Pragmatic and cognitive elements in literary machine translation. An assessment of an excerpt from J. Polzin’s Brood translated with Google, DeepL, and Microsoft.

The development of neural MT has made it possible to realistically test its application to literary texts, traditionally reserved to human translation because of their inherent creativity. While recent research (Toral 2015; Toral et al. 2018; González, Toral & Guerberof 2019, among others) shows both MT achievements and its potential for literary translation, this article discusses linguistic and ethical issues from the point of view of a practising literary translator, and analyses a section from an in-progress translation project (English into Italian) translated with Google Translate, DeepL and Microsoft Translator. The theoretical discussion focuses on the role of co-text and context – hence the pragmatic dimension “because both pragmatics and fiction are interested in how human relationships are 'encoded in the structure of language'” (Morini 2019: 194); the challenges posed by the inherently “literary mind” of humans, who use metaphors, ambiguity and iconicity (Boase-Beier 2018); and the contention that translation skills are best taught using literary texts because their volume and complexity best represents cross-cultural human practices (Tymoczko 2014). The three machine-translated versions of an excerpt from a contemporary American novel are analysed with reference to both the pragmatic and cognitive aspects mentioned above and to the amount and quality of the post-editing required.