Across the Levels of Analysis: Explaining Predictive Processing in Humans Requires More Than Machine-Estimated Probabilities

TL;DR

This paper critiques and extends claims about language models and language processing, emphasizing the need for multi-level analysis beyond machine-estimated probabilities, and suggests future integration with psycholinguistic models.

Contribution

It challenges existing claims by applying Marr's levels of analysis and proposes combining LLMs with psycholinguistic models for deeper understanding.

Findings

Predicting linguistic information is not solely based on machine probabilities.

Current claims about LLMs' role in psycholinguistics are incomplete.

Future research should integrate LLMs with psycholinguistic models.

Abstract

Under the lens of Marr's levels of analysis, we critique and extend two claims about language models (LMs) and language processing: first, that predicting upcoming linguistic information based on context is central to language processing, and second, that many advances in psycholinguistics would be impossible without large language models (LLMs). We further outline future directions that combine the strengths of LLMs with psycholinguistic models.