Higher-order Comparisons of Sentence Encoder Representations

TL;DR

This paper applies Representational Similarity Analysis (RSA) to compare language encoder representations with human eye-tracking data, revealing new insights into how models relate to human processing difficulty.

Contribution

It introduces RSA as a tool for comparing language models and human data, demonstrating its effectiveness in interpretability and establishing new model-human correspondence insights.

Findings

RSA reveals correspondence between language encoders and eye-tracking data.

RSA offers a transparent, sample-efficient comparison method.

The study uncovers previously unknown links between models and human processing.

Abstract

Representational Similarity Analysis (RSA) is a technique developed by neuroscientists for comparing activity patterns of different measurement modalities (e.g., fMRI, electrophysiology, behavior). As a framework, RSA has several advantages over existing approaches to interpretation of language encoders based on probing or diagnostic classification: namely, it does not require large training samples, is not prone to overfitting, and it enables a more transparent comparison between the representational geometries of different models and modalities. We demonstrate the utility of RSA by establishing a previously unknown correspondence between widely-employed pretrained language encoders and human processing difficulty via eye-tracking data, showcasing its potential in the interpretability toolbox for neural models