Inferring Processing Times from a Corpus of Conversations
Nigel Ward · 2005
One of the ultimate goals of psycholinguistics is to know what goes on in the mind of a language user, including what all the processes are, how they fit together, and how they operate in everyday language use. Psycholinguistics research today, however, mostly focuses one or another specific aspect of language processing, more or less in isolation, in controlled, unnatural tasks. This paper proposes a new research technique that addresses these problems. 2 The Proposed Technique Measuring processing times requires, of course, the ability to determine both the start and end point of processing. The end point is simply the point at which the output of the response begins. The difficulty is determining the start point. In controlled experiments this start point is the point at which the stimulus is given, or at which the “go ” signal is given. In free conversation things are of course less simple, but it turns out that, for one special case, there is a kind of “go ” signal in conversation. This is the cue which the speaker gives when he wants to receive back-channel feedback, also sometimes called listener responses. These are, roughly, the expressions which a hearer produces during the speaker’s turn. In English, typical expressions used in back-channel feedback include yeah, uh-huh, hm, right and okay. In both Japanese and English this cue, or at least one of the cues, is a region of low pitch (Ward 1996; Ward 1997). The current best characterization of the cue is given as a predictive rule in Figure