Description:

Speech perception is often viewed as a challenge for the listener due to the lack of acoustically invariant cues. A large body of work has demonstrated that listeners use prediction or expectation to resolve potentially ambiguous signals. The task of segmenting a continuous speech stream into the discrete percept of words is similarly challenging for listeners because speech does not typically consist of pauses at word boundaries. Coarticulation in normal conversation can result in ambiguous stretches of speech that could be interpreted in a number of ways. For example, a sentence like “Don must see the harbor or boats” could be produced in such a way that there is not a pause between harbor and or, which would result in an ambiguous stretch of speech that could be perceived as either “Don must see the harbor or boats” or “Don must see the harbor boats.” Both are syntactically and semantically acceptable, and both could be consistent with the acoustic information the listener receives in a reduced utterance. How, then, does the listener determine whether they have heard or in the utterance?
In this talk, I will present a series of studies examining expectation or prediction in spoken word recognition, as well as how experience modifies these predictions. Specifically, I will discuss the role of speaking rate in spoken word recognition, as well as interactions between speaking rate and other cues used in spoken word recognition. I will present evidence that listeners use speaking rate to make predictions about the material they will encounter down the speech stream, and use these predictions to disambiguate acoustically unclear material. I will conclude by addressing some recent work that suggests that non-native speakers utilize speaking rate in ways that differ from native speakers, and will describe current projects designed to better understand how expectation and prediction shift as a function of speaker and listener language backgrounds.