word candidates.

We believe that these real time syntactic properties are valuable to speech recognition system，as these utterances will improve speech recogniser's predicting power in identifying unknown speech words and narrowing down the search scope from candidate lists.These utterance patterns can be obtained through the large amount of data extraction.