The Evaluation of Translations of Three Persian Systems of Machine Translation, Based on Catfords’ Shifts

Abstract

Since in the process of machine translation the source author is a human but the target translator is a machine, and up to now the human performance has been structurally and lexically more developed than that of the machines, sampling of human’s translation is more profitable for machine translation. This adherence makes the system ready to encounter the precise grammatical strategies of the source text efficiently. Emphasizing this conforming, the authors of this study try to scrutinize the performance of three systems of MT (Pars, Padide, and Google), translating from English into Persian, to see to what extent they could observe this point. Systems' translations are examined on the basis of Catford's shifts and are then compared to corresponding human translations. Then a statistical analysis of machines' performances in contrast to human performance is done; it is revealed that although Google's performance is much better than that of Pars and Padide, these systems still need to be improved to efficiently cope with complicated structures.