CLewR: Curriculum Learning with Restarts for Machine Translation Preference Learning

TL;DR

This paper introduces CLewR, a curriculum learning strategy with restarts, to improve multilingual machine translation by mitigating forgetting of easy examples, showing consistent gains across models.

Contribution

The paper proposes a novel curriculum learning method with restarts (CLewR) that enhances preference learning in machine translation by repeatedly emphasizing easy examples.

Findings

CLewR improves translation performance across multiple models.

Consistent gains observed with various preference optimization techniques.

Code is publicly available at the provided GitHub link.

Abstract

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.