ELSC Seminar Series

“Large Language Models and human language: is deeper understanding just around the corner?”

Large Language Models – an excellent engineering feat without a doubt – have also been claimed to be adequate scientific models of human language behavior and its neural underpinnings. How do we evaluate this claim? Thus far, hardly any evaluation criteria have been proposed. I discuss some sketchy proposals of such criteria, in light of which I examine some leading LLMs and their application to neurolinguistics. To forecast, the currently available empirical record falls short of supporting the claim that LLMs can be viewed as an adequate scientific theories of brain/language relations in humans. Is there hope in sight? My answer: yes, but for that, formal linguistic models must be incorporated into the already rich computational apparatus.