Sparse autoencoders help understand language model features better
Sparse Autoencoders for Mechanistic Interpretability in NLP: A Critical Survey of Recoverability, Interpretability, and Utility
Computation and Language
Summary
Understanding how language models work inside is important but hard. This paper looks at a specific method, sparse autoencoders, which try to break down model signals into simple parts that people can understand. The authors reviewed many studies to see how well these methods work and how useful and accurate their explanations are. They organize the findings, point out disagreements, and suggest better ways to test and improve these methods.
What this means in practice
- •For nlp engineers: Improve debugging of language models by decomposing activations into interpretable sparse features to understand model decisions better.
- •For machine learning auditors: Use surveys synthesizing sparse autoencoder research to assess the faithfulness and semantic clarity of model explanations in audits.
A survey. It maps existing work.
Authors
Salima Lamsiyah, Adelaide Danilov, Oleksandr Marchenko Breneur, Aria Nourbakhsh
Abstract
Sparse autoencoders (SAEs) have emerged as a promising approach to mechanistic interpretability by decomposing language-model activations into sparse, potentially interpretable features. However, differences in design assumptions, evaluation objectives, and experimental settings complicate conclusions about their effectiveness. We present a critical survey of 138 studies on SAEs in natural language processing. Our contributions are fourfold: (1) we propose a task-grounded taxonomy connecting dictionary design, feature operations, and NLP functions; (2) we examine how sparsity allocation, geometry, data coverage, and inference affect feature recoverability; (3) we assess evidence for semantic interpretability, mechanistic faithfulness, and behavioral utility; and (4) we synthesize findings across multilingual processing, retrieval, reasoning, classification, and behavioral regulation, highlighting the roles of objectives, baselines, supervision, and intervention pipelines. Through comparative analyses of published results, we reconcile apparent disagreements, identify unresolved questions, and propose directions for more rigorous evaluation. An anonymous research register https://anonymous.4open.science/r/sae-research-register-C331/register.html accompanies the survey.