{"slug": "language-specialized-multi-teacher-on-policy-distillation-for-multilingual-llm", "title": "Language-Specialized Multi-Teacher On-Policy Distillation for Multilingual LLM-Based ASR", "summary": "Researchers propose Language-Specialized Multi-Teacher On-Policy Distillation (LS-MOPD), a method that uses language-specialized teachers optimized via reinforcement learning and integrates their expertise into a generalist multilingual student through language routing and token-level distillation. In experiments on benchmarks covering Mandarin, Mandarin subdialects, Cantonese, and English, LS-MOPD outperformed RL baselines and surpassed the performance of the best-performing RL teachers, suggesting it can generalize beyond all teachers in multilingual ASR.", "body_md": "arXiv:2608.03610v1 Announce Type: new\nAbstract: Modern LLM-based ASR systems have established multilingual capability as a standard feature, leveraging large-scale multilingual corpora and LLMs' cross-lingual knowledge to achieve competitive performance across multilingual benchmarks. However, joint modeling of languages with heterogeneous acoustic, phonological, and lexical characteristics inevitably introduces optimization conflicts, undermining language-wise specialization. To address this challenge, we propose Language-Specialized Multi-Teacher On-Policy Distillation (LS-MOPD), which decouples language-specific knowledge acquisition from multilingual capability integration: language-specialized teachers are independently optimized via reinforcement learning (RL), after which their expertise is integrated into a generalist multilingual student through language routing and token-level multi-teacher distillation, thereby reducing direct cross-lingual optimization conflicts. We further explore two acoustic-prefix configurations, static and dynamic, to examine how teacher--student prefix consistency influences the efficacy of on-policy distillation. Experiments on benchmarks covering Mandarin, Mandarin subdialects, Cantonese, and English demonstrate that LS-MOPD substantially outperforms RL baselines and consistently surpasses the empirical performance envelope defined by best-performing RL teachers, revealing its potential to generalize beyond all teachers in multilingual ASR.", "url": "https://wpnews.pro/news/language-specialized-multi-teacher-on-policy-distillation-for-multilingual-llm", "canonical_source": "https://www.machinebrief.com/news/language-specialized-multi-teacher-on-policy-distillation-fo-7h15", "published_at": "2026-08-05 04:00:00+00:00", "updated_at": "2026-08-05 06:35:59.944128+00:00", "lang": "en", "topics": ["machine-learning", "artificial-intelligence", "natural-language-processing"], "entities": ["arXiv"], "alternates": {"html": "https://wpnews.pro/news/language-specialized-multi-teacher-on-policy-distillation-for-multilingual-llm", "markdown": "https://wpnews.pro/news/language-specialized-multi-teacher-on-policy-distillation-for-multilingual-llm.md", "text": "https://wpnews.pro/news/language-specialized-multi-teacher-on-policy-distillation-for-multilingual-llm.txt", "jsonld": "https://wpnews.pro/news/language-specialized-multi-teacher-on-policy-distillation-for-multilingual-llm.jsonld"}}