Chinese utterance segmentation in spoken language translation

Abstract

This paper presents an approach to segmenting Chinese utterances for a spoken language translation (SLT) system in which Chinese speech is the source input. We propose this approach as a supplement to the function of sentence boundary detection in speech recognition, in order to identify the boundaries of simple sentences and fixed expressions within the speech recognition results of a Chinese input utterance. In this approach, the plausible boundaries of split units are determined using several methods, including keyword detection, pattern matching, and syntactic analysis. Preliminary experimental results have shown that this approach is helpful in improving the performance of SLT systems.