Align to the Pivot: Dual Alignment with Self-Feedback for Multilingual Math Reasoning

2026-01-25

AI Summary

This work addresses the performance degradation of large language models in multilingual mathematical reasoning, particularly for low-resource languages, stemming from misalignment between language understanding and reasoning capabilities. To mitigate this, the authors propose PASMR, a novel approach that introduces a pivot-language-based dual-alignment self-feedback mechanism. The method first translates the target-language problem into a high-resource pivot language to align reasoning patterns, then leverages the pivot-language reasoning output to supervise the reasoning process in the target language—all without requiring external answers or reward signals. By integrating multilingual translation, cross-lingual reasoning transfer, and self-supervised fine-tuning, PASMR substantially enhances overall performance on multilingual math reasoning tasks, with especially pronounced gains for low-resource languages.

Technology Category

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

Application Category

Search and Retrieval-Augmented AI: Multilingual and cross-lingual Web searchUser Modeling, Personalization and Recommendation: Large Language Models (LLM) for user modeling and recommendationSemantics and Knowledge: Methods to enhance, augment, integrate or synergize semantic models such as knowledge graphs and LLMs

Abstract

Despite the impressive reasoning abilities demonstrated by large language models (LLMs), empirical evidence indicates that they are not language agnostic as expected, leading to performance declines in multilingual settings, especially for low-resource languages. We attribute the decline to the model's inconsistent multilingual understanding and reasoning alignment. To address this, we present Pivot-Aligned Self-Feedback Multilingual Reasoning (PASMR), aiming to improve the alignment of multilingual math reasoning abilities in LLMs. This approach designates the model's primary language as the pivot language. During training, the model first translates questions into the pivot language to facilitate better alignment of reasoning patterns. The reasoning process in the target language is then supervised by the pivot language's reasoning answers, thereby establishing a cross-lingual self-feedback mechanism without relying on external correct answers or reward models. Extensive experimental results demonstrate that our method enhances both the model's understanding of questions and its reasoning capabilities, leading to notable task improvements.

Problem

• multilingual reasoning
• language alignment
• low-resource languages
• mathematical reasoning
• large language models

Innovation

• multilingual reasoning
• pivot language
• self-feedback
• cross-lingual alignment
• mathematical reasoning