Disentangling Syntax and Semantics in the Brain with Deep Networks

TL;DR

This study introduces a framework to disentangle and analyze the neural representations of syntax and semantics in the brain using deep language models and fMRI data, revealing shared neural substrates.

Contribution

It proposes a taxonomy and statistical method to decompose language model activations and brain activity into distinct linguistic classes, challenging previous modular assumptions.

Findings

Compositional representations involve widespread cortical areas.

Syntax and semantics share a common neural substrate.

The framework isolates distributed linguistic representations in brain activity.

Abstract

The activations of language transformers like GPT-2 have been shown to linearly map onto brain activity during speech comprehension. However, the nature of these activations remains largely unknown and presumably conflate distinct linguistic classes. Here, we propose a taxonomy to factorize the high-dimensional activations of language models into four combinatorial classes: lexical, compositional, syntactic, and semantic representations. We then introduce a statistical method to decompose, through the lens of GPT-2's activations, the brain activity of 345 subjects recorded with functional magnetic resonance imaging (fMRI) during the listening of ~4.6 hours of narrated text. The results highlight two findings. First, compositional representations recruit a more widespread cortical network than lexical ones, and encompass the bilateral temporal, parietal and prefrontal cortices. Second, contrary to previous claims, syntax and semantics are not associated with separated modules, but, instead, appear to share a common and distributed neural substrate. Overall, this study introduces a versatile framework to isolate, in the brain activity, the distributed representations of linguistic constructs.