EnAnchored-X2X: English-Anchored Optimization for Many-to-Many Translation
Fuente:
arXiv
Saved in:
| Main Authors: | , , , , , |
|---|---|
| Format: | Preprint |
| Published: |
2025
|
| Subjects: | |
| Online Access: | |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| _version_ | 1866911173163614208 |
|---|---|
| author | Yang, Sen Bao, Yu Lu, Yu Chen, Jiajun Huang, Shujian Cheng, Shanbo |
| author_facet | Yang, Sen Bao, Yu Lu, Yu Chen, Jiajun Huang, Shujian Cheng, Shanbo |
| contents | Large language models (LLMs) have demonstrated strong machine translation capabilities for English-centric language pairs but underperform in direct non-English (x2x) translation. This work addresses this limitation through a synthetic data generation framework that leverages models' established English-to-x (en2x) capabilities. By extending English parallel corpora into omnidirectional datasets and developing an English-referenced quality evaluation proxy, we enable effective collection of high-quality x2x training data. Combined with preference-based optimization, our method achieves significant improvement across 72 x2x directions for widely used LLMs, while generalizing to enhance en2x performance. The results demonstrate that strategic exploitation of English-centric strengths can bootstrap comprehensive multilingual translation capabilities in LLMs. We release codes, datasets, and model checkpoints at https://github.com/NJUNLP/EAX |
| format | Preprint |
| id |
arxiv_https___arxiv_org_abs_2509_19770 |
| institution | arXiv |
| publishDate | 2025 |
| record_format | arxiv |
| spellingShingle | EnAnchored-X2X: English-Anchored Optimization for Many-to-Many Translation Yang, Sen Bao, Yu Lu, Yu Chen, Jiajun Huang, Shujian Cheng, Shanbo Computation and Language Large language models (LLMs) have demonstrated strong machine translation capabilities for English-centric language pairs but underperform in direct non-English (x2x) translation. This work addresses this limitation through a synthetic data generation framework that leverages models' established English-to-x (en2x) capabilities. By extending English parallel corpora into omnidirectional datasets and developing an English-referenced quality evaluation proxy, we enable effective collection of high-quality x2x training data. Combined with preference-based optimization, our method achieves significant improvement across 72 x2x directions for widely used LLMs, while generalizing to enhance en2x performance. The results demonstrate that strategic exploitation of English-centric strengths can bootstrap comprehensive multilingual translation capabilities in LLMs. We release codes, datasets, and model checkpoints at https://github.com/NJUNLP/EAX |
| title | EnAnchored-X2X: English-Anchored Optimization for Many-to-Many Translation |
| topic | Computation and Language |
| url | https://arxiv.org/abs/2509.19770 |