Synapse
⌘+K
Synapse
PulseExploreClubsResearchersJournals
Instagram
HomeClubsExplore
October 9, 2025Open Access

Parallel Scaling Law: Unveiling Reasoning Generalization through A Cross-Linguistic Perspective

View Full Paper
Ask AI
Bookmark
Share

Authors

WYWen YangWuhan UniversityJWJunhong WuHainan Normal UniversityCLChong LiGansu Agricultural University

Discussion

Loading...

Member takes

Implication

This analysis reveals variations in cross-lingual transferability in LRMs, suggesting implications for multilingual reasoning development.

Key Points

  • Cross-lingual transferability in LRMs varies by model and language, complicating reasoning generalization.
  • Evaluation metrics show that initial English capabilities negatively impact reasoning performance in other languages.
  • A parallel training study indicates a predictable scaling law, revealing insights into reasoning generalization across languages.
  • Identifying the monolingual generalization gap highlights limitations in English-centric LRMs for multilingual contexts.

Cite This Study

Yang et al. (2025) studied this question.

synapsesocial.com/papers/68e7d631bd66d359be6266dahttps://doi.org/10.48550/arxiv.2510.02272
View Full Paper
Ask AI
Bookmark
Share

Also Consider

Synapse has enriched 5 closely related papers on similar clinical questions. Consider them for comparative context:

  1. 1Large Language Models Are Cross-Lingual Knowledge-Free Reasoners2024 · 1 citations
  2. 2Best-of-L: Cross-Lingual Reward Modeling for Mathematical Reasoning2025
  3. 3An Empirical Study on Reasoning and Generalization in Large Language Models2026
  4. 4Scaling Up RL: Unlocking Diverse Reasoning in LLMs via Prolonged Training2025
  5. 5When Models Reason in Your Language: Controlling Thinking Language Comes at the Cost of Accuracy2025