The Research Seminar Series in Philosophy invites you to a seminar with Jessica Heine, "An Empiricist Account of Semantic Understanding".
Abstract:
Despite knowing that string S is a sentence of language L, I may lack any semantic understanding of S. Perhaps, if asked what S means, I can repeat, “S”. However, this proves nothing, as I cannot use S to intentionally express P without some semantic understanding of S.
Here’s an intuitive way to capture this lack of semantic understanding: my attitudes, beliefs, dispositions, etc., with regard to S fail to rule out any possible meanings of S. For any proposition P that S could express, it’s consistent with all I seem to understand about S that S expresses P in L. If I have any semantic understanding of S, my attitudes toward S should be sufficiently substantive to allow for the possibility of misunderstanding. If you understand S in language L, your understanding should rule out meanings of S insofar as there are some propositions Q such that, if S expressed Q in L, you would misunderstand S.
This paper develops this intuitive idea into a precise account of semantic understanding. Putnam showed that people can use S to intentionally express P, even when their subjective understanding does not rule out all incorrect meanings of S. I defend a graded account of semantic understanding, beginning with a minimum threshold required for intentional language use, and culminating in “perfect understanding”. Putnam’s classic subjects, and most users of most language, exist somewhere between these bounds.
Non-perceiving agents cannot meet this minimal threshold, regardless of conversational competence. My account thus provides an argument that current LLMs lack any semantic understanding of natural language, even if they are conscious. More broadly, my account generates the empiricist conclusion that knowledge attained via symbolic strings necessarily relies on a foundation of knowledge about how the world perceptually seems to the knower.
