Comparing machine translation and human translation for South African languages

Abstract

Machine translation (MT) has been available for several years but has only recently begun tobe considered viable, particularly in the context of indigenous South African languages.Although the quality of an MT remains inferior to that of a human translation (HT), MT systemshave gained popularity, making some significant contributions to translation studies. This studyexplored the similarities and differences between the two modes of translation. A case studyapproach based on a qualitative research method was used. The source data for this study arean MT and an HT of an article titled “Stellenbosch University to offer academic andpsychosocial support to students”, written by Ntwaagae Seleka and published on News24. Asa computer-aided translation tool, Autshumato Machine Translation Web Service (MTWS)produced the MT from English into Setswana as a target text. Meanwhile, a Setswana master’sstudent from the Tshwane University of Technology manually translated the source text. Thefindings show that the human translator outperformed the MTWS in providing high-qualitytranslation. The MTWS was unable to offer a better translation in terms of case sensitivity andterminological inconsistencies. It also mistranslated, adding and omitting words that changedthe intended meaning and leaving certain words untranslated. Human cognitive competency,intelligence, and flexibility enable human translators to deal with such translation problems toprovide high-quality outputs. The initiatives discussed in this study show that even though theMTWS operates as a useful translation tool with the capacity to instantly translate a largenumber of documents, its output is not yet capable of replacing an HT when translating intoindigenous South African languages. Post-editing of the MTWS outputs is alwaysrecommended.