Get Started
Home
Topics
Search
Library
Research questionHow can multilingual representation sharing be measured without confusing anisotropy with genuine cross-lingual structure?Different metrics can assign different amounts of sharing to the same multilingual representations. When representations cluster in a narrow region of the embedding space, geometric effects can make a measurement look like evidence of cross-lingual structure.
AI
Evaluation & Benchmarks
Machine Learning
Mechanistic Interpretability
Natural Language Processing
Research Paper
Latest papersRecent research connected to this question, newest first.A Systematic Comparison of Multilingual Interpretability Methods Reveals Anisotropy-Driven FailuresThe study compares CKA, ANC, GMM dominance per token, and ILO across 21 base models from five families, ranging from 125M to 14B parameters, and examines their relationships with transfer on five downstream tasks. It also analyzes anisotropy and controls for model size, model family, and per-task variation; only ILO's reported correlation with transfer survives those controls.research paper · Sep 4, 2026
Related questions
How can sparse autoencoder features be shared across language models without per-model retraining?When does weight-space merging preserve translation quality across models with shared versus different target languages?How can multilingual question answering remain consistent across languages without erasing culturally appropriate differences?How can we compare feature contributions to language-model representations when those features are correlated?