2. Context and framing

Tsui opened the session by situating the discussion within broader global developments. The ubiquity of AI-mediated communication, she noted, calls for a fundamental re-examination of the ideological, pedagogical, and epistemological bases of EMI. English’s dominance as a global academic language rests on historical, economic, and symbolic capital. GenAI technologies, while offering unprecedented linguistic accessibility, may simultaneously erode the symbolic value by equalizing linguistic competence through automation. The panel’s purpose was to ask how EMI might evolve within this new ecology. Will English continue to serve as a vehicle of disciplinary knowledge and intercultural exchange when AI can mediate virtually every act of writing and comprehension? How can educators ensure that AI augments, rather than replaces, the intellectual work of thinking and learning through language? The conversation was structured around the following four guiding questions, each addressed by a panelist and followed by audience discussion.

3. Will English continue to be the lingua franca in academia, or will EMI be rendered superfluous by GenAI? (Amy Tsui)

Tsui’s contribution examined the changing ecology of English in higher education using data from multiple sources, including global language monitoring reports (e.g., British Council & Studyportals; EF English Proficiency Index), market intelligence on international student recruitment (ICEF Monitor), cross-national statistical databases (Statista), and media reports from Asia and Europe.

Recently, an article from The Economist (2024) titled “Why China is losing interest in English” analyzed the waning of “English fever” (yingyu re) that had been rampant in China and across Asia for several decades (see also Tsui, Reference Tsui and Tsui2021). The article argued that English learning in China is no longer a priority for business professionals or students, because of strained relations with major English-speaking countries and rapid advances in AI-based translation tools. Supporting this observation, Tsui cited data from the Chinese Ministry of Education on approved undergraduate programs (2018–2022), showing that 109 universities discontinued 28 foreign-language-related majors and 21 universities discontinued English majors. This trend intensified between 2023 and 2025, with additional universities suspending foreign-language programs, including English. For example, in late 2023, the prestigious Chinese University of Science and Technology cancelled its English major, after which more than 100 universities implemented similar changes (Zou, Reference Zou2025). The report further noted that graduates in English language and translation experienced lower employment rates, salaries, and job satisfaction than graduates from other disciplines.

Tsui observed that similar trends are evident in Japan and South Korea, where English fever was also pronounced a decade ago. In South Korea, media headlines such as “Korean students’ motivation to study English drops as AI tools rise” (Moon, Reference Moon2025) reflect growing concern. A nationwide survey conducted in 2024 by the South Korean Ministry of Education involving middle and high school students (n = 27,606) found that only 61.3% described themselves as “highly motivated” to study English.

Declining motivation in China, Japan, and South Korea is further reflected in Education First’s English Proficiency Index (2025). Between 2019 and 2025, China’s proficiency level fell from “medium” to “low,” while Japan’s declined from “low” to “very low.” Their global rankings dropped sharply, from 40 to 86 for China and from 53 to 96 for Japan. South Korea’s decline was less dramatic, falling from 37 to 48.Footnote ¹

Against this shifting landscape, Tsui examined whether English will continue to function as the global academic lingua franca, or whether it might eventually be displaced by Chinese – a claim periodically advanced over the past two decades. A decade ago, the view of EMI as an “unstoppable train that has already left the station” (Macaro, Reference Macaro2019, p. 232) was supported by the rapid growth of English-taught programs (ETPs), which numbered over 157,000 at bachelor’s and master’s levels in 2021, with further growth projected (Neghina & Agnew, Reference Neghina and Agnew2021). Would these predictions be borne out in the current context?

To address this question, Tsui presented two sets of data. First, statistics on the world’s most spoken languages showed that Chinese and English have comparable numbers of speakers: approximately 1.18 billion Chinese speakers and 1.53 billion English speakers worldwide. However, data on language use in digital domains revealed a stark contrast: 49.4% of global web content is in English, compared with only 1.2% in Chinese (Galan, Reference Galan2025). Second, compared with 2021, the global supply of ETPs increased by 22%, from 157,278 to 183,262 at all levels, with particularly strong projected growth at the doctoral level (British Council & Studyportal Analytics Consulting Team, 2024).

Tsui also highlighted changing student mobility patterns, noting a shift from predominantly international mobility to increasingly intra-regional mobility. The number of ETPs grew by 48% between 2019 and 2024. Within Asia, Japan, South Korea, and Taiwan have expanded EMI provision and set new targets for attracting international students. In Europe, Italy, Spain, Germany, France, Portugal, and Switzerland have emerged as highly searched study destinations. According to ICEF Monitor, global student mobility has expanded beyond the “Big Four” (UK, US, Canada, and Australia) to the “Big Ten” and even the “Big Fourteen” (ICEF Podcast, 2025). Tsui argued that as student mobility continues to grow, communication contexts will become increasingly multilingual and multicultural, reinforcing the need for a shared lingua franca.

Tsui concluded that the evidence strongly suggests English will remain the lingua franca of global communication and academia. Far from rendering English obsolete, GenAI and digital technologies – given their reliance on English as a primary mediational language – are likely to accelerate rather than diminish the global spread of English.

4. How far are GenAI tools adequate in helping students cope with the challenge of expressing disciplinary reasoning in a foreign language? (Maria Kuteeva)

Kuteeva’s contribution to the panel was based on her ongoing research conducted in the context of undergraduate ETPs in Sweden, e.g., in the fields of Business, Politics, and Teacher Education. Higher education in Swedish universities requires a great deal of independent study and critical engagement by an increasingly heterogeneous, multilingual student population (Kuteeva, Reference Kuteeva2023). In this context, GenAI tools have affordances to help students in EMI contexts, for example:

1. (1) A sounding board for disciplinary reasoning and argumentation

Recent interviews with students taking ETPs indicate that they use GenAI tools to unpack complex ideas, to provide further examples, or to generate exam questions with different levels of difficulty. They also use ChatGPT as a sounding board for brainstorming and refining arguments and for clarifying whatever is unclear in the instructional materials. ChatGPT is also useful for creating clear instructions, e.g., how to run a brainstorming session and discuss ideas with peers, which can be helpful for group projects.

1. (2) Experimenting with multimodal ways of engaging with academic content

Students may use various AI tools to listen to texts, as some find it easier to understand when they hear academic content. There are also tools, e.g., NotebookLM, that can convert a traditional genre, e.g., a research article (a common type of compulsory course reading), to a more accessible one, e.g., a podcast. When Kuteeva interviewed a group of Ukrainian refugee students taking ETPs in Sweden, they found it very helpful to engage with academic content in this dialogic format. In this sense, both EMI and GenAI have the potential to support access and social integration for various minoritized groups.

At the same time, engagement with GenAI output poses a number of challenges:

1. (1) Entanglement between disciplinary and digital literacy

A meaningful engagement with GenAI depends on students’ disciplinary and digital literacies. Some students are tech-savvy and use GenAI critically to clarify complex ideas, brainstorm, and generate examples, while others may rely too heavily on low-quality output generated through a zero-shot prompt (e.g., Kuteeva & Andersson, Reference Kuteeva and Andersson2025). Despite high levels of digitalization in compulsory education, there is a mismatch between assumptions about “digital natives” and students’ actual digital literacy. Familiarity with smartphones and apps does not automatically translate into the ability to work in complex digital environments. Learning of disciplinary literacy in ETPs should go hand in hand with the development of critical GenAI literacy.

1. (2) De facto replacement of traditional learning materials, e.g., textbooks and other assigned literature

Many students rely heavily on lecture slides and grow used to acquiring knowledge in bullet points. Some students report asking ChatGPT to elaborate on slides to explain “what the professor means,” even though they usually have assigned readings in the form of a textbook and/or research articles. Thus, Large Language Models (LLMs) are quietly replacing textbooks as the main introductory source of disciplinary knowledge, although it is hard to check the quality of output and sources generated in response to students’ queries. This practice signals a shift in how disciplinary knowledge, including disciplinary reasoning and argumentation, is being accessed and mediated.

1. (3) Limitations with argumentation and context

“I spent several hours last night arguing with ChatGPT,” one student confessed. This simple statement crystallizes both the affordances and limitations of LLMs with regard to discipline-specific argumentative skills. Although an LLM can act as a sounding board to generate and refine ideas, it often lacks human insight into specific contexts and individual positions, mainly due to its design as a statistical predictor (Bender et al., Reference Bender, Gebru, McMillan-Major and Shmitchell2021). Guided by hidden prompts, ChatGPT tends to avoid controversial subjects and often presents epistemically weak arguments, although the language is idiomatic and structurally correct. On the positive side, the student in question was getting frustrated with the bot’s replies, which helped her develop her own arguments.

To conclude, GenAI can support comprehension, experimentation, and idea development in EMI settings, making disciplinary learning more accessible, e.g., for minoritized groups such as war refugees. At the same time, while LLMs are useful for language-related support, they remain limited in terms of epistemic depth and critical insight, which are essential for disciplinary literacy.

5. What challenges do students face when using GenAI tools to support learning through EMI? How can they move beyond superficial reliance and develop disciplinary engagement and metalinguistic awareness? (Jack Pun)

Drawing on a series of empirical studies, Pun examined how GenAI is transforming the dynamics of EMI classrooms. Across these contexts, GenAI functions as a third actor in the traditional teacher–student dyad, mediating access to disciplinary content while simultaneously complicating the process of language and knowledge co-construction (Bannister et al., Reference Bannister, Curle and Pun2026). Pun identified three recurring challenges emerging from his classroom-based and multimodal research:

1. (1) Overreliance and superficial comprehension – Students often use AI-generated summaries or explanations without interrogating their epistemic validity. Fluent phrasing and grammatical accuracy can mask conceptual misunderstandings, leading to the illusion of mastery. This “surface fluency effect” risks eroding students’ disciplinary language awareness and their capacity for independent reasoning.

2. (2) Limited epistemic and metalinguistic awareness – From multimodal analyses of Hong Kong secondary students’ drawings and written reflections, Pun found that many learners perceived ChatGPT as a “scientist” that conducts experiments, rather than a probabilistic text generator. This misconception illustrates a lack of epistemic literacy – the ability to evaluate how AI constructs knowledge claims, distinguish scientific evidence from linguistic plausibility, and read AI-generated texts critically.

3. (3) Contextual misalignment and ethical uncertainty – AI tools often fail to capture discipline-specific conventions such as hedging, multimodal reasoning, and evaluative stance. Students struggle to adapt AI outputs to local academic norms and remain uncertain about ethical disclosure and citation. Institutions, too, face ambiguity in defining acceptable AI use within EMI assessment.

Despite these challenges, Pun highlighted how GenAI can serve as a pedagogical partner when embedded within explicit-reflective EMI pedagogy (Moorhouse et al., Reference Moorhouse, Pun and Curle2026). His “AI-mediated peer discourse” approach covers classroom strategies where students critically engage with AI outputs. In one example, science students compared AI-generated hypotheses with authentic abstracts, identifying differences in modality, evidence use, and authorial stance. Through guided reflection and co-editing, students developed greater sensitivity to disciplinary reasoning and genre conventions (Cheung et al., Reference Cheung, Pun, Kenneth-Li and Mai2025).

Pun further demonstrated how teacher agency determines the meaningfulness of AI integration. In a multi-case study of Chinese EMI teachers (Jia & Pun, Reference Jia and Pun2025), those who exercised ecological agency – balancing professional experience, contextual judgment, and pedagogical adaptation – used ChatGPT to simplify complex texts, generate contextualized examples, and design collaborative activities without relinquishing pedagogical control. Teachers who lacked such agency tended to disengage after encountering unreliable or decontextualized AI outputs. These findings underscore that AI does not replace pedagogy; rather, it amplifies the need for critical mediation and teacher expertise.

Finally, Pun called for cultivating AI and EMI literacies jointly. Effective EMI pedagogy in the AI era must develop students’ metalinguistic agency: the ability to question, adapt, and transform AI-produced language, as well as their epistemic literacy: the capacity to evaluate the sources, justification, and reliability of AI-generated knowledge. He argued that teachers and institutions should embed these dimensions into curricula through explicit-reflective instruction, multimodal reading interventions, and collaborative assessment design (Pun et al., Reference Pun, Cheung and Kenneth-Li2025). Thus, Pun emphasized that GenAI offers support, not replacement for EMI pedagogy. The central task for educators is to help students move from dependence to disciplinary authorship – using AI not as a linguistic crutch, but as a catalyst for deeper engagement with knowledge, language, and inquiry.

6. How might teachers help students move beyond uncritical use of GenAI to become autonomous learners of disciplinary literacy in English? (John Airey)

In his presentation, Airey argued that the rapid uptake of GenAI tools such as ChatGPT has merely intensified long-standing debates about literacy and learning in higher education. In EMI contexts in particular, the fact that GenAI now offers students immediate access to fluent academic English raises concerns that language production will become decoupled from disciplinary understanding. This disconnect has been termed “discourse imitation” and can be seen in situations where students produce on-the-face-of-it disciplinary correct accounts, but do not understand the thinking that lies behind the language they use (Airey & Linder, Reference Airey and Linder2009). Recognizing this problem, there have been calls by some to limit the use of GenAI (de Fine Licht, Reference de Fine Licht2024). For example, a study by Tayan et al. (Reference Tayan, Hassan, Khankan and Askool2024) found that a sizable minority of prestige universities have implemented explicit bans or prohibitions on the use of GenAI tools in coursework or assessment contexts, with unauthorized use treated as academic misconduct unless specifically permitted by instructors.

However, like Pun before him, Airey believes the way forward lies not in prohibiting GenAI use but rather in understanding how teachers can help students move beyond an uncritical reliance on such tools toward the development of disciplinary literacy.

Airey began by problematizing the term “disciplinary literacy.” Taking a broader approach than that of Shanahan and Shanahan (Reference Shanahan and Shanahan2012), he defines this as “The ability to appropriately participate in the communicative practices of the discipline” (Airey, Reference Airey2011, p. 3). Airey (Reference Airey2020) visualizes this disciplinary literacy in the form of a triangle. Here, the three apexes represent the three settings where he argues disciplinary literacy can be used: the academy, the workplace, and society. Each discipline can be positioned somewhere within the triangle depending on the emphasis placed on disciplinary literacy for use in these three settings. The separate triangles problematize the role of L1, L2, and other modes of representation across these domains. The point made was that in EMI classrooms, academic assessment often privileges L2 English output, even though disciplinary reasoning in the workplace or society may involve both multilingual interaction and multimodal practices such as visual representation or mathematical problem-solving. Airey’s model thus challenges the assumption that disciplinary literacy is simply equivalent to polished English writing, reminding us that disciplinary meaning-making frequently occurs across languages and modalities and depends on the context and purpose at hand. From this perspective, one of the major challenges for EMI is the development of disciplinary literacy in L1 when teaching is in English.

Practically, Airey argued that GenAI in higher education requires four shifts in teacher practice:

• • From correction to knowledge construction

• • From product to process

• • From written to oral assessment

• • From outlawed to classroom integration.

First, since GenAI can eliminate surface-level language errors, instructional focus needs to move away from the production of a final product toward evaluating reasoning and disciplinary knowledge claims. Second, pedagogy needs to shift from product to process, emphasizing drafting, reflection, redrafting, and documenting the process of GenAI use in order to make disciplinary thinking visible (Wingate, Reference Wingate2015). Third, teachers should shift from written to oral and other forms of assessment. This is not simply a response to the inability to evaluate student understanding from AI-written submissions, but also reflects the communicative demands of workplaces and societal participation, where disciplinary knowledge is often negotiated orally, multilingually, and multimodally. Finally, building on the three apexes of his disciplinary literacy triangle, Airey suggested that lecturers must move from outlawing GenAI to integrating it into their classrooms. Here, we need to demonstrate for students how they can use GenAI both critically and ethically in ways that support disciplinary learning across academic, professional, and societal contexts.

7. Concluding remarks

The ETRA 2025 panel discussion highlighted how GenAI is reshaping the linguistic, pedagogical, and epistemic landscape of higher education. Across the four contributions, a common theme emerged: while GenAI is transforming how language and knowledge are accessed, produced, and mediated, it does not render EMI obsolete. Rather, it reconfigures the conditions under which disciplinary knowledge is learned and communicated through English. As the panel concluded, “The question is not whether AI will replace English, but whether educators can guide students to use both AI and English as instruments for knowledge construction.” The main educational challenge, then, is critical engagement with AI by both students and teachers in EMI contexts.

The panel highlighted both the affordances and limitations of GenAI for disciplinary learning in EMI contexts. On the one hand, AI tools can support comprehension, idea generation, and multimodal engagement with academic content, potentially widening access for diverse and minoritized student populations. On the other hand, these technologies risk fostering superficial engagement with disciplinary knowledge when students rely uncritically on fluent but epistemically weak AI-generated texts. As several of the panelists emphasized, the apparent linguistic competence of AI outputs can obscure conceptual gaps, creating an illusion of mastery that challenges traditional approaches to language and content assessment.

For teachers, EMI pedagogy must shift its focus from evaluating linguistic products to making the processes of knowledge construction visible – through reflective engagement with AI outputs, collaborative inquiry, and diverse forms of assessment. GenAI does not replace pedagogy; rather, it amplifies the need for pedagogical expertise and critical guidance. Effective EMI pedagogy in the AI era, therefore, requires explicit-reflective approaches that cultivate both epistemic literacy – students’ ability to evaluate knowledge claims – and metalinguistic agency – the capacity to question and reshape AI-produced language.

The role of English as a key medium of scholarly communication will increasingly depend on how educators harness AI to cultivate reflective, multilingual, and ethically aware learners. In this light, EMI’s next frontier lies not in resisting technological change but in leading it – anchoring language education within a broader vision of human–AI co-intelligence that values critical inquiry, intercultural understanding, and epistemic humility.