CREAMT: Creativity and narrative engagement of literary texts translated by translators and NMT

CREAMT: Creativity and narrative engagement of literary texts translated by translators and NMT Ana Guerberof Arenas University of Surrey/ University of Groningen a.guerberof.arenas@rug.nl Antonio Toral University of Groningen a.toral.ruiz@rug.nl Abstract We present here the EU-funded project CREAMT that seeks to understand what is meant by creativity in different translation modalities, e.g. machine translation, post- editing or professional translation. Focus- ing on the textual elements that determine creativity in translated literary texts and the reader experience, CREAMT uses a novel, interdisciplinary approach to assess how effective machine translation is in lit- erary translation considering creativity in translation and the ultimate user: the reader. 1 Introduction Research has shed some light on the usability of machine translation (MT) in literary texts (Toral, Wieling, and Way 2018), showing that MT might help literary translators when it comes to productivity. At the same time, translators’ perception is that the “more creative” the literary text, the less useful MT is (Moorkens et al. 2018). But can we quantify the creativity in texts translated by humans as opposed to those produced with the aid of machines? And, since one of the aims of the translation of a literary text is to preserve the reading experience of the original, what is the reader’s experience when faced with machine-translated texts? Do users exposed to different translation modalities have different reading experiences? To provide answers to these questions, the CREAMT is articulated in two main axes with a two-year duration. The first axis proposes to © 2022 The authors. This article is licensed under a Crea- tive Commons 3.0 licence, no derivative works, attribution, CCBY-ND. identify creative shifts (see section 2.2) while the second axis seeks to identify reader’s narrative engagement and gather data on enjoyment and translation reception. 2 First axis We translated two stories: Murder in the Mall by Sherwin B. Nuland (1995) was translated into Catalan for a pilot project and 2BR02B by Kurt Vonnegut (1999) was translated into Catalan and Dutch for the main experiment. 2.1 Translation Process The conditions human translation (HT) and post- editing (PE) were processed by two professional literary translators. To reduce the effect of the translator, each professional translated and post- edited 50% of each modality. The MT condition was based on the output of state-of-the-art literary-adapted neural MT systems based on the transformer architecture (Vaswani et al. 2017) trained to translate from English to Catalan (Toral, Oliver, and Ribas- Bellestín 2020) and to Dutch (Toral, van Cranenburgh, and Nutters 2021). The training data did not contain the text used for the experiment nor any by these authors. 2.2 Creativity The source text (ST) was first annotated for units of creative potential (e.g. metaphors, wordplay and puns, comparisons). A team of five professional reviewers annotated the target texts (TT) as either reproduction, omission, or creative shift (Bayer-Hohenwarter 2011). The creative shifts could be 1) modification (i.e. ST is modified for the target culture), 2) concretisation (i.e. ST is replaced by a more concrete example in the TT) and 3) abstraction …

Content truncated to 500 words