İngilizce-Türkçe nöral makine çevirisi için son-düzenleme becerileri

This study examines the performance of Neural Machine Translation (NMT) in Turkish as well as the post-editing skills that may be required particularly for Turkish post-editing. A machine translation (MT) test is performed with five online NMT systems, namely Google Translate, Bing Translator, Yandex.Translate, Reverso, and Watson Language Translator. For this purpose, two technical, two legal, and two medical texts are machine- translated. The raw MT outputs are then assessed using the SCATE MT error taxonomy. The results are analyzed from a post-editing standpoint, demonstrating which mistake types are most frequently amended in the post- editing process. Furthermore, the abilities required to correct faults in these areas are highlighted. The results reveal that agglutinative nature and orthography rules are essential elements of post-editing in Turkish. The three fluency categories that involve the greatest number of errors are grammar, lexicon, and orthography, and their most erroneous subcategories are found to be word form, extra word, and missing word for grammar; lexical choice for lexicon; and punctuation, capitalization, and spelling for orthography. In terms of accuracy, omission, mistranslation, and mechanical categories are the three most erroneous categories. For their subcategories, word sense disambiguation is dominant for mistranslation, and for mechanical, capitalization and other come to the forefront. According to the findings, the necessary Turkish post-editing skills are a thorough understanding of the use and role of Turkish suffixes, grammar and orthography rules, terminology, and lexicology.