Large Language Models Behave (Almost) As Rational Speech Actors: Insights From Metaphor Understanding

Gaia Carenini, Louis Bodot, Luca Bischetti, Walter Schaeken, Valentina Bambini

Keywords: LLMs; Bayesian reasoning; Pragmatics

TL;DR: Investigating mathematically the pragmatics behaviour of GPT when understanding metaphors

Abstract:

What are the inner workings of large language models? Can they perform pragmatic inference? This paper attempts to characterize from a mathematical angle the processes of large language models involved in metaphor understanding. Specifically, we show that GPT2-XL model’s reasoning mechanisms can be well predicted within the Rational Speech Act framework for metaphor understanding, which has
already been used to grasp the principles of human pragmatic inference in dealing with figurative language. Our research contributes to the field of explainability and interpretability of large language models and highlights the usefulness of adopting a Bayesian model of human cognition to gain insights into the pragmatics of conversational agents.

Submission Number: 24