Do Machines have a (Theory of) Mind? An investigation across Large Language Models and Humans

This thesis investigates whether Large Language Models (LLMs) possess a form of Theory of Mind (ToM), the capacity to infer the thoughts, desires, and intentions of others. The research combines computational and psycholinguistic methods, comparing model performance with that of human speakers through multiple evaluation techniques, while critically addressing the methodological and theoretical challenges that characterize this field. The work is organized along two main axes: on one hand, the direct evaluation of LLMs' pragmatic competence, with particular focus on indirect speech acts, conversational implicatures, and irony; on the other, the study of how models' internal representations align with human neural signals during the comprehension of figurative language.