Wednesday, 2pm-4pm
First session on Feb 3.
Zoom information Links will be communicated by email
Course Description
Prerequisites:
This is an advanced seminar targeting graduate students (MA and PhD) in cognitive science, linguistics, philosophy and related fields. We will expect
Familiarity with logic or linguistic semantics
High-school level knowledge of probability theory
If you’re not sure you satisfy these prerequisites, please contact the instructors to check.
Course Description
Linguistic pragmatics and the psychology of reasoning are different yet related fields. On the one hand, linguistic pragmatics studies a specific type of reasoning, namely reasoning about speakers’ communicative intentions. On the other hand, the psychology of reasoning investigates how humans derive conclusions from various pieces of information, almost always provided in a linguistic format, and some deviations from normative logic can be explained in terms of linguistic processes that interfere with deductive reasoning. In both domains, the most classical approach is to model human reasoning as a kind of deductive reasoning. However, more recently, several theories have treated both pragmatic inferences and general-purpose reasoning as involving probabilistic inferences in an essential way. This view has led to new perspectives on the core questions and phenomena of both fields.
The goal of this course is to introduce students to probabilistic models of reasoning and pragmatics, and to discuss recent proposals in this domain. We will focus, in particular, on the role of Bayesian reasoning in pragmatics and general-purpose reasoning, as well as the influence of prior probabilities in both domains.
The seminar will consist of lectures and discussions of papers and student projects.
Specifically, we will focus on:
Psychology of reasoning
We will study three frameworks for general-purpose reasoning that make substantive and distinct commitments about the role of probabilities. We will start with the heuristics and biases paradigm of Kahneman and Tversky, which argues that humans avoid reasoning with and about probabilities and instead deal in plausibly more ecologically motivated notions such as stereotypicality. We will review in detail very striking empirical studies on the conjunction fallacy, base-rate neglect, and related phenomena. We will then move on to what is known as the New Paradigm of research on reasoning (e.g. work by Oaksford and Chater), which holds that the functional aim of human reasoning is not to match logical validity, but rather probabilistic validity as a model of reasoning under uncertainty. For this topic, we will focus on the semantics of conditionals and on empirical studies on reasoning with conditionals. The third topic will be Bayesian confirmation-theoretic approaches to reasoning: the idea that human reasoning is a process of testing hypotheses taking into consideration explanatory adequacy rather than just maximization of posterior probabilities. Throughout these three topics, we will regularly examine the question of interpretation-based alternatives to general-purpose-reasoning accounts of the data.
Probabilistic Pragmatics
We will mostly discuss the Rational Speech Act (in the RSA) model of pragmatics, and compare it to more classical approaches. The RSA framework is a game-theoretic framework to pragmatics, which characterize the strategies of speakers and listeners as a function of each other. It views message choice as being governed by a tradeoff between cost and informativity, where the informativity of a message is equated with its surprisal value to the listener, which itself depends on the priors that speakers and listeners are assumed to share. The listener, on the other hand, is modeled as a Bayesian agent who draws infers from the message she hears. One guiding question will be the influence of prior probabilistic information on message choice and interpretation. We will discuss quantity implicatures, vagueness, and non-literal interpretations, and the division of labor between semantics and pragmatics. Students will be encouraged to implement various models to simulate their predictions.