Colloquium | Talk

Rachel Ostrand (IBM Research, New York) – Linguistic alignment: What is it good for?

Abstract:

Linguistic alignment: What is it good for?

Speakers engage in linguistic alignment – modifying characteristics of their own language production to converge upon or match those of their interlocutor’s – in many conversational contexts. In particular, speakers sometimes engage in partner-specific alignment, in which they align to characteristics of their current partner’s language, as contrasted with alignment to the statistics of the overall linguistic environment, and thus speakers may tailor their speech differently to each of multiple partners. But what determines when or whether speakers engage in partner-specific alignment?

In this talk, I will discuss my work investigating the factors that influence whether a speaker aligns in a partner-specific vs. a partner-general manner. I’ll discuss experiments which investigate characteristics of listener(s) which might affect the speaker’s expectations of the listener’s comprehension ability, including comparing native vs. non-native comprehenders, human vs. chatbot comprehenders, and comprehenders who behaviorally demonstrate high vs. low comprehension ability. I’ll also discuss whether speakers align at multiple linguistic levels within the same conversation. Finally, I will discuss experiments which investigate partner-specificity in comprehension and production within the same paradigm, to see if people align their production any time they detect systematic differences in their partner’s own production, or align more selectively. Across experiments, types of partners, and linguistic levels, speakers are much more likely to align partner-specifically when they expect such alignment will help their listener’s comprehension, based either on assumptions derived from the listener’s identity, or observations of the listener’s comprehension behavior, and not merely anytime they detect systematic variability in their interlocutor’s language use. Together, this evidence suggests that speakers only engage in partner-specific alignment when there is a communicative benefit for doing so; and do not align partner-specifically automatically, or for purely social reasons.