Presenting a quality estimation model of English to Farsi machine translator using transfer learning

Nowadays the evaluation of machine translation without reference translation is of great importance as one of the research areas of machine translation . One of the challenges in this field, especially for languages with few sources, is the lack of suitable training data . For this purpose, it is possible to use neural network based methods that have been previously trained on multilingual language models and estimate the translation quality for a pair of new languages using transfer learning . In this article, the quality of an English Persian test set is estimated in this way. Also, a set of educational data for the English Persian language pair has been prepared, and appropriate pre processing has been done on it, and the existing multilingual model has been fine tuned with that data . The use of these training data has improved the Pearson correlation with the test set by 29