Rethinking Evaluation of Sparse Autoencoders through the Representation of Polysemous Words

TL;DR

This paper introduces a new evaluation framework for sparse autoencoders that assesses their ability to produce interpretable monosemantic features, especially for polysemous words, revealing insights into their internal mechanisms and limitations.

Contribution

It proposes a suite of semantic evaluation metrics for SAEs, highlighting their strengths and weaknesses in capturing polysemy, and offers insights into how LLMs distinguish multiple meanings within words.

Findings

SAEs optimized for MSE-L0 may reduce interpretability.

Deeper layers and Attention modules help distinguish polysemy.

Semantic evaluation reveals limitations of traditional SAE metrics.

Abstract

Sparse autoencoders (SAEs) have gained a lot of attention as a promising tool to improve the interpretability of large language models (LLMs) by mapping the complex superposition of polysemantic neurons into monosemantic features and composing a sparse dictionary of words. However, traditional performance metrics like Mean Squared Error and L0 sparsity ignore the evaluation of the semantic representational power of SAEs -- whether they can acquire interpretable monosemantic features while preserving the semantic relationship of words. For instance, it is not obvious whether a learned sparse feature could distinguish different meanings in one word. In this paper, we propose a suite of evaluations for SAEs to analyze the quality of monosemantic features by focusing on polysemous words. Our findings reveal that SAEs developed to improve the MSE-L0 Pareto frontier may confuse interpretability, which does not necessarily enhance the extraction of monosemantic features. The analysis of SAEs with polysemous words can also figure out the internal mechanism of LLMs; deeper layers and the Attention module contribute to distinguishing polysemy in a word. Our semantics focused evaluation offers new insights into the polysemy and the existing SAE objective and contributes to the development of more practical SAEs.