Dissociating language and thought in large language models

TL;DR

This paper distinguishes between formal linguistic knowledge and functional language use in large language models, highlighting their strengths and limitations, and suggests that mastering both requires specialized mechanisms.

Contribution

It introduces a framework grounded in neuroscience to evaluate LLMs' linguistic and cognitive capabilities, emphasizing the need for separate mechanisms for formal and functional competence.

Findings

LLMs excel at formal linguistic competence

Performance on functional tasks is inconsistent and often requires fine-tuning

Separate mechanisms may be necessary for formal and functional language use

Abstract

Large Language Models (LLMs) have come closest among all models to date to mastering human language, yet opinions about their linguistic and cognitive capabilities remain split. Here, we evaluate LLMs using a distinction between formal linguistic competence -- knowledge of linguistic rules and patterns -- and functional linguistic competence -- understanding and using language in the world. We ground this distinction in human neuroscience, which has shown that formal and functional competence rely on different neural mechanisms. Although LLMs are surprisingly good at formal competence, their performance on functional competence tasks remains spotty and often requires specialized fine-tuning and/or coupling with external modules. We posit that models that use language in human-like ways would need to master both of these competence types, which, in turn, could require the emergence of mechanisms specialized for formal linguistic competence, distinct from functional competence.