Language as a Latent Variable for Reasoning Optimization

• URL: http://arxiv.org/abs/2604.21593v1
• Date: Thu, 23 Apr 2026 12:19:14 GMT
• Title: Language as a Latent Variable for Reasoning Optimization
• Abstract summary: As LLMs reduce English-centric bias, a surprising trend emerges: non-English responses sometimes outperform English on reasoning tasks.<n>We hypothesize that language functions as a latent variable that structurally modulates the model's internal inference pathways.<n>We propose polyGRPO, an RL framework that treats language variation as an implicit exploration signal.
• Score: 45.35129925776798
• License: http://arxiv.org/licenses/nonexclusive-distrib/1.0/
• Abstract: As LLMs reduce English-centric bias, a surprising trend emerges: non-English responses sometimes outperform English on reasoning tasks. We hypothesize that language functions as a latent variable that structurally modulates the model's internal inference pathways, rather than merely serving as an output medium. To test this, we conducted a Polyglot Thinking Experiment, in which models were prompted to solve identical problems under language-constrained and language-unconstrained conditions. Results show that non-English responses often achieve higher accuracy, and the best performance frequently occur when language is unconstrained, suggesting that multilinguality broadens the model's latent reasoning space. Based on this insight, we propose polyGRPO (Polyglot Group Relative Policy Optimization), an RL framework that treats language variation as an implicit exploration signal. It generates polyglot preference data online under language-constrained and unconstrained conditions, optimizing the policy with respect to both answer accuracy and reasoning structure. Trained on only 18.1K multilingual math problems without chain-of-thought annotations, polyGRPO improves the base model (Qwen2.5-7B-Instruct) by 6.72% absolute accuracy on four English reasoning testset and 6.89% in their multilingual benchmark. Remarkably, it is the only method that surpasses the base LLM on English commonsense reasoning task (4.9%), despite being trained solely on math data-highlighting its strong cross-task generalization. Further analysis reveals that treating language as a latent variable expands the model's latent reasoning space, yielding consistent and generalizable improvements in reasoning performance.

Related papers

• Building Multilingual Bridges: Data Mixing as the Pillar of Generalization for In-Language Reasoning [75.46846056993418]
  Reasoning language models primarily reason in English regardless of the language they are prompted in.<n>This is inaccessible for non-English-speaking users, risks losing the intent of the original question, and forgoes knowledge more readily expressed in the target language.<n>In this work, we advance L2 reasoning, the ability of a model to reason consistently in the language of the user's prompt, thus building an in-language bridge between the prompt and the answer.
  arXiv  Detail & Related papers  (2026-09-09T16:56:25Z)
• AdaMame: A Training Recipe for Adaptive Multilingual Reasoning [32.31269959716733]
  Large Reasoning Models (LRMs) often fail to reason in the language of the query.<n>AdaMame is a two-stage training recipe for multilingual mathematical reasoning.
  arXiv  Detail & Related papers  (2026-06-13T03:22:35Z)
• Align to the Pivot: Dual Alignment with Self-Feedback for Multilingual Math Reasoning [71.4175109189942]
  We present Pivot-Aligned Self-Feedback Multilingual Reasoning (PASMR)<n>This approach designates the model's primary language as the pivot language.<n>It establishes a cross-lingual self-feedback mechanism without relying on external correct answers or reward models.
  arXiv  Detail & Related papers  (2026-01-25T03:20:00Z)
• Aligning Multilingual Reasoning with Verifiable Semantics from a High-Resource Expert Model [13.788758077632432]
  We introduce Pivot-Based Reinforcement Learning with Semantically Verifiable Rewards.<n>This framework enhances multilingual reasoning by circumventing the need for human-annotated data in target languages.<n>We show that our method significantly narrows the performance gap between English and other languages.
  arXiv  Detail & Related papers  (2025-09-29T22:03:11Z)
• 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)
• Cross-Lingual Consistency: A Novel Inference Framework for Advancing Reasoning in Large Language Models [10.231866835957538]
  Chain-of-thought (CoT) has emerged as a critical mechanism for enhancing reasoning capabilities in large language models (LLMs)<n>We propose the Cross-Lingual Consistency (CLC) framework, which integrates multilingual reasoning paths through majority voting to elevate LLMs' reasoning capabilities.<n> Empirical evaluations on the CMATH dataset reveal CLC's superiority over the conventional self-consistency method.
  arXiv  Detail & Related papers  (2025-04-02T16:09:39Z)
• Assessing Agentic Large Language Models in Multilingual National Bias [31.67058518564021]
  Cross-language disparities in reasoning-based recommendations remain largely unexplored.<n>This study is the first to address this gap.<n>We investigate multilingual bias in state-of-the-art LLMs by analyzing their responses to decision-making tasks across multiple languages.
  arXiv  Detail & Related papers  (2025-02-25T08:07:42Z)
• SLAM: Towards Efficient Multilingual Reasoning via Selective Language Alignment [78.4550589538805]
  We propose an efficient multilingual reasoning alignment approach that precisely identifies and fine-tunes the layers responsible for handling multilingualism.<n> Experimental results show that our method, SLAM, only tunes 6 layers' feed-forward sub-layers including 6.5-8% of all parameters within 7B and 13B LLMs.
  arXiv  Detail & Related papers  (2025-01-07T10:29:43Z)
• The Power of Question Translation Training in Multilingual Reasoning: Broadened Scope and Deepened Insights [108.40766216456413]
  We propose a question alignment framework to bridge the gap between large language models' English and non-English performance. Experiment results show it can boost multilingual performance across diverse reasoning scenarios, model families, and sizes. We analyze representation space, generated response and data scales, and reveal how question translation training strengthens language alignment within LLMs.
  arXiv  Detail & Related papers  (2024-05-02T14:49:50Z)
• Breaking Language Barriers in Multilingual Mathematical Reasoning: Insights and Observations [59.056367787688146]
  This paper pioneers exploring and training powerful Multilingual Math Reasoning (xMR) LLMs. We construct the first multilingual math reasoning instruction dataset, MGSM8KInstruct, encompassing ten distinct languages. By utilizing translation, we construct the first multilingual math reasoning instruction dataset, MGSM8KInstruct, encompassing ten distinct languages.
  arXiv  Detail & Related papers  (2023-10-31T08:09:20Z)

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.