Domain-Specific Text Generation for Machine Translation. (arXiv:2208.05909v1 [cs.CL])

Preservation of domain knowledge from the source to target is crucial in any
translation workflow. It is common in the translation industry to receive
highly specialized projects, where there is hardly any parallel in-domain data.
In such scenarios where there is insufficient in-domain data to fine-tune
Machine Translation (MT) models, producing translations that are consistent
with the relevant context is challenging. In this work, we propose a novel
approach to domain adaptation leveraging state-of-the-art pretrained language
models (LMs) for domain-specific …

arxiv generation machine machine translation text text generation translation