Machine translation in marketing texts: Achieving persuasion in post-editing

Due to the demand for fast turnaround, the practice of machine translation with post-editing (MTPE) in creative texts has become increasingly common. While MT systems have been conditioned to translate literary texts and reduce the human input of post-editing (Van de Cruys, 2020), ways to post-edit marketing texts, another genre of creative texts, remain under-researched. This presentation reveals the types of post-editing required in the transcreation of marketing texts, as compared to those in the translation of technical texts, i.e., a non-creative genre. Two parallel English-Chinese corpora, one marketing texts and the other technical texts, are compiled. MT results are generated in Google Translate API, one of the most used MT engines (Savenkov, 2019). PE distance (percentage of modifications by comparing the number of word count) is calculated and the types of post-editing are categorised in an error taxonomy adapted from Moorkens (2018) and Benetello (2018). While both corpora share similarities regarding syntactic and contextual post-edits, the marketing corpus also requires creative post-edits. In areas where the intended effect – persuasion – is required, though the translation is correct, MT fails to achieve this effect. This presentation attempts to answer two questions: where persuasion is needed in marketing texts and what types of creative post-editing are needed. The answers can inform future fine-tuning of MT systems. They also encourage further investigation on cognitive effort in creative post-editing using eye-tracking and keystroke-logging tools, and the NASA Task Load Index (Sun and Shreve, 2014) for translators’ self-assessment.