In her doctoral project, Paola Ruffo studied literary translators’ self-imaging strategies and their relationship to technology. The CALT researcher resumes the state of the art.
News reports on how Artificial Intelligence (AI) can now write, translate, and it’s going to steal our jobs very soon are almost ubiquitous these days. However, while some of these new technological developments are indeed impressive and their impact on the job market shouldn’t be underestimated, the truth is often considerably more complex than what transpires from these articles. In fact, on closer inspection, the performance of these systems might still be limited outside the realm of carefully-crafted examples, while some of these claims have been refuted altogether.
The Vertical Leap
As far as literary translation is concerned, until very recently, the common assumption was that translation technologies, such as Computer-Aided Translation (CAT) tools and machine translation (MT), are not suitable to deal with the specific challenges posed by creative texts. However, this is slowly starting to change, as both industry practices, researchers’ focus, and practitioners’ workflows shift alongside rapidly-evolving technology. This brings to the fore deeper questions related, for example, to literary translators’ (new?) role and status in this ever-changing technological landscape, the effect of AI on translation processes and products, as well as copyright and ethical issues.
Based on the Kollektive-Intelligenz experiments, it would seem that Searle’s Chinese Room argument still holds for literary translators, as machine translation, (more or less) subtly but surely, reveals to them its lack of understanding. The machine has learned from the data, but it has not experienced the data, similar to how education differs from culture. As Chiara Valerio puts it in La matematica è politica (Mathematics is political), “l’istruzione è orizzontale e collettiva, la cultura invece verticale e individuale” (“education is horizontal and collective, culture, instead, is vertical and individual”). So far, AI-powered technology, no matter how sophisticated, seems to struggle or outright fail to make this vertical leap, rather moving on a horizontal plane made up of accumulated collective data. This, however, doesn’t mean its output is useless, just as education by itself is not useless. However, one could argue that at the core of literary translation is the ability to combine personal experiences with knowledge, contextualize information and form opinions to produce meaning, as well as navigating with dexterity between deductive and inductive reasoning.
Literary Translators’ Professionalization and Self-Imaging Strategies
In this respect, research on literary translators’ professionalization discourse has highlighted how their self-imaging strategies place emphasis on characteristics that are hard to define in terms of professionalization, such as vocation, creativity and passion. As Heino (2020) puts it, literary translators tend to prioritise social and cultural capital over economic capital. Their positioning in the field of translation is rooted in what Voinova and Shlesinger (2013: 41) call “a strange outsiderness”. Furthermore, according to Sela-Sheffy (2008), literary translators adopt three main idealised personae to affirm their professional identities, namely (1) custodian of language, (2) cultural ambassador and innovator, and (3) artist.
Similar findings emerged from my doctoral project on literary translators’ self-imaging strategies and their relationship to technology. In fact, for the 150 literary translators who answered the questionnaire, writing skills, passion, creativity, and artistic sensitivity were deemed essential to being a literary translator. In particular, respondents’ focus on personal qualities and skills, experiences and backgrounds, resulted in a portrait of literary translators as unique and irreplaceable. In the words of one respondent: “nothing can help the absence of inborn or acquired feeling for the subtleties of a given [literary] text”. By contrast, 73% of questionnaire respondents thought that those outside the literary translation profession do not view it the same way as they do. In particular, 50% mentioned outsiders being either unaware or unappreciative of what the job entails.
Translators’ Attitudes Towards and Special Use of Technology
The questionnaire also enquired about literary translators’ attitudes towards and use of technology. Overall, 49% of respondents expressed positive feelings towards technology in general (Ruffo 2022). This being said, positive attitudes could mainly be linked to generic tools (e.g., internet search, online dictionaries, word processors, digital glossaries etc. ) and Translation Memories (TMs), while the more negative attitudes were reserved almost exclusively to MT and AI. Overall, any technology perceived to try and “bypass the human understanding of language and its nuances in order to save costs” was considered unappealing, as were any “attempts to push the boundaries of technology within an essentially contemplative profession which requires an unfashionable degree of isolation and respect for experience”. Interestingly, 25% of respondents mentioned using CAT tools for literary translation.
Finally, the study found that attitudes towards technology are linked to age and translation technology training. In particular, those under 25 had more positive attitudes towards the use of technology in translation, while those who had received either academic or non-academic training in translation technology were both more positive towards technology and more confident using it. Other recent surveys of literary translators’ use of technology uncovered how they have little knowledge of specialized translation technology and make little use of it (Daems 2022). Additionally, when they do, the tendency is to adapt it to their specific needs rather than using them as intended by developers (Slessor 2020).
Literary Translation, A Recent Object of Study
It is only over the past decade that the idea of exploring the use of translation technology for creative texts has gained traction in academic research. While most of it so far has focused on the potential of MT and post-editing (e.g., Genzel et al. 2010; Greene et al. 2010; Voigt and Jurafsky 2012; Jones and Irvine 2013; Toral and Way 2014, 2015a, 2015b, 2018; Besacier and Schwartz 2015; Tezcan et al. 2019; Murchú 2019; Toral et al. 2020), an increasing number of studies is now starting to directly involve literary translators to explore their needs, perceptions and interaction with translation technology.
An example of this is Moorkens et al. 2018, where literary translators were asked to share their feedback after performing tasks involving translation from scratch and post-editing of MT output: participants favoured translation from scratch, despite the latter requiring less time. In another experiment involving post-editing of literary translation, Kenny and Winters (2020) found that “the translator’s voice is somewhat dampened in his post-editing work”. The effects of MT could also be seen in Guerberof-Arenas and Toral (2023) when comparing the reception of a fictional story translated from English into Catalan and Dutch using MT, post-editing of MT output, and translation from scratch. Here, human translation achieved higher levels of engagement, enjoyment and translation reception for Catalan readers, while Dutch readers preferred the post-editing output, leading the authors to believe that factors such as language status in one’s society might also play a role in how different types of translations are received. Furthermore, in response to most studies focusing on time- and cost-saving aspects of MT and post-editing, some researchers have started to explore the potential of alternative workflows, including the use of corpus linguistics and text-visualization tools (Youdale 2019), and CAT tools and TMs (Horenberg 2019, Youdale and Rothwell 2022).
New Ways For Poetry To Exist
Ultimately, we are only starting to scratch the surface of the potential of translation technology for literary translation. In particular, the extent to which AI-powered tools can enhance the literary translation workflow, and the specific ways in which using these tools can affect, for example, translators’ working conditions, translation quality and translation reception remain to be explored. Furthermore, literary-translation-specific factors, such as literary translators’ self-imaging strategies, their limited training or knowledge of available tools, and the original ways in which they adapt existing technology to their needs, should be an integral part of how both researchers and developers approach the study and introduction of new technologies in the profession. Finally, the challenge remains for researchers to offer a balanced counter-narrative to the claims being made in regards to the capacities of these new technologies, and to make sure that the goal of technological development is “not to imitate existing poetry, but to find new ways for poetry to exist” (Parrish, 2015).
DUAL-T, the EU-funded project I am currently working on as part of the Language and Translation Technology team (LT3) at Ghent University, aims at tackling some of these questions by actively involving literary translators in the testing of three different workflows for the language pair English-Dutch involving, respectively, (1) a word processor, (2) a CAT tool, and (3) an MT post-editing platform. In addition to user testing via keystroke logging and screen capturing, questionnaires, in-depth interviews and focus groups will be conducted before and after the translation tasks to assess participants’ perceptions of each workflow. I am going to start recruiting participants soon, so if you would like to take part, feel free to get in touch.
Paola Ruffo is a researcher in the field of Computer-Aided Literary Translation (CALT). She is currently working on her Marie Skłodowska–Curie Postdoctoral Fellowship project Developing User-centred Approaches to Technological Innovation in Literary Translation (DUAL-T) at Ghent University. You can follow her on twitter @traduzionemille. The German translation of her article is to be found here at the Aggregate Intelligence website.
Picture credits: flashmovie