Answers are grounded in this paper and cite their sources.
Generative AI has made fluent language easier to produce, but harder to interpret as evidence of learning. In language education, learners can now use ChatGPT, machine translation, automated writing feedback, grammar checkers and post-editing tools to generate, revise, translate and polish texts within minutes. Recent work on language assessment in the age of generative AI argues that the field has been disrupted in relation to reliability, validity, fairness and practicality across classroom-based, institutional and large-scale assessment contexts (Chuang and Yan, 2025). The problem is therefore not simply that AI can produce acceptable language. The more urgent assessment question is what teachers can legitimately infer from a fluent final text. This question matters beyond specialist translation programmes. In foreign language, additional language, EAP, EAL and multilingual classrooms, teachers often use language products as evidence of learning: translated sentences, bilingual explanations, targetlanguage summaries, AI-revised drafts, reflective commentaries or polished academic paragraphs. In translation classrooms, final texts are also used to assess accuracy, fluency, register, risk awareness and cultural mediation. Across these contexts, generative AI and machine translation destabilise the evidential relationship between text and learning. A polished text may appear competent, but it may no longer show how much linguistic, communicative, cultural or ethical judgement the learner exercised.The key implication is that assessment should no longer treat polished language as a transparent proxy for learning. Wider higher education debates are similarly moving beyond simply catching AI use towards assessment designs that make authentic learning, process transparency and assessable judgement visible (Kickbusch et al., 2025). This issue is particularly important for language-intensive assessment, because language is not only a product to be corrected; it is also a medium through which learners show meaning, identity, cultural positioning and disciplinary understanding. This Opinion argues that AI-mediated language, translation and multilingual assessment should move from product verification towards accountable judgement. I use this term to refer to learners' capacity to explain, evidence and take responsibility for decisions made when AI has shaped a language task, translation task or multilingual text. Accountable judgement builds on evaluative judgement. Evaluative judgement concerns learners' capability to judge the quality of their own and others' work (Tai et al., 2018), including productive engagement with generative AI outputs and processes (Bearman et al., 2024). Accountable judgement adds a further requirement: learners must make visible the reasoning behind their decisions, disclose how AI contributed, and take responsibility for the linguistic, communicative, cultural and ethical consequences of the final text. Thus, evaluative judgement concerns judging quality, whereas accountable judgement foregrounds the responsibility to justify and evidence that judgement when agency is distributed between learner and AI. The central question is not simply: Did AI produce this? It is: Can the learner explain how meaning was produced, revised, culturally mediated and responsibly carried? The issue is not only authorship. It is meaning responsibility. AI-mediated fluency: potentials and assessment limits Generative AI and neural machine translation can offer clear benefits in language and translation classrooms. In foreign and additional language learning, AI can provide vocabulary support, reformulations, grammar suggestions and alternative registers. Work on neural machine translation in language teaching shows its potential as a learning resource when learners are guided to compare, evaluate and reflect on output rather than accept it passively (Klimova et al., 2023). In translation pedagogy, AI and machine translation have made post-editing, comparison and revision central to translator education, requiring learners to identify risk and decide whether to accept, modify or reject machinegenerated suggestions (Nitzke et al., 2019;Ehrensberger-Dow et al., 2023).However, AI-supported fluency creates a specific assessment problem. In a language classroom, a sentence may be grammatically accurate but pragmatically inappropriate: too direct, too formal, culturally awkward or unsuitable for the relationship between speakers. In a translation classroom, a fluent AI-supported target text may hide unresolved problems of register, cultural mediation and responsibility. The assessment question should therefore shift from whether the final product is accurate or fluent to why particular AI suggestions were accepted, rejected or revised, and what cultural meaning was preserved, adapted, explained or resisted. In assessment terms, accountable judgement therefore requires learners to substantiate the linguistic and communicative decisions underlying an AIsupported text, rather than relying on the fluency of the final product as evidence of competence.From feedback uptake to accountable judgement Recent discussion of AI feedback in writing instruction has shifted attention from the quality of feedback to learners' engagement with feedback. Helm and Hesse (2026), for example, argue that AI feedback should not be treated as an endpoint, but as a catalyst for classroom communication and revision. For language and translation assessment, this argument can be pushed further: feedback uptake is not enough unless it becomes judgement uptake, the process through which learners transform AI suggestions, teacher comments, peer feedback or automated corrections into their own explainable decisions. Boud and Molloy (2013) frame feedback as a learning process rather than information transmission, and this view becomes more important when feedback is instant, abundant and partly automated. From an ecological perspective, feedback literacy involves more than the uptake of information; it is shaped by the contextual resources and interactions through which learners engage with feedback (Chong, 2021). In AI-mediated assessment, the proposed shift towards judgement uptake places particular emphasis on learners' capacity to explain how that engagement informs their linguistic and communicative decisions. Guo and Wang (2024) similarly show the need to rethink teacher feedback in EFL writing when ChatGPT enters the feedback ecology. The issue is whether learners can evaluate feedback, select what is useful, reject what is inappropriate and explain what has changed in their own judgement. Accountable judgement therefore differs from simple compliance with feedback: students are not only expected to revise, but to account for why a revision is linguistically appropriate, communicatively effective, culturally responsible and ethically disclosed. Revision trails, decision logs, reflective commentaries, annotated drafts, oral justifications and bilingual explanations can contribute assessment evidence, rather than serving as optional appendices. This responsibility for explaining decisions also extends to the cultural meanings negotiated through AI-mediated language and translation.The question Who owns the text? is too narrow for AI-mediated language and translation assessment. A more significant question is: Who carries the meaning? AI can generate words, but it does not carry educational, cultural or ethical responsibility for those words. Translation is not only linguistic transfer; it is cultural mediation. Baker (2018), Katan and Taibi (2021), and Venuti (2018) show, in different ways, that translators work within relations of culture, visibility, power and interpretive choice. These relations do not disappear when AI participates in the task. AI-mediated translation is therefore not only a linguistic task but also a cultural judgement task. A machine-generated text may be accurate at the level of words and sentences while still failing to carry humour, politeness, relational meaning, local knowledge, historical association, ideological positioning or minority-language visibility. Cultural mediation cannot be reduced to fluency or naturalness. It requires learners to decide what should be retained, adapted, explained, softened, resisted or made visible for a particular audience. These decisions are not merely stylistic; they are part of the learner's accountable judgement. In cultural mediation, such accountability entails articulating the interpretive grounds for linguistic and cultural choices and considering their implications for the intended audience.Assessment should make this cultural work visible. In a language classroom, learners might explain why one expression is more polite, more relational or more suitable for a particular audience than another. In a translation classroom, students might justify why a cultural term should be retained, glossed, adapted or deliberately made visible. In EAP, EAL and multilingual classrooms, learners might show how AI feedback was interpreted through their existing linguistic resources. Such evidence is especially important for multilingual equity. An assessment design that recognises only polished final products may miss precisely the judgement that multilingual learners are well placed to show. Taken together, these linguistic, communicative and cultural considerations inform the framework proposed in the following section.A framework for accountable judgement I propose accountable judgement as a principle for AI-mediated language, translation and multilingual assessment. Its purpose is to make learning visible through the learner's explanation of choices, risks, revisions and responsibilities. Accountable judgement therefore combines the evaluative capacity to recognise quality with two additional demands that become particularly important in AI-mediated assessment: the mediation of cultural meaning and the responsibility to explain and evidence decisions when AI has contributed to the task. This framework repositions AI use as an assessable learning process rather than an invisible shortcut. It asks learners to compare AI outputs, identify linguistic and cultural risks, revise the text, justify key decisions and provide evidence of process. It also reframes academic integrity. Disclosure remains important, but disclosure alone does not prove learning. A student may honestly declare AI use and still fail to understand the final text. Responsibility requires explanation.Discussion and practical implications AI can support language learning, translation practice, feedback generation, academic writing and revision. The assessment challenge arises when AI-supported fluency stands in for learner judgement. Wider research on generative AI and higher education assessment similarly suggests that assessment must be redesigned around integrity, responsibility and learning processes rather than detection alone (Moorhouse et al., 2023;Xia et al., 2024).For language teachers, accountable judgement means asking learners to explain why an AIsuggested expression is appropriate or inappropriate for a particular audience, register, relationship or communicative purpose. For translation teachers, it means asking students to compare AI outputs, identify cultural and linguistic risks, justify post-editing decisions and document how much AI shaped the final text. For EAP and academic writing teachers, it means asking students to show how AI feedback was interpreted, accepted, rejected or transformed during revision. For assessment designers and academic integrity teams, it means shifting policy conversations from How do we detect AI? to What evidence shows that learning, judgement and responsibility have taken place?This can be operationalised through small assessment changes. A language-writing task might require students to explain three AI-suggested changes they accepted, two they rejected and one cultural or pragmatic issue they had to resolve. A translation assignment might include an AI-output comparison, a risk log, a cultural mediation note and a final reflective commentary. A multilingual task might allow bilingual commentary or oral explanation as legitimate evidence of judgement. Together, these forms of process evidence may make learner judgement more visible. However, the evidential value of written decision logs and reflective commentaries requires careful consideration, since such accounts may themselves be generated by AI. Situated assessment activities, classroom dialogue and selective oral justification may provide complementary evidence of learners' reasoning, although no single mode can independently establish authenticity.Implementing these forms of process evidence also raises practical questions of feasibility, particularly in larger classes, where reviewing detailed decision logs or reflective commentaries may increase assessor workload. A proportionate approach could ask learners to justify a small number of consequential decisions, use short structured templates, or combine written evidence with selective oral explanation. Such a design would retain meaningful evidence of learner judgement while keeping the assessment manageable in practice. Expectations for such evidence should also be responsive to learners' age, language proficiency and developing AI literacy. Structured prompts, bilingual commentary and supported oral reflection may provide appropriate scaffolding, particularly where learners are not yet able to articulate complex decisions independently. This differentiation is relevant not only to the assessment of writing but also to the ways in which AI mediates planning, composing and revision.An important next step is to evaluate accountable judgement empirically. Future research should examine whether the proposed process evidence provides valid and reliable evidence of learner judgement, how feasible the framework is across different class sizes and assessment contexts, and how it operates across linguistically diverse learner groups.Studies could also compare forms of evidence such as decision logs, annotated drafts, bilingual commentary and oral justification. Particular attention should be paid to validity, fairness and scalability. The central question is not simply: Did AI produce this? It is: Can the learner explain, revise, contextualise and take responsibility for the meaning produced? AI can help with words, but culture is not just words. The learner may not own every word in a narrow authorship sense, but the learner must carry the judgement, cultural mediation and responsibility behind the meaning. That is where learning becomes visible.Author contributions JY: Conceptualisation, writing -original draft, writing -review and editing.
No takes yet. Share an insight, caveat, or question.
Jun Yan (2026) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: