A formal framework proposal: for a hybrid neurosymbolic semantics

Abstract

With the advancement of Large Language Models (LLMs), such as popular chatbots, a new set of challenges and opportunities for formal semantics emerges. While LLMs demonstrate an impressive capacity to generate and process natural language, the way they represent and manipulate meaning opens up a fruitful topic of active research. The gap between the deep latent semantics of LLMs and traditional symbolic formalisms of formal semantics is an open problem. The question of how to extract and formalize semantic knowledge from LLMs, and how to use formal semantics to control and explain the behavior of these models, is an emerging area that itself synthesizes an open problem.