Skip to main content
EMT Blog
  • News article
  • 9 January 2025
  • Directorate-General for Translation
  • 11 min read

Pushing the boundaries of Digital Translatology: some notes on creativity, pragmatics and discourse

By Oliver Czulo, full professor at the Institute of Applied Linguistics and Translatology, Leipzig University

Photo from Google DeepMind on Pexels: https://www.pexels.com/de-de/foto/abstrakt-technologie-forschung-digital-18069696

In the following, I want to highlight three areas of research which I believe to be fruitful to Digital Translatology if expanded. I understand Translatology (with a big T) here in a German tradition, encompassing translation and interpreting in its very many variants, and, at least in my view, with a more pronounced empirical focus. ‘The digital’ is seen both in a horizontal perspective, i. e. regarding the many areas of translation and interpreting that have incorporated digital tools one way or another, and a vertical perspective, i. e. those fields that are deeply or natively digital such as machine translation or video game localization.

None of the three areas of research I will touch upon here are completely new, but I would argue that they are understudied and suffer from a lack of operationalisation. Furthermore, they present new potential for the field of studying ‘the digital’ beyond such questions as

  • How should we evaluate machine translations?
  • What are good guidelines for post-editing?
  • How do machines affect translation/interpreting production?
  • Are machines or humans better at translating/subtitling/...?

etc., the types of which have been predominantly asked in the recent years. These mostly come from the intersection of Translatology and Machine Translation; from my experience, we mostly associate the latter field with ‘the digital’ in translation. The rather narrow focus of these questions centres around text production and text quality. What it leaves out, in my view, is a wide range of possibilities to both (a) tackle ‘the digital’ with concepts that are strong in the humanities but haven’t been applied much to Digital Translatology and (b) include concepts typical for human-centred translation research which at the same time may profit from a ‘digital’ perspective, e. g., in terms of operationalisation.

Creativity

Machine translation has now also reached literary translation which, considering the recent advances, was to be expected. While a few well-translated sentences do not yet make a good narration, this does shed a new light on traditional translation tenets. Literary translation is still often viewed as the epitome of creativity and is depicted as some sort of ‘most human translation’. This is surprising as, after all, there are various other genres which generally require creative translation such as marketing or music.[1] ‘The digital’ has brought in further such genres such as video game localization or has strengthened other such as subtitling. And it has been recognized that other specialized texts may need their own creative approaches, e. g., to tackle metaphoricity or bridge terminological gaps. (To be clear: Literary translation, to me, is just as much a specialization in translation as all the others, and including those in interpreting; literary language is thus specialized language the translation of which requires specific training and/or background knowledge.)

Research on literary machine translation, however, exposes a long-standing shortcoming: We are still lacking a good operationalisation of what creativity in translation (and interpreting) actually is. Guerberof-Arenas and Toral (2022) offer an operationalisation partly based on rather standard evaluation of the translations using the Multidimensional Quality Metric, partly on the analysis of translations of pre-annotated source segments which (may) require creative solutions in the target text. Few to none of the examples discussed in their paper would strike me as (potentially) exclusive to literature, though. Creativity in translation, so I would conclude from Guerberof-Arenas and Toral’s operationalization, requires both routined and inventive problem solving. Their research method does not, however, allow for the conclusion that literary texts are solely and always on top of the creativity pyramid (and, I believe, the authors do not intend to express that), as the texts used in the study are pre-selected from the genre of literature. It would be interesting to see how well texts from other domains would fare in terms of creativity when judged by the same method. It remains to be confirmed at this point what Guerberof-Arenas and Toral state themselves: Creativity remains an elusive concept.

Pragmatics

But wait, isn’t functional translation theory built on the very notion of pragmatics? Indeed, it is, but it is text centric in nature: Following the analysis of functional aspects of a source text, target text translation decisions are made resp. evaluated. There are various approaches to this type of analysis (e.g., House 2015; Nord 2009), but what falls somewhat outside of it are the many pragmatic phenomena such as information structure or text deixis which can occur (at least to a degree) independently of the text function. At the same time, as House and Kádár (2021, 2) point out, the linguistic formalisation of pragmatic phenomena has been lacking, inhibiting cross-cultural research into (linguistic) pragmatics.

Recent developments in the frame semantics community offer such a formalisation: Czulo and colleagues (2020), building on early remarks by Fillmore (1982), propose to use the methods of frame semantics, so far mostly applied to (lexical) semantics, in order to describe knowledge structures related to language use.[2] Triesch-Herrmann and Czulo (2024) make use of pragmatic frames for translation analysis by the example of English-German translations into the adverb bekanntlich (roughly ‘as known’). They assume the adverb to evoke the frame Gemeinsames_Vorwissen[3] (‘Common_ground’), i. e., a knowledge structure on how to refer to Content the knowledge of which is shared by the Cognizers, i. e. both senders and addressees in the given communication situation. The authors uncover that in DGT translations from English into German there is a non-prototypical semantic use of the adverb relating to the frame Gewahrsein_Status[4] (‘Awareness_status’), i. e., to cases in which a sender expresses their status of familiarity with respect to some information, not including the addressees. This use cannot be attributed to effects such as normalization or shining through, but may be some sort of EU speak.

The variation of pragmatic vs. semantic use of the adverb bekanntlich cannot be modelled by means of standard valency theory, as the adverb does not have valency in the traditional sense. As the pragmatic frames are linked to forms, they can be automatically annotated and thus lend themselves to a more automated analysis as is both useful and customary in machine translation, a field in which, as Valdeón (2023, 10ff.) points out, research on pragmatics is lacking. The various new possibilities that are opened up by this operationalisation, of course, do not only pertain to machine translation, but also to other areas of ‘the digital’ in translation and interpreting.

Discourse

As stated earlier, the dominant analysis unit in Translatology is the text: Its functional properties in a communicative setting, so the theory, determine key translation choices. What does not seem to have received as much attention in Translatology as in Linguistics, however, are cross-textual patterns of language realisation which facilitate certain perspectives, and are subject to rules of what is (usually) said how, and who says and can say what – altogether a rough working definition of discourse (in a Foucault’ian sense). These patterns can be carried over from one language community to another by means of translation, but this transfer may elude a simple text-based analysis. Sometimes, there may not even be ‘the one (piece of a) translation’ that would exhibit a direct translation of such a pattern.

An example of this is studied by Bisiada and colleagues (2023) with regard to the translation of discourse patterns around the originally (US-)English concept #MeToo on (back then) Twitter into German and Spanish. Using frame semantics, the authors first analyse the event and role structure of what is usually seen as #MeToo in its original US discourse: sexualized aggression which is (publicly) addressed, with standard values such as, prototypically, women as victims of male aggression. They find, i. a., that some far right-wing actors on German-language Twitter perverted the standard structure into one in which ‘blonde’ German women are attacked by ‘foreign-looking’ men; the aggression, so the supposed event structure to be derived from the tweets, is first covered up by ‘left-wing’ or ‘woke’ agents but then publicly disclosed.[5] While there does not seem to be ‘the tweet’ which is a direct translation of an English template for the German tweets, we see both the translation of source text patterns in a structurally close fashion as well as right-wing audience specific adaptations of these patterns.

In the realm of ‘the digital’, this kind of research may become more relevant with multilingual language models generating texts. Similarly to the grammatical ‘accent’ of such multilingual models as described, e. g., by Papadimitriou and colleagues (2023), there may be residues of prototypical anglophone patterns in the text generated in other languages. When I asked ChatGPT 3.5 in German to write a story about a man in his mid-40s who had a rough start in life, ChatGPT produced a German text in which it came up with a character named “David Davidson” who, i. a., gets into a fight between gangs and catches a bullet. In a country that doesn’t have a gang culture even remotely close to that of the USA and has very strict gun regulations such as Germany, this does not seem to be a prototypical story element, but rather inspired by what at least TV and movies suggest to be more US-typical. We may thus be looking at another type of ‘accent’, where the transference of patterns lies neither in shining through of formal language structures nor of story elements in direct text translations, but in the use of trans-textual patterns not typical for the target culture inspired by source texts such as, in the case of ChatGPT, predominantly English training material. An operationalisation of describing, e. g., standard elements of a concept or story such as presented by Bisiada and colleagues allows for modelling such differences between languages beyond a prosaic description.

New avenues for Digital Translatology

While the kinds of text-centric questions around production and quality cited in the introduction may not lose relevance to Digital Translatology in the nearer future, it is my conviction that we need to move well beyond those. The areas of research very briefly touched upon here not only present new interesting opportunities of research, but also represent topics and approaches in which humanities research has a strong and rich tradition. Making use of these, we can contribute to the study of ‘the digital’ and can benefit from new, digital perspectives on them. This list of areas of research is, of course, not exclusive. Research in these and similar other directions will certainly involve looking into other fields such as Computational Linguistics (not so much the NLP-heavy part, but the part on formal modelling of language) or Corpus Linguistics in order to learn more about the possibilities of operationalisation, and looking into other humanities and social sciences for updates on areas of research such as creativity, pragmatics or discourse. This opens many doors for fruitful exchange and for Digital Translatology to contribute its part to this research.

References


Bisiada, Mario, Oliver Czulo, und Eleonore Schmitt. 2023. „#MeToo in drei Sprachen: Qualitative Analyse von Konzepten und Diskursmustern im Englischen, Deutschen und Spanischen anhand von Twitter“. Deutsche Sprache, Nr. 1 (März), 5. https://doi.org/10.37307/j.1868-775X.2023.01.05.

Czulo, Oliver, Alexander Ziem, und Tiago Timponi Torrent. 2020. „Beyond lexical semantic frames: notes on pragmatic frames“. In Towards a global, multilingual FrameNet, herausgegeben von Tiago Timponi Torrent, Collin Baker, Oliver Czulo, Kyoko Ohara, und Miriam Petruck, 1–7. Proceedings of the LREC 2020. Marseille: Association for Computational Linguistics. https://aclanthology.org/2020.framenet-1.1/.

Fillmore, Charles J. 1982. „Frame semantics“. In Linguistics in the Morning Calm, 111–37. Seoul, South Korea: Hanshin Publishing.

Guerberof-Arenas, Ana, und Antonio Toral. 2022. „Creativity in translation: Machine translation as a constraint for literary texts“. Translation Spaces 11 (2): 184–212.

House, Juliane. 2015. Translation Quality Assessment: Past and Present. London: Routledge.

House, Juliane, und Dániel Z. Kádár. 2021. Cross-Cultural Pragmatics. Cambridge New York, NY Port Melbourne New Delhi Singapore: Cambridge University Press. https://doi.org/10.1017/9781108954587.

Matsumoto, Yoshiko. 2010. „Interactional Frames and Grammatical Descriptions: The Case of Japanese Noun-Modifying Constructions“. Constructions and Frames 2 (2): 135–57. https://doi.org/10.1075/cf.2.2.01mat.

Nord, Christiane. 2009. Textanalyse und Übersetzen: theoretische Grundlagen, Methode und didaktische Anwendung einer übersetzungsrelevanten Textanalyse. 4., Überarb. Aufl. Tübingen: Groos.

Ohara, Kyoko. 2018. „Relations between frames and constructions: A proposal from the Japanese FrameNet constructicon“. In Constructicography: Constructicon development across languages, herausgegeben von Benjamin Lyngfelt, Lars Borin, Kyoko Ohara, und Tiago Timponi Torrent, 141–64. Amsterdam; Philadelphia: Benjamins. https://doi.org/10.1075/cal.22.05oha.

Papadimitriou, Isabel, Kezia Lopez, und Dan Jurafsky. 2023. „Multilingual BERT Has an Accent: Evaluating English Influences on Fluency in Multilingual Models“. In Findings of the Association for Computational Linguistics: EACL 2023, 1194–1200. Dubrovnik, Croatia: Association for Computational Linguistics. https://doi.org/10.18653/v1/2023.findings-eacl.89.

Triesch-Herrmann, Susanne, und Oliver Czulo. 2024. „A frame-based analysis of the pragmatics and semantics of “bekanntlich” in English-German translation“. trans-kom Zeitschrift für Translationswissenschaft und Fachkommunikation 17 (1): 130–46.

Valdeón, Roberto A. 2023. „Automated Translation and Pragmatic Force: A Discussion from the Perspective of Intercultural Pragmatics“. Babel. Revue Internationale de La Traduction / International Journal of Translation, Juli. https://doi.org/10.1075/babel.00328.val.


[1] While this is reflected and acknowledged by such concepts as transcreation, my personal experience is that this acknowledgement has not yet fully permeated research and practice.

[2] The discussion around pragmatic frames is not new, but has been rather dormant since Fillmore’s first remarks, with few exceptions such as (Matsumoto 2010; Ohara 2018).

[3] https://framenet-constructicon.hhu.de/framenet/frame?id=1569 (last accessed 2024-12-28)

[4] https://framenet-constructicon.hhu.de/framenet/frame?id=728 (last accessed 2024-12-28)

[5] As reported by colleagues, this pattern is not exclusive to German and not to Twitter, but it shows up prominently in the Twitter data analyzed by the authors of the study.

Details

Publication date
9 January 2025
Author
Directorate-General for Translation
Department
Directorate-General for Translation
Language
  • English
  • German
EMT Category
  • Translation technology