Sparse Autoencoders Can Capture Language-Specific Concepts Across Diverse Languages
This work addresses the problem of understanding multilingual mechanisms in LLMs for researchers, offering an incremental improvement by applying sparse autoencoders to identify language-specific features.
The authors tackled the challenge of isolating language-specific units in multilingual large language models by introducing SAE-LAPE, a method using sparse autoencoders and feature activation probability, which identified interpretable language-specific features in middle to final layers that achieved language identification performance comparable to fastText.
Understanding the multilingual mechanisms of large language models (LLMs) provides insight into how they process different languages, yet this remains challenging. Existing studies often focus on individual neurons, but their polysemantic nature makes it difficult to isolate language-specific units from cross-lingual representations. To address this, we explore sparse autoencoders (SAEs) for their ability to learn monosemantic features that represent concrete and abstract concepts across languages in LLMs. While some of these features are language-independent, the presence of language-specific features remains underexplored. In this work, we introduce SAE-LAPE, a method based on feature activation probability, to identify language-specific features within the feed-forward network. We find that many such features predominantly appear in the middle to final layers of the model and are interpretable. These features influence the model's multilingual performance and language output and can be used for language identification with performance comparable to fastText along with more interpretability. Our code is available at https://github.com/LyzanderAndrylie/language-specific-features