XDLM: Cross-lingual Diffusion Language Model for Machine Translation
This addresses a gap in cross-lingual diffusion models for machine translation, offering a novel method but appears incremental in scope.
The paper tackles the problem of applying diffusion models to cross-lingual machine translation, proposing XDLM with pretraining and fine-tuning stages, and reports outperforming diffusion and Transformer baselines on benchmarks.
Recently, diffusion models have excelled in image generation tasks and have also been applied to neural language processing (NLP) for controllable text generation. However, the application of diffusion models in a cross-lingual setting is less unexplored. Additionally, while pretraining with diffusion models has been studied within a single language, the potential of cross-lingual pretraining remains understudied. To address these gaps, we propose XDLM, a novel Cross-lingual diffusion model for machine translation, consisting of pretraining and fine-tuning stages. In the pretraining stage, we propose TLDM, a new training objective for mastering the mapping between different languages; in the fine-tuning stage, we build up the translation system based on the pretrained model. We evaluate the result on several machine translation benchmarks and outperformed both diffusion and Transformer baselines.