Paraphrase-Aligned Machine Translation

TL;DR

This work addresses the divergence between translated text and native fluency by aligning source sentences to target-language structures through paraphrase before translation. It introduces ParaAlign Translator, a two-phase approach that first creates paraphrase-aligned data via back-translation and then fine-tunes LLaMA-3-8B with LoRA to paraphrase and align inputs prior to translation. The method yields consistent improvements over strong baselines in both high-resource (English–Chinese, English–German) and low-resource (Hebrew, Swahili) scenarios, approaching or surpassing a much larger model while using a fraction of the data. These results highlight a data-efficient strategy for enhancing translation quality and fluency, with implications for deploying effective MT systems on resource-constrained ecosystems.

Abstract

Large Language Models (LLMs) have demonstrated significant capabilities in machine translation. However, their translation quality is sometimes questioned, as the generated outputs may deviate from expressions typically used by native speakers. These deviations often arise from differences in sentence structure between language systems. To address this issue, we propose ParaAlign Translator, a method that fine-tunes LLMs to paraphrase sentences, aligning their structures with those of the target language systems. This approach improves the performance of subsequent translations. Experimental results demonstrate that the proposed method enhances the LLaMA-3-8B model's performance in both resource-rich and low-resource scenarios and achieves parity with or surpassing the much larger LLaMA-3-70B model.

Figures (1)

• Figure 1: Relationship between the number of paraphrased sentence pairs and performances.