Large Reasoning Models Are (Not Yet) Multilingual Latent Reasoners

2026-01-06

arXiv.org

AI Summary

This study investigates the distribution and mechanistic consistency of latent reasoning capabilities in large reasoning models across multilingual settings. Focusing on 11 languages, the authors systematically evaluate model performance on low-resource languages and complex tasks through chain-of-thought truncation, hidden state analysis, and cross-lingual representation similarity metrics. The work reveals, for the first time, a pronounced imbalance in multilingual latent reasoning abilities: high-resource languages exhibit stronger reasoning capacities, while low-resource languages show weaker performance, with reasoning signals generally attenuated across all languages on challenging tasks. Despite this disparity, internal reasoning pathways across languages demonstrate high representational alignment with an English-centric pattern, suggesting a consistent underlying reasoning mechanism that transcends linguistic boundaries.

Technology Category

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

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
• User Modeling, Personalization and Recommendation: Large Language Models (LLM) for user modeling and recommendation

Abstract

Large reasoning models (LRMs) achieve strong performance on mathematical reasoning tasks, often attributed to their capability to generate explicit chain-of-thought (CoT) explanations. However, recent work shows that LRMs often arrive at the correct answer before completing these textual reasoning steps, indicating the presence of latent reasoning -- internal, non-verbal computation encoded in hidden states. While this phenomenon has been explored in English, its multilingual behavior remains largely unknown. In this paper, we conduct a systematic investigation of multilingual latent reasoning in LRMs across 11 languages. Using a truncation-based strategy, we examine how the correct answer emerges as the model is given only partial reasoning traces, allowing us to measure stepwise latent prediction formation. Our results reveal clear evidence of multilingual latent reasoning, though unevenly: strong in resource-rich languages, weaker in low-resource ones, and broadly less observable on harder benchmarks. To understand whether these differences reflect distinct internal mechanisms, we further perform representational analyses. Despite surface-level disparities, we find that the internal evolution of predictions is highly consistent across languages and broadly aligns with English -- a pattern suggesting an English-centered latent reasoning pathway.

Problem

• multilingual latent reasoning
• large reasoning models
• chain-of-thought
• low-resource languages
• hidden states

Innovation

• multilingual latent reasoning
• large reasoning models
• chain-of-thought truncation
• representational analysis
• cross-lingual alignment