Listener Adaptation to Variable Use of Uncertainty Expressions

Speakers exhibit considerable variability in their productions at all levels of linguistic representation. Listeners deal with such variability by adapting to it and updating phonetic (e.g., Kleinschmidt and Jaeger, 2015), syntactic (e.g., Kamide, 2012; Fine et al., 2013), and semantic/pragmatic (Yildirim et al., 2016) expectations. We build upon the work in semantic/pragmatic adaptation and investigate how listeners adapt to variable uses of uncertainty expressions such as might and probably. I will present three experiments that investigate speaker-specific production and comprehension of uncertainty expressions. These experiments provide evidence that a) speakers vary in their use of uncertainty expressions to describe future events with different probabilities of the event happening; b) listeners quickly adapt their expectations on how a specific speaker uses these expressions after a short exposure period; and c) this change in expectation also translates to changes in interpretation.

I will further present preliminary modeling results. We model production and comprehension of uncertainty expressions within the Rational Speech Act framework and we model the adaptation process as Bayesian belief updating of the model parameters.