First International Conference on Spoken Language Processing (ICSLP 90)

Kobe, Japan
November 18-22, 1990

Process and Connectionist Models of Speech Perception

Dominic W. Massaro

Program in Experimental Psychology, University of California, Santa Cruz, CA, USA

Connectionist models offer an alternative frame-work to existing process models for the study of speech perception. The empirical domain consists of the integration of auditory and visual speech in bimodal speech perception. Two classes of connectionist models of speech perception have been evaluated and tested. The classes are interactive-activation models and feed- forward models. Empirical results indicate that while several sources of information simultaneously influence speech perception, these representations of the sources remain independent of one another. This independence is strong evidence against interactive activation in speech perception. Performance in speech identification experiments can be described adequately by a connectionist model with only input and output layers. Given the similarities between this model and a successful process model, we currently have two adequate theories of speech perception by human observers.

Full Paper

Bibliographic reference.  Massaro, Dominic W. (1990): "Process and connectionist models of speech perception", In ICSLP-1990, 577-580.