All Bark and No Bite: Rogue Dimensions in Transformer Language Models Obscure Representational Quality

TL;DR

This paper reveals that a few rogue dimensions dominate similarity measures in contextual language models like BERT and GPT-2, misleading analysis, and shows simple postprocessing can correct this issue to better understand model representations.

Contribution

It identifies the impact of rogue dimensions on similarity measures in contextual models and demonstrates that standardization can mitigate this problem, improving interpretability.

Findings

A small number of rogue dimensions dominate similarity measures.

Standardization corrects for rogue dimensions and reveals true model representations.

Rogue dimensions cause a mismatch between similarity measures and model behavior.

Abstract

Similarity measures are a vital tool for understanding how language models represent and process language. Standard representational similarity measures such as cosine similarity and Euclidean distance have been successfully used in static word embedding models to understand how words cluster in semantic space. Recently, these measures have been applied to embeddings from contextualized models such as BERT and GPT-2. In this work, we call into question the informativity of such measures for contextualized language models. We find that a small number of rogue dimensions, often just 1-3, dominate these measures. Moreover, we find a striking mismatch between the dimensions that dominate similarity measures and those which are important to the behavior of the model. We show that simple postprocessing techniques such as standardization are able to correct for rogue dimensions and reveal underlying representational quality. We argue that accounting for rogue dimensions is essential for any similarity-based analysis of contextual language models.