Volume 70, Issue 1 - Themenheft „Die KI-gestützte Übersetzungsbranche“ Special Issue “The AI-based translation industry”

This article is concerned with the following question: Which competences are required by stakeholders who are tasked with implementing current generative artificial intelligence (AI) technologies (primarily in the form of uni- and multimodal large language models) in language industry workflows? To set the scene, the article first discusses briefly the ongoing AI saturation of the language industry. It then proceeds to discuss the concept of artificial intelligence literacy and shows how this concept has been operationalised for language industry contexts via a domain-specific AI literacy framework for translation, interpreting and specialised communication. In this context, the article also discusses potential AI literacy roles and competence levels that have been proposed in the literature and briefly sketches how these may be related to the domain-specific AI literacy framework. Then, the article zooms in on the implementation dimension of the framework. For each subdimension of this dimension, it proposes a draft competence descriptor and discusses in detail the relevant aspects of generative AI implementation subsumed under these subdimensions from a competence perspective. The article concludes with a brief outlook on how domain-specific AI literacy may help stakeholders navigate the further evolution of the language industry under the impact of high-performing AI technologies.

Product investigation, i. e., trying out a technical product, is a necessary activity in order to post-edit AI-based translations of information products such as user manuals in a product-safe manner. Product investigation can be understood as a form of embodiment (Zlatev 2013). An example, analysed on the basis of three models devised in the field of specialized 1 communication studies (Schubert 2007; Holste 2024; Schenda 2024), is used to illustrate the relevance of product investigation for the post-editing of information products in mono- and multilingual technical documentation. First, the architecture of hybrid systems – a combination of rule-based and AI-based systems – is briefly outlined (Krüger 2023) and current trends towards hybrid systems in the mono- and multilingual technical documentation process are pointed out. In the following, an example illustrates how product investigation may enhance post-editing, thus indicating that translators actually handling the product can access knowledge which remains inaccessible to AI/machine translators. To conclude, the article outlines possibilities for empirical research into product investigating as part of the post-editing process. The article is intended as a prelude to more in-depth empirical research (Heidrich et al. i. V.), such as that envisaged in the planned ‘UX4tech-kom’ research lab at TH Köln – University of Applied Sciences.

Generative AI, like other technological milestones in the language industry preceding it, such as translation memories and neural machine translation, has the potential to substantially change the ways in which translators work and how they are trained. Recent publications (Penet et al. forthcoming; Pym/Hao 2025) have suggested concrete approaches to utilizing ChatGPT in translator training contexts, yet empirical research on how it is actually used remains scant. This article presents the results of a small-scale exploratory study that gauges how MA students of translation (English-German) utilize ChatGPT in conjunction with other translation tools and resources. Triangulated think-aloud, screen recording and questionnaire data shed light on such phenomena as prompting behavior, the phases and textual levels of ChatGPT application, attitudes regarding its efficacy and usability and its impact on cognitive processing. Findings suggest ChatGPT, while not having a detrimental impact on cognitive processing as such, remains somewhat misunderstood and underutilized as an assistive technology distinct from machine translation, despite the important role this technology will inevitably play in shaping AI literacy in the language industry (Krüger 2023).

This article analyses and compares two texts produced via neural machine translation (NMT). It describes potential problems concerning lexical, syntactical and textual aspects and potential strategies for legal translation from English to German with reference to a blog text on criminal law authored by a US law firm. The text is part of the teaching material in a legal translation course of the Master’s program in Multilingual Specialized Communication and Specialized Translation (MA FKÜ) at the Institute of Translation and Multilingual Communication (ITMK) at the TH Köln – University of Applied Sciences. For this purpose, translations produced by DeepL and ChatGPT (more specifically the didactic interface used at the TH Köln, THKI GPT-Lab) are analysed with a focus on specific legal questions and the issue of legal comparison which is fundamental to legal translation. Finally, this comparison leads to a recommendation on the use of these tools.

Although it is less technologized than other language services, (partial) automation has also been witnessed in literary translation. Since 2017, neural machine translation (NMT) systems such as DeepL and, subsequently, large language models (LLMs) such as ChatGPT have been drastically changing, if not revolutionizing, the language industry. LLMs have larger context windows than NMT systems and allow for translations to be adapted to various reception situations. For instance, a text can be simultaneously translated and stylistically raised or lowered, and its literary form can be adapted in the target language. LLMs can also be useful for translating neologisms and other playful or creative textual elements. Despite the advantages of AI-based translation processes, some AI critics want literary translation to remain the sole preserve of humans. They speak of the impending destruction of the literary industry, a sensitive ecosystem, or of human creativity being connected to the desire to answer profound questions, which only humans are capable of, in the process of intellectual debate. Taking account of these conflicting positions, this contribution suggests that artificial intelligence can only transform literary translation in a sustainable way if humans are placed at the center of translation-relevant technological advancements.

This article traces the attitudes of national (German-speaking) and international employee associations in the field of audiovisual translation towards the development, training and use of AI in the language industry. By collecting and examining key statements in positioning papers, it is shown that the employees who unite in such creative associations are less concerned with the use of AI in the context of the multi-step translation process. Rather, they denounce the legal situation surrounding the use of copyrighted material in the development, training and subsequent use of AI applications. Copyright infringement has already occurred and continues to occur as questions around copyright are still insufficiently clarified. The collection of arguments in this contribution is followed by a look at the solutions provided by the associations for building a sustainable AVT industry that does not shy away from the use of AI solutions but also clearly defines the rights and obligations of the stakeholders involved.

In this paper, we analyse how English compound terms presented in specialized texts are rendered in the German target text by linking this transfer to the translationese phenomena of explicitation, implicitation and equivalence. We also use eye-tracking data to measure the cognitive effort that is related to the applied translation strategy. Our results indicate that when translating compound terms, translators usually prefer to choose the structural equivalent, i.e., a German compound. In addition, we found that explicitating a compound requires more cognitive effort than translating the compound with a structural equivalent, whereas choosing an implicitation reduces the cognitive effort for the translator.

This article investigates linguistic and aesthetic strategies employed by non-binary authors in contemporary German-language literature, with a specific focus on the challenges they present for translation. A brief survey of prevalent practices in German and Italian gender-fair language provides background for two representative case studies: Sasha Marianna Salzmann’s Außer sich (2017) and Kim de l’Horizon’s Blutbuch (2022). These texts illustrate the distinct strategies used by two acclaimed non-binary authors to depict gender-nonconforming characters. The analysis then turns to the Italian translations of both novels – Fuori di sé (2019) and Perché sono da sempre un corso d’acqua (2023) – to assess critically how the original German-language strategies adopted to challenge cis-normativity are conveyed (or lost) in the target language. By foregrounding the complexities of non-binary representation, this research underscores the importance of linguistic and cultural sensitivity in literary translation, emphasizing the need for translators to engage deeply with text, cultural context, and author intention, as well as to be mindful of non-binary identities when navigating the constraints of a target language.

This paper challenges the blanket rejection of the passive voice, advocating for its nuanced use and removal from the list of features commonly associated with poor or convoluted writing. Instead, it recognizes the passive voice as a valuable tool for crafting clear, comprehensible prose, particularly in plain language contexts. The paper reviews experimental evidence suggesting that while passive constructions may be slightly more challenging to produce orally and act out—especially for children, low-literacy individuals, and non-native speakers under time pressure—they are not inherently more difficult to comprehend than active constructions. In fact, passive sentences can enhance written comprehension when used appropriately: for example, when time pressure is absent, the information structure justifies their use, and the content does not need to be remembered or acted out in real-life situations. The findings underscore the importance of calibrating the use of the passive voice depending on the communicative context. Perhaps it is time for plain language advocates to reconsider the passive voice and lift the ban on its use.

To underscore the shift from linguistic to cognitive approaches in metaphor translation within Translation Studies, this paper analyzes a specialized corpus of metaphor translations in Sandalwood Death, utilizing cognitive framework procedures. Cognitive theory is posited as capable of penetrating the core of metaphor, a cognitive device crucial for meaning-making, and translation, which is inherently a cognitive activity. According to the Discourse Dynamics Framework (DDF), metaphor is interconnected across various dimensions, enhancing the understanding of metaphor within discourse. This study, therefore, examines the English translation of metaphors in the modern Chinese novel Sandalwood Death through the lens of the DDF and Schmidt’s cognitive typology, with the aim of proposing a multi-level model for interpreting metaphors and addressing translation challenges in complex discourses.

When comparing an ancient original text with its translations into a modern language, one may be surprised by the considerable variety of terms chosen by different translators in target texts, which is manifested not only in a lexical variety between synonyms, but also in a certain type of lexical variety that can be subjectively influenced by translators’ interpretations. With regard to the latter, this study analyzes to what extent this variety can affect the depiction of the Viking that is transmitted through translations into Modern English, deviating from what the original texts may be transmitting in Old English. For this purpose, the present study undertakes a qualitative and comparative analysis of thirteen different translations of poetry anthologies in Old English with texts in which the Vikings are mentioned, such as The Battle of Maldon, The Battle of Brunnanburh, Widsith, the Sermo Lupi ad Anglos (Wulfstan), as well as the Passion of Saint Edmund (Ælfric). Among the conclusions it is worth highlighting, in addition to a great lack of lexical uniformity, a generalization of terms such as Viking and pirate in translations, which can distort the image of the Viking in the original texts.

This article assesses the performance of four neural machine translation engines using the DQF-MQM methodology, focusing on the challenges inherent in using machine translation while working with resource-limited or minority languages such as Galician. The study examines translations between German and Galician in both directions, specifically within the context of touristic texts. The research considers different categories of errors, including accuracy, fluency and cultural references, to provide a comprehensive evaluation of the performance of the engines. By highlighting the pedagogical importance of these findings, the study aims to contribute to the training of students and future translators, equipping them with the skills needed to navigate the complexities of working with lesser-known languages. Furthermore, the results are intended to stimulate discussion on the wider implications of using machine translation in such language pairs, thereby enriching the ongoing discourse on the application of machine translation in diverse linguistic contexts.