Machine translation and its use as a lexicographic tool when writing digital texts: an analysis of real working documents

Abstract

Using real working documents as a basis for research, the aim of this paper is to compare the lexical accuracy of bilingual dictionaries versus machine translation engines that apply neural networks, when employed by non-professionals to translate texts. We put three such machine translation engines to the test (Google Translate, Bing Microsoft Translator and DeepL), and evaluate their performance when translating a selection of real working documents. We focus particular attention on lexical errors originated by machine translation engines that compromise the reader’s understanding of the final text. Finally, we compare the results obtained for the same tasks when making use of bilingual dictionaries in place of MT engines. The conclusion reached is that when translating reasonably complex texts, the output generated by MT engines is unsatisfactory: lexicographic data obtained from sundry bilingual dictionaries online generally offers more precise and comprehensive lexicographic information to satisfy the user’s communicative needs.