Title: Modeling truth-value judgments as a rational speech act
Speaker: Prof. Gregory Scontras (University of California, Irvine)
Time: 11/2 (Thursday) 10:00 AM
Online talk: https://meet.google.com/chm-qkzh-nsp
Abstract: Research into children’s understanding of language commonly employs truth-value judgments: whether a sentence can truthfully describe a given scenario. On the basis of such judgments, researchers have concluded that young children perform quite differently from adults when it comes to understanding ambiguous utterances with multiple potential meanings. However, subtle changes to the truth-value judgment task setup make children more adult-like. I summarize key results from the literature on child ambiguity resolution, noting three factors that affect children’s disambiguation behavior. One of these factors concerns children’s processing ability: how easy it is to access the different grammatical interpretations. The other two factors concern children’s ability to manage the pragmatic context: understanding what the topic of conversation is, and modulating expectations about the world being described. In an attempt to identify the role of each factor in language understanding, I then formally articulate a computational cognitive model within the Bayesian Rational Speech Act framework that specifies the role of each of these three factors in providing truth-value judgments. The results suggest that pragmatic factors may play a larger role than grammatical processing factors in explaining children’s observed non-adult-like behavior, and the computational model offers a hypothesis as to why that’s so: pragmatic factors have a larger impact on informativity, which serves as the ultimate arbiter of utterance endorsement in the truth-value judgment task. I close with a discussion of recent adult behavioral data further supporting the model’s predictions.