The Multilingual Mind : A Survey of Multilingual Reasoning in Language Models (2502.09457v1)

Abstract: While reasoning and multilingual capabilities in LLMs (LMs) have achieved remarkable progress in recent years, their integration into a unified paradigm, multilingual reasoning, is at a nascent stage. Multilingual reasoning requires LLMs to handle logical reasoning across languages while addressing misalignment, biases, and challenges in low-resource settings. This survey provides the first in-depth review of multilingual reasoning in LMs. In this survey, we provide a systematic overview of existing methods that leverage LMs for multilingual reasoning, specifically outlining the challenges, motivations, and foundational aspects of applying LLMs to reason across diverse languages. We provide an overview of the standard data resources used for training multilingual reasoning in LMs and the evaluation benchmarks employed to assess their multilingual capabilities. Next, we analyze various state-of-the-art methods and their performance on these benchmarks. Finally, we explore future research opportunities to improve multilingual reasoning in LMs, focusing on enhancing their ability to handle diverse languages and complex reasoning tasks.

• The paper surveys the emerging field of multilingual reasoning in Language Models, covering motivations, challenges, methodologies, data resources, and evaluation benchmarks.
• It examines the standard data resources and evaluation benchmarks used to train and assess LMs for multilingual reasoning tasks.
• The survey analyzes state-of-the-art methods' performance and explores future research directions to enhance multilingual reasoning capabilities in LMs.

Survey of Multilingual Reasoning in LLMs

The paper "The Multilingual Mind : A Survey of Multilingual Reasoning in LLMs" (2502.09457) provides an in-depth review of multilingual reasoning in LLMs (LMs). The survey addresses the challenges, motivations, and foundational aspects of using LMs to reason across diverse languages. It also examines standard data resources for training multilingual reasoning in LMs and the evaluation benchmarks used to assess their multilingual capabilities. The paper analyzes state-of-the-art methods and their performance on these benchmarks and explores future research opportunities.

Core Focus and Motivation

The central theme of this survey revolves around the integration of reasoning and multilingual capabilities into a unified paradigm, referred to as multilingual reasoning. The authors emphasize that while LMs have demonstrated notable advancements in both reasoning and multilingual understanding individually, the combined field is still in its early stages. Multilingual reasoning involves enabling LMs to perform logical reasoning across multiple languages, addressing issues such as misalignment, biases, and limitations in low-resource languages. This survey aims to provide a structured overview of existing methodologies that employ LMs for multilingual reasoning, thereby highlighting the challenges, underlying motivations, and fundamental elements of applying these models across diverse linguistic contexts.

Data Resources and Evaluation Benchmarks

A significant component of the survey is the examination of data resources and evaluation benchmarks employed in the context of multilingual reasoning. These resources play a crucial role in training and assessing the capabilities of LMs in handling multilingual reasoning tasks. The survey identifies and discusses the standard datasets and benchmarks used in the field, providing insights into their characteristics, strengths, and limitations. By analyzing these resources, the authors aim to offer a comprehensive understanding of the current landscape of data and evaluation methodologies in multilingual reasoning.

State-of-the-Art Methods and Performance Analysis

The survey includes a detailed analysis of state-of-the-art methods for multilingual reasoning in LMs. Various approaches are scrutinized, with a focus on their methodologies and performance across different benchmarks. The authors delve into the technical aspects of these methods, evaluating their effectiveness in addressing the challenges associated with multilingual reasoning. This analysis provides a comparative perspective on the strengths and weaknesses of different approaches, contributing to a deeper understanding of the current state of the art in the field.

Future Research Directions

In the concluding section, the survey explores potential future research directions aimed at enhancing multilingual reasoning in LMs. The authors identify key areas for improvement, with a particular emphasis on enhancing the ability of LMs to handle diverse languages and complex reasoning tasks. These research opportunities are geared toward addressing the existing limitations and pushing the boundaries of multilingual reasoning in LMs. The proposed directions are intended to stimulate further investigation and development in the field, fostering advancements in the capabilities of LMs to reason effectively across multiple languages.

In summary, this survey paper comprehensively reviews the emerging field of multilingual reasoning in LMs, covering motivations, challenges, methodologies, data resources, evaluation benchmarks, and future research opportunities. It serves as a valuable resource for researchers and practitioners interested in the intersection of reasoning and multilingual capabilities in LLMs.