# AdaMCoT: Rethinking Cross-Lingual Factual Reasoning through Adaptive Multilingual Chain-of-Thought

Source: [AdaMCoT: Rethinking Cross-Lingual Factual Reasoning through Adaptive Multilingual Chain-of-Thought](<https://www.sota2.com/research/paper/68fa7a601e209131cea82b5f>)

Collection index: [llms.txt](<https://www.sota2.com/research/llms.txt>)

Data updated: 2026-04-16T05:00:04.683Z

Authors: Weihua Zheng, Xin Huang, Zhengyuan Liu, Tarun Kumar Vangani, Bowei Zou, Xiyan Tao, Yuhao Wu, Ai Ti Aw, Nancy F. Chen, Roy Ka-Wei Lee

Published: 2025-01-27T15:48:57.000Z

## Abstract

Large language models (LLMs) have shown impressive multilingual capabilities through pretraining on diverse corpora. Although these models show strong reasoning abilities, their performance varies significantly between languages due to the imbalanced distribution of training data. Existing approaches using sample-level translation for extensive multilingual pretraining and cross-lingual tuning face scalability challenges and often fail to capture nuanced reasoning processes across languages. In this paper, we introduce AdaMCOT (Adaptive Multilingual Chain-of-Thought), a framework that enhances multilingual factual reasoning by dynamically routing thought processes in intermediary "thinking languages" before generating target-language responses. AdaMCOT leverages a language-agnostic core and incorporates an adaptive, reward-based mechanism for selecting optimal reasoning pathways without requiring additional pretraining. Our comprehensive evaluation across multiple benchmarks demonstrates substantial improvements in both factual reasoning quality and cross-lingual consistency, with particularly strong performance gains in low-resource language settings. An in-depth analysis of the model's hidden states and semantic space further elucidates the underlying mechanism of our method. The results suggest that adaptive reasoning paths can effectively bridge the performance gap between high and low-resource languages while maintaining cultural and linguistic nuances.

## Recorded benchmark results

A research result applies to its stated dataset, split, metric and evaluation configuration. Compare like-for-like settings, preserve the metric's unit and direction, and distinguish paper-reported results from independent reproduction. The best result recorded here is not a claim of exhaustive, current global coverage.

- [Multilingual Factual Reasoning on mTruthfulQA](<https://www.sota2.com/research/sota/multilingual-factual-reasoning-on-mtruthfulqa.md>): Multilingual Factual Reasoning; mTruthfulQA; Score (ar): 54.46.

- [Multilingual Logical Reasoning on CrossMMLU](<https://www.sota2.com/research/sota/multilingual-logical-reasoning-on-crossmmlu.md>): Multilingual Logical Reasoning; CrossMMLU; Accuracy (English): 84.7.

- [Multilingual Logical Reasoning on CrossLogiQA](<https://www.sota2.com/research/sota/multilingual-logical-reasoning-on-crosslogiqa.md>): Multilingual Logical Reasoning; CrossLogiQA; Accuracy (en): 75.

- [Open-ended Question Answering on CrossAlpaca-Eval zh 2.0](<https://www.sota2.com/research/sota/open-ended-question-answering-on-crossalpaca-eval-zh-2-0.md>): Open-ended Question Answering; CrossAlpaca-Eval zh 2.0; GPT-4o Score: 8.53.

- [Open-ended Question Answering on CrossAlpaca-Eval en 2.0](<https://www.sota2.com/research/sota/open-ended-question-answering-on-crossalpaca-eval-en-2-0.md>): Open-ended Question Answering; CrossAlpaca-Eval en 2.0; GPT-4o Score: 8.58.

## Original sources and implementation

[Original paper](<https://arxiv.org/abs/2501.16154>)

[PDF](<https://arxiv.org/pdf/2501.16154.pdf>)