CLewR: Curriculum Learning with Restarts for Machine Translation Preference Learning
Fuente:
arXiv
Saved in:
| Main Authors: | , , |
|---|---|
| Format: | Preprint |
| Published: |
2026
|
| Subjects: | |
| Online Access: | |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| _version_ | 1866913040022110208 |
|---|---|
| author | Dragomir, Alexandra Brad, Florin Ionescu, Radu Tudor |
| author_facet | Dragomir, Alexandra Brad, Florin Ionescu, Radu Tudor |
| contents | Large language models (LLMs) have demonstrated competitive performance in zero-shot multilingual machine translation (MT). Some follow-up works further improved MT performance via preference optimization, but they leave a key aspect largely underexplored: the order in which data samples are given during training. We address this topic by integrating curriculum learning into various state-of-the-art preference optimization algorithms to boost MT performance. We introduce a novel curriculum learning strategy with restarts (CLewR), which reiterates easy-to-hard curriculum multiple times during training to effectively mitigate the catastrophic forgetting of easy examples. We demonstrate consistent gains across several model families (Gemma2, Qwen2.5, Llama3.1) and preference optimization techniques. We publicly release our code at https://github.com/alexandra-dragomir/CLewR. |
| format | Preprint |
| id |
arxiv_https___arxiv_org_abs_2601_05858 |
| institution | arXiv |
| publishDate | 2026 |
| record_format | arxiv |
| spellingShingle | CLewR: Curriculum Learning with Restarts for Machine Translation Preference Learning Dragomir, Alexandra Brad, Florin Ionescu, Radu Tudor Computation and Language Artificial Intelligence Machine Learning Large language models (LLMs) have demonstrated competitive performance in zero-shot multilingual machine translation (MT). Some follow-up works further improved MT performance via preference optimization, but they leave a key aspect largely underexplored: the order in which data samples are given during training. We address this topic by integrating curriculum learning into various state-of-the-art preference optimization algorithms to boost MT performance. We introduce a novel curriculum learning strategy with restarts (CLewR), which reiterates easy-to-hard curriculum multiple times during training to effectively mitigate the catastrophic forgetting of easy examples. We demonstrate consistent gains across several model families (Gemma2, Qwen2.5, Llama3.1) and preference optimization techniques. We publicly release our code at https://github.com/alexandra-dragomir/CLewR. |
| title | CLewR: Curriculum Learning with Restarts for Machine Translation Preference Learning |
| topic | Computation and Language Artificial Intelligence Machine Learning |
| url | https://arxiv.org/abs/2601.05858 |