What Makes Two Language Models Think Alike?

TL;DR

This paper introduces metric-learning encoding models (MLEMs) to compare how different language models represent linguistic features, providing transparent insights into their similarities and differences.

Contribution

The paper presents a novel, feature-based method (MLEMs) for comparing neural representations across models, enhancing interpretability and extendability to other domains.

Findings

MLEMs identify specific linguistic features responsible for model similarities and differences.

The approach offers transparent, symbolic comparisons of neural representations.

Method can be extended to speech, vision, and human brain studies.

Abstract

Do architectural differences significantly affect the way models represent and process language? We propose a new approach, based on metric-learning encoding models (MLEMs), as a first step to answer this question. The approach provides a feature-based comparison of how any two layers of any two models represent linguistic information. We apply the method to BERT, GPT-2 and Mamba. Unlike previous methods, MLEMs offer a transparent comparison, by identifying the specific linguistic features responsible for similarities and differences. More generally, the method uses formal, symbolic descriptions of a domain, and use these to compare neural representations. As such, the approach can straightforwardly be extended to other domains, such as speech and vision, and to other neural systems, including human brains.