Listeners' adaptation to unreliable intonation is speaker-sensitive.

Variable linguistic environments require the ability to quickly update expectations and behavior including speech comprehension. This adaptive capacity is key to understanding how listeners successfully recognize speaker intentions in light of the ubiquitous variability in speech. The present study investigates how listeners' real-time sentence comprehension adapts to speaker-specific prosodic variability. In two forced choice mouse tracking experiments, listeners had to identify a visual referent guided by pre-recorded instructions. When exposed to a speaker that uses unconventional pitch accent placement, listeners discard intonational information for that speaker, but keep using intonation to resolve the referential ambiguity for another speaker that places pitch accents conventionally. These results show for the first time that intonationally guided sentence comprehension adapts in a speaker-sensitive way. The study further provides valuable first insights into the temporal unfolding of this adaptation process. Listeners first attribute unconventional patterns to the context, thus discarding the informational value of intonation for both speakers. After sufficient evidence, however, listeners start attributing unexpected patterns to only the unconventional speaker. Materials, data, and scripts can be retrieved here: http://osf.io/fdpg4.