When Does Monolingual Data Help Multilingual Translation: The Role of Domain and Model Scale

arXiv:2305.14124v3 Announce Type: replace
Abstract: Multilingual machine translation (MMT), trained on a mixture of parallel and monolingual data, is key for improving translation in low-resource language pairs. However, the literature offers conflicting results on the performance of different methods of including monolingual data. To resolve this, we examine how denoising autoencoding (DAE) and backtranslation (BT) impact MMT under different data conditions and model scales. Unlike prior studies, we use a realistic dataset of 100 translation directions and consider many domain …

abstract arxiv cs.cl data domain however improving key language literature low machine machine translation multilingual performance results role scale translation type