Teaching translation students about data in the age of generative AI

Chapter 4 Teaching translation students about data in the age of generative AI Lynne Bowker Université Laval, Canada AI translation tools are now a key part of translation education, but many educators are searching for effective ways to teach the essentials of these tools to students with no background in computer science. This chapter explains why corpora make a good entry point to learning about AI translation tools, and it explores how science communication techniques such as framing, analogies and visualization can be used to help translation educators and students come to grips with data and machine learning. 1 Introduction Though translator education often takes place in an arts or humanities faculty, technology has been a vital part of the translation profession, and hence of translator education programmes, for at least thirty years. Over this period, translators and translation students have shown remarkable resilience as they adapt to new tools and new technology-based ways of working. However, this does not mean that teaching technologies to translation students is easy, and educating the educators can be a particular challenge (Bowker 2023, Kenny 2020). As the pace of new tool releases gets faster, it can be difficult for translator educators to know where to begin. This chapter proposes that teaching translation students about data is one of the key building blocks in preparing them to use translation technology effectively. In the case of translation, data frequently takes the form of texts organised into corpora. This chapter therefore begins with a brief review of corpora and corpus-based tools, noting how AI translation tools have influenced Lynne Bowker. 2026. Teaching translation students about data in the age of generative AI. in JC Penet, Joss Moorkens & Masaru Yamada (eds.), Teaching translation in the age of generative AI: New paradigm, new learning?, 67–85. Berlin: Language Science Press. DOI: 10.5281/zenodo.17641070 Lynne Bowker the nature of corpora. Next, the chapter outlines why a science communication approach, rather than traditional scientific communication, could be useful in the context of translation technology education. Three science communication techniques – framing, analogies and visualisation – are combined with examples to demonstrate how science communication could be adapted to teach translation students about data-related topics in the age of AI. 2 Corpora and corpus-based translation tools 2.1 The changing characteristics of corpora in the age of AI Over the past few decades, translators have seen the introduction and integration of a range of different tools, including concordancers (Zanettin 2023), term extractors (Korkontzelos & Ananiadou 2022), translation memory systems (Melby & Wright 2023), machine translation systems (Way 2020), and most recently, generative AI (GenAI) systems (Siu 2024). The various tools have become increasingly sophisticated with regard to their capabilities and their underlying architecture, but a common feature for all of the tools named above is that their core functionality revolves around processing data in the form of texts. Collections of texts are usually referred to as corpora, and these can take different forms depending on the nature of the texts and the way that they are organised (McEnery 2022). Translation tools that process corpora are frequently described as corpus-based or data-driven tools (Wang et al. 2022). As pointed out by Isabelle et al. (1993: 205), “existing translations contain more solutions to more translation problems than any other available resource”. Therefore, one type of corpus that has been used very often by translation tools is the bilingual parallel corpus (Simard 2020). In this type of corpus, a collection of source texts are aligned – usually at sentence level – with their counterpart target texts. In other words, each sentence in the source text is linked to its corresponding translation in the target text. Some translation tools might also use monolingual corpora of original texts in the source and/or target language to act as a linguistic model for that language. For students who have already learned about concordancers, term extractors or translation memory systems, the notion of a corpus is already familiar because the corpus is a very visible resource in such tools. When using these tools, the translator or translation student often has a hand in creating the corpus or may need to upload the corpus that has been provided by the client or educator. Owing to their familiarity, corpora make a good starting point for learning about neural machine translation or GenAI tools. 68 4 Teaching translation students about data in the age of generative AI According to Bowker & Pearson (2002: 9), A corpus can be described as a large collection of authentic texts that have been gathered in electronic form according to a specific set of criteria. There are four important characteristics to note here: ‘authentic’, ‘electronic’, ‘large’ and ‘specific criteria’. While the general notion of a corpus remained relatively stable in the period before AI translation tools appeared, the introduction of these tools has ushered in some changes with regard to the features of corpora. Therefore, it is important for translation students to understand how AI tools have influenced and altered the nature of corpora. 2.2 Machine-readable form The need for the corpus to be in electronic or machine-readable form has not changed. Indeed, we could say it is more important than ever since AI translation tools take on an even greater degree of corpus processing than do tools such as concordancers or translation memory systems. While concordancers and translation memory systems conduct pattern matching and then sort and display information for the tool users to interpret, AI translation tools go further by attempting to interpret the results and present fully formed translation solutions. 2.3 Size Corpora are used to reveal linguistic patterns, which only become apparent when there are multiple examples of a given linguistic phenomenon. Therefore, another feature of corpora is that they are usually very large collections of text. However, our understanding of what constitutes “large” has evolved over time. The first generation of corpora created in the 1960s contained hundreds of thousands of words and were mainly consulted by linguists (McEnery 2022). These linguists used corpus analysis tools (e.g. concordancers) to help them sort and display the text data, but the linguists were still responsible for interpreting it. Today, the corpora used to power AI tools contain hundreds of billions of words (Hughes 2023). This is in large part because these AI tools do not understand text in the way that people do, and they need a much larger number of examples in order to predict patterns with confidence. However, this need for extremely large corpora is influencing other characteristics of corpora, such as the texts selected for inclusion (see Section 2.4) and sometimes even the authenticity of the texts (see Section 2.5). 69 Lynne Bowker The availability of machine-readable texts can differ dramatically from one language to another, creating disparities with regard to an AI tool’s performance in different languages. A high-resource situation occurs when it is relatively straightforward to gather a large number of high-quality resources for a given language or language pair. For instance, both English and French are widely used languages, and there is a lot of translation activity between these two languages. As a result, it is not too difficult to compile monolingual and bilingual parallel corpora for these languages, and they are thus referred to as high-resource languages. In contrast, a low-resource situation can occur when languages (or language varieties) are less widely used, such as some of the Indigenous languages of the Americas or the variety of French used in Canada. For languages or language varieties of limited diffusion, it can be more challenging to build a large corpus. Moreover, even if two languages have a large number of speakers (e.g. Russian and Hindi), there may not be a lot of translation activity between them, making it hard to create bilingual parallel corpora for this language pair. Therefore, languages, language varieties or language pairs for which there are few corpora available are described as being low resource. 2.4 Specific criteria As emphasised by McEnery (2022), in order to be most useful, a corpus cannot consist of texts that have been gathered at random or in a purely opportunistic way. Rather, the texts in a corpus are selected because they correspond to specific criteria and are representative of a larger set of texts with those characteristics. One clear criterion in the context of translation is that the texts should be of high quality. Beyond this, there are many different options for designing a corpus depending on its intended purpose, but the key point here is that the choice of which texts to include is motivated. For example, in the context of a corpus to be used for translation, it could be important to select texts that are on a given topic, of a certain text type or register, or from a particular time period. Evidence of the importance of corpus design can be seen in the way that translators construct translation memory databases, such as by creating different databases (or adding relevant metadata) for different domains or for different clients (e.g. to respect their preferred terminology or house style). In this way, they can restrict a search to texts that have specific features (Melby & Wright 2023). Likewise, neural machine translation tools are known to achieve better quality when the corpus is adapted for a specific domain (Chu & Wang 2018). However, as noted in section Section 2.3, AI translation tools need to have an enormous number of texts in the corpus. As a result, it can be challenging to 70 4 Teaching translation students about data in the age of generative AI achieve the necessary size while trying to be selective about the content. The low-resource situation described in section Section 2.3 can be further complicated when it comes to finding certain text types or texts on very specialised topics in less widely used or translated languages and language varieties. A consequence of not having a large enough high-quality corpus is that the AI tool does not have enough good examples to draw on, and so the tool’s performance may be poorer in low-resource situations (Way 2025). If lower quality texts are included in the corpus in order to increase the size, then the tool may generate low-quality translations. The implications of insufficient quantity and quality of texts in training corpora used by AI translation tools is discussed in more detail in sections Section 4.2 and Section 4.3. 2.5 Authenticity Finally, the need to have authentic texts in the corpus used to be sacrosanct. For instance, McEnery (2022) describes a corpus as “a large body of linguistic evidence composed of attested language use” (494) and “a collection of naturally occurring language data” (495). In the case of translation, there is a desire to have high-quality data, which means using texts that have been translated by language professionals. As observed by Kenny (2011: 2), the reason that the developers of translation tools use corpora of human translations to train their systems is because such corpora are assumed to contain good answers to translation problems; and they are assumed to contain good answers precisely because they contain translations performed by human beings. One consequence of the need for bigger and bigger bilingual parallel corpora for training AI translation tools has been that such authentic high-quality human translated texts have become increasingly valuable commodities. This in turn has raised many ethical questions about ownership of translation data and permission to use it, prompting an explicit need to discuss these issues with translation students. Moorkens (2022) contains a detailed examination of such ethical issues, along with suggestions for how these can be integrated into translator education (see Section 4.1). Another response to the potential shortage of text needed for constructing very large corpora for use with AI translation tools has been to set aside the longestablished tradition of using authentic data and to explore the use of synthetic data. In the context of translation, synthetic data is created by using a machine translation tool to translate additional texts, and then adding these machinetranslated texts to the corpus (Sennrich et al. 2016). The quality of synthetic data can vary, and while the need for additional data tends to be for low-resource 71 Lynne Bowker languages, these are the languages for which machine translation already underperforms. Therefore, using a low-performance tool to generate more translations in that same language is likely to result in poor quality text. If this poor quality text is then used to further train the tool, there is a risk that using synthetic data could perpetuate a cycle of mediocrity. This section has elaborated why understanding corpora is important for translation students, and has outlined some of the ways that corpora are changing in the age of AI. The following section moves beyond what students need to know about translation data (i.e., corpora) to consider how educators can communicate this information effectively. 3 From scientific communication to science communication 3.1 Scientific communication Scientific communication (sometimes called internal science communication) is a type of expert-to-expert communication (Hanauska 2019). It occurs when one subject field specialist addresses another, with both parties having a deep knowledge of the complex matter in question. Academics are typically very comfortable with this type of communication since they spend much of their time discussing their research with their peers. Even as part of their teaching, they are normally communicating information directly from their own field, helping their students to acquire the necessary expertise to become specialists in their own right. However, the arrival of AI tools – more than any other technology – has changed that situation for many translation educators. As noted in the introduction, technology is not new either to the translation profession or to translator education programmes. Yet as pointed out by Kenny (2020), when it comes to translation technology, identifying what to teach has often been easier than working out how to teach it. As the field of translation became more technologised, some educators admitted to feeling challenged by the demands of keeping up with technology (e.g. Kenny 2007, Marshman & Bowker 2012), but this responsibility lay primarily with those educators who had chosen to specialise in technology. Technology-related teaching was often restricted to a core course on translation technologies, where a large focus was on learning how to use the tools, which offered an array of sophisticated features (Bowker 2023). AI translation tools are different in several respects. Firstly, the user interface is comparatively simple – often requiring the user to do little more than select 72 4 Teaching translation students about data in the age of generative AI the source and target languages and paste or type a text. In some cases, the AI tool may even be called automatically by another tool. Therefore, the “howto” aspect of translation technology teaching is no longer a principal focus. A more important difference, however, is that the technology behind the interface is far from simple. As Kenny (2018) points out, older generations of technology were relatively transparent and comprehensible in their inner workings, while AI translation tools are more opaque in that the people using them – and even the people developing them – cannot always understand how the tool arrives at its proposals or output. In the field of AI more broadly, this opacity has led to a push for more explainable AI (XAI) (Ridley 2025), while machine translation researchers are also beginning to work on explainable aspects of this technology (e.g. Lankford et al. 2023). In the meantime, AI translation tools are now starting to appear in courses across the translation education curriculum rather than solely in core technology courses, and translator educators are seeking ways to help themselves and their students understand how these tools can affect translation processes and products, and how to use the tools responsibly. In these circumstances, scientific communication is not feasible since translation educators and students are not computer scientists. The technology behind neural machine translation and GenAI tools is highly sophisticated. The tools may appear to be simple because the user interface is simple, but behind the scenes, there are very complex artificial neural networks, word embeddings, vectors, transformers, and more. It is not necessary for most translation educators or students to understand all the details of how an artificial neural network operates in order to use AI-based translation tools.1However, understanding the role of data (i.e., corpora) in a data-driven AI translation tool is valuable and can help translation students to be more informed and responsible users of this type of technology. At the very least, understanding the role of data is a useful first step, and the approach proposed in section 4 is intended as way to introduce data on programmes aimed at educating students who will work primarily as language 1This is currently true for many translation-oriented jobs and translation education programmes. However, as Briva-Iglesias & O’Brien (2022) point out, new jobs are emerging that straddle the boundary between translator and technologist and that require a deeper grounding in technologies. To address this changing market, new education programmes are also emerging that place more emphasis on technological skills, such as the MSc in Translation Technology at Dublin City University in Ireland, and the MSc in Multilingual Digital Communication at McGill University in Canada. In these more technology-oriented programmes, a deeper understanding of the underlying architectures is both required of the educators and offered to the students, and data literacy training may also be deeper and more technical, such as the approach described by Hackenbuchner & Krüger (2023) as part of the DataLitMT project. 73 Lynne Bowker professionals rather than as technology specialists. But if scientific communication is not a feasible approach, can science communication help? 3.2 Science communication When translation students hear people talking about AI-based translation tools using terms such as transformers or word embeddings, they may feel intimidated and lack confidence in their ability to use these tools effectively. Therefore, when introducing these tools, translation educators may find it helpful to draw on science communication techniques. As pointed out by Burns et al. (2003), science communication is tricky to define precisely, and there are myriad competing and overlapping descriptions of this concept in the literature. However, there is general agreement that the essential goal is to make expert knowledge accessible to non-experts. This is why this activity is sometimes described more specifically as external science communication (Hanauska 2019). In the case of translation students, while they are developing expertise in translation, they are not experts in AI or machine learning. Academics in all disciplines are increasingly encouraged to engage in science communication to share the results of their research with a wider audience, which could include policy makers, participants in research studies, funding agencies and even the general public (e.g. National Academies of Sciences, Engineering, and Medicine 2017). Yet while science communication is strongly encouraged, many academics receive little training in this area, even though it is a complex and skilled task (Borowiec 2023). Fortunately, tips, guides and recommendations are emerging to help fill this gap (e.g. Borowiec 2023, Cooke et al. 2017, Henville 2020). From these, we can glean several suggestions that can be integrated into teaching to help explain data-related concepts to translation students as a foundation for learning about AI translation tools. Frame your message: Frame your message in terms that are accessible, relatable and meaningful for your audience. Framing is not about marketing your point of view but about finding a way to actively engage the audience with an issue by showing them why they should care about it. Make the abstract concrete and connect unfamiliar concepts to familiar ones: Use metaphors or analogies that connect at the level of underlying commonalities to help the audience understand an unfamiliar or abstract concept in relation to a familiar and concrete one. 74 4 Teaching translation students about data in the age of generative AI Visualise the content: Use visual aids (e.g. images, concept maps, graphs) to support your message. Making the information available in different modes or formats can help to crystallise the ideas in the mind of your audience. The following section considers how these three science communication techniques could be implemented to help translation educators teach concepts related to data. 4 Applying science communication techniques to teach translation students about data 4.1 Framing the message about data It was noted in section 2 that corpora can be a good entry point for introducing data-driven AI tools to translation students because they have likely been exposed to corpora when using other tools (e.g. concordancers, term extractors, translation memory systems). Corpora are less immediately visible in the more automated AI translation tools, but this makes it all the more important to actively discuss them. As Kenny (2011) emphasises, corpus-based tools cannot function without corpora. If no corpus is present, the tool itself has nothing to offer as a translation aid. This is a very important point to share with translation students because it demonstrates the essential contribution of translators. Of course the tool developers play a key role in both programming the tools and determining how corpus data can be processed effectively, but the data itself – the translated text – is indispensable, and this fact is often glossed over when the tools are described or promoted. Yet as part of translator education, it is important to make students aware of the value of corpora because this helps them to understand their own worth as language professionals in addition to understanding how corpus-based or data-driven tools work, and where the potential pitfalls might lie. As mentioned in section Section 2.5, translation corpora have become valuable resources, leading to some questionable practices. Framing the discussion of AI translation tools as part of a broader issue of professional practice, ethics, and digital citizenship, rather than simply as an instrumental means of making their job easier, can help translation students to recognise not only the value of their work, but also the fact that technology does not exist in a vacuum. Discussions about the role of data in data-driven tools can encourage translation students to reflect on how AI translation technology can impact tool users and end users of translated products in various ways. It also empowers language professionals to 75 Lynne Bowker Burns, T. W., D. J. O’Connor & S. M. Stocklmayer. 2003. Science communication: A contemporary definition. Public Understanding of Science 12(2). 183–202. DOI: 10.1177/09636625030122004. Chu, Chenhui & Rui Wang. 2018. A survey of domain adaptation for neural machine translation. In Proceedings of the 27th international conference on computational linguistics, 1304–1319. Santa Fe, USA: Association for Computational Linguistics. Cooke, Steven J., Austin J. Gallagher, Natalie M. Sopinka, Vivian M. Nguyen, Rachel A. Skubel, Neil Hammerschlag, Sarah Boon, Nathan Young & Andy J. Danylchuk. 2017. Considerations for effective science communication. Facets 2. 233–248. DOI: 10.1139/facets-2016-0055. Ehrensberger-Dow, Maureen, Alice Delorme Benites & Caroline Lehr. 2023. A new role for translators and trainers: MT literacy consultants. The Interpreter and Translator Trainer 17(3). 393–411. DOI: 10.1080/1750399X.2023.2237328. Flach, Peter. 2012. Machine learning. Cambridge, UK: Cambridge University Press. Hackenbuchner, Janiça & Ralph Krüger. 2023. DataLitMT – teaching data literacy in the context of machine translation literacy. In Mary Nurminen, Judith Brenner, Maarit Koponen, Sirku Latomaa, Mikhail Mikhailov, Frederike Schierl, Tharindu Ranasinghe, Eva Vanmassenhove, Sergi Alvarez Vidal, Nora Aranberri, Mara Nunziatini, Carla Parra Escartín, Mikel Forcada, Maja Popovic, Carolina Scarton & Helena Moníz (eds.), Proceedings of the 24th annual conference of the European Association for Machine Translation, 285–293. Tampere, Finland: European Association for Machine Translation. Hanauska, Monika. 2019. Historical aspects of external science communication. In Annette Leßmöllmann, Marcelo Dascal & Thomas Gloning (eds.), Science communication, 585–600. Berlin, Germany: De Gruyter Mouton. DOI: 10.1515/ 9783110255522-027. Henville, Letitia. 2020. Effective science communication deconstructed. University Affairs. https://universityaffairs.ca/career-advice/ask-dr-editor/effectivescience-communication-deconstructed/. Hook, Daniel. 2023. The lone banana problem. Digital Science. https: // www. digitalscience.com/tldr/article/thelonebananaproblemorthenewprogramming-speaking-ai/. Hughes, Alex. 2023. ChatGPT: everything you need to know about OpenAI’s GPT-4 tool. https://www.sciencefocus.com/future-technology/gpt-3. Isabelle, Pierre, Marc Dymetman, George Foster, Jean-Marc Jutras, Elliott Macklovitch, François Perrault, Xiaobo Ren & Michel Simard. 1993. Translation analysis and translation automation. In Proceedings of the fifth interna82 4 Teaching translation students about data in the age of generative AI tional conference on theoretical and methodological issues in machine translation of natural languages: Association for computational linguistics, 201–217. Kenny, Dorothy. 2007. Translation memories and parallel corpora: Challenges for the translation trainer. In Dorothy Kenny & Kyongjoo Ryou (eds.), Across boundaries: International perspectives on translation studies, 192–208. Newcastle-Upon-Tyne, UK: Cambridge Scholars Publishing. Kenny, Dorothy. 2011. The ethics of machine translation. In Sybille Ferner (ed.), Proceedings of the 20th New Zealand Society of Translators and Interpreters national conference. Auckland, New Zealand: New Zealand Society of Translators & Interpreters. Kenny, Dorothy. 2018. Machine translation. In Piers Rawling & Philip Wilson (eds.), The Routledge handbook of translation and philosophy, 428–445. London, UK: Routledge. Kenny, Dorothy. 2020. Technology and translator training. In Minako O’Hagan (ed.), The Routledge handbook of translation and technology, 498–515. London, UK: Routledge. Koponen, Maarit & Mary Nurminen. 2024. Risk management for content delivery via raw machine translation. In Marion Winters, Sharon Deane-Cox & Ursula Böser (eds.), Translation, interpretation and technological change, 111–135. London, UK: Bloomsbury. Korkontzelos, Ioannis & Sophia Ananiadou. 2022. Term extraction. In Ruslan Mitkov (ed.), The Oxford handbook of computational linguistics, 2nd edn., 991– 1012. Oxford, UK: Oxford University Press. Lankford, Séamus, Haithem Afli & Andy Way. 2023. adaptNMT: An open-source, language-agnostic development environment for neural machine translation. Language Resources and Evaluation 57. 1671–1696. DOI: 10.1007/s10579-02309671-2. Marshman, Elizabeth & Lynne Bowker. 2012. Translation technologies as seen through the eyes of educators and students: Harmonizing views with the help of a centralized teaching and learning resource. In Séverine HubscherDavidson & Michał Borodo (eds.), Global trends in translator and interpreter training, 69–95. London, UK: Continuum. McEnery, Tony. 2022. Corpora. In Ruslan Mitkov (ed.), The Oxford handbook of computational linguistics, 2nd edn., 494–507. Oxford, UK: Oxford University Press. Melby, Alan K. & Sue Ellen Wright. 2023. Translation memory. In Chan Sinwai (ed.), Routledge encyclopedia of translation technology, 2nd edn., 766–781. London, UK: Routledge. 83 Lynne Bowker Moorkens, Joss. 2022. Ethics and machine translation. In Dorothy Kenny (ed.), Machine translation for everyone, 121–140. Berlin, Germany: Language Science Press. Narla, Akhila, Brett Kuprel, Kavita Sarin, Roberto Novoa & Justin Ko. 2018. Automated classification of skin lesions: From pixels to practice. Journal of Investigative Dermatology 138(10). 2108–2110. DOI: 10.1016/j.jid.2018.06.175. National Academies of Sciences, Engineering, and Medicine. 2017. Communicating science effectively: A research agenda. Washington DC, USA: The National Academies Press. DOI: 10.17226/23674. https://nap.nationalacademies.org/ catalog/23674/communicating-science-effectively-a-research-agenda. Ribeiro, Marco T., Sameer Singh & Carlos Guestrin. 2016. Why should I trust you? Explaining the predictions of any classifier. In Proceedings of the 22nd ACM SIGKDD International Conference on Knowledge Discovery and Data Mining: Association for computing machinery, 1135–1144. New York, USA. DOI: 10.1145/ 2939672.2939778. Ridley, Michael. 2025. Human-centered explainable artificial intelligence: An annual review of information science and technology (ARIST) paper. Journal of the Association for Information Science and Technology 76(1). 98–120. DOI: 10.1002/asi.24889. Roberts, Michael, Derek Driggs, Matthew Thorpe, Julian Gilbey, Michael Yeung, Stephan Ursprung, Angelica I. Aviles-Rivero, Christian Etmann, Cathal McCague, Lucian Beer, Jonathan R. Weir-McCall, Zhongzhao Teng, AixCovnet Effrossyni Gkrania-Klostas, James H. F. Rudd, Evis Sala & CarolaBibiane Schönlieb. 2021. Common pitfalls and recommendations for using machine learning to detect and prognosticate for COVID-19 using chest radiographs and CT scans. Nature Machine Intelligence 3. 199–217. DOI: 10.1038/ s42256-021-00307-0. Sennrich, Rico, Barry Haddow & Alexandra Birch. 2016. Improving neural machine translation models with monolingual data. In Proceedings of the 54th annual meeting of the Association for Computational Linguistics (volume 1): Association for computational linguistics, 86–96. Simard, Michel. 2020. Building and using parallel text for translation. In Minako O’Hagan (ed.), The Routledge handbook of translation and technology, 78–90. London, UK: Routledge. Siu, Sai Cheong. 2024. Revolutionising translation with AI: Unravelling neural machine translation and generative pre-trained large language models. In Yuhong Peng, Huihui Huang & Defeng Li (eds.), New advances in translation technology, 29–54. Singapore: Springer. DOI: 10.1007/978-981-97-2958-6_3. 84 4 Teaching translation students about data in the age of generative AI Vanmassenhove, Eva, Christian Hardmeier & Andy Way. 2018. Getting gender right in neural machine translation. In Proceedings of the 2018 Conference on Empirical Methods in Natural Language Processing, 3003–3008. Brussels, Belgium: Association for Computational Linguistics. Wang, Haifeng, Hua Wu, Zhongjun He, Liang Huang & Kenneth W. Church. 2022. Progress in machine translation. Engineering 18. 143–153. DOI: 10.1016/j.eng. 2021.03.023. Way, Andy. 2013. Traditional and emerging use-cases for machine translation. In Proceedings of Translating and the Computer 35. London, UK: ASLIB. https: //aclanthology.org/2013.tc-1.12/. Way, Andy. 2020. Machine translation: Where are we at today? In Erik Angelone, Maureen Ehrensberger-Dow & Gary Massey (eds.), The Bloomsbury companion to language industry studies, 311–332. London, UK: Bloomsbury. Way, Andy. 2025. What does the future hold for translation technologies? In Stefan Baumgarten & Michael Tieber (eds.), The Routledge handbook of translation technology and society, 448–461. London, UK: Routledge. Zanettin, Federico. 2023. Concordancing. In Chan Sin-wai (ed.), Routledge encyclopedia of translation technology, 2nd edn., 498–511. London, UK: Routledge. 85