Cognitive Modeling with Scaffolded LLMs: A Case Study of Referential Expression Generation

Polina Tsvilodub, Michael Franke, Fausto Carcassi

Keywords: cognitive modeling, referential expressions, language generation, neuro-symbolic models, LLMs

Abstract: To what extent can LLMs be used as part of a cognitive model of language generation? In this paper, we approach this question by exploring a neuro-symbolic implementation of an algorithmic cognitive model of referential expression generation by Dale & Reiter (1995). The symbolic task analysis implementing the generation as an iterative procedure scaffolds symbolic and gpt-3.5-turbo-based modules. We compare this implementation to an ablated model and a one-shot LLM-only baseline on the A3DS dataset (Tsvilodub & Franke, 2023) and find that our hybrid approach is at the same time cognitively plausible and performs well in complex contexts, while allowing for more open-ended modeling of language generation in a larger domain.

Submission Number: 56