Cohesive devices, as the explicit language-specific markers that signal cohesion within a text, are critical in legal translation, where precision and logic are valued. Nevertheless, as cohesive devices and the relevant grammar and conventions vary across languages, they may create challenges for neural machine translation (NMT) systems. This study investigates legal translation between English and Chinese, focusing on two types of referential cohesive devices (RCDs) – personals and demonstratives – in a self-built corpus of English and Chinese legislative texts that are non-translated, human-translated and machine-translated by the three NMT systems Baidu Translate, DeepL Translator and Google Translate. The findings suggest that the frequencies of specific RCDs differ significantly between translated and non-translated texts and between human-translated and machine-translated texts. However, the extent of these differences varies with the direction of translation, and not all differences between translated and non-translated texts can be attributed to source language interference. The findings also suggest how a translated RCD may influence the grammaticality, accuracy, clarity of expression and logical flow of a translation. Some final recommendations are provided in light of these findings.
Yang Xu (Tue,) studied this question.