Generative AI as a facilitator of deliberate practice in translator training

Chapter 2 Generative AI as a facilitator of deliberate practice in translator training Erik Angelone TH Köln University of Applied Sciences, Germany Advancement along an expertise trajectory, whether in training or professional contexts, stems from translators having ample opportunity to engage in deliberate practice, defined in the expertise studies literature as strategically designed activities to improve performance (Ericsson et al. 1993). In order for practice to be deliberate, a number of core conditions need to be met, including, among others, intrinsic learner motivation, focused and self-directed performance monitoring, informative (and relatively immediate) feedback, tasks being structured and undertaken at an appropriate difficulty level, and learners having opportunities to correct errors (Shreve 2006). In translator training contexts, complex, heterogeneous learner profiles, particularly when it comes to knowledge, skills, and competences (EMT 2022), pose inherent challenges in attempts to make sure these conditions are adequately met. In addressing these challenges, and in optimising opportunities for deliberate practice in general, Generative AI offers pedagogical value as a vehicle to guide self-directed learning. This chapter will discuss how generative AI can be used to facilitate deliberate practice in translator training, where students engage in self-directed metacognitive activity to reflect on and assess facets of their own performance, and, in doing so, acquire and advance their expertise. 1 Introduction Technological advancement, most recently in the realm of generative AI, is rapidly changing the roles and responsibilities of professional translators. It is also re-shaping the competences they need to possess to find success in the language industry, as reflected in the recently updated EMT Translation Competence Framework (EMT 2022). In times of seemingly perpetual change and a Erik Angelone. 2026. Generative AI as a facilitator of deliberate practice in translator training. In JC Penet, Joss Moorkens & Masaru Yamada (eds.), Teaching translation in the age of generative AI: New paradigm, new learning?, 27–47. Berlin: Language Science Press. DOI: 10.5281/zenodo.17641066 Erik Angelone corresponding need to adapt (Angelone 2023), one aspect regarding competence has remained fairly constant, namely the need for translators to critically reflect on, assess, and optimise their own performance. This capacity is articulated in several of the sub-competences pertaining to overarching ‘translation competence’ in the EMT framework, such as translators being able to “analyse a source document, identify potential textual and cognitive difficulties and assess the strategies and resources needed to reformulate it”, and “analyse and justify their translation solutions and choices” (EMT 2022: 8). It is also manifest in the domain of personal and interpersonal competence as the ability to “manage workload, cognitive load, stress and critical professional situations”, and “continuously selfevaluate, update and develop competences and skills through personal strategies and collaborative learning” (EMT 2022: 10). Competence, which can broadly be defined as the set of knowledge and skills needed to successfully translate, is closely tied to expertise, which, depending on research paradigm and theoretical foundation, can be described as ‘consistently superior performance’ in a given task domain (Ericsson & Charness 1994), or ‘optimal performance’ across varied, but interrelated task domains (Hatano & Inagaki 1986). Performance, when it comes to successful translation, involves not only quality (such as identifying and mitigating errors or successfully adhering to established project specifications), but also productivity (such as leveraging assistive technologies to expedite output or utilising ergonomically sound approaches when facing time constraints). Expertise should be regarded as something developmental and incremental (Shreve 2018) rather than a desirable ‘end state’ in the sense of an ‘expert’ translator. With this in mind, training for translation expertise acquisition and advancement has an important place from the outset. In order to advance along an expertise trajectory, for example from ‘novice’, to ‘advanced beginner’, to ‘competent’, to ‘proficient’, to ‘expert’ (Dreyfus & Dreyfus 1986), experience alone will likely not suffice. Instead, translators should intentionally seek opportunities to engage in ‘deliberate practice’, which can be defined as “individualized training activities especially designed by a coach or teacher to improve specific aspects of an individual’s performance through repetition and successive refinement” (Ericsson & Lehmann 1996: 278–279). Deliberate practice necessitates moving beyond “plateaus where one is comfortable and confident” (Horn & Masunaga 2006: 601). Unlike the somewhat predictable activity and approaches that translators might be inclined to encounter in their daily work, deliberate practice involves “exploring alternative methods with unknown reliability” (Ericsson et al. 1993: 368). Indeed, common industry constraints on time, budget, infrastructure, and risk-taking in general make deliberate practice 28 2 Generative AI as a facilitator of deliberate practice in translator training something distinct from ‘work’. It is more of an external, upfront investment that can ultimately optimise work performance. 2 Core conditions of deliberate practice In order for practice to be deliberate, a number of important core conditions need to be met. Translation Studies scholars have outlined these as they pertain to the task of translation (Shreve 2006). Perhaps most important is the condition of translators receiving immediate, informative feedback on their performance. “In the absence of adequate feedback, efficient learning is impossible and improvement minimal” (Ericsson et al. 1993: 367). Informative feedback ideally encompasses aspects of translation that are transferable and applicable across tasks. For example, feedback on the efficacy or inefficacy of a given external resource or assistive technology is likely more informative than feedback on a misspelled word that the translator might encounter in the context of one given translation but never again. As a second condition, translators need to have opportunities to correct errors. In pedagogical contexts, this condition is often met through workshopping, where students submit and receive feedback on drafts, followed by the opportunity to submit revisions in which errors brought to their attention can be corrected. Learning is particularly enhanced when error correction stems from discovery-based learning, where errors are annotated, but deliberately not fully spelled out by trainers. The very presence of a trainer is yet another core condition of deliberate practice. The trainer, or ‘coach’, is responsible not only for providing informative feedback, but also for establishing individualised learning objectives (Miller et al. 2020) and designing tasks based on firm understanding of the learner’s pre-existing knowledge. Some proponents of deliberate practiceoriented training suggest that individualised supervision is ultimately preferred over group-based instruction (Ericsson et al. 1993: 367). The trainer also plays a pivotal role in fulfilling another core dimension of deliberate practice, namely making sure the translator is engaging in tasks of an appropriate difficulty level. Stagnation, due to strict adherence to the tried and true, can result in a plateau effect that stands in the way of expertise advancement. Translators may encounter this when working on tasks of a uniform difficulty level (same source text length, same readability level, same set of external resources, same project specifications, same assessment parameters, etc.). Trainers can push trainees to work outside of their comfort zones and help determine “when transitions to more complex and challenging tasks are appropriate” (Ericsson et al. 1993: 367). 29 Erik Angelone Deliberate practice hinges on the learner’s intrinsic motivation, another core condition. Translators looking to advance along an expertise trajectory need to embrace working outside of their comfort zones and have a firm belief that deliberate practice will ultimately improve their performance. Deliberate practice takes immense time and effort. The oft-debated 10,000-hour ‘rule’ to becoming an ‘expert’ (Gladwell 2011) is often cited in attempts to quantify just how much time and effort are required. This would amount to “two and a half years of sustained effortful practice every day for five hours” (Shreve 2019: 173–174), raising questions regarding feasibility of this ‘rule’ and its place in the deliberate practice model. Deliberate practice calls for the learner’s commitment to “conscious performance monitoring” (Horn & Masunaga 2006: 601) and engaging in full concentration, as opposed to “mindless, routine performance” (Ericsson 2006: 692). This takes self-discipline and learner dedication to honing metacognitive capacities. Strategic scaffolding by the trainer, along with trainer/trainee dedication to the central ideas of cognitive constructivism (Piaget 1952), are instrumental in bolstering learner metacognition and performance monitoring processes. Figure 1 provides an overview of the core conditions of deliberate practice outlined in this chapter. It does not represent an exhaustive list of all conditions mentioned in the Expertise Studies literature, but rather focuses on those conditions taken up in Translation Studies to date. 3 Challenges to the implementation of deliberate practice It is worth noting here that empirical research on the benefits of deliberate practice on translation performance is still quite scant. This dearth can at least be partly explained by a series of inherent challenges associated with its implementation in formal training contexts. As mentioned in the previous section, several proponents of deliberate practice suggest that its benefits are best realised in contexts involving learners working one on one with individual trainers and in the absence of a pre-set curriculum (Ericsson et al. 1993: 367). This type of design does not readily align with the fashion in which translators are usually trained for a number of different reasons, starting with the financial constraint of needing to hire a personal trainer. Another constraint potentially standing in the way of meeting several of the core conditions of deliberate practice is the degree of learner heterogeneity commonly found in classroom-based training contexts. Students of translation often have widely varying levels of competence and experience, not to mention di30 2 Generative AI as a facilitator of deliberate practice in translator training immediate, informative feedback error correction opportunities "trainer" presence appropriate difficulty level intrinsic motivation conscious performance monitoring Core conditions of deliberate practice Figure 1: Core conditions of deliberate practice verse learning needs and interests. This can make it difficult for trainers to establish truly individualised learning objectives and corresponding tasks to meet them. The problem is exacerbated when student enrolments are high. Feedback, arguably the most important dimension of deliberate practice, often becomes less detailed and, out of necessity, much less immediate. Peer feedback and selffeedback activities can help address this gap, but immediacy, as a criterion for practice to be deliberate, often remains very difficult to obtain. Beyond time constraints, the deliberate practice condition of informative, immediate feedback might not be met if the kind of feedback being provided is not transferable in the sense of applying to future translation tasks. Oftentimes, assessment rubrics are used to mark up errors in accordance with various textual levels, such as grammar, word choice, and syntax. If the feedback given pertains to patterns along these lines that are applicable across tasks, it could be regarded as truly informative. If, on the other hand, feedback simply consists of marking up one-off errors pertaining to items that the translator may never encounter 31 Erik Angelone again in future translations, such as an isolated collocation error, it is relatively shallow and not particularly informative. Learner heterogeneity in large cohort settings also presents challenges when it comes to meeting the deliberate practice condition of making sure tasks are being undertaken at an appropriate level of difficulty. Translation practice courses are often informed by trainer intuition and prediction of appropriate difficulty level, in turn based on such facets as level of study, intake examination results, various cognitive process metrics (Sun & Shreve 2014), or previous student performance and corresponding ‘rich points’ (PACTE 2011). However, studies have shown a danger of misalignment between perceived or predicted problems and what actually proves to be problematic when translating (Angelone 2018). What is assumed to be difficult is actually not necessarily so, making attempts at predicting and setting an appropriate difficulty level challenging, particularly across a wide range of students with diverging needs. Furthermore, the idea of having students work outside of their comfort zone, or at the periphery of what they can realistically accomplish, for the purpose of advancing expertise may contradict approaches to translator training that embrace the predictability of staying within the learner’s comfort zone. According to deliberate practice guidelines, the learner’s intrinsic motivation needs to be constant. Having translation students work outside of their comfort zone in pedagogical contexts, particularly when grades are involved, runs the risk of hampering such motivation. The outcomes of deliberate practice would need to be gauged using metrics beyond formal grades, with a focus on helping learners become more self-reflective translators. Intrinsic motivation would come not so much from getting good grades or doing well in a given course, but rather from seeing the benefits of putting in hard work and embracing difficulty in order to become a better translator. 4 Generative AI as a facilitator of deliberate practice Against the backdrop of the conditions of deliberate practice presented in Section 2 and the constraints potentially standing in its way, as outlined in Section 3, the question remains: how can we best go about facilitating deliberate practice in translator training for purposes of expertise acquisition and advancement? In particular, how can we establish the requisite highly individualised, ‘coach’- oriented approach at the heart of deliberate practice? The pedagogical features of generative AI, in providing real-time feedback and prompt-driven interaction in a user-centred fashion, would seem to hold potential in this regard, as will be illustrated through a series of concrete scenarios in Section 6. 32 2 Generative AI as a facilitator of deliberate practice in translator training At the time of writing, we are still witnessing the dawn of generative AI as a vehicle for optimising translation, both in professional and pedagogical contexts. Its increasingly ubiquitous integration by LSP companies in project workflows has drawn attention to a need for artificial intelligence (AI) literacy (Krüger 2023), alongside MT literacy and data literacy in a broad sense. AI literacy can be defined as “a set of competencies that enables individuals to critically evaluate AI technologies, communicate and collaborate effectively with AI, and use AI as a tool online, at home, and in the workplace” (Long & Magerko 2020:2). In conjunction with discussions of translator performance, much of the discourse on AI intelligence to date has focused on how generative AI can be leveraged to facilitate human-in-the-loop translation, with an emphasis on the translation product. Outside of several of the chapters in this volume, relatively little discussion has been dedicated to the potential benefits of generative AI as a conversational agent, focusing less on generating translated content, and more so on enabling translators to reflect on their performance and engage in translation tasks to facilitate deliberate practice. Extending on sociocultural learning theories (Vygotsky 1965), some have come to regard a generative AI tool like ChatGPT as a ‘more knowledgeable other’ (MKO), in essence taking on the role of a personalised trainer that can “lead the learner from the zone of current development to the zone of proximal development – the space where one cannot quite master a content/task of their own, but they can with the help of an expert” (Stojanov 2023: 2). As an MKO, ChatGPT can address the aforementioned conditions of deliberate practice, particularly the presence of a one-on-one personalised trainer. Through real-time responses to learner prompts, ChatGPT ensures the immediacy of feedback that is so difficult to obtain in a larger enrolment classroom-based translator training environment. The learner’s intrinsic motivation is likely to be heightened when training is personalised and self-driven, rooted in immediate, informative feedback, and interactive, driven by their own prompts in relation to aspects of their own performance. 5 Generative AI as a scaffold for self-directed learning The utilisation of generative AI for purposes of deliberate practice calls for the learner to partake in self-directed learning (SDL), where “individuals take the initiative, with or without the help of others, in diagnosing their learning needs, setting learning goals, identifying resources, choosing appropriate learning strategies, and evaluating their learning outcomes” (Knowles 1975: 18). In this case, 33 Erik Angelone generative AI helps scaffold learning in line with learner prompts. Translation trainers well-versed in the conditions of deliberate practice can provide learners with valuable information on the nature of prompts they should enter. However, self-directed learners need to be “able, ready, and willing to prepare, execute and complete learning independently” (Jossberger et al. 2010: 419). The need for learner independence does not make the trainer superfluous, but rather shifts the focus of assignments undertaken and how they are assessed. Models of self-directed learning bear very close resemblance to the deliberate practice model. One such model that is widely cited in the literature consists of three closely interrelated dimensions: 1) self-management, 2) self-monitoring, and 3) motivation (Garrison 1997). Self-management involves the learner establishing concrete learning goals and managing learning resources to achieve these goals. In other words, they take control, deciding on the tasks in which they will engage. From a deliberate practice perspective, through strategic prompts, the translator can leverage generative AI to annotate errors in their translations. Generated annotations could then serve as a framework for the translator to selfdiscover the nature of the errors. Generative AI could then be prompted to provide similar translation tasks, with the goal of engaging the learner in deliberate practice centred around a certain error pattern (such as avoiding false cognates, erroneous literal translation, or problematic translationese at a syntactic level). Beyond error detection and mitigation, the translator can also use generative AI prompting to self-manage the difficulty level of the tasks they are undertaking. Section 6 provides more concrete scenarios and descriptions along these lines. Self-monitoring, the second component of Garrison’s model, pertains to the learner’s metacognitive processes. As an important dimension of self-directed learning, self-monitoring “requires learners to take responsibility to construct meanings” (Garrison 1997: 24). Through interactive feedback, generative AI can shed valuable light to help learners identify salient features of the translation task on which to focus their attention. For example, translators can enter a prompt asking generative AI to annotate source content that could be anticipated or predicted to present challenges in translation. Over the past decade, screen recording has found a place in process-oriented translator training for purposes of fostering self-monitoring and to enhance learner metacognition based on documentation of translation behaviors suggesting problems, including pausing, information retrieval, and revision (Angelone 2019). At present, generative AI tools like ChatGPT do not offer functionality where translators can upload screen recordings of their work for purposes of receiving analytic feedback at a granular level. Given recent advancements in this technology however, such as APIs that can provide 34 2 Generative AI as a facilitator of deliberate practice in translator training automated video summarisation, it is quite likely that utilisation of generative AI tools for such purposes is not too far away. The third component of Garrison’s self-directed learning model, motivation, directly aligns with motivation as a core condition of deliberate practice. Whereas deliberate practice regards motivation as the learner embracing challenge through an inherent desire to become better, Garrison draws attention to the importance of motivation for purposes of staying on task. The conversational interface of generative AI tools such as ChatGPT requires active participation on behalf of the learner, and, thereby, a heightened need to stay on task. However, without the physical presence of actual trainers or peers, staying on task, and motivation in general, is not a given. ChatGPT is not inclined to openly praise the performance of the learner to pique interest and motivation. Indeed, at present, the ChatGPT interface itself is quite basic, lacking any structural or discourse elements that might lend themselves well to inherently facilitating learner motivation. It will be interesting to see if this changes over time, perhaps in line with empirical user experience studies. 6 Application scenarios This chapter will now provide a series of concrete scenarios to illustrate how generative AI can be used to facilitate each of the core dimensions of deliberate practice put forward in Section 2 and to help translators advance along an expertise trajectory. The examples will be based on interaction with ChatGPT based on GPT-4o,1given the relative ubiquity and popularity of this particular generative AI tool at the time of writing. This focus on ChatGPT for illustrative purposes does not discredit the growing range of other generative AI tools available for use in a similar fashion. The translation scenario being used for purposes of contextualisation is the German-English translation of web content from a German private health insurance company,2translated for an international, English-speaking audience for informative purposes. The English translation was generated using DeepL.3 Figure 2 provides side-by-side alignment of the source and target content. As a point of departure, and to underscore the importance of strategic prompting when it comes to utilising GenAI for facilitating deliberate practice, (2) de1https://openai.com/index/gpt-4/ 2https://www.huk.de/gesundheit-vorsorge-vermoegen/krankenversicherung/ krankenvollversicherung.html#huk 3https://www.deepl.com/en/translator 35 Erik Angelone (11) “Provide me with a German source text to translate in this same domain that is a little more difficult than my previous translation”. (12) Generation of a slightly more difficult translation task Ihre private Krankenversicherung der HUK-COBURG Warum sich eine private Krankenversicherung lohnt • Umfangreicher Schutz – Unsere Tarife bieten umfassende Leistungen, die weit über das hinausgehen, was gesetzliche Krankenkassen abdecken. • Schnelle Terminvergabe – Bevorzugte Behandlung bei Fachärzten und kürzere Wartezeiten. • Hohe Flexibilität – Wählen Sie aus verschiedenen Selbstbeteiligungsstufen und Tarifoptionen, die sich Ihren individuellen Bedürfnissen anpassen. • Präventionsangebote – Zugang zu exklusiven Gesundheitsprogrammen und Vorsorgeuntersuchungen. • Internationaler Schutz – Weltweite Absicherung für Reisen und Aufenthalte im Ausland. • Transparente Beiträge – Keine versteckten Kosten, klare und nachvollziehbare Prämienberechnung. Establishing an appropriate difficulty level often proves to be a daunting task for trainers in formal translation classroom settings. It is interesting to see how ChatGPT goes about describing what exactly it is that makes this task more difficult when prompted to do so, as reflected in (13). (13) ChatGPT’s description of what makes the proposed task more difficult This translation task is more difficult because it includes: •Complex Vocabulary and Phrasing: The German text uses more specific and technical terms related to insurance, such as "umfangreicher Schutz," "Selbstbeteiligungsstufen," and "Prämienberechnung." Translating these terms accurately requires a good understanding of insurance terminology in both languages. •Longer Sentences and More Detailed Content: The sentences are longer and contain more detailed information, which requires careful attention to maintain clarity and coherence in English. 42 2 Generative AI as a facilitator of deliberate practice in translator training •Varied Benefits and Features: The text lists multiple benefits and features of the insurance, each with its own nuances. Ensuring that each point is clearly and accurately translated without losing meaning or context adds to the complexity. •Idiomatic Expressions: Certain phrases in German, such as "weit über das hinausgehen" and "bevorzugte Behandlung," might not have direct equivalents in English, requiring the translator to find idiomatic expressions that convey the same meaning. At a quick glance, much of what is deemed to be “more difficult” is somewhat generic (more complex field-specific terminology, a longer text, potential lexical gaps, etc.), as opposed to personalised for the individual translator based on previous performance. This condition of deliberate practice may very well prove to be the most challenging for not only human trainers, but also ChatGPT to meet. Perhaps ChatGPT’s efficacy in doing so hinges on the translator entering multiple translations as part of the prompt. These translations could then serve as a corpus of ‘big data’ for training the system to base what is being proposed as more difficult on concrete patterns (errors, text attributes, etc.). 7.4 Intrinsic motivation The extent to which learners are intrinsically motivated to partake in the scenarios outlined in this chapter is intertwined with the extent to which they value and trust ChatGPT as a ‘trainer’ and the extent to which they embrace self-directed learning. Both can take some time and necessitate adjustment, particularly in the case of translation students who might be used to interacting with peers and instructors. As previously mentioned, intrinsic motivation in the context of self-directed, generative AI-based deliberate practice will likely stem first and foremost from the level of personalisation and one-on-one interaction the technology enables. Learners are very much in the driver’s seat. It is their prompting and content input that serves as a catalyst for the entire deliberate practice process. Nevertheless, intrinsic motivation could quickly falter in instances where generative AI makes errors or provides unhelpful feedback. To maintain motivation, translators would need to look beyond such isolated instances and see the bigger picture of how the technology can foster their expertise acquisition on the whole. Depending on learning styles and preferences, some learners may even find motivation at the idea of being able to openly question and challenge the 43 Erik Angelone non-human ‘trainer’ in such contexts, and would perhaps be less inclined to do so in contexts involving face-to-face instruction involving human trainees and trainers. 7.5 Conscious performance monitoring Successful deliberate practice calls for the learner to be consciously engaged in reflection on their performance. In the absence of an actual human trainer and face-to-face learning, it is basically up to the learner to make sure they are on task and actively engaged in such monitoring. This important condition of deliberate practice may prove to be difficult for ChatGPT to facilitate in a fashion that is inherently advantageous in relation to what one might find in a face-to-face learning environment. When prompted to describe ways in which it can facilitate learners’ conscious performance monitoring when utilised as a sole ‘trainer’, ChatGPT proposed such strategies as programming the tool to periodically ask users for feedback on their performance, yet also points out that it could not track detailed usage analytics or goal progression. Instead, it can guide users on how to track their usage manually. In other words, the onus is truly on the learner when it comes to meeting this condition of deliberate practice. 8 Conclusion Whereas many of the current discussions surrounding generative AI involve its potential place in translation project workflows and in the generation of translated content, this chapter advocates for leveraging its pedagogical features as an MKO and conversational agent in facilitating deliberate practice and expertise acquisition for human translators. Its primary advantages in this capacity lie in its ability to generate immediate, informative feedback in a highly personalised fashion. As far as the nature of feedback is concerned, a tool like ChatGPT is equally efficacious at annotation for purposes of error correction and description for purposes of facilitating nuanced understanding and error mitigation. It can readily provide translators with deliberate practice in relation to a given text attribute or error type, with an eye towards transferability across tasks. Personalised feedback, one-on-one training, and ongoing interaction through learner prompt entry and tool response can foster intrinsic motivation. The core deliberate practice conditions of making sure tasks are undertaken at an appropriate difficulty level and that learners engage in conscious performance monitoring, at least at the time of writing, are still somewhat of a challenge for ChatGPT, yet both can be addressed through strategic prompting. 44 2 Generative AI as a facilitator of deliberate practice in translator training Truly experimental studies on the benefits of self-directed, deliberate practice in relation to more traditional learning approaches are still limited in scope, for the main reason that the core conditions of deliberate practice have been quite hard to address to date (Miller et al. 2020: 8). This is not something specific to translator training, but training in general, in all situations where learner heterogeneity, time constraints, and the absence of one-on-one training opportunities emerge as roadblocks. ChatGPT offers untapped potential to be a game changer in this regard and will hopefully draw renewed interest in both pedagogical and empirical research on deliberate practice and its place in translation expertise acquisition. References Angelone, Erik. 2018. Reconceptualizing problems in translation using triangulated process and product data. In Riitta Jääskeläinen & Isabel Lacruz (eds.), Innovation and expansion in translation process research, 17–36. Amsterdam, Netherlands: John Benjamins. Angelone, Erik. 2019. Process-oriented assessment of problems and errors in translation: Expanding horizons through screen recording. In Elsa HuertasBarros, Sonia Vandepitte & Emilia Iglesias-Fernández (eds.), Quality assurance and assessment practices in translation and interpreting, 179–198. Hershey, USA: IGI Global. Angelone, Erik. 2023. Weaving adaptive expertise into translator training. In Gary Massey, Elsa Huertas-Barros & David Katan (eds.), The human translator in the 2020s, 60–73. London, UK: Routledge. Dreyfus, Hubert & Stuart Dreyfus. 1986. Mind over machine: The power of human intuition and expertise in the era of the computer. Oxford, UK: Basil Blackwell. EMT. 2022. European master’s in translation competence framework 2022. Tech. rep. Brussels: European Commission. 1–12. https://commission.europa.eu/ system/files/2022-11/emt_competence_fwk_2022_en.pdf. Ericsson, K. Anders. 2006. The influence of experience and deliberate practice on the development of superior expert performance. In The Cambridge handbook of expertise and expert performance, 683–703. Cambridge, UK: Cambridge University Press. DOI: 10.1017/CBO9780511816796.038. Ericsson, K. Anders & Neil Charness. 1994. Expert performance: Its structure and acquisition. American Psychologist 49. 725–747. Ericsson, K. Anders, Ralf T. Krampe & Clemens Tesch-Römer. 1993. The role of deliberate practice in the acquisition of expert performance. Psychological Review 100(3). 363–406. DOI: 10.1037/0033-295X.100.3.363. 45 Erik Angelone Ericsson, K. Anders & Andreas C. Lehmann. 1996. Expert and exceptional performance: Evidence of maximal adaptation to task constraints. Annual Review of Psychology 47. 273–305. DOI: 10.1146/annurev.psych.47.1.273. Garrison, D. R. 1997. Self-directed learning: Toward a comprehensive model. Adult Education Quarterly 48(1). 18–33. DOI: 10.1177/074171369704800103. Gladwell, Malcolm. 2011. Outliers: The story of success. New York, USA: Back Bay Books. Hatano, Giyoo & Kayako Inagaki. 1986. Two courses of expertise. In Harold Stevenson, Hiroshi Azuma & Kenji Hakuta (eds.), Child development and education in Japan, 262–272. New York, USA: W. Y. Freeman & Co. Horn, John & Hiromi Masunaga. 2006. A merging theory of expertise and intelligence. In The Cambridge handbook of expertise and expert performance, 587–611. Cambridge, UK: Cambridge University Press. DOI: 10.1017/CBO9780511816796. 034. Jossberger, Helen, Saskia Brand-Gruwel, Henny Boshuizen & Margje van de Wiel. 2010. The challenge of self-directed and self-regulated learning in vocational education: A theoretical analysis and synthesis of requirements. Journal of Vocational Education & Training 62(4). 415–440. DOI: 10.1080/13636820.2010. 523479. Knowles, Malcolm. 1975. Self-directed learning: A guide for learners and teachers. Chicago, USA: Follett. Krüger, Ralph. 2023. Artificial intelligence literacy for the language industry – with particular emphasis on recent large language models such as GPT-4. Lebende Sprachen 68(2). 283–330. DOI: 10.1515/les-2023-0024. Long, Duri & Brian Magerko. 2020. What is AI literacy? Competencies and design considerations. In Proceedings of the 2020 CHI Conference on Human Factors in Computing Systems (CHI ’20), 1–16. Honolulu, USA: Association for Computing Machinery. DOI: 10.1145/3313831.3376727. https://doi.org/10.1145/3313831. 3376727. Miller, Scott D., Daryl Chow, Bruce E. Wampold, Mark A. Hubble, A. C. Del Re, Cynthia Maeschalck & Susanne Bargmann. 2020. To be or not to be (an expert)? Revisiting the role of deliberate practice in improving performance. High Ability Studies 31(1). 5–15. DOI: 10.1080/13598139.2018.1519410. https: //doi.org/10.1080/13598139.2018.1519410. PACTE. 2011. Results of the validation of the PACTE translation competence model: Translation problems and translation competence. In Cecilia Alvstad, Adelina Hild & Elisabet Tiselius (eds.), Methods and strategies of process research: Integrative approaches to translation studies, 317–343. Amsterdam, Netherlands: John Benjamins. 46 2 Generative AI as a facilitator of deliberate practice in translator training Piaget, Jean. 1952. The origins of intelligence in children. New York, USA: International Universities Press. Shreve, Gregory. 2006. The deliberate practice: Translation and expertise. Journal of Translation Studies 9(1). 27–42. Shreve, Gregory. 2018. Levels of explanation and translation expertise. Hermes 57. 97–108. Shreve, Gregory. 2019. Professional translator development from an expertise perspective. In Erik Angelone, Maureen Ehrensberger-Dow & Gary Massey (eds.), The Bloomsbury companion to language industry studies. London, UK: Bloomsbury Academic. Stojanov, Ana. 2023. Learning with ChatGPT 3.5 as a more knowledgeable other: An autoethnographic study. International Journal of Educational Technology in Higher Education 20(1). 35. DOI: 10.1186/s41239-023-00404-7. Sun, Sanjun & Gregory Shreve. 2014. Measuring translation difficulty: An empirical study. Target 26(1). 98–127. Vygotsky, Lev. 1965. Thought and language. Cambridge, USA: MIT Press. 47