Tracing Multilingual Representations in LLMs with Cross-Layer Transcoders
This research addresses the problem of understanding and improving multilingual alignment in LLMs for developers and researchers, though it is incremental as it builds on existing analysis methods.
The study investigated how multilingual LLMs internally represent different languages, finding evidence for pivot language representations where early layers are nearly identical across languages and later layers handle language-specific decoding. By analyzing attribution graphs and intervening on language features, they showed that decoding relies on high-frequency features in final layers, which can be manipulated to substitute languages in outputs.
Multilingual Large Language Models (LLMs) can process many languages, yet how they internally represent this diversity remains unclear. Do they form shared multilingual representations with language-specific decoding, and if so, why does performance still favor the dominant training language? To address this, we train a series of LLMs on different mixtures of multilingual data and analyze their internal mechanisms using cross-layer transcoders (CLT) and attribution graphs. Our results provide strong evidence for pivot language representations: the model employs nearly identical representations across languages, while language-specific decoding emerges in later layers. Attribution analyses reveal that decoding relies in part on a small set of high-frequency language features in the final layers, which linearly read out language identity from the first layers in the model. By intervening on these features, we can suppress one language and substitute another in the model's outputs. Finally, we study how the dominant training language influences these mechanisms across attribution graphs and decoding pathways. We argue that understanding this pivot-language mechanism is crucial for improving multilingual alignment in LLMs.