Multi-Domain Adaptation in Neural Machine Translation Through Multidimensional Tagging

02/19/2021
by   Emmanouil Stergiadis, et al.
0

Many modern Neural Machine Translation (NMT) systems are trained on nonhomogeneous datasets with several distinct dimensions of variation (e.g. domain, source, generation method, style, etc.). We describe and empirically evaluate multidimensional tagging (MDT), a simple yet effective method for passing sentence-level information to the model. Our human and BLEU evaluation results show that MDT can be applied to the problem of multi-domain adaptation and significantly reduce training costs without sacrificing the translation quality on any of the constituent domains.

READ FULL TEXT

Please sign up or login with your details

Forgot password? Click here to reset