The Reasoning Lingua Franca: A Double-Edged Sword for Multilingual AI

2025-10-23

AI Summary

Large reasoning models (LRMs) exhibit a pervasive “English-default” bias in multilingual reasoning, undermining cultural contextual understanding and interpretability. This work presents the first systematic investigation of cross-lingual cognitive reasoning behavior, analyzing inference pathways across languages on the MGSM and GPQA Diamond benchmarks. We find that while English-based reasoning significantly improves overall accuracy—especially on complex tasks—it introduces a novel error class: *translation drift*, wherein critical semantic content from the source language is distorted during translation into English prior to reasoning. Our analysis reveals a fundamental tension between linguistic translation and reasoning capability, demonstrating that accuracy alone is insufficient for evaluating multilingual reasoning. We argue that semantic fidelity must be jointly optimized with correctness, and provide cognitive-level evidence supporting the development of truly multilingual-native reasoning models—models that reason natively in diverse languages without mandatory English mediation.

Technology Category

• Natural Language Processing: Machine Translation, Multilinguality, Cross-Lingual NLP
• Knowledge Representation and Reasoning: Computational Complexity of Reasoning
• Cognitive Modeling & Cognitive Systems: Conceptual Inference and Reasoning

Application Category

• Search and Retrieval-Augmented AI: Multilingual and cross-lingual Web search
• Web Mining and Content Analysis: Mining multimedia, multimodal, multilingual, cross-lingual Web data
• Semantics and Knowledge: Data modeling to support human-machine intelligence, including LLMs agents, intelligent system behavior, explanations, and user-friendly interactions

Abstract

Large Reasoning Models (LRMs) achieve strong performance on mathematical, scientific, and other question-answering tasks, but their multilingual reasoning abilities remain underexplored. When presented with non-English questions, LRMs often default to reasoning in English, raising concerns about interpretability and the handling of linguistic and cultural nuances. We systematically compare an LRM's reasoning in English versus the language of the question. Our evaluation spans two tasks: MGSM and GPQA Diamond. Beyond measuring answer accuracy, we also analyze cognitive attributes in the reasoning traces. We find that English reasoning traces exhibit a substantially higher presence of these cognitive behaviors, and that reasoning in English generally yields higher final-answer accuracy, with the performance gap increasing as tasks become more complex. However, this English-centric strategy is susceptible to a key failure mode - getting "Lost in Translation," where translation steps lead to errors that would have been avoided by question's language reasoning.

Problem

• Exploring multilingual reasoning abilities in Large Reasoning Models
• Analyzing performance gap between English and native language reasoning
• Identifying translation-induced errors in non-English question processing

Innovation

• Evaluated multilingual reasoning performance across languages
• Analyzed cognitive behaviors in reasoning traces systematically
• Identified translation-induced errors in English-centric reasoning