Cross-lingual Representation Learning via Centroid Intervention Fusion

arXiv:2608.26357v1 Announce Type: new Abstract: Large language models (LLMs) exhibit uneven multilingual performance, especially when dealing with low-resource languages. Inference-time intervention offers a lightweight way to improve cross-lingual transfer by modifying the hidden states produced by the LLMs during the forward pass, without updating model parameters. However, existing cross-lingual intervention methods typically learn separate projections from source to target languages, which limits scalability and prevents knowledge sharing across languages. We propose Centroid Intervention Fusion (CIF), a projection fusion framework that consolidates multiple multilingual intervention projections into a single language-shared operator. Across multilingual commonsense reasoning, natural language inference, factual editing, and machine translation benchmarks, CIF outperforms the strongest prior pairwise intervention baseline by up to +3.378 pp on average across four model backbones, while supporting performance gains for low resource languages. The code is available at https://github.com/VRCMF/CIF.git.

Related

• Reasoning over Grammar: Can Synthetic Linguistic Reasoning Traces Enhance Low-Resource Machine Translation?
• Heterogeneity in Formal Linguistic Competence of Language Models: Is Data the Real Bottleneck?
• Exploring Continual Fine-Tuning for Enhancing Language Ability in Large Language Model

Source: arXiv cs.CL | 2026-08-28