The Development of Psychological Models for Speech Recognition.

Abstract : The report has two principal sections. In Section I the theoretical issues relating to the development of syntax recognition routines based on psychological models of human speech recognition are discussed and the relevant psychological literature reviewed. The research reported in this section deals with attempts to relate various syntactic variables to measures of the perceptual complexity of sentences. The results of the research indicate that (1) analysis by synthesis routines are probably not appropriate as models of the system employed by human speakers for speech recognition, (2) the complexity of sentences is not related in any direct way to the number of operations required by a grammar to produce them, and (3) both the lexical structure of verbs and the relation of certain other lexical formatives to transformational operations of the grammar are significantly related to the ease of understanding sentences. Section II of the report deals with research concerning the perceptual segmentation strategies employed by speakers in their analysis of continuous speech. The report details the development of a particular investigative technique and its application to the determination of segmentation strategies. The research reported indicates that speech signals are initially analyzed into segments that approximate the clause. The experiments reviewed explore the relationship between the constituent structure of sentences and this clause-like segementation.