From monolingual virtual exchange to AI-supported multilingualism: a critical perspective

1 Introduction

Virtual Exchange (VE) has become a central strategy for internationalization in higher education. It offers students opportunities for intercultural collaboration without the financial, environmental, and logistical costs of physical mobility (O'Dowd, 2023). Most VE initiatives aim to develop 21st-century skills such as digital literacy, problem-solving, and cross-cultural collaboration (O'Dowd and Beaven, 2019). Students from diverse linguistic and cultural contexts engage in online group work, discussions, and reflection. Since such learning projects do not require students to travel, VE is frequently presented as an inclusive and accessible pedagogical approach.

While universities frequently promote VE as “inherently inclusive” (Beelen and Jones, 2015), research suggests that participation is often shaped by unequal access to linguistic resources (Hauck et al., 2025; Lin et al., 2026). Drawing on the logic of neoliberal internationalization (Marginson, 2022), many VE initiatives prioritize individual competence development and employability outcomes, while overlooking how unequal access to linguistic resources constrains meaningful engagement. In such contexts, proficiency in English functions as symbolic capital (Bourdieu, 1991), positioning some participants as more legitimate interlocutors than others. Differences in linguistic proficiency can influence students' emotions, sense of confidence, and participation (Háhn, 2021). As a result, VE risks reproducing the very exclusions it seeks to overcome.

Advances in generative artificial intelligence (AI) and translation technologies are increasingly seen as offering new possibilities for addressing such challenges. Tools such as real-time translation, automated captioning, and AI-assisted text generation may enable participants to communicate across languages more flexibly. In this paper, I use the term AI-supported multilingualism to refer to the use of these technologies within VE to support interaction among participants with different language backgrounds.

However, the integration of AI into VE also raises important questions. While AI tools can facilitate communication, they also mediate how meaning is negotiated. A tension could emerge between the affordances of AI and the interactional processes. I refer to this tension as the paradox of communicative friction: the possibility that technologies designed to streamline communication may also reduce the need for more in-depth interactional and language work.

To address this tension, the paper brings together insights from Critical Virtual Exchange (Hauck, 2023), relational approaches to learning (Helm, 2024), and translanguaging theory (García and Wei, 2014), alongside the concept of linguistic justice. It also draws on ethical perspectives on encountering the Other in digitally mediated environments. Rather than positioning AI as either a solution or a problem, the paper adopts a critical but constructive perspective, considering how AI-mediated communication may both expand and reshape multilingual interaction in VE.

2 Toward critical virtual exchange and relational learning

Critical Virtual Exchange (CVE) offers an alternative orientation by positioning VE as a socially accountable practice grounded in decolonial awareness and social justice (Hauck, 2023). CVE encourages students to critically examine how technology, language, and power intersect to shape participation and inclusion. (Hauck 2023) defines CVE as a socially accountable practice that goes beyond skill acquisition. Within this framing, VE shifts from competence accumulation to reciprocal engagement and ethical responsibility. Such an approach is in contrast with the neoliberal and instrumentalist tendencies of mainstream VE, which often focuses on task completion, badges and employability outcomes.

As a teacher, I have observed that students frequently experience such VE designs as overly task-oriented, limiting meaningful dialogue. Participants often divide labor rather than engaging in sustained relational exchange. This suggests the need for pedagogical designs in which the pragmatic outcomes (or “outputs”) are not central to the online encounters and VE projects are seen as more than opportunities for personal development and skill practice. VE design should foreground the value of encountering and engaging with the Other so that students do not aim only for self-development in these projects.

In contrast, relational learning emphasizes that the value of exchange lies not in extracting skills from the Other, but in engaging with participants' histories, positionalities, and situated knowledges (Helm, 2024; Icaza and Vázquez, 2018). This approach prioritizes reciprocity and mutuality over employability, challenging the instrumental focus of traditional VE designs, where interaction can become superficial as participants avoid confrontation to maintain a false sense of harmony (O'Dowd, 2019).

Table 1 below contrasts instrumental and relational approaches to VE.

Table 1. Instrumental and relational orientations in VE.

By adopting a relational approach, VE can shift from a tool of internationalization to a transformative pedagogical practice that places multilingualism and critical reflexivity at its core (Helm, 2024).

3 Language, participation, and the “linguistically deficient self”

Many collaborative online international learning initiatives operate within monolingual frameworks, most often using English as the lingua franca (Hauck, 2023; Kokkonen and Natri, 2025). It is important to acknowledge that VE is not uniformly English-dominant. Researchers have increasingly explored multilingual approaches that encourage participants to draw on diverse linguistic resources and challenge monolingual assumptions. Nevertheless, English continues to function as a principal medium of communication in many VE contexts (Hauck, 2023; Hauck et al., 2025; Kokkonen and Natri, 2025; Lin et al., 2026), making questions of linguistic justice, participation, and communicative authority particularly salient.

While a shared language can facilitate communication, monolingual norms often position other languages (and their speakers) as peripheral. In my own VE research (Háhn, 2021), students with weaker English skills consistently reported feelings of anxiety, hesitation, and reduced agency during collaboration. They often internalized deficit narratives. Many also described feeling “othered” by peers who were more fluent. Moreover, linguistic insecurity frequently led to self-othering, where participants positioned themselves as inferior in relation to imagined fluent peers. Interestingly, more fluent participants sometimes adopted paternalistic roles, expressing concern for “weaker” speakers while implicitly reinforcing hierarchies of competence. Even when interactions were intended to be equitable, monolingual assumptions shaped participation.

This “linguistically deficient self” reflects broader processes of othering (Kristeva, 1991; Riggins, 1997; Bhabha, 1994), where identity is constructed relationally through perceived linguistic hierarchy. According to (Piller 2017), linguistic proficiency acts as symbolic capital in intercultural communication. Since fluency in a language shapes participation and authority (Piller, 2017), VE should critically examine monolingual assumptions and seek ways to reduce the impact of linguistic hierarchies on participation.

Importantly, linguistic inequality in VE should not be understood solely as a difference in language proficiency. Rather, it is shaped through the ways participants perceive, evaluate, and position themselves and others in relation to language. As such, linguistic hierarchies are relationally and discursively constructed, influencing participation, confidence, and a sense of belonging.

4 AI tools as pedagogical scaffolds for linguistic equity

Moving beyond predominantly monolingual approaches to VE may help address some of the challenges discussed above. The growing availability of AI-supported multilingual communication may open new possibilities for addressing linguistic asymmetries in VE.

AI tools such as translation, captioning, and generative support may reduce linguistic barriers and redistribute communicative responsibility. Empirical evidence suggests that VE participants already use multimodal and multilingual strategies, such as images, audio, and code-switching, to overcome linguistic barriers (Pacheco et al., 2021; Darling and Dervin, 2023; Yeh et al., 2024; Tzirides, 2026). AI can strengthen these practices by adding further scaffolding for interaction (see Table 2). Most of the affordances outlined in Table 2 remain pedagogical possibilities rather than established practices in the VE literature. Although emerging studies have begun to explore AI-supported multilingual and multimodal interaction (e.g., Tzirides, 2026), further empirical research is needed to evaluate how these affordances shape participation.

Table 2. Potential AI affordances for multilingual participation in virtual exchange.

AI may reduce communicative pressure in lingua franca environments by shifting emphasis from linguistic accuracy to meaning-making. For example, real-time translation can support participation in live discussions, while AI-assisted paraphrasing can help students refine ideas prior to interaction. The use of AI can encourage students to use different languages, experiment with translations, engage in AI-supported interactions, switch between languages, and create multilingual content.

Recent work provides an emerging example of this possibility: (Tzirides 2026) examined a GenAI-supported virtual exchange in which students used AI image generation and translation tools to co-construct visual representations of cultural themes, illustrating how AI can become part of multilingual and multimodal meaning-making. In this study, GenAI was conceptualized as a mediational tool that can support translanguaging (García and Wei, 2014), enabling students to draw on multiple linguistic and semiotic resources for meaning-making.

However, AI does not remove linguistic inequality. Because AI systems are trained on uneven datasets, they often reproduce dominant language hierarchies, particularly privileging English. This raises concerns about technological solutionism and epistemic homogenization. Technological solutionism is the assumption that communicative inequalities can be resolved primarily through technical mediation. A further risk concerns epistemic and discursive homogenization. AI-generated outputs often normalize standardized academic discourse and may reduce the diversity of rhetorical and linguistic expression. Without critical pedagogical framing, AI may standardize communication. Ethical concerns such as data privacy, surveillance, and algorithmic bias further complicate AI integration in VE.

Recent discussions also suggest that AI may reshape VE itself by reducing the need for negotiated meaning-making (Hackett and Dawson, 2026). This introduces a tension between accessibility and the pedagogical value of communicative friction. (Hackett and Dawson 2026) argue that AI-mediated communication may reduce the need for linguistic negotiation and interactional repair work that traditionally underpin intercultural learning. From this perspective, AI does not simply enhance VE; it potentially reconfigures its epistemic and pedagogical foundations.

This raises a central paradox: VE depends, at least in part, on communicative friction, including moments of misunderstanding, repair, and negotiation across linguistic and cultural difference. Yet, AI systems are explicitly designed to minimize such friction through translation, prediction, and linguistic smoothing. While this can increase accessibility and participation, it may also diminish the very interactional conditions through which relational and intercultural learning emerges.

Valuing communicative friction does not mean preserving linguistic barriers or treating difficulty as inherently desirable. AI should reduce exclusionary barriers while preserving opportunities for meaningful negotiation, reflection, and engagement with difference. The question is therefore not whether communication should be difficult, but which barriers should be removed and which forms of interactional work remain educationally valuable.

Consequently, the pedagogical task is to design VE environments where AI functions as a scaffold for multilingual practices and reflection, rather than a substitute for interactional engagement. This requires making mediation visible: students should not only use AI tools, but also critically examine how these tools reshape meaning-making, participation, linguistic hierarchies, and perceived linguistic competence.

5 Ethical interculturality

Beyond the technical and linguistic, we must consider the profound ethical dimension of the digital encounter. Applying the philosophy of (Levinas 1969), we can conceptualize VE as an ethical event. Levinas posits that the Self is not an independent, knowing subject but is instead constituted through its fundamental responsibility to the other. The digitally mediated encounter evokes an ethical demand, an obligation that precedes knowledge, judgment, or linguistic competence (van der Merwe and de Voss, 2008).

However, the current digital landscape risks undermining this ethical relationship. Technological solutionism (Ferri, 2018) threatens to reduce the other to a data point or a translated output. If we use AI merely to mask the language barrier and create a seamless, frictionless interaction, we risk sanitizing the encounter and removing the unpredictability and vulnerability that are essential for true ethical responsibility (Helm, 2024; Ferri, 2018). Ethical interculturality requires that we do not just learn to engage “properly” with others, but that we learn to respond to their alterity without imposing our own normative, native-speakerist frameworks.

The central challenge in AI-supported VE is to ensure that technology does not become a barrier to ethical responsibility. As (Levinas 1969) argued, ethics must precede knowledge. If AI is used to reach a state of linguistic mastery, it may hide the very vulnerability of the Other that demands our ethical response. We must resist the urge to use technology for perfection and instead use it to foster humility and care in communication. The goal of VE is not to eliminate the difficulty of understanding the Other, but to remain responsible for the Other even, and especially, in the face of misunderstanding (Levinas, 1969; Ferri, 2018; Helm, 2024).

6 Discussion

I propose that AI does not simply enhance VE but reconfigures the conditions under which intercultural learning occurs. While AI tools expand access and reduce communicative barriers, they can simultaneously reshape participation, authority, and epistemic relations. As (Hackett and Dawson 2026) suggest, generative AI may lower the need for linguistic negotiation that traditionally supports intercultural learning. This creates a paradox: the more effectively AI supports seamless communication, the more it may weaken the very conditions through which relational learning is enacted.

This tension becomes particularly significant when viewed through a relational and ethical lens. If VE is understood not merely as skill acquisition but as an encounter with alterity (Levinas, 1969; Helm, 2024), then the reduction of communicative difficulty raises an important question: what happens to ethical responsibility when linguistic vulnerability is technologically mediated or removed? If VE is intended to be a relational experience, then the design goal cannot be frictionless communication. Instead, it must be the cultivation of reflective engagement with difference, including linguistic, epistemic, and cultural difference.

From this perspective, Hackett and Dawson's (2026) concern can be reframed not as a rejection of AI in VE, but as a design warning. The challenge is not to preserve linguistic difficulty for its own sake, but to ensure that AI-supported communication does not eliminate the reflective and negotiated dimensions that enable intercultural learning.

The concept of linguistic justice provides a useful orientation for this rethinking. Rather than using AI to eliminate linguistic asymmetries, VE design should foreground the negotiation of these asymmetries as a site of learning. This requires making AI mediation visible and open to critique, so that students and educators can interrogate how translation, prediction, and language generation shape participation.

On this basis, this paper outlines three guiding principles for AI-informed VE design:

• Mediated communication as an object of reflection: AI tools should be explicitly acknowledged, examined, and problematized within VE activities.
• Relational pedagogy over communicative efficiency: VE tasks should prioritize dialogue, negotiation, and responsiveness over task completion or output optimization.
• Recognition of multilingual epistemologies: learners' full linguistic repertoires should be treated as legitimate and productive resources for meaning-making.

Taken together, these principles reposition AI not as a solution to linguistic inequality but as a catalyst for rethinking how linguistic difference is pedagogically and ethically engaged. For example, VE tasks might explicitly incorporate AI-assisted multilingual dialogue while requiring students to reflect on how translation shapes meaning and participation.

In conclusion, I argue that the future of virtual exchange depends on resisting the assumption that communicative friction is a problem to be eliminated. Instead, friction must be understood as a constitutive element of relational learning. The challenge for AI-supported VE is therefore not to resolve difference, but to sustain meaningful engagement with it under changing socio-technical conditions.

Generative AI statement

The author(s) declared that Generative AI was used in the creation of this manuscript for language editing and stylistic refinement of the manuscript. No generative AI tools were used for content generation. All AI-assisted outputs were critically reviewed and verified by the author, who takes full responsibility for the content. Example prompts are provided in the Supplementary Material.

Any alternative text (alt text) provided alongside figures in this article has been generated by Frontiers with the support of artificial intelligence and reasonable efforts have been made to ensure accuracy, including review by the authors wherever possible. If you identify any issues, please contact us.

Publisher's note

All claims expressed in this article are solely those of the authors and do not necessarily represent those of their affiliated organizations, or those of the publisher, the editors and the reviewers. Any product that may be evaluated in this article, or claim that may be made by its manufacturer, is not guaranteed or endorsed by the publisher.

Supplementary material

The Supplementary Material for this article can be found online at: https://www.frontiersin.org/articles/10.3389/fcomp.2026.1898906/full#supplementary-material

Supplementary material Supplementary material at the source

Footnotes

Edited by: Maria Victoria Soulé, University of Cyprus, Cyprus

Reviewed by: Goh Yingsoon, MARA University of Technology, Malaysia