Crosslingual On-Policy Self-Distillation for Multilingual Reasoning

• URL: http://arxiv.org/abs/2605.09548v1
• Date: Sun, 10 May 2026 14:06:09 GMT
• Title: Crosslingual On-Policy Self-Distillation for Multilingual Reasoning
• Abstract summary: Crosslingual On-Policy Self-Distillation (COPSD) transfers a model's own high-resource reasoning behavior to low-resource languages.<n>Experiments on 17 low-resource African languages show that COPSD consistently improves low-resource mathematical reasoning across model sizes.
• Score: 48.68444770923683
• License: http://arxiv.org/licenses/nonexclusive-distrib/1.0/
• Abstract: Large language models (LLMs) have achieved remarkable progress in mathematical reasoning, but this ability is not equally accessible across languages. Especially low-resource languages exhibit much lower reasoning performance. To address this, we propose Crosslingual On-Policy Self-Distillation (COPSD), which transfers a model's own high-resource reasoning behavior to low-resource languages. COPSD uses the same model as student and teacher: the student sees only the low-resource problem, while the teacher receives privileged crosslingual context, including the problem translation and reference solution in English. Training minimizes full-distribution token-level divergence on the student's own rollouts, providing dense supervision while avoiding the sparsity and instability of outcome-only reinforcement learning (RL). Experiments on 17 low-resource African languages show that COPSD consistently improves low-resource mathematical reasoning across model sizes and substantially outperforms Group Relative Policy Optimization (GRPO). Further analyses show that COPSD improves answer-format adherence, strengthens test-time scaling, and generalizes to harder multilingual reasoning benchmarks, with especially large gains for lower-resource languages. We make our code and data available at: https://github.com/cisnlp/COPSD.

Related papers

• Translation or Recitation? Calibrating Evaluation Scores for Machine Translation of Extremely Low-Resource Languages [39.985923582735936]
  We introduce the FRED Difficulty Metrics, which include the Fertility Ratio (F), Retrieval Proxy (R), Pre-training Exposure (E) and Corpus Diversity (D)<n>These metrics reveal that a significant portion of result variability is explained by train-test overlap and pre-training exposure rather than model capability.
  arXiv  Detail & Related papers  (2026-03-26T09:20:17Z)
• Low-Resource Dialect Adaptation of Large Language Models: A French Dialect Case-Study [0.043496401697112684]
  We study the use of continual pre-training (CPT) for dialect learning under tight data and compute budgets.<n>We adapt three large language models to the Qu'ebec French dialect using a very small dataset and benchmark them on the COLE suite.<n>Our experiments demonstrate an improvement on the minority dialect benchmarks with minimal regression on the prestige language benchmarks with under 1% of model parameters updated.
  arXiv  Detail & Related papers  (2025-10-26T16:49:06Z)
• Relic: Enhancing Reward Model Generalization for Low-Resource Indic Languages with Few-Shot Examples [58.55904048776596]
  Most open-source multilingual reward models are primarily trained on preference datasets in high-resource languages.<n>We propose RELIC, a novel in-context learning framework for reward modeling in low-resource Indic languages.
  arXiv  Detail & Related papers  (2025-06-19T17:56:16Z)
• Cross-lingual Collapse: How Language-Centric Foundation Models Shape Reasoning in Large Language Models [44.94287386776289]
  We identify textbfCross-lingual Collapse, a systematic drift in which a multilingual language model reverts to its dominant pre-training language.<n>Our experiments reveal three key findings: (i) GRPO rapidly amplifies pre-training language imbalances, leading to the erosion of low-resource languages within just a few hundred updates; (ii) language consistency reward mitigates this drift but does so at the expense of an almost 5 - 10 pp drop in accuracy.
  arXiv  Detail & Related papers  (2025-06-06T08:08:48Z)
• Language Matters: How Do Multilingual Input and Reasoning Paths Affect Large Reasoning Models? [59.970391602080205]
  Despite multilingual training, LRMs tend to default to reasoning in high-resource languages at test time.<n>Cultural reasoning degrades performance on reasoning tasks but benefits cultural tasks, while safety evaluations exhibit language-specific behavior.
  arXiv  Detail & Related papers  (2025-05-23T02:46:18Z)
• When Less Language is More: Language-Reasoning Disentanglement Makes LLMs Better Multilingual Reasoners [111.50503126693444]
  We show that language-specific ablation consistently boosts multilingual reasoning performance.<n>Compared to post-training, our training-free ablation achieves comparable or superior results with minimal computational overhead.
  arXiv  Detail & Related papers  (2025-05-21T08:35:05Z)
• Scaling Test-time Compute for Low-resource Languages: Multilingual Reasoning in LLMs [3.9530780161144667]
  We investigate the multilingual mechanism by which Large Language Models internally operate in a latent space biased toward their inherently dominant language.<n>We train models to generate the chain-of-thought (CoT) in English while outputting the final response in the target language, given input in the low-resource language.<n>Our experiments demonstrate that this approach, named English-Pivoted CoT Training, outperforms other baselines, with up to 28.33% improvement.
  arXiv  Detail & Related papers  (2025-04-02T16:58:36Z)
• Enhancing Code Generation for Low-Resource Languages: No Silver Bullet [55.39571645315926]
  Large Language Models (LLMs) rely on large and diverse datasets to learn syntax, semantics, and usage patterns of programming languages.<n>For low-resource languages, the limited availability of such data hampers the models' ability to generalize effectively.<n>We present an empirical study investigating the effectiveness of several approaches for boosting LLMs' performance on low-resource languages.
  arXiv  Detail & Related papers  (2025-01-31T12:23:28Z)
• Challenges in Adapting Multilingual LLMs to Low-Resource Languages using LoRA PEFT Tuning [0.4194295877935868]
  This study investigates the effects of Low-Rank Adaptation (LoRA) -Efficient Fine-Tuning (PEFT) on multilingual Gemma models for Marathi.<n>Using a translated dataset with 52,000 instruction-response pairs, our findings reveal that while evaluation performance decline post-fine-tuning, manual assessments frequently suggest that the fine-tuned models outperform their original counterparts.
  arXiv  Detail & Related papers  (2024-11-27T18:14:38Z)
• SMILE: Speech Meta In-Context Learning for Low-Resource Language Automatic Speech Recognition [55.2480439325792]
  Speech Meta In-Context LEarning (SMILE) is an innovative framework that combines meta-learning with speech in-context learning (SICL)<n>We show that SMILE consistently outperforms baseline methods in training-free few-shot multilingual ASR tasks.
  arXiv  Detail & Related papers  (2024-09-16T16:04:16Z)

This list is automatically generated from the titles and abstracts of the papers in this site.

This site does not guarantee the quality of this site (including all information) and is not responsible for any consequences.