Machine Translation Evaluation: Analysis of English-Persian Translation of Health and Beauty Brochures[Latin Thesis]

Fatemeh Karimi

This study was set out to evaluate errors of the two popular translation machines "Google Translate" and "Bing" in the translation of health/beauty related texts, brochures, leaflets, manuals, and laboratory reports (English to Persian - Persian to English). To this end, 50 brochures (25 Persian, and 25 English), 50 manuals (25 Persian, and 25 English), and 50 laboratory reports (25 Persian, and 25 English) were chosen from the health/beauty field. All the errors made by Google Translate and Bing collected, counted, analized and categorized by researcher according to the model presented by Vilar, Xu, D’Haro, and Ney (2006). Then, errors which were categorized in the first stage, were compared to each other in order to find the most and the least number of their frequency. Finally, the results show Google Translate with 968 errors is 7.67% more valid and reliable than Bing with 1129 errors because of fewer errors and, the quality of translation in Persian to English with 873 errors is better than English to Persian with 1224 errors, “Incorrect Words” is the most frequent and “Missing Words” is the least frequent error made by both chosen machine translations