Speech Perception as a Cognitive Process: The Interactive Activation Model.

摘要

In this paper we describe several attempts to model speech perception in terms of a processing system in which knowledge and processing is distributed over large numbers of highly interactive -- but computationally primitive -- elements, all working in parallel to jointly determine the result of the perceptual process. We begin by discussing the properties of speech which we feel demand a parallel interactive processing system, and then review previous attempts to model speech perception, both psycholinguistic and machine-based. We then present the results of a computer simulation of one version of an interactive activation model of speech, based loosely on Marslen-Wilson's COHORT model. One virtue of the model is that it is capable of word recognition and phonemic restoration without depending on preliminary segmentation of the input into phonemes. However, this version of the model has several deficiencies -- among them are excessive sensitivity to speech rate and excessive dependence on accurate information about the beginnings of words. To address some of these deficiencies, we describe an alternative called the TRACE model. In this version of the model, interactive activation processes take place within a structure which serves as a dynamic working memory. This structure permits the model to capture contextural influences in which the perception of a portion of the input stream is influenced by what follows it as well as what preceeds it in the speech signal. (Author)