Research seminars in linguistics and cognition

Thursday, April 4, 2019 - 14:00
Reasoning with disjunctions as a form of hypothesis testing

When: thursday 4 april 2019, 14:00 – 16:00
Where: Aula A, San Niccolò, Via Roma 56, Siena

 

Seminar: "Reasoning with disjunctions as a form of hypothesis testing"

 

Salvador Mascarenhas (Ecole Normale Supérieure/DEC - Paris)

 

Abstract
The idea that human reasoning is best modeled by rational (Bayesian) update procedures resting on a representational system with probability measures has gained great currency in the psychology of reasoning over the past twenty years (e.g. Oaksford and Chater, 2007). Conversely, the popularity of research paradigms that approach human reasoning within the mold of a model-theoretic formal system without probability measures has decreased. I argue that elements from both approaches are necessary to understand a large class of fallacious inference patterns involving disjunction and disjunction-like elements. For example, from "Mary met every king or every queen of Europe" and "Mary met the king of Spain" subjects overwhelmingly assent to the fallacious conclusion that "Mary met every king of Europe". I argue that this is because the disjunctive premise puts forth two alternatives and invites the reasoner to pick one, while the second premise provides evidence in favor of one alternative rather than the other. I extend the account Crupi et al. (2008) give of the conjunction fallacy in terms of hypothesis testing: the hypothesis "Mary met every king of Europe" is a better theory of the evidence "Mary met the king of Spain" than its competing hypothesis "Mary met every queen of Europe". Additionally, I show that the conjunction fallacy (Tversky & Kahneman, 1983) is in a concrete way a special case of these inferences with disjunction. The view of human reasoners as hypothesis testers rather than posterior-probability maximizers generalizes to a very broad class of compelling fallacies.

 

More concretely, I give a reconceptualization of the Erotetic Theory of Reasoning of Koralus and Mascarenhas (2013), a kind of Mental Models Theory (Johnson-Laird, 1983), in terms of Bayesian confirmation theory, where the hypotheses under consideration are determined by disjunctive premises interpreted as questions, along the lines of Inquisitive Semantics (Groenendijk, 2008; Mascarenhas, 2009). The resulting picture of human reasoning is by construction aligned with well-supported theories of linguistic interpretation, and it illustrates the need for both probabilistic tools (confirmation theory) and model-theoretic tools (non-classical semantics for disjunction) in an account of mental representations.