Do LLMs Need Inherent Reasoning Before Reinforcement Learning? A Study in Korean Self-Correction

2026-01-09

arXiv.org

AI Summary

This work addresses the limited reasoning and self-correction capabilities of large language models in low-resource languages such as Korean, where direct application of reinforcement learning yields marginal gains. The authors propose that effective reinforcement learning first requires aligning the model’s internal reasoning mechanisms with Korean inputs through targeted fine-tuning. To this end, they introduce a fine-tuning strategy that activates Korean-specific neurons in early layers and construct the first Korean–English code-switched self-correction dataset. Experimental results demonstrate significant performance improvements in both mathematical reasoning and self-correction tasks, revealing that multilingual reasoning hinges on activating—rather than injecting—language-specific knowledge. These findings underscore the critical role of internal reasoning alignment in unlocking the potential of reinforcement learning for low-resource languages.

Technology Category

• Natural Language Processing: (Large) Language Models
• Machine Learning: Reinforcement Learning
• Knowledge Representation and Reasoning: Knowledge Representation Languages

Application Category

• Search and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for ranking
• Semantics and Knowledge: Methods to enhance, augment, integrate or synergize semantic models such as knowledge graphs and LLMs
• User Modeling, Personalization and Recommendation: Large Language Models (LLM) for user modeling and recommendation

Abstract

Large Language Models (LLMs) demonstrate strong reasoning and self-correction abilities in high-resource languages like English, but their performance remains limited in low-resource languages such as Korean. In this study, we investigate whether reinforcement learning (RL) can enhance Korean reasoning abilities to a degree comparable to English. Our findings reveal that RL alone yields limited improvements when applied to models lacking inherent Korean reasoning capabilities. To address this, we explore several fine-tuning strategies and show that aligning the model's internal reasoning processes with Korean inputs-particularly by tuning Korean-specific neurons in early layers-is key to unlocking RL's effectiveness. We introduce a self-correction code-switching dataset to facilitate this alignment and observe significant performance gains in both mathematical reasoning and self-correction tasks. Ultimately, we conclude that the crucial factor in multilingual reasoning enhancement is not injecting new linguistic knowledge, but effectively eliciting and aligning existing reasoning capabilities. Our study provides a new perspective on how internal translation and neuron-level tuning contribute to multilingual reasoning alignment in LLMs.

Problem

• Large Language Models
• Low-resource Languages
• Reasoning
• Reinforcement Learning
• Self-correction

Innovation

• reasoning alignment
• neuron-level tuning
• code-switching dataset
• multilingual LLMs
• self-correction