Machine translation quality depends on reasoning language and length

When Does Reasoning Help in Machine Translation? A Hierarchical Analysis of LRM Reasoning Traces

Computation and Language

Summary

Machine translation systems sometimes use step-by-step reasoning to translate text, but it’s not always clear when this helps or hurts translation quality. The authors studied different models and found that the best way to reason depends on the specific model, the length of the reasoning does not always improve results, and certain common patterns appear in how reasoning happens. They created a new method called Hierarchical Meta-Summarization to find these patterns without pre-set categories. Their findings suggest that translation systems should use reasoning carefully based on the model, length, and reasoning patterns instead of always encouraging more reasoning.

What this means in practice

Authors

Yuxiang Liu, Jiaming Luo, Eleftheria Briakou, Colin Cherry

Abstract

Large Reasoning Models increasingly use intermediate traces for machine translation, but it remains unclear when such reasoning helps or hurts. We analyze reasoning traces across models, languages, domains, and datasets, focusing on reasoning language, length, and structure. We find that the best reasoning language is model-specific, reasoning length has a non-monotonic relationship with quality, and traces exhibit recurring functional patterns. To uncover these patterns, we introduce Hierarchical Meta-Summarization (HMS), a scalable framework that induces coarse- and fine-grained reasoning structures without predefined taxonomies. HMS reveals a shared organization--understanding/planning, translating/drafting, and refining/verifying--alongside domain-specific variation. Our results suggest that MT reasoning should be controlled in a model-aware, length-aware, and pattern-aware manner rather than uniformly encouraged.
View PDFOpen arXiv