LaCoMSA: Language-Consistency Multilingual Self-Alignment with Latent Representation Rewarding
Year: 2026
Type: conference-paper
Abstract: Large Language Models (LLMs) have achieved impressive performance yet remain inconsistent across languages, often defaulting to highresource outputs such as English.Existing multilingual alignment methods mitigate these issues through preference optimization but rely on external supervision, such as translation systems or English-biased signal.We propose Multilingual Self-Alignment (MSA), a targeted preference optimization framework that leverages an LLM's own latent representations as intrinsic supervision signals, rewarding lower-resource language outputs based on their alignment with high-resource (English) counterparts in the "semantic hub".We further introduce Language-Consistency MSA (La-CoMSA), which augments MSA with a finallayer language-consistency factor to prevent off-target generation.Integrated with Direct Preference Optimization, LaCoMSA improves a Llama 3 8B-based model multilingual win rates by up to 6.8% absolute (55.0%relatively) on X-AlpacaEval and achieves consistent gains across benchmarks and models.Our findings demonstrate that LaCoMSA can serve as an effective and scalable mechanism, opening a new venue toward multilingual self-alignment.
Language: English
Subfield: Artificial Intelligence
Field: Computer Science
Domain: Physical Sciences
Open Access status: hybrid
OpenAlex ID: W7140091342