INVESTIGATION OF MACHINE TRANSLATION APPLICABILITY

Abstract

The study evaluates the text complexity metrics applicability in assessing text fitness for machine translation (MT). Possible approaches to such assessments are discussed. The experiment used a corpus of research papers translated from Russian into English by both a professional translator and the DeepL MT engine. It is shown that the existing metrics cannot be used to forecast the MT quality since the MT output may be unusable even for low-complexity source texts. It is also proved that MT is unsuitable for almost 50% of the specialized text to be translated.

Keywords

• machine translation quality
• text complexity metrics
• editing distance
• cat tools
• machine translation applicability
• качество машинного перевода
• показатели сложности текста
• расстояние редактирования
• cat-инструменты
• применимость машинного перевода