What Makes Good Multilingual Reasoning? Disentangling Reasoning Traces with Measurable Features

TL;DR

This paper investigates what characterizes effective multilingual reasoning in large models, revealing that features associated with reasoning vary across languages and challenging English-centric assumptions.

Contribution

It defines measurable reasoning features, uses autoencoders to discover latent reasoning concepts, and evaluates their effectiveness across multiple languages and models.

Findings

Most features positively correlate with accuracy

Association strength varies across languages

English-centric reward designs may be insufficient

Abstract

Large Reasoning Models (LRMs) still exhibit large performance gaps between English and other languages, yet much current work assumes these gaps can be closed simply by making reasoning in every language resemble English reasoning. This work challenges this assumption by asking instead: what actually characterizes effective reasoning in multilingual settings, and to what extent do English-derived reasoning features genuinely help in other languages? We first define a suite of measurable reasoning features spanning multilingual alignment, reasoning step, and reasoning flow aspects of reasoning traces, and use logistic regression to quantify how each feature associates with final answer accuracy. We further train sparse autoencoders over multilingual traces to automatically discover latent reasoning concepts that instantiate or extend these features. Finally, we use the features as test-time selection policies to examine whether they can steer models toward stronger multilingual reasoning. Across two mathematical reasoning benchmarks, four LRMs, and 10 languages, we find that most features are positively associated with accuracy, but the strength of association varies considerably across languages and can even reverse in some. Our findings challenge English-centric reward designs and point toward adaptive objectives that accommodate language-specific reasoning patterns, with concrete implications for multilingual benchmark and reward design.