Steering beyond Language: Improving Multilingual Training-Free Text Embeddings via Paired-Residual Subspace Removal
This work proposes Paired-Residual Subspace Removal (PRSR), a training-free inference-time intervention that steers LLM representations toward a more language-invariant semantic space and demonstrates that PRSR consistently improves the performance of existing training-free baselines in both monolingual and cross-lingu...