An Oblivious Approach to Machine Translation Quality Estimation

Machine translation (MT) is being used by millions of people daily, and therefore evaluating the quality such systems an important task. While human expert evaluation MT output remains most accurate method, it not scalable any means. Automatic procedures that perform task Translation Quality Estimation (MT-QE) are typically trained on a large corpus source–target sentence pairs, which labeled with judgment scores. Furthermore, test set drawn from same distribution as train. However, recently, interest in low-resource unsupervised MT-QE has gained momentum. In this paper, we define study further restriction setting call oblivious MT-QE. Besides having no access scores, algorithm to text’s distribution. We propose system based new notion cohesiveness introduce. tested our standard competition datasets for various language pairs. all cases, performance was comparable non-oblivious baseline provided organizers. Our results suggest reasonable can be carried out even restrictive setting.