Release:2017, Vol. 3. №3
About the authors:Maria A. Kunilovskaya, Cand. Sci. (Philol.), Associate Professor, Department of English Philology and Translation, Institute of Philology and Journalism, University of Tyumen; firstname.lastname@example.org
This paper sets out to develop a method of learner translation quality assessment, based on error statistics. It is used to rank student translations in descending order by their relative quality. It means that each translation is assessed in comparison with other translations of the same text. The resulting rating can be converted into absolute values such as grades or number of points earned if necessary. The authors use expert assessment conducted within a holistic approach as a quality standard. In this research the rank order of multiple translations to the same source is effected as a statistical approximation to the consensus rating produced for the same texts from independent ratings by three experts. This paper has a brief description of the error typology and its technical implementation, as well as of the corpus project behind this research. The agreement statistics used in the experiment show that the best results are achieved, if the error categories, discussed in the paper, are accounted for in the multiple sort of translations in the following order: 1) critical, 2) content and 3) total. The authors suggest that it makes sense to manually adjust the ranking to the number of annotated “good translational solutions”. Finally, the authors outline the current and perspective uses of the learner translator corpus and its error-annotated part in translator education.