Ko-MuSR: A Multistep Soft Reasoning Benchmark for LLMs Capable of Understanding Korean

TL;DR

Ko-MuSR is a comprehensive Korean reasoning benchmark that evaluates multistep reasoning in long narratives, revealing cross-lingual generalization in large language models and the effectiveness of advanced prompting strategies.

Contribution

It introduces Ko-MuSR, the first detailed benchmark for multistep reasoning in Korean, and demonstrates how prompting strategies improve model performance.

Findings

Multilingual models outperform Korean-specific models in Korean reasoning tasks.

Prompting strategies significantly boost reasoning accuracy, nearing human performance.

Cross-lingual reasoning capabilities are evident in large language models.

Abstract

We present Ko-MuSR, the first benchmark to comprehensively evaluate multistep, soft reasoning in long Korean narratives while minimizing data contamination. Built following MuSR, Ko-MuSR features fully Korean narratives, reasoning chains, and multiple-choice questions verified by human annotators for logical consistency and answerability. Evaluations of four large language models -- two multilingual and two Korean-specialized -- show that multilingual models outperform Korean-focused ones even in Korean reasoning tasks, indicating cross-lingual generalization of reasoning ability. Carefully designed prompting strategies, which combine few-shot examples, reasoning traces, and task-specific hints, further boost accuracy, approaching human-level performance. Ko-MuSR offers a solid foundation for advancing Korean NLP by enabling systematic evaluation of long-context reasoning and prompting strategies.