并行缩放定律：从跨语言视角揭示推理泛化能力
Parallel Scaling Law: Unveiling Reasoning Generalization through A Cross-Linguistic Perspective
AI总结：
该研究从跨语言视角探究大型推理模型的推理泛化问题，提出并行缩放定律等三项关键发现，揭示跨语言推理迁移的幂律规律，为开发语言无关的推理模型提供参考。
摘要：
强化后训练（Reinforcement Post-Training, RPT）的最新进展显著提升了大型推理模型（Large Reasoning Models, LRMs）的能力，引发了人们对基于强化学习的推理泛化的更多关注。现有研究主要聚焦于探究其在不同任务或模态间的泛化性，本研究则提出了一种全新的跨语言视角来研究推理泛化问题，这引出了一个关键问题：通过英文RPT获得的推理能力能否有效迁移到其他语言？\n为解答该问题，我们在多语言推理基准上系统评估了以英文为中心的LRM，并提出了一种量化跨语言迁移性的指标。研究发现，跨语言迁移性因初始模型、目标语言和训练范式的不同而存在显著差异。通过干预研究我们发现，初始英文能力更强的模型往往过度依赖英文特有的模式，导致跨语言泛化能力下降。\n为解决这一问题，我们开展了全面的并行训练研究，实验结果得出三项关键发现：第一是“并行跃升（Parallel Leap）”，即从单语言训练仅增加一种并行语言时，性能会出现大幅跃升；第二是可预测的“并行缩放定律（Parallel Scaling Law）”，表明跨语言推理迁移效果与训练所用并行语言的数量呈幂律关系。此外，我们将实际单语言性能与幂律预测值之间的差异定义为“单语言泛化差距（Monolingual Generalization Gap）”，这表明以英文为中心的LRM无法实现完全的跨语言泛化。本研究对LRM的推理能力与人类认知相似的假设提出了质疑，为开发更具语言无关性的LRM提供了关键见解。
英文摘要：
Recent advancements in Reinforcement Post-Training (RPT) have significantly enhanced the capabilities of Large Reasoning Models (LRMs), sparking increased interest in the generalization of RL-based reasoning. While existing work has primarily focused on investigating its generalization across tasks or modalities, this study proposes a novel cross-linguistic perspective to investigate reasoning generalization. This raises a crucial question: $\textit{Does the reasoning capability achieved from English RPT effectively transfer to other languages?}$ We address this by systematically evaluating English-centric LRMs on multilingual reasoning benchmarks and introducing a metric to quantify cross-lingual transferability. Our findings reveal that cross-lingual transferability varies significantly across initial model, target language, and training paradigm. Through interventional studies, we find that models with stronger initial English capabilities tend to over-rely on English-specific patterns, leading to diminished cross-lingual generalization. To address this, we conduct a thorough parallel training study. Experimental results yield three key findings: $\textbf{First-Parallel Leap}$, a substantial leap in performance when transitioning from monolingual to just a single parallel language, and a predictable $\textbf{Parallel Scaling Law}$, revealing that cross-lingual reasoning transfer follows a power-law with the number of training parallel languages. Moreover, we identify the discrepancy between actual monolingual performance and the power-law prediction as $\textbf{Monolingual Generalization Gap}$, indicating that English-centric LRMs fail to fully generalize across languages. Our study challenges the assumption that LRM reasoning mirrors human cognition, providing critical insights for the development of more language-agnostic LRMs.
机构、总结、中文摘要均由AI分析而来，可能存在错误，请以论文正文为准。
