Brain Score Tracks Shared Properties of Languages: Evidence from Many Natural Languages and Structured Sequences

TL;DR

This study evaluates how language models trained on diverse natural and structured data correlate with human brain activity, revealing shared structural properties but questioning the specificity of Brain Score as a measure of human-like processing.

Contribution

The paper demonstrates that Brain Score performance is similar across models trained on various natural and structured datasets, highlighting shared structural features.

Findings

Models trained on different natural languages have similar Brain Score performance.

Structured data like the genome and code also achieve comparable Brain Score results.

High Brain Score does not necessarily indicate human-like language processing.

Abstract

Recent breakthroughs in language models (LMs) using neural networks have raised the question: how similar are these models' processing to human language processing? Results using a framework called Brain Score (BS) -- predicting fMRI activations during reading from LM activations -- have been used to argue for a high degree of similarity. To understand this similarity, we conduct experiments by training LMs on various types of input data and evaluate them on BS. We find that models trained on various natural languages from many different language families have very similar BS performance. LMs trained on other structured data -- the human genome, Python, and pure hierarchical structure (nested parentheses) -- also perform reasonably well and close to natural languages in some cases. These findings suggest that BS can highlight language models' ability to extract common structure across natural languages, but that the metric may not be sensitive enough to allow us to infer human-like processing from a high BS score alone.