Seminars

In natural languages, biological constraints push toward cross-linguistic homogeneity while linguistic, cultural, and historical processes promote language diversification. I focus in my talk on the effects of these opposing forces on sign languages. Research in movement science has shed light on the nature and effects of neural and motor constraints on hand use. Analyses of 33 languages revealed that signs exhibited the same form of adaptation to biological constraints found in tasks for which the hand naturally evolved (for example, grasping). Cross-linguistic comparisons also showed that signs vary cross-linguistically under the effects of linguistic, cultural, and historical processes. Their effects could thus emerge even without departing from the demands of biological constraints. New sign languages born in the last decades represent an exceptional opportunity for understanding the development of sign languages. Signs recorded at different ages of the new languages showed that biological constraints predicted sign change during language development.

Online measures of word recognition such as eyetracking have revolutionized the way we understand phonological processing in young children. However, most work in this area has only tested listeners exposed to a single dialect of a single language. In this talk I will discuss the influence of natural exposure to multiple dialects - with different distributions of phonological contrasts - on toddler’s online word recognition. Contrary to some earlier reports, we find no evidence that exposure to multiple language varieties has long-lasting detrimental effects on toddler’s word recognition efficiency. We take a glass- is-half-full view to explaining the impact of multi-accent exposure on development: We suggest that previously reported effects of accent exposure could be best captured in a developmental framework where children by 2 years of age have developed speech processing strategies best suited for their environment, and are adept at flexibly handling accents beyond a socially dominant language variety.

The evolution of vocal learning — an important component of speech and language development — in theory allows for increased communicative flexibility, compared with animals whose vocalizations are not learned. Songbirds, the most species-rich group of vocal learners, exhibit great diversity in terms of song structure and processes of inheritance. Relatively little is known about songbird vocal interactions — or how birds may learn how to sing (as opposed to simply what). In this talk, I describe research examining the vocal interactions of the dominant laboratory songbird, the male zebra finch. I show that while zebra finch song has been productively exploited for its simplicity and stereotypy as a model of sensorimotor learning and control, these gregarious, nonterritorial animals possess complex vocal repertoires that are dynamically expressed in social interactions, reminiscent of alignment phenomena that pervade human communication at multiple linguistic levels. I argue that a comparative-species perspective can offer new insights into the emergence of dynamic communication during development.

In this talk, I describe a model trained to represent conversational dynamics by predicting prosodic and language output based on autoregressive input from the target speaker and their interlocutor. Separate attention mechanisms for each speaker allow us to examine which of the interlocutor's turns are considered influential for the prediction of a target turn. Using this novel approach, we formulate and evaluate several hypotheses about how a speaker's conversational speech influences that of their interlocutor (often called “entrainment”), and discuss how these attention scores can be used to sketch the discourse structure of a dialogue. These findings extend previous work on entrainment by expanding the hypothesis space to search for nonlocal and nonlinear dynamics.