LinguaLens: Towards Interpreting Linguistic Mechanisms of Large Language Models via Sparse Auto-Encoder

TL;DR

LinguaLens introduces a comprehensive framework using Sparse Auto-Encoders to analyze and interpret the linguistic mechanisms within large language models across multiple languages and linguistic features.

Contribution

This work presents a novel systematic approach for analyzing linguistic mechanisms in LLMs, including a large-scale counterfactual dataset and insights into cross-layer and cross-lingual representations.

Findings

Intrinsic linguistic representations in LLMs identified

Patterns of cross-layer and cross-lingual distribution uncovered

Demonstrates potential for controlling model outputs

Abstract

Large language models (LLMs) demonstrate exceptional performance on tasks requiring complex linguistic abilities, such as reference disambiguation and metaphor recognition/generation. Although LLMs possess impressive capabilities, their internal mechanisms for processing and representing linguistic knowledge remain largely opaque. Prior research on linguistic mechanisms is limited by coarse granularity, limited analysis scale, and narrow focus. In this study, we propose LinguaLens, a systematic and comprehensive framework for analyzing the linguistic mechanisms of large language models, based on Sparse Auto-Encoders (SAEs). We extract a broad set of Chinese and English linguistic features across four dimensions (morphology, syntax, semantics, and pragmatics). By employing counterfactual methods, we construct a large-scale counterfactual dataset of linguistic features for mechanism analysis. Our findings reveal intrinsic representations of linguistic knowledge in LLMs, uncover patterns of cross-layer and cross-lingual distribution, and demonstrate the potential to control model outputs. This work provides a systematic suite of resources and methods for studying linguistic mechanisms, offers strong evidence that LLMs possess genuine linguistic knowledge, and lays the foundation for more interpretable and controllable language modeling in future research.