Self-Improving Multilingual Long Reasoning via Translation-Reasoning Integrated Training

Fuente: arXiv
Saved in:
Bibliographic Details
Main Authors: Liu, Junxiao, Wang, Zhijun, Li, Yixiao, Lai, Zhejian, Huang, Liqian, Huang, Xin, Han, Xue, Feng, Junlan, Huang, Shujian
Format: Preprint
Published: 2026
Subjects:
Online Access:
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866914308489740288
author Liu, Junxiao
Wang, Zhijun
Li, Yixiao
Lai, Zhejian
Huang, Liqian
Huang, Xin
Han, Xue
Feng, Junlan
Huang, Shujian
author_facet Liu, Junxiao
Wang, Zhijun
Li, Yixiao
Lai, Zhejian
Huang, Liqian
Huang, Xin
Han, Xue
Feng, Junlan
Huang, Shujian
contents Long reasoning models often struggle in multilingual settings: they tend to reason in English for non-English questions; when constrained to reasoning in the question language, accuracies drop substantially. The struggle is caused by the limited abilities for both multilingual question understanding and multilingual reasoning. To address both problems, we propose TRIT (Translation-Reasoning Integrated Training), a self-improving framework that integrates the training of translation into multilingual reasoning. Without external feedback or additional multilingual data, our method jointly enhances multilingual question understanding and response generation. On MMATH, our method outperforms multiple baselines by an average of 7 percentage points, improving both answer correctness and language consistency. Further analysis reveals that integrating translation training improves cross-lingual question alignment by over 10 percentage points and enhances translation quality for both mathematical questions and general-domain text, with gains up to 8.4 COMET points on FLORES-200.
format Preprint
id arxiv_https___arxiv_org_abs_2602_05940
institution arXiv
publishDate 2026
record_format arxiv
spellingShingle Self-Improving Multilingual Long Reasoning via Translation-Reasoning Integrated Training
Liu, Junxiao
Wang, Zhijun
Li, Yixiao
Lai, Zhejian
Huang, Liqian
Huang, Xin
Han, Xue
Feng, Junlan
Huang, Shujian
Computation and Language
cs.CL
Long reasoning models often struggle in multilingual settings: they tend to reason in English for non-English questions; when constrained to reasoning in the question language, accuracies drop substantially. The struggle is caused by the limited abilities for both multilingual question understanding and multilingual reasoning. To address both problems, we propose TRIT (Translation-Reasoning Integrated Training), a self-improving framework that integrates the training of translation into multilingual reasoning. Without external feedback or additional multilingual data, our method jointly enhances multilingual question understanding and response generation. On MMATH, our method outperforms multiple baselines by an average of 7 percentage points, improving both answer correctness and language consistency. Further analysis reveals that integrating translation training improves cross-lingual question alignment by over 10 percentage points and enhances translation quality for both mathematical questions and general-domain text, with gains up to 8.4 COMET points on FLORES-200.
title Self-Improving Multilingual Long Reasoning via Translation-Reasoning Integrated Training
topic Computation and Language
cs.CL
url https://arxiv.org/abs/2602.05940