Mechanisms Underlying the Production of Language
T. Florian Jaeger
My research focuses on human language production and comprehension, with an emphasis on production. In particular, I am interested in questions about the use of probabilistic information in language processing. To what extent do speaker and comprehenders use probabilistic information to sufficiently optimize the processes involved in production and comprehension? How detailed are language users' estimations of probabilistic event, and what information precisely do we keep track of? What are the processes that are affected by probabilities? For example, why are frequent and predictable words produced shorter—is this some type of articulatory training effect or do speakers modulate their utterances (including the length of words) based on their assumption of what's easier to understand for their audience?
More generally, speakers often have a choice how between different variants to express the same message. These variants can differ in phonetic details (like word length, articulatory detail), the choice of words, the order of words, prosodic phrasing, etc. What determines speakers' choice in those cases? Are considerations about their audience part of their choice (e.g. to guarantee successful information transfer)?
To study these questions my lab (under development) uses both corpus studies and laboratory experiments. Corpora of spontaneous speech have the advantage that speakers are recorded under rather natural conditions (compared to even the most natural conditions in a lab), but they don't yield balanced data. We use modern statistical modeling to tackle this problem by controlling for the influence of many factors at the same time. Nevertheless, laboratory experiments sometimes become necessary to study language users under very specific circumstances. We do this by using standard methods such as self-paced reading, picture descriptions, etc. I am also interested in the simultaneous study of speakers and hearers (in dialogue situations) using simultaneous eye-tracking of both conversation partners.