Teaching subtitling in the times of generative AI

Chapter 9 Teaching subtitling in the times of generative AI David Orrego-Carmona University of Warwick, United Kingdom and University of the Free State, South Africa This chapter explores the integration of Generative Artificial Intelligence (GenAI) into subtitler training and its impact on translator education. Drawing on teaching experiences and research insights, it presents strategies for designing and delivering training courses that prepare future subtitling professionals to work effectively in an AI-enhanced industry. Subtitling is a multifaceted translation practice that requires technical, linguistic and cultural skills. Recent developments in GenAI are reshaping how these skills are taught and applied. Based on realistic scenarios and exploratory teaching methods, the chapter examines how Large Language Models (LLMs) can support different stages of the subtitling process through inquisitive integration. The discussion encompasses practical approaches to integrating GenAI tools into the classroom, drawing on examples related to the translation of cultural references and template creation. The chapter adopts a hands-on problem-solving approach to training that encourages students to evaluate technological possibilities whilst developing foundational knowledge. Through examples from teaching practice, it shows how comparing different AI solutions and assessing their suitability for specific tasks helps students make informed decisions about implementing automated solutions. This approach positions students as active agents in their learning process while helping them understand the potential and limitations of automation. Critically examining the role of educators in this changing landscape, the chapter advocates for training that prepares adaptable professionals who can navigate technological developments whilst maintaining high standards and advocating for sustainable working conditions. More broadly, it contributes to discussions about providing students with the necessary tools and knowledge to shape sustainable careers in an increasingly automated media localisation industry. David Orrego-Carmona. 2026. Teaching subtitling in the times of generative AI. in JC Penet, Joss Moorkens & Masaru Yamada (eds.), Teaching translation in the age of generative AI: New paradigm, new learning?, 167–190. Berlin: Language Science Press. DOI: 10.5281/zenodo.17641080 David Orrego-Carmona 1 Introduction This chapter addresses the question of integrating Generative Artificial Intelligence (GenAI) into subtitler training. Audiovisual translation (AVT) is a technology-driven field, and the reconfiguration of subtitling processes aptly encapsulates how accelerated GenAI-influenced changes affect education. Subtitlers require technical, linguistic and transfer skills, as well as market awareness, to operate efficiently in the industry. Training professional subtitlers who can engage effectively with the industry involves “taking into account the linguacultural dimension as well as the technological possibilities and the market reality” (Díaz-Cintas & Remael 2021: 62). Until relatively recently, technological tools in subtitling workflows focused on technical aspects such as synchronisation and segmentation, and sometimes automatic speech recognition (ASR) (Georgakopoulou 2020). The field is experiencing exponential growth in the implementation of machine translation (MT) and GenAI solutions (Slator 2024), pushed forward by Language Service Companies (LSCs) to speed up processes and address what they have called a talent crunch (Estopace 2017, Papercup 2022), although subtitlers attribute this so-called shortage to declining rates of remuneration rather than an actual lack of available professionals (Orrego-Carmona 2024). The expansion of the media localisation industry has placed subtitling as one of the most active sectors of the market, resulting in growing interest from LSCs, increasing demands on production networks and pressure on the workforce. This rapid transformation presents unique challenges for subtitling education. A survey conducted by Slator (2024) showed that different types of captioning and subtitling services involving AI are commonly offered by Language Service Providers. Both trainers and trainees must continuously adapt their knowledge and skills as technologies evolve. The traditional approach of mastering established tools and workflows needs to give way to developing adaptability and critical evaluation skills for emerging technologies. This chapter explores how we can structuresubtitling training to address these challenges while maintaining high professional standards. Drawing on teaching experiences and research insights, I present strategies for designing and delivering training courses that empower future subtitling professionals to work effectively in an AI-shaped industry. Understanding the conditions, constraints and opportunities at every stage allows future professionals to make informed decisions and supports their communication and negotiation skills. The discussion encompasses practical approaches to integrating GenAI tools into the classroom 168 9 Teaching subtitling in the times of generative AI Figure 1: Results of a survey by Slator from May 2024 that asked 129 Language Service Providers which language AI services their company offers within a project-based approach (Mitchell-Schuitevoerder 2020), exploring conversational Large Language Models (LLMs), and implementing problem-solving activities to develop critical evaluation skills. The chapter adopts a hands-on problem-solving approach developed through teaching practice, equipping future graduates to respond to the rapidly changing demands of the media localisation industry. The proposed activities intentionally focus on non-conventional subtitler tasks to demonstrate the wide-ranging application of AI technologies and encourage a broader reflection on AI integration. The chapter examines how educators can keep up with technological developments and guide students to harness AI’s potential while fostering a deep understanding of linguistic features, cultural contexts, and ethical considerations intrinsic to subtitling, automation, and the media industry. More broadly, this chapter contributes to discussions about providing students with the necessary tools and knowledge to shape a sustainable career. 2 Evolving landscape of subtitling training Training subtitlers requires addressing the technical requirements of the practice related to synchronisation and spotting while balancing the linguistic and cultural requirements. Traditional subtitling workflows for single language pairs followed a relatively straightforward process: transcription of the source content, translation, synchronisation (or spotting) of subtitles with the audiovisual content, and quality control (QC) (Díaz-Cintas & Remael 2021). Each stage required specific skills and was typically carried out sequentially, with subtitlers working on standalone workstations using subtitling software. The globalisation of the media industry has led to the implementation of template-based workflows (Georgakopoulou 2019). In this model, LSCs create a master file, usually in English, which serves as the basis for pivot translations into multiple target languages (Figure 2) and streamlines the multilingual subtitling 169 David Orrego-Carmona process (Valdez et al. 2023). Templates are essential for the multilingual operations of transnational platforms. Netflix (n.d.) defines subtitle templates as “an edited, positioned, researched, annotated and checked subtitle file, timed to shot and audio, matching the source language of the associated content”. These templates include not only the text but also technical parameters and notes that support translations into multiple languages. For example, in producing subtitles for Netflix’s successful Korean show “Squid Game”, an English template serves as the pivot for translations into multiple languages. A template for this show would include not only the script and timing but also essential cultural annotations. Critical elements that would require annotation include Korean honorific forms (indicating relationships between characters), cultural references (such as the children’s games that structure the plot), and social hierarchies expressed through linguistic choice. For instance, the relationship between the two characters who share a childhood friendship requires careful annotation as their use of informal language reflects their familiarity, while their interactions with other characters follow strict social hierarchies reflected in formal language patterns. These annotations in the English template help translators working into other languages understand and preserve these nuanced relationships and cultural elements in their target versions. In some cases, these templates are locked and do not allow translators to alter the timings of the subtitles. Thus, subsequent translations need to adapt to the synchronisation and reading speed of the pivot language. This approach streamlines the production process but also transforms the subtitler’s role from having control over technical and linguistic aspects to focusing primarily on translation within pre-established constraints. Template creation Template QC Translation into target language(s) Translation QC Product delivery Figure 2: Template-based translation process The subtitling landscape is experiencing an unprecedented transformation. As Szarkowska & Jankowska (2024) observe, workflows are becoming increasingly automated. They suggest that manual spotting and transcription may soon become obsolete, replaced by automatic speech-recognition-based spotting. This coincides with Bolaños García-Escribano’s view that these changes are “perhaps leading to scenarios where spotting will be increasingly automatised and linguists will serve as language engineers” (2025: 19). In current post-editing 170 9 Teaching subtitling in the times of generative AI processes, AI tools in the form of ASR and MT reshape traditional practices and embed revision stages to accommodate the human verification of the automatic output (Figure 3). These workflows typically involve two distinct post-editing phases: transcription post-editing, where subtitlers verify and correct automatically generated transcripts from ASR systems, and translation post-editing, where they refine machine-translated content. This process transformation is further accelerated by the shift from standalone workstations to cloud-based environments that integrate translation memories, QC automation and MT and transfer the control of these resources to the LSCs. Automatic timed transcription Transcr. PE MT PE Quality control Delivery Figure 3: Post-editing (PE) production processing using ASR and MT The implementation of AI tools for technical aspects has a long tradition in subtitling software and subtitlers tend to be more acquainted with them. Features such as importing and segmenting scripts, automatic timing and synchronisation, shot-change identification and automated QC are standard in subtitling workflows. Well-established tools, such as Ooona, EZTitles and ZOOsubs integrate these by default, with some even allowing customisation. Core competences in subtitling foster the mastering of technical aspects such as synchronisation and segmentation, while developing the linguistic and cultural awareness needed to produce high-quality translations. However, the current industry landscape demands additional skills. Subtitlers now need to understand and critically assess the output of multiple automated systems, from ASR to MT, when engaging with post-editing workflows. Post-editing AVT requires professionals to coordinate an increasing number of sources of information and evaluate them before deciding to keep, edit or fully replace automatically generated content. The industry’s push towards automation responds to multiple factors (OrregoCarmona 2024). The expansion of streaming platforms has increased the demand for subtitled content, with companies requiring fast turnaround times for multiple language combinations. LSCs argue that a talent shortage constrains their ability to meet these growing demands, leading them to implement hybrid and automated solutions (Iyuno SDI Group 2022, Marking 2022). However, professional associations disagree with this assessment, denying that low payment and poor working conditions, rather than a talent shortage, are the primary concerns (AVTE 2023a). 171 David Orrego-Carmona The changes in the production processes require a substantial revision of how we approach subtitler training. The focus must shift from teaching specific technical and translation skills to developing adaptable professionals who can critically engage with evolving technologies while maintaining high-quality standards. Like in other areas of translator training, programmes need to address new technological requirements (ASR, MT, automated spotting, LLMs), industry demands, critical assessment skills for AI-generated outputs, and professional development needs in a rapidly changing environment. The challenge for training programmes is to balance translation skills and technical proficiency with critical awareness. Students need to understand not only how to use new tools but also how to evaluate their suitability for different projects, markets and contexts. This involves developing what Tipton (2024) calls digital reflexivity: the ability to identify challenges in the external environment, evaluate available resources and implement appropriate strategies. Such an approach helps students navigate the changing landscape while maintaining professional standards and developing sustainable careers. 3 Designing AI-enhanced training and digital reflexivity Digital reflexivity refers to students’ ability to evaluate challenges and resources, and implement solutions. This can be illustrated through subtitling project work where students must make informed decisions at every stage. For instance, when creating subtitles, students first analyse their source product’s translation needs, including the technical, cinematographic, linguistic and cultural aspects. They then evaluate available tools and resources, from subtitling software to AI solutions, justifying how these align with project requirements. Through this process, students learn to identify potential challenges in the external environment (such as technical constraints or cultural references), assess their resources (both technological and knowledge-based), and implement appropriate strategies to bridge any gaps. This reflective practice manifests in practical decisions such as: • Selecting appropriate software based on project needs • Deciding which stages of the workflow can benefit from automation • Evaluating when to use AI tools and when human expertise is crucial • Documenting and justifying translation decisions 172 9 Teaching subtitling in the times of generative AI Reflective practice extends beyond technical decision-making to include selfawareness about how technological changes affect professional identity, work satisfaction, and career expectations. Students need opportunities to examine their own responses to AI integration, both excitement about new possibilities and concerns about professional autonomy, and consider how these evolving conditions align with their personal values and career goals. The resulting commentary demonstrates how students develop digital reflexivity by connecting theoretical knowledge with practical implementation whilst maintaining professional standards. Training future subtitling professionals requires a comprehensive approach that recognises this multifaceted nature of the profession and the impact of technological developments on established practices. Building on the reflective practice outlined above, encouraging students to explore technical aspects (synchronisation, segmentation, formatting), linguistic aspects (translation, cultural adaptation, register), and technological aspects (AI tool evaluation, workflow integration), while examining the potential integration of automation into workflows should provide them with opportunities to develop their skills, systematically evaluate the performance of LLMs, ASR, MT and other AI systems, and reflect on how these tools affect their professional identity and work satisfaction. As an example of this type of reflective practice, this section focuses on the use of LLMs for the translation of cultural references to show how subtitling processes can be augmented through the critical use of automation tools (See O’Brien 2024 on augmentation in translation). 3.1 Contrasting and assessing translations for subtitles Conversational LLMs and other AI tools can be integrated early on in the training process. A common exercise used in introductory subtitling classes is to share examples of cultural adaptations trainers and students find in the subtitles they encounter. After selecting some examples, students reflect on the aspects relevant to the translation of cultural references in subtitling (See Díaz-Cintas & Remael 2021). These examples can be analysed using LLMs to allow students to assess subtitling decisions and understand the adaptation process, contrasting their developing knowledge of subtitling against the decisions of other subtitlers and the explanations provided by LLMs. The example (1) below shows the transcript of the subtitles from Anatomy of a Fall (dir. Triet, 2023) shown on Amazon Prime in the UK. The original version of the film is multilingual, including dialogues in English, French and German. 173 David Orrego-Carmona (1) Anatomy of a Fall, 00:39:08:00-00:39:25:00 a. French: Trois éléments en particulier : un rapport d’expertise sur des projections de sang de M. Maleski retrouvées au domicile du couple, et dont la direction semble indiquer qu’il a reçu un coup violent à la tête alors qu’il se trouvait sur le balcon du 2ème étage. b. English: Three elements in particular. Analyses of Mr Maleski’s blood spatter found at the couple’s domicile indicate he may have received a blunt force blow to the head when he was on the third-floor balcony. Having access to the source and target versions of the text, students can use LLMs as agents to assess how the systems would respond to the tasks and how subtitling, as a practice, is depicted in the output. Example (2) shows the prompt used and the responses provided by three LLMs. (2) Prompt: You are acting as an agent supporting the task of translating subtitles. This text is part of the script of the film Anatomy of a Fall. Translate it and adapt it to subtitles for the original film in English: Trois éléments en particulier : un rapport d’expertise sur des projections de sang de M. Maleski retrouvées au domicile du couple, et dont la direction semble indiquer qu’il a reçu un coup violent à la tête alors qu’il se trouvait sur le balcon du 2ème étage. a. ChatGPT (4o mini): Subtitle Adaptation: "Three key elements: an expert report on blood spatter belonging to Mr. Maleski found in the couple's home." "The blood pattern suggests he was struck violently on the head while on the second-floor balcony." Let me know if this fits well or if any adjustments are needed! b. Copilot, GPT4: Sure, here’s the translation and adaptation for subtitles: “Three key elements: an expert report on Mr. Maleski’s blood splatter found at the couple’s home, indicating he was 174 9 Teaching subtitling in the times of generative AI struck violently on the head while on the second-floor balcony.” Does this work for your needs? c. Claude 3.5 Sonnet: Let me help you with the subtitle translation and adaptation, focusing on clarity and proper subtitle format. Since this appears to be part of evidence being discussed in the film, I'll format it appropriately for subtitling: Three elements in particular: a forensic report on Mr. Maleski's blood spatter found in the couple's home, where the direction appears to indicate he received a violent blow to the head while on the second-floor balcony. The text is broken down into readable chunks, each line keeping under the standard 42-character limit for subtitles. This maintains readability and follows standard subtitle timing practices. Would you like me to adjust the line breaks or pacing differently? The results of this initial test indicate disparities across the systems. Apart from providing a translation, Claude (2c) provides context both about the scene and the nature of subtitling. The resulting translation also displays the limitation of subtitles (segmentation, character limitation, timing and readability), providing students with an opportunity to reflect on subtitling guidelines and instructions. Having access to the official translation, students can ask the systems to compare the two versions. When asked to contrast the original translation against the new proposed translation, Claude again stands out by displaying contextual awareness (3). For instance, it states that the original version’s decision to translate domicile as ‘domicile’ instead of ‘home’ reflects language that is ‘more formal/legal, fitting for a courtroom scene’. (3) Prompt: The translation of the subtitles in English reads as below. Can you explain the differences from your translation? “Three elements in particular. Analyses of Mr Maleski’s blood spatter found at the couple’s domicile indicate he may have received a blunt force blow to the head when he was on the third-floor balcony.” 175 David Orrego-Carmona QC and documentation offer additional opportunities for the integration of AI tools in self-assessment. Students can use LLMs to identify inconsistencies and evaluate the potential reception of their decisions. They can use the system to extract records of their choices, evaluate consistency, and gauge the rationale behind their translations as decoded by the system. This can prove particularly useful for team projects, where groups need to coordinate the translation of multiple episodes. Working together, students can learn to develop strategies that ensure consistency across episodes, preparing them for real-world scenarios where large teams collaborate in localisation projects. Recent developments in media localisation and AI point to further changes in how subtitles are created and integrated into audiovisual content (Slator 2024). The emergence of LLMs capable of processing multimodal input will likely continue transforming subtitling workflows. In the near future, it is possible these systems will efficiently identify speakers, analyse context based on visual elements and suggest timing based on scene composition. Some of these tools are in development but not yet deployed. Even if these technologies are not currently integrated into media localisation processes at scale, preparing students to understand upcoming multimodal developments is essential. The expansion of accessibility services, such as audio description, and the development of automatic dubbing suggest professionals will still need to analyse interactions between different meaning-making modes and make decisions considering their target audiences. Activities and courses that support multimodal analysis and integrate AI tools while maintaining human agency in decision-making encourage future professionals to think strategically about subtitling workflows and their role in an evolving industry. 4 Tailoring workflows in subtitling training The proposed integration of AI tools into subtitling training moves beyond teaching students how to use specific systems to promote an understanding of the considerations relevant to the products, processes and producers. It fosters a problem-solving mindset that enables students to develop flexible, contextappropriate and industry-informed workflows. By encouraging students to critically engage with different tools and approaches, they can develop the foundational knowledge and adaptability needed in the current industry landscape. Further, the continuous reference to academic knowledge and industry expertise empowers students to move swiftly between sectors. 182 9 Teaching subtitling in the times of generative AI A key component of this approach is allowing students to design their own subtitling projects and take ownership of all project-relevant decisions: from material selection to the tailoring of the translation situation. Adopting a projectbased approach to translator training (Kiraly 2006, Mitchell-Schuitevoerder 2020) and rather than working with pre-defined parameters, students select the audiovisual products to be translated, define the production conditions, select the most suitable guidelines and reflect on the reception requirements that frame their translations. This autonomous decision-making process helps students understand how different factors – audience needs, technical constraints, cultural considerations, and technological tools – influence workflow design. For instance, a student might choose to subtitle a documentary aimed at a specialist audience, requiring careful consideration of terminology management and potentially benefiting from AI tools for quality assurance. Another might work on a comedy clip where cultural references and wordplay demand more human intervention and support for creative solutions. Classes can also integrate spaces to share and discuss the students’ decisions along the way: short progress presentations or online discussion forums to provide progress updates can help students learn from each other. Through these self-directed projects, students learn to: • Evaluate which stages of the workflow can benefit from automation • Assess the suitability of different AI tools for specific tasks • Understand when human expertise is most critical and when automation provides the most advantages • Translate their products, reflecting on specific translation needs • Design QC processes that combine automated and manual checks • Document their decision-making process for future reference The flexibility to explore different workflows bridges academic expertise with industry developments in meaningful ways. While academic approaches emphasise systematic analysis and an in-depth understanding of subtitling principles, industry practices focus on efficiency, market relevance and technological innovation. By allowing students to design their workflows, this independent project-based approach creates opportunities for them to understand how theoretical principles support informed decision-making in practical scenarios. Students learn to apply academic research on reading speeds (Kruger et al. 2022, 183 David Orrego-Carmona Szarkowska & Bogucka 2019, Szarkowska et al. 2024) and segmentation (GerberMorón et al. 2018) to evaluate the output of automated systems, or draw on translation theory to assess MT suggestions. This integration of perspectives helps students develop digital reflexivity (Tipton 2024: 86) to support industry practices. As LSCs continue to deploy new automated solutions that challenge traditional workflows and companies develop more tools, students need to engage critically with the process, resort to solid foundational knowledge and be able to swiftly integrate into new technological production environments. Having a sound theoretical basis allows students to critically assess these developments and adapt their practice accordingly rather than following prescribed procedures. This proposal promotes collaborative learning environments that benefit both students and trainers. When students are empowered to test different tools and assess their performance, they develop the confidence to access new knowledge autonomously. Rather than positioning tutors as the sole source of information, the responsibility of engaging with emerging technologies and industry practices is shared between tutors and students (Kiraly 2003, 2006). This approach recognises that trainers themselves are navigating a rapidly evolving landscape and need to continuously update their skills. The resulting dynamic supports trainers’ continuing professional development as they learn alongside students, fostering an environment of mutual growth that better reflects the constant technological changes in the industry. 5 Ethical considerations The implementation of automated solutions and, in particular, the popularisation of LLMs has made discussions about the ethical dimensions of these practices essential (Moorkens 2020, 2024). According to the Statement on Generative Artificial Intelligence published by AVTE (AudioVisual Translators Europe, the European Federation of National Associations of Audiovisual Translators), the use of AI in creative translation does not respond to creative needs but an intention to increase profit “while compromising the livelihood of professional translators and offering audiences subpar products” (AVTE 2023b: 5). Training activities must be designed and implemented with careful consideration of ethical implications. Acknowledging that engaging with AI tools is now essential for employability in the industry, but it is also important to recognise and address the ethical and social issues of LLMs. Firstly, the use of copyrightprotected materials in LLM training raises concerns about intellectual property rights. Creative industries, such as audiovisual translation, have been partic184 9 Teaching subtitling in the times of generative AI ularly active in denouncing the deployment of LLMs as an exploitative practice whose impact is incremented by the deployment of AI-powered production models (AVTE 2023b, ATRAE 2021, En chair et en Os 2024). Secondly, and more importantly, training should acknowledge how the implementation of AI tools might affect working conditions in the industry (AVTE 2023a, Karakanta et al. 2022, Koponen et al. 2020). Professional associations have pointed out that the push for post-editing workflows involving AST, MT and LLMs comes with a demand for reduced rates and increased workload for translators, changing working conditions and potentially affecting job satisfaction (do Carmo 2020, Sakamoto et al. 2024). Ensuring the employability of graduates demands training that is aware and responsive to the deployment of AI (EMT 2022), but education requires a critical evaluation of these tools and production models to equip students with the knowledge and skills to evaluate the impact of societal, professional and ecological sustainability (Moorkens et al. 2024). These concerns resonate with broader discussions about the ethical implications of AI development, including issues of bias in training data (Aka et al. 2021, Johnson et al. 2022), the environmental impact of LLMs (Luccioni et al. 2023), and the risk of perpetuating cultural hegemonies through automation and standardisation (Bender et al. 2021). Ensuring that students are aware of the capabilities and limitations of the systems is also an ethical task as training programmes have a responsibility to prepare students for the requirements of the market. LSCs are steadily implementing AI solutions into their workflows (Orrego-Carmona 2024, Slator 2024) and graduates need hands-on experience with these tools to successfully integrate into the profession. They need to be able to assess when the systems are suitable or not for a task. This will only be achieved through familiarisation. Further, when the systems are deemed suitable, it should also be future trained professionals who decide how to deploy them to ensure they are augmenting, not hindering, the capabilities of other professionals (O’Brien 2024). Activities should foster critical discussions about the impact of AI on both the profession and society at large, encouraging students to consider how these tools can support rather than replace human expertise and develop the practical competences the market demands. This critical engagement must include the reflective practice that allows students to assess how different technological implementations affect their work satisfaction and align with their professional values. The goal is to prepare students to engage with AI tools ethically and strategically, understanding both their potential and limitations, while developing the agency to advocate for fair working conditions and responsible AI implementation in an increasingly automated industry. 185 David Orrego-Carmona 6 Concluding remarks Training subtitlers in the age of GenAI requires balancing subtitling-specific skills with awareness about the market and understanding media localisation practices. This type of training encompasses, firstly, ensuring graduates develop the skills they need to enter the industry, which requires a thorough understanding of subtitling principles, workflows and tools. Secondly, the increasing implementation of automated solutions to educate reflective professionals who can adapt to rapidly changing production settings. And lastly, training that fosters ethical awareness to support responsible engagement with technology and advocates for sustainable professional practices. Future-proofing subtitling education requires integrating emerging technologies while maintaining core professional competences. The problem-solving approach presented in this chapter responds to these needs by creating spaces for students to explore technological possibilities and develop foundational knowledge. Engaging with LLMs and other GenAI tools allows students to understand the potential and limitations of automation. By comparing different solutions and evaluating their suitability for specific tasks, students learn to make informed decisions about implementing automated solutions. The ability to identify specific limitations through comparative analysis of diverging outputs is a transferable skill that will benefit students in virtually any future career involving AI systems. Further, this approach encourages them to think strategically about their role in an evolving industry. The activities discussed demonstrate how subtitling training can integrate technological developments maintaining a focus on quality and professional standards. These activities show students how to harness GenAI tools to support their work rather than allowing automation to dictate their practice. The goal is to prepare adaptable professionals who can navigate technological changes, maintain high standards and advocate for sustainable working conditions. The continuous development of GenAI tools suggests the media localisation industry will keep evolving and graduates need to be ready to assess, critique and deploy GenAI-powered solutions. This requires careful consideration of how automation affects subtitling processes, fostering critical engagement with emerging technologies and promoting ethical awareness. Only by addressing these aspects can training programmes prepare professionals who will contribute to developing a sustainable media localisation industry. 186 9 Teaching subtitling in the times of generative AI References Aka, Osman, Ken Burke, Alex Bauerle, Christina Greer & Margaret Mitchell. 2021. Measuring model biases in the absence of ground truth. In Proceedings of the 2021 AAAI/ACM Conference on AI, Ethics, and Society (AIES ’21), 327–335. Virtual Event, USA: Association for Computing Machinery. DOI: 10.1145/3461702. 3462557. ATRAE. 2021. Comunicado sobre la posedición. Asociación de Traducción y Adaptación Audiovisual. https://atrae.org/comunicado-sobre-la-posedicion/. AVTE. 2023a. 2023 AVTE survey on audiovisual translators’ working conditions. AudioVisual Translators Europe. https:/ / prezi . com/ i /a83j _ udsgdtj / avte - survey-2022-23/. AVTE. 2023b. AVTE statement on generative artificial intelligence. AudioVisual Translators Europe. https://avteurope.org/avte-statementongenerativeartificial-intelligence. Bender, Emily M., Timnit Gebru, Angelina McMillan-Major & Shmargaret Shmitchell. 2021. On the dangers of stochastic parrots: Can language models be too big? In Proceedings of the 2021 ACM Conference on Fairness, Accountability, and Transparency (FAccT ’21), 610–623. Virtual Event, Canada: Association for Computing Machinery. DOI: 10.1145/3442188.3445922. Bolaños García-Escribano, Alejandro. 2025. Practices, education and technology in audiovisual translation. London, UK: Routledge. Díaz-Cintas, Jorge & Aline Remael. 2021. Subtitling: Concepts and practices. London, UK: Routledge. do Carmo, Félix. 2020. “Time is money” and the value of translation. Translation Spaces 9(1). 35–57. DOI: 10.1075/ts.00020.car. 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. En chair et en Os. 2024. Littérature, cinéma, presse, jeux vidéo: non à des traductions sans âme. https://enchairetenos.org/. Estopace, Eden. 2017. Talent crunch hits media localization as Amazon, Netflix accelerate market growth. https : / / slator . com / talent - crunch - hits - media - localization-as-amazon-netflix-accelerate-market-growth/. Georgakopoulou, Panayota. 2019. Template files: The holy grail of subtitling. Journal of Audiovisual Translation 2(2). 137–160. Georgakopoulou, Panayota. 2020. The faces of audiovisual translation. In The many faces of translation: From video games to the Vatican (DG TRAD conference 187 David Orrego-Carmona 2019), 16–33. European Union. DOI: doi/10.2861/898773. https://op.europa.eu/ s/y4PS. Gerber-Morón, Olivia, Agnieszka Szarkowska & Bencie Woll. 2018. The impact of text segmentation on subtitle reading. Journal of Eye Movement Research 11(4). 1–18. DOI: 10.16910/jemr.11.4.2. Iyuno SDI Group. 2022. A simplified look at the subtitling production process. https: / / iyuno . com / news / white - papers / a - simplified - look - at - the - subtitling - production-process. Johnson, Rebecca L, Giada Pistilli, Natalia Menédez-González, Leslye Denisse Dias Duran, Enrico Panai, Julija Kalpokiene & Donald Jay Bertulfo. 2022. The ghost in the machine has an American accent: Value conflict in GPT-3. arXiv. DOI: 10.48550/ARXIV.2203.07785. Karakanta, Alina, Luisa Bentivogli, Mauro Cettolo, Matteo Negri & Marco Turchi. 2022. Post-editing in automatic subtitling: A subtitlers’ perspective. In Proceedings of the 23rd Annual Conference of the European Association for Machine Translation, 261–270. Ghent, Belgium: European Association for Machine Translation. https://aclanthology.org/2022.eamt-1.29. Kiraly, Don. 2003. From instruction to collaborative construction: A passing fad or the promise of a paradigm shift in translator education? In Brian James Baer & Geoffrey S. Koby (eds.), Beyond the ivory tower. rethinking translation pedagogy, 3–27. Amsterdam, Netherlands: John Benjamins. DOI: 10.1075/ata. xii.03kir. Kiraly, Don. 2006. Project-based learning: A case for situated translation. Meta 50(4). 1098–1111. DOI: 10.7202/012063ar. Koponen, Maarit, Umut Sulubacak, Kaisa Vitikainen & Jörg Tiedemann. 2020. MT for subtitling: User evaluation of post-editing productivity. In Proceedings of the 22nd Annual Conference of the European Association for Machine Translation, 115–124. Lisboa, Portugal: European Association for Machine Translation. https://aclanthology.org/2020.eamt-1.13. Kruger, Jan-Louis, Natalia Wisniewska & Sixin Liao. 2022. Why subtitle speed matters: Evidence from word skipping and rereading. Applied Psycholinguistics 43(1). 211–236. DOI: 10.1017/S0142716421000503. Lommel, Arle. 2020. Augmented translation: are we there yet? https://csa-research. com/Blogs-Events/Blog/augmented-translation-2020. Luccioni, Sasha, Christopher Akiki, Margaret Mitchell & Yacine Jernite. 2023. Stable bias: Evaluating societal representations in diffusion models. In A. Oh, T. Naumann, A. Globerson, K. Saenko, M. Hardt & S. Levine (eds.), Advances in neural information processing systems, vol. 36, 56338–56351. Curran Asso188 9 Teaching subtitling in the times of generative AI ciates. https : / / proceedings . neurips . cc / paper _ files / paper / 2023 / file / b01153e7112b347d8ed54f317840d8af-Paper-Datasets_and_Benchmarks.pdf. Marking, Marion. 2022. Iyuno-SDI and Disney on overcoming the challenges of localizing at global scale. https://slator.com/iyuno-sdi-disney-on-overcomingthe-challenges-of-localizing-at-global-scale/. Mitchell-Schuitevoerder, Rosemary. 2020. A project-based approach to translation technology. London, UK: Routledge. Moorkens, Joss. 2020. “A tiny cog in a large machine”: Digital Taylorism in the translation industry. Translation Spaces 9(1). 12–34. DOI: 10.1075/ts.00019.moo. Moorkens, Joss. 2024. ‘I am not a number’: On quantification and algorithmic norms in translation. Perspectives 32(3). 477–492. DOI: 10.1080/0907676X.2023. 2278536. Moorkens, Joss, Sheila Castilho, Federico Gaspari, Antonio Toral & Maja Popović. 2024. Proposal for a triple bottom line for translation automation and sustainability: An editorial position paper. Journal of Specialised Translation (41). 2– 25. DOI: 10.26034/cm.jostrans.2024.4706. Netflix. n.d. Timed text style guide: subtitle templates. https : / / partnerhelp . netflixstudios.com/hc/en-us/articles/219375728-Timed-Text-Style-GuideSubtitle-Templates. O’Brien, Sharon. 2024. Human-centered augmented translation: Against antagonistic dualisms. Perspectives 32(3). 391–406. DOI: 10.1080 /0907676X.2023. 2247423. Orrego-Carmona, David. 2024. Placing human agency in the AI-powered media localisation industry. Journal of Audiovisual Translation 7(2). 1–21. Oziemblewska, Magdalena & Agnieszka Szarkowska. 2022. The quality of templates in subtitling. A survey on current market practices and changing subtitler competences. Perspectives 30(3). 432–453. DOI: 10.1080/0907676X.2020. 1791919. Papercup. 2022. Translator talent crunch leads to investment in machine learning. https://www.papercup.com/blog/translatortalentcrunch-leadstoinvestment-in-machine-learning. Sakamoto, Akiko, Darren Van Laar, Joss Moorkens & Félix Do Carmo. 2024. Measuring translators’ quality of working life and their career motivation: Conceptual and methodological aspects. Translation Spaces 13(1). 54–77. DOI: 10.1075/ ts.23026.sak. Slator. 2024. Slator pro guide: Audiovisual translation. https://slator.com/proguide-audiovisual-translation/. 189 David Orrego-Carmona Szarkowska, Agnieszka & Lidia Bogucka. 2019. Six-second rule revisited: An eyetracking study on the impact of speech rate and language proficiency on subtitle reading. Translation, Cognition & Behavior 2(1). 101–124. DOI: 10.1075/tcb. 00022.sza. Szarkowska, Agnieszka & Anna Jankowska. 2024. Introducing audiovisual translation. 1st edn. London, UK: Routledge. Szarkowska, Agnieszka, Valentina Ragni, David Orrego-Carmona, Sharon Black, Sonia Szkriba, Jan-Louis Kruger, Krzysztof Krejtz & Breno Silva. 2024. Impact of video and subtitle speed on subtitle reading: An eye-tracking replication study. Journal of Audiovisual Translation 7(1). DOI: 10.47476/jat.v7i1.2024.283. Tipton, Rebecca. 2024. The Routledge guide to teaching ethics in translation and interpreting education. London, UK: Routledge. Valdez, Susana, Hanna Pięta, Ester Torres-Simón & Rita Menezes. 2023. Subtitlers’ beliefs about pivot templates. Target 35(3). 426–454. DOI: 10.1075/target. 00019.val. 190