Interpreting speech in context.
Mark A. Pitt · The Journal of the Acoustical Society of America · 2010
One of the challenges in understanding how humans recognize spoken language is to reconcile the robustness of verbal communication with the many forms of ambiguity and variability in the speech signal. How much of the discrepancy is due to our incomplete understanding of the acoustics of speech (i.e., what is the critical information) versus the recruitment of mental processes (e.g., memory) to aid recognition? My approach to these explanations has been to study the processing of words that have been degraded in controlled ways (e.g., digital manipulation) and by using conversation-style speech that contains naturally produced ambiguities. Recognition is measured in isolation and in sentences using a variety of tasks whose purpose is to measure the source, magnitude, and time course of ambiguity resolution. I will present an overview of this work in the context of a few experimental settings, highlighting methodological issues along the way. Together, the findings suggest that word recognition is a highly fluid process that depends on the rapid integration of multiple sources of information.