Semantics circle: New evidence for a Bayesian approach to gradable expressions

Abstract

In probabilistic approaches to the interpretation of gradable adjectives, a sentence such as 'Gloria is tall' expresses the proposition that h>s, where h and s are two independent random variables that represent, respectively, Gloria's height (h) and the threshold for tallness (s). In the most simple model (in the spirit of Egré, Spector, Mortier & Verheyen 2023), upon hearing 'Gloria is tall', the listener updates their joint distribution over (h,s) by conditioning on h > s, which leads them to increase their estimate of h.

I will first present experimental evidence for this general approach when applied to approximative expressions ('around 40'), based on joint work with Adèle Mortier, Paul Egré and Steven Verheyen. 

I will then discuss two additional, previously unnoticed predictions of this simple model in the case of gradable adjectives, based on joint work with Emile Enguehard and Carlotta Cascino.

- Prediction 1: Processing a sentence such as “Helen, Gloria, and Sue are tall”  sjoujld lead to a lower estimate of Gloria’s height than just processing “Gloria is tall.”  

- Prediction 2 : A sentence such as “Helen is not tall and Gloria is tall” should yield a higher estimate of Gloria’s height compared to simply “Gloria is tall.” 

Prediction 1 is shared  both by the most simple model and the more complex, pragmatic model of  Lassiter  & Goodman  (2017). The second model, however,  is not committed to Prediction 2. I will present experimental results that strongly support Prediction 1 but not Prediction 2. These findings provide evidence for Bayesian models of gradable adjective interpretation, and support more specifically models in which reasoning about thresholds involves informativity considerations, as in Lassiter & Goodman (2017).

References

- Egré, P., Spector, B., Mortier, A., & Verheyen, S. (2023). On the optimality of vagueness: “around”, “between” and the Gricean maxims. Linguistics and Philosophy, 46 (5), 1075–1130.

- Lassiter, D., & Goodman, N. D. (2017). Adjectival vagueness in a Bayesian model of interpretation. Synthese, 194, 3801–383

Code of Conduct for ZAS events: The ZAS is committed to fair, respectful, and professional interaction at its events. Therefore, please observe the Code of Conduct for this event.