Towards Better Understanding of Program-of-Thought Reasoning in Cross-Lingual and Multilingual Environments

TL;DR

This paper investigates Program-of-Thought prompting for multilingual reasoning in large language models, demonstrating that fine-tuning improves reasoning and answer accuracy across languages by separating reasoning from execution.

Contribution

It introduces a framework to evaluate PoT prompting in multilingual contexts and shows that fine-tuning enhances reasoning quality and answer correctness.

Findings

PoT fine-tuning outperforms CoT in multilingual reasoning

Reasoning quality strongly correlates with answer accuracy

Separating reasoning from execution benefits multilingual LLM performance

Abstract

Multi-step reasoning is essential for large language models (LLMs), yet multilingual performance remains challenging. While Chain-of-Thought (CoT) prompting improves reasoning, it struggles with non-English languages due to the entanglement of reasoning and execution. Program-of-Thought (PoT) prompting separates reasoning from execution, offering a promising alternative but shifting the challenge to generating programs from non-English questions. We propose a framework to evaluate PoT by separating multilingual reasoning from code execution to examine (i) the impact of fine-tuning on question-reasoning alignment and (ii) how reasoning quality affects answer correctness. Our findings demonstrate that PoT fine-tuning substantially enhances multilingual reasoning, outperforming CoT fine-tuned models. We further demonstrate a strong correlation between reasoning quality (measured through code quality) and answer accuracy, highlighting its potential as a test-time performance improvement heuristic.