University of Rochester
The adaptive speaker: Dynamic millisecond adjustments to speech as a function of communicative goals?
T. Florian Jaeger (University of Rochester) about work by Esteban Buz (University of Rochester) and Scott Seyfarth (UCSD)
One of the large unresolved debates in research on language production centers around the extent to which the various stages involved in the linguistic encoding of messages are (im)permeable to communicative goals. On the one hand, languages exhibit many properties that are communicatively efficient and unexpected by chance. On the other hand, production behaviors that appear to serve communicative goals can be side-effects of the time course of linguistic planning.
I’ll focus on articulation and specifically, the question of whether speakers can dynamically adjust the pronunciation of words depending on their expected a priori perceptual confusability. I’ll present on our recent studies on context-driven hyper-articulation, such as increasing the voice onset timing of the voiceless /p/ in “pin”, when a voiced competitor (“bin”) is contextually present. I further demonstrate a) why even tiny changes to the signal can be communicatively rational, once the noisiness of the motor plant is taken into account and b) that speakers dynamically adapt even micro aspects of subsequent productions based on feedback from interlocutors.
These and related results make an interpretation of hyper-articulation purely in terms of non-communicative processes implausible. Rather, it seems that speakers can monitor the effect of their productions on their interlocutors and adapt their subsequent productions accordingly. A pressing question for future research is thus to determine how increased time pressure or attentional load affects this ability (e.g., during conversational speech). Time permitting, I discuss what our result add to existing proposal that refer to internal or external self-monitoring, as well as the perception-production loop operating between interlocutors.
T. Florian Jaeger is an Associate Professor in the Departments of Brain and Cognitive Sciences, Computer Science, and Linguistics at the University of Rochester. He is the Director of the Center for Language Sciences. He received his PhD in Linguistics at Stanford University in 2006 and an M.A. in Linguistics and Computer Science from Humboldt University in 2001. He is an Alfred P. Sloan Research Fellow and NSF CAREER Recipient. His research seeks to understand the inference and learning processes underlying robust human language processing and these processes shape linguistic representations over time. This work is funded by the National Institutes of Health and the National Science Foundation.