Rethinking Higher Education/Chapter 4

Language: EN · ZH · EN-ZH · ← Book

The End of Translation: How AI is Transforming Cross-Cultural Communication in Education

Martin Woesler

Hunan Normal University

Abstract

The rise of neural machine translation (NMT) and generative AI has fundamentally altered the landscape of cross-cultural communication and language education. This article examines the transformation wrought by AI-powered translation tools — including DeepL, ChatGPT, and Chinese platforms such as Baidu Translate and Youdao — on translation practice, the translation labour market, and language pedagogy. Drawing on recent comparative studies and labour market data, we document a dramatic contraction in professional translation demand alongside an expansion of machine translation post-editing (MTPE) workflows. We argue that this shift necessitates a new educational paradigm centred on „machine translation literacy„ rather than traditional translation competence. Through a systematic comparison of European and Chinese responses to this transformation, we identify divergent regulatory approaches and converging pedagogical challenges. We conclude that the „end of translation“ is not the end of cross-cultural understanding but rather a reorientation of human expertise toward what remains irreducibly human: cultural nuance, ethical judgement, and literary sensibility.

Keywords: machine translation, AI in education, DeepL, ChatGPT, post-editing, translation literacy, EU-China comparison, language education

1. Introduction

On 30 May 2024, as part of the Jean Monnet Centre of Excellence lecture series „Digitalization in China and Europe“ at Hunan Normal University, the author delivered a lecture entitled „The End of Translation.“ The title was deliberately provocative. Translation, one of humanity’s oldest intellectual activities, appears to face an existential challenge from artificial intelligence. Neural machine translation systems now process billions of words daily with accuracy levels that would have seemed inconceivable a decade ago. The International Monetary Fund reported in January 2026 that its translator and interpreter staff had declined from 200 to approximately 50, with the remainder increasingly reassigned to quality control of machine-generated output (CNN 2026). A 2024 survey by the Chartered Institute of Linguists (CIOL) found that over 70 percent of freelance translators reported decreased work volumes (CIOL 2024).

Yet the provocative framing demands qualification. What is ending is not translation as an intellectual and cultural activity but rather translation as a routine professional service performed primarily by humans. What is emerging is a new ecology of human-machine collaboration in which the human role shifts from producing translations to evaluating, refining, and contextualizing machine output. For education — particularly for language departments and translation programmes in both European and Chinese universities — this shift poses urgent questions. What should we teach students who will enter a profession that looks radically different from its twentieth-century predecessor? How do we cultivate the distinctly human competencies that machines cannot replicate? And how do different educational systems — the European and the Chinese — respond to these challenges?

This article addresses these questions by examining the current state of AI translation technology, its labour market consequences, the emerging concept of „machine translation literacy,“ the role of post-editing in education, and the divergent responses of European and Chinese institutions. It contributes to the broader anthology project by connecting the technological dimension of AI in education with the ethical framework developed in the companion chapter on Appropriateness Theory (Woesler, this volume) and the empirical findings on AI-assisted language learning (Woesler, this volume).

2. Historical Context: Translation and Technology

Before examining the current AI-driven transformation, it is useful to situate it within the longer history of technology’s relationship with translation. Translation has always been mediated by technology in the broadest sense — from the invention of writing itself, which made translation possible as a textual practice, to the printing press, which created mass demand for translated works, to the typewriter and word processor, which transformed the translator’s physical workflow.

The computational approach to translation dates to Warren Weaver’s 1949 memorandum, which proposed applying information theory and cryptographic techniques to the „translation problem.“ The subsequent decades of rule-based machine translation (RBMT) and statistical machine translation (SMT) produced systems that were useful for gisting — obtaining a rough understanding of a foreign text — but inadequate for producing publishable translations. The neural revolution of the mid-2010s, beginning with the attention mechanism (Bahdanau et al. 2014) and culminating in the Transformer architecture (Vaswani et al. 2017), changed this fundamentally. For the first time, machine output was often indistinguishable from competent human translation for routine texts.

The historical parallel most often invoked is the impact of photography on painting. Photography did not eliminate painting but it did eliminate painting’s monopoly on visual representation, freeing painters to explore dimensions of visual experience — abstraction, expressionism, conceptual art — that photography could not capture. Similarly, AI translation may not eliminate human translation but it is eliminating human translation’s monopoly on cross-linguistic communication, potentially freeing human translators to focus on dimensions of cross-cultural meaning-making that machines cannot replicate.

3. The AI Translation Landscape: Tools and Capabilities

3.1 Neural Machine Translation vs. Generative AI

The distinction between dedicated neural machine translation (NMT) systems and generative AI (GenAI) tools is crucial for understanding the current landscape. As Ohashi (2024) demonstrate in their systematic comparison, NMT systems such as Google Translate and DeepL are purpose-built for translation, trained on parallel corpora, and optimized for fluency and adequacy in specific language pairs. GenAI tools such as ChatGPT, Claude, and China‘s Ernie Bot handle translation as one capability among many, leveraging broader linguistic understanding but lacking the specialized training data of dedicated NMT systems.

This distinction has practical consequences. A multidimensional comparison of ChatGPT, Google Translate, and DeepL on Chinese tourism texts found that ChatGPT outperformed dedicated NMT across metrics of fidelity, fluency, cultural sensitivity, and persuasiveness — particularly when given culturally tailored prompts (Chen et al. 2025). However, ChatGPT occasionally introduced semantic shifts absent from the source text, a phenomenon that dedicated NMT systems avoid more reliably. Sun, R. (2024), evaluating Chinese-to-English literary translation through the lens of implied subjects, found that ChatGPT handled the implicit subject structures of contemporary Chinese prose (specifically Yu Qiuyu’s essays) with greater sophistication than DeepL, suggesting that GenAI’s broader contextual understanding compensates for its lack of specialized translation training in literary domains.

2.2 Chinese AI Translation Platforms

The Chinese AI translation ecosystem deserves separate attention, as it operates largely independently of the Western ecosystem. Baidu Translate (百度翻译) supports 203 languages, including rare dialects and classical Chinese, using NMT technology with context-aware sentence-level processing. Youdao (有道翻译), developed by NetEase, released its „Ziyue“ (子曰) educational large language model in 2023, followed by the Ziyue-o1 reasoning model in 2025, integrating translation capabilities with deep dictionary and study tool functions across approximately 107 languages. Tencent Translate and iFlytek’s systems similarly combine translation with broader educational and business applications.

A comparative quality assessment of Youdao AI Translate, Baidu Translate, and Tencent Translate for Chinese music historical texts using Likert scaling evaluation found significant variation in domain-specific accuracy, with none achieving consistent expert-level quality in specialized cultural content (Zhang et al. 2025). This finding underscores a recurring theme: AI translation excels at general-purpose communication but struggles with domain-specific, culturally embedded, and stylistically demanding texts.

2.3 The State of the Art: Strengths and Persistent Weaknesses

Current AI translation systems handle routine informational texts — news articles, business correspondence, technical documentation, user interfaces — at levels approaching or matching average human translators. The quality gap narrows further when post-editing is factored in. However, systematic weaknesses persist in several domains:

Literary translation, where voice, rhythm, ambiguity, and cultural resonance are constitutive rather than incidental. Machine translation of poetry, for instance, remains largely inadequate, as documented by ongoing debates in the publishing industry about whether AI models are „advanced enough to translate literature“ (The Markup 2025).

Humour, irony, and sarcasm, which depend on shared cultural knowledge, contextual inference, and intentional violation of linguistic expectations. Machines can identify some patterns of irony through training data, but they cannot understand why something is funny or how irony functions rhetorically.

Culturally embedded expressions that lack direct equivalents — not merely idioms (which NMT systems increasingly handle through pattern matching) but conceptual frameworks that reflect distinct worldviews. The Chinese philosophical term 仁 (rén), for example, resists translation into any single English word („benevolence,“ „humaneness,“ „goodness,“ „humanity“) because each option foregrounds different aspects of a concept that in Chinese encompasses all of them simultaneously.

Ethical and political sensitivity, where translation choices carry consequences beyond linguistic accuracy. The rendering of politically charged terms — „developing country,“ „human rights,“ „democracy“ — involves judgements that AI systems make based on statistical patterns in training data rather than ethical deliberation.

3.3 A Note on Evaluation Methodology

The evaluation of AI translation quality remains contested. Traditional metrics such as BLEU (Bilingual Evaluation Understudy) and more recent neural metrics like COMET and BERTScore capture different aspects of translation quality and sometimes yield contradictory results. Human evaluation, while considered the gold standard, is expensive, time-consuming, and subject to evaluator bias. The growing practice of using AI systems to evaluate other AI systems’ translations raises questions of circularity.

For educational contexts, the most relevant evaluation criterion is not abstract quality but fitness for purpose. A machine translation used to help a student understand a foreign-language text has different quality requirements than one used for publication, legal proceedings, or diplomatic communication. This pragmatic perspective — which resonates with the Skopos Theory tradition in translation studies and with the Appropriateness framework proposed elsewhere in this volume — suggests that the question „Is machine translation good enough?“ can only be answered in relation to a specific purpose, context, and set of consequences.

4. Labour Market Transformation

4.1 The Contraction of Translation Demand

The impact of AI on the translation profession has been severe and rapid. Research by the Centre for Economic Policy Research (CEPR) using variation in Google Translate adoption across US local labour markets found that „areas with higher adoption experienced a decline in translator employment“ and that „improvements in machine translation have reduced the demand for foreign language skills in general“ (Frey & Llanos-Paredes 2025). Over three-quarters of professional translators surveyed in 2024 expected generative AI to adversely affect their future income.

The International Monetary Fund’s reduction of its translator staff from 200 to approximately 50 represents a concrete institutional example (CNN 2026). Duolingo laid off approximately 10 percent of its translation contractors in January 2024, shifting to AI-led content production. Translation agencies have been „increasingly switching to a business model revolving around MTPE, slashing rates and often compromising the quality of the final product“ (The Markup 2025).

These figures must be interpreted with care. The translation industry encompasses a wide range of activities — from certified legal translation to website localization to literary translation to conference interpreting — and the impact of AI varies dramatically across these sub-sectors. The Bureau of Labor Statistics projects continued demand for interpreters and translators in the United States through 2032, driven by globalization and immigration, even as the nature of the work changes. What is contracting is not demand for cross-linguistic communication but demand for the traditional model of human-only translation of routine texts.

4.2 The Rise of Post-Editing

The decline in traditional translation employment has been partially offset by the growth of machine translation post-editing (MTPE) as a professional activity. Post-editors review, correct, and refine machine-generated translations rather than translating from scratch. This represents a fundamental shift in the translator’s role — from author to editor, from creator to quality controller.

Research on ChatGPT-4o’s potential for augmenting post-editing found that it „can complement human expertise in post-editing“ but „cannot provide completely accurate translations without human intervention,“ and that full integration would „significantly reduce costs, time, and effort“ (Chen 2025). Jiang, Wei, and Al-Shaibani (2025), studying NMT post-editing of Chinese intangible cultural heritage texts into English, documented how MTPE workflows can handle specialized cultural content while demonstrating that human expertise remains essential for culturally sensitive material.

However, the post-editing model raises its own concerns. The practice of „post-editing fatigue“ — cognitive exhaustion from constantly evaluating and correcting machine output rather than engaging in creative translation — has been documented among professional translators (The Markup 2025). Some argue that post-editing trains the human mind to think like a machine rather than developing the cultural and literary sensitivity that distinguishes expert human translation.

4.3 Differentiated Impact

The labour market impact is not uniform across translation domains. Routine business, technical, and administrative translation has been most affected, with AI systems handling these tasks at acceptable quality with minimal human intervention. Literary, legal, and diplomatic translation — domains where precision, nuance, and cultural sensitivity are paramount — have been less affected, though AI tools are increasingly used as first-draft generators even in these fields.

The geographic distribution of impact also varies. In Europe, where the translation profession is relatively well-organized and regulated, the transition has been accompanied by professional debate and some institutional responses. In China, where the translation market is larger but more fragmented, the shift has been faster and less mediated by professional organizations. Chinese translation companies have adopted AI tools aggressively, driven by market competition and the availability of domestic AI platforms optimized for Chinese language pairs.

5. Machine Translation Literacy: A New Educational Paradigm

5.1 Defining MT Literacy

The concept of „machine translation literacy“ was developed by Bowker and Buitrago Ciro (2019) and further elaborated by Bowker (2023) in De-mystifying Translation: Introducing Translation to Non-translators. MT literacy encompasses several competencies: understanding how MT systems work (architecturally, not just functionally), understanding how MT systems are used in practice, appreciating the wider social and economic implications of MT, evaluating the MT-friendliness of source texts, creating or modifying texts for better MT output, and modifying MT output for quality and appropriateness.

This framework shifts the educational focus from translation competence (the ability to translate) to translation literacy (the ability to work effectively with translation technologies). The shift is analogous to the broader educational transformation from information production to information evaluation that characterizes the digital age.

5.2 Student Preparedness: Current Gaps

A study investigating novice translation students’ AI literacy in translation education found that students „frequently mention terms like ‘big data,’ ‘deep learning,’ and ‘neural network’ but exhibit little knowledge of what these words mean or how they relate to AI translation tools“ (Zhang et al. 2025). This gap between superficial familiarity and genuine understanding characterizes the current state of translation education in many institutions.

A survey of university students’ perceptions of using generative AI in translation practices found that GenAI tools offer advantages in „enhancing translation efficiency, quality, learning, and practice,“ fostering a positive outlook — but that this positive outlook may reflect uncritical acceptance rather than informed evaluation (Zhang et al. 2025). The challenge for educators is to cultivate critical engagement with AI tools without either demonizing or uncritically celebrating them.

Research on machine translation in English language teaching found that Google Translate is the most popular MT tool among students, followed by DeepL and ChatGPT, primarily used for reading comprehension, grammar checking, and writing assistance (ELT Journal 2024). A significant finding was that „teachers tend to view MT as ‘a foe’ while students see it as ‘a friend’„ — a generational divide that complicates pedagogical responses.

5.3 Towards a New Professional Role

The emergence of MT literacy as an educational paradigm has led to proposals for new professional roles. Ehrensberger-Dow, Delorme Benites, and Lehr (2023) proposed the role of „MT literacy consultant“ — a professional who bridges the gap between MT technology and pedagogical or organizational needs (Ehrensberger-Dow 2023). This role involves not only technical competence but also the ability to assess institutional needs, design MT integration strategies, and train end-users in effective MT use.

A comprehensive protocol for integrating human-AI collaboration into translation education was proposed in Ren et al. (2025), outlining assessment, diagnosis, and strategy development approaches that position human-AI collaboration as a core pedagogical framework rather than an add-on to traditional translation training.

6. Post-Editing in the Curriculum

6.1 MTPE as Core Competency

The integration of MTPE into translation curricula is no longer optional but essential. Students entering the translation profession will almost certainly work with machine-generated first drafts rather than translating from blank pages. Curricula must therefore develop competencies in:

Error detection and classification — identifying the types of errors that NMT and GenAI systems characteristically produce (omissions, additions, mistranslations, register errors, cultural inappropriateness).

Efficient editing strategies — distinguishing between „light post-editing„ (minimal corrections for gist comprehension) and „full post-editing“ (thorough revision to publication quality), and knowing when each is appropriate.

Source text evaluation — assessing whether a text is suitable for MT or requires human translation, based on factors such as domain specificity, cultural embeddedness, register requirements, and the consequences of error.

Quality assessment — developing frameworks for evaluating MT output systematically, moving beyond impressionistic judgement to structured evaluation criteria.

6.2 A Taxonomy of Machine Translation Errors

Effective post-editing requires a systematic understanding of the types of errors that AI translation systems characteristically produce. Based on the research literature and the author’s own experience with Chinese-English and Chinese-German machine translation, we can identify the following recurring error categories:

Omission errors: NMT systems sometimes skip portions of the source text, particularly subordinate clauses, parenthetical remarks, or culturally specific explanations. This is especially problematic in Chinese-to-English translation, where the absence of explicit grammatical markers (no articles, no obligatory plural marking, no inflected verb forms) means that the machine must infer structural relationships that may lead to simplification.

Addition errors: Conversely, GenAI systems — particularly ChatGPT — occasionally insert information not present in the source text, drawing on their training data to „complete“ what they perceive as incomplete information. This „hallucination“ problem, well-documented in other GenAI applications, manifests in translation as the insertion of explanatory phrases, implicit cultural knowledge, or inferred logical connections.

Register errors: Machine translation systems struggle with register — the level of formality, technicality, or intimacy appropriate to a given text type. A formal diplomatic communiqué translated at the register level of casual conversation, or a children’s story rendered in academic prose, may be semantically accurate but pragmatically inappropriate.

Cultural substitution errors: When faced with culturally specific references, AI systems sometimes substitute target-culture equivalents rather than preserving source-culture specificity. This is a manifestation of the domestication tendency that Venuti (1995) critiqued in human translation, now automated and therefore more pervasive.

Coherence errors: While individual sentences may be accurately translated, the coherence of longer texts — the logical flow of argument, the maintenance of thematic threads, the consistency of terminology — may be compromised by sentence-level processing that loses sight of discourse-level structure.

Understanding these error types is essential for effective post-editing and should form a core component of MTPE curricula.

6.3 Challenges in MTPE Education

Teaching MTPE effectively poses several pedagogical challenges. Students must first develop sufficient translation competence to recognize machine errors — one cannot effectively edit a translation one could not have produced or at least evaluated independently. This creates a paradox: the very competencies that MTPE seems to render obsolete (deep language knowledge, cultural understanding, stylistic sensitivity) are precisely those needed to perform MTPE well.

Furthermore, MTPE training risks reducing translation education to a technical exercise in error correction, neglecting the creative, cultural, and ethical dimensions that distinguish translation from other forms of text processing. Educators must find ways to develop both technical MTPE skills and the broader humanistic competencies that give those skills meaning and direction.

A promising pedagogical approach involves structured comparison exercises in which students evaluate the same source text translated by multiple systems (NMT, GenAI, and human translator) and analyze the characteristic strengths and weaknesses of each. Such exercises develop critical evaluation skills, build awareness of error types, and cultivate the metalinguistic awareness that distinguishes an informed post-editor from a passive corrector of machine output.

7. European and Chinese Responses Compared

7.1 The European Approach

European responses to the AI translation revolution have been shaped by several factors: the EU’s strong tradition of multilingualism (24 official languages), the relatively mature professionalization of translation through organizations such as the European Master’s in Translation (EMT) network, and the regulatory framework provided by the EU AI Act.

The EMT network has begun revising its competence framework to incorporate AI-related skills, though the pace of institutional change lags behind technological development. Several European universities have introduced dedicated MTPE courses or modules within existing translation programmes. The EU AI Act, which classifies certain educational AI applications as high-risk (see the companion chapter on AI Ethics in Education, Woesler, this volume), provides a regulatory context that encourages careful, ethically informed integration of AI tools rather than uncritical adoption.

A notable European development is the growing emphasis on „AI literacy„ as a general educational requirement. The EU AI Act mandates AI literacy training for deployers of AI systems, with this obligation having entered into force in February 2025. While this mandate targets organizations rather than students directly, it creates institutional incentives for universities to develop AI literacy curricula that include translation-related competencies.

The European Master’s in Translation (EMT) network, which coordinates translation training across more than 80 European universities, has begun a systematic review of its competence framework in light of AI developments. The 2024 review process identified machine translation literacy, post-editing skills, and AI ethics as priority areas for curriculum integration. However, the pace of implementation varies dramatically across member institutions — some have already restructured their programmes around human-AI collaboration, while others continue to teach translation skills as if the technological landscape had not fundamentally changed.

In Germany specifically, the Conference of University Translation Departments (Konferenz der Universitätstranslationsinstitute) has debated the extent to which traditional translation examinations — which typically require students to translate unseen texts under time pressure without technological aids — remain appropriate as assessment instruments. The emerging consensus favours a dual approach: retaining elements of traditional translation competence assessment while adding new evaluation forms that test post-editing ability, MT output evaluation, and critical reflection on AI tools.

France has taken a somewhat different approach, with the Société française des traducteurs (SFT) publishing guidelines for the integration of AI tools into professional practice that have influenced university curricula. These guidelines emphasize the translator’s evolving role as a „language technology expert“ who can advise clients on the appropriate use of machine and human translation for different purposes.

7.2 The Chinese Approach

China‘s response has been characterized by rapid institutional adoption, state-level policy direction, and the availability of domestic AI platforms. The Chinese Ministry of Education mandated AI education in all primary and secondary schools from September 2025 — primary schools focus on AI literacy and exposure, junior high schools on logic and critical thinking, and senior high schools on applied innovation and algorithm design. At the university level, Chinese institutions have been quicker than many European counterparts to integrate AI tools into language and translation curricula, driven partly by market competition and partly by state encouragement of AI adoption.

The domestic AI ecosystem — Baidu Translate, Youdao with its Ziyue educational model, Tencent Translate, iFlytek — provides Chinese students and educators with AI tools specifically optimized for Chinese language pairs and integrated with broader educational platforms. This contrasts with the European reliance on primarily Western commercial tools (DeepL, Google Translate, ChatGPT), which are optimized for European language combinations and may not serve Chinese-European language pairs as effectively.

However, China‘s approach also faces distinctive challenges. The restriction of access to global AI tools such as ChatGPT, Claude, and Google Gemini means that Chinese students and researchers work primarily with domestic alternatives, which may limit their exposure to the full range of AI translation capabilities and approaches. The emphasis on rapid adoption may also come at the expense of the critical evaluation and ethical reflection that European institutions, influenced by the AI Act framework, are beginning to prioritize.

A further dimension of the Chinese response is the integration of AI translation with broader national strategies. The State Council’s „New Generation Artificial Intelligence Development Plan“ (2017) and „Education Modernization 2035“ both frame AI capabilities — including translation — as strategic national competencies. This means that the adoption of AI translation tools in Chinese education is not merely a pedagogical decision but a contribution to national technological development. Chinese universities consequently face different incentive structures than their European counterparts: where European institutions may be cautious about AI adoption due to concerns about academic integrity and human skill degradation, Chinese institutions face institutional pressure to demonstrate AI integration as evidence of modernization and alignment with national policy.

The integration of AI translation into China‘s college English education system illustrates this dynamic. The College English Curriculum Requirements (2024 revision) explicitly include „the ability to use AI tools for cross-cultural communication“ as a graduate competency. Several major Chinese universities — including Peking University, Fudan University, and Hunan Normal University — have introduced modules on AI-assisted translation and communication within their foreign language programmes, often taught by faculty who are themselves learning to navigate the new technological landscape.

7.3 Convergent Challenges

Despite these differences, European and Chinese institutions face convergent challenges:

Curriculum lag — both systems struggle to update curricula at the pace of technological change, with accreditation processes and institutional inertia slowing adaptation.

Faculty preparedness — in both contexts, many translation faculty lack personal experience with AI tools and are uncertain about how to integrate them effectively into their teaching.

Assessment design — traditional translation examinations that test the ability to translate texts from scratch become problematic when students have access to AI tools, necessitating new assessment approaches that evaluate MT literacy, critical evaluation, and post-editing competence rather than raw translation ability.

Balancing technical and humanistic education — both systems must navigate the tension between training students for the immediate demands of the MTPE-dominated market and developing the broader cultural, ethical, and creative competencies that distinguish human translators from machines.

8. Beyond Machine Translation: What Remains Irreducibly Human

8.1 Literary and Cultural Translation

The domain where human translators remain most clearly indispensable is literary and cultural translation. Literary translation involves not merely converting meaning from one language to another but recreating voice, rhythm, ambiguity, humour, irony, and aesthetic effect in a different linguistic and cultural context. A machine can translate the words of a poem; it cannot translate the poem.

This is not merely an assertion of humanistic faith against technological progress. It reflects the fundamental difference between pattern matching (which AI excels at) and meaning-making (which requires embodied cultural experience, emotional understanding, and aesthetic judgement). When Emily Wilson translates Homer’s Odyssey or Pevear and Volokhonsky retranslate Dostoevsky, they bring not only linguistic expertise but personal interpretive vision — a quality that distinguishes great translation from competent rendering.

8.2 Ethical Judgement in Translation

The companion chapter on Appropriateness Theory (Woesler, this volume) argues that translation cannot be evaluated purely on linguistic criteria but must be assessed within an ethical framework that considers the conditions of production, the historical context, and the potential consequences of translation choices. This ethical dimension — the ability to recognize when a translation choice may cause harm, when cultural sensitivity demands deviation from literal accuracy, when political context requires careful framing — remains beyond the current capabilities of AI systems.

AI translation systems make choices based on statistical patterns in training data. They cannot exercise ethical judgement in the sense of weighing competing values and accepting responsibility for the consequences of those choices. The post-editor who reviews machine output and decides that a particular rendering, while linguistically accurate, is culturally inappropriate or politically sensitive exercises a form of human agency that no current AI system can replicate.

8.3 The „Complementarity Thesis“ Extended

The empirical study in this volume (Woesler, this volume) found that AI and human instruction serve „different, complementary functions“ in language learning — what we termed the Complementarity Thesis. This thesis extends naturally to translation: AI and human translators serve different, complementary functions. AI excels at speed, consistency, coverage, and the handling of routine informational texts. Human translators excel at cultural interpretation, literary creativity, ethical judgement, and the handling of texts where precision and nuance carry high stakes.

The future of translation is neither the replacement of humans by machines nor the preservation of traditional human-only translation practice. It is a collaborative ecology in which the division of labour between human and machine is continually renegotiated as AI capabilities evolve and as our understanding of what constitutes „good translation“ deepens.

8.4 The Case of Chinese Literature in Translation

A concrete illustration of the limits of AI translation can be drawn from the author’s own experience translating Chinese literary texts — including works by Lu Xun and classical Chinese novels — into German and English. Consider the opening line of Lu Xun’s 狂人日记 (Diary of a Madman, 1918): „今天晚上，很好的月光.“ A machine translation produces something like „Tonight, very good moonlight“ — grammatically adequate but aesthetically dead. The challenge for the human translator is to convey not only the semantic content but the deliberately flat, unsettling tone that establishes the narrator’s mental state: the juxtaposition of ordinary observation („very good moonlight“) with the extraordinary context of emerging madness.

Similarly, the translation of classical Chinese poetry — where a single character may carry multiple simultaneous meanings, where tonal patterns create musical structures, and where cultural allusions operate on several levels — remains fundamentally beyond current AI capabilities. Tang dynasty poetry, with its strict formal requirements (律诗 lüshi: regulated verse with tonal patterns, parallelism, and compressed imagery), resists not only machine translation but any translation that does not involve deep engagement with Chinese literary tradition, Buddhist and Daoist philosophical frameworks, and the specific historical contexts of individual poets.

These examples are not mere curiosities but illustrations of a general principle: the more a text depends on cultural embeddedness, aesthetic effect, and multi-layered meaning, the less adequate AI translation becomes. This principle has direct implications for education, suggesting that the study of literary translation — far from being an obsolete specialization — may become the most distinctively human and therefore most valuable component of translation education.

9. Conclusion

The „end of translation“ is a misleading phrase if understood literally. What is ending is translation as a routine professional service performed predominantly by humans. What is emerging is a new ecology of human-machine collaboration that demands new competencies, new professional roles, new curricula, and new forms of ethical reflection.

For education, this transformation demands nothing less than a reconceptualization of what it means to teach and learn translation. Machine translation literacy — encompassing technical understanding, critical evaluation, effective post-editing, and ethical judgement — must become central to language and translation curricula. At the same time, the humanistic competencies that machines cannot replicate — cultural sensitivity, literary sensibility, ethical reasoning, creative interpretation — must be preserved and strengthened rather than abandoned as obsolete.

The comparison of European and Chinese responses reveals a productive tension between the EU’s cautious, regulation-informed approach and China‘s rapid, state-supported adoption of AI tools. Neither approach is sufficient alone. The EU’s emphasis on ethical frameworks and critical evaluation provides necessary guardrails for responsible AI integration. China’s speed and scale of adoption generate practical experience and institutional adaptation that the EU can learn from. A synthesis of both approaches — combining ethical rigour with practical agility — offers the most promising path forward.

Several practical recommendations emerge from this analysis. First, translation and language programmes should introduce machine translation literacy as a core competency from the first semester, ensuring that students develop critical evaluation skills alongside traditional language competencies. Second, MTPE should be taught not merely as a technical skill but within a framework that includes error taxonomy, quality assessment methodology, and ethical reflection on the consequences of translation choices. Third, literary and cultural translation should be preserved and strengthened as the domain where human expertise remains most clearly irreplaceable — and most clearly valuable in an AI-saturated landscape. Fourth, EU-China academic cooperation in translation education should be intensified, leveraging the complementary strengths of the European emphasis on regulation and ethical frameworks with the Chinese emphasis on practical AI integration and scale.

For the broader field of educational digitalization, the „end of translation“ case study offers a cautionary tale and an encouraging precedent. Cautionary, because the speed and scale of AI disruption can outpace institutional adaptation, leaving students inadequately prepared for a transformed professional landscape. Encouraging, because the transformation does not eliminate the need for human expertise but redirects it — from routine execution to critical evaluation, creative interpretation, and ethical judgement. These are precisely the competencies that higher education is best positioned to develop.

As this anthology demonstrates through its companion chapters on AI-assisted language learning, alternative learning forms, AI ethics, and the university of the future, the transformation wrought by AI in education is not a single phenomenon but a constellation of interconnected changes that must be addressed holistically. The end of translation as we knew it is, simultaneously, the beginning of something new — a form of cross-cultural communication in which human and artificial intelligence collaborate to bridge the gaps between languages, cultures, and worldviews.

Acknowledgments

This research was supported by the Jean Monnet Centre of Excellence „EU-Studies Centre: Digitalization in Europe and China„ (EUSC-DEC), funded by the European Union under Grant Agreement No. 101126782. Views and opinions expressed are those of the author only and do not necessarily reflect those of the European Union.

References

Algaraady, J., & Mahyoob, M. (2025). Exploring ChatGPT’s potential for augmenting post-editing in machine translation across multiple domains: challenges and opportunities. Frontiers in Artificial Intelligence, 8, 1526293.

Bowker, L. (2023). De-mystifying Translation: Introducing Translation to Non-translators. Routledge.

Bowker, L., & Buitrago Ciro, J. (2019). Machine Translation and Global Research: Towards Improved Machine Translation Literacy in the Scholarly Community. Emerald Publishing.

Chen, S., & Lin, Y. (2025). A multidimensional comparison of ChatGPT, Google Translate, and DeepL in Chinese tourism texts translation: fidelity, fluency, cultural sensitivity, and persuasiveness. Frontiers in Artificial Intelligence, 8, 1619489.

CIOL. (2024). Freelance translators and interpreters: Work volumes survey. Chartered Institute of Linguists.

CNN. (2026, January 23). Translation and language jobs face automation as AI transforms the industry.

Ehrensberger-Dow, M., Delorme Benites, A., & Lehr, C. (2023). A new role for translators and trainers: MT literacy consultants. The Interpreter and Translator Trainer, 17(3), 393–411.

Frey, C. B., & Llanos-Paredes, D. (2025). Lost in translation: AI’s impact on translators and foreign language skills. CEPR/VoxEU.

Kirchhoff, P. (2024). Machine translation in English language teaching. ELT Journal, 78(4), 393–400.

Jiang, L., Wei, B., & Al-Shaibani, G. K. S. (2025). Effective neural machine translation with human post-editing of Chinese intangible cultural heritage corpus into English. SAGE Open.

Ohashi, L. (2024). AI in language education: The impact of machine translation and ChatGPT. In P. Ilic, I. Casebourne, & R. Wegerif (Eds.), Artificial Intelligence in Education: The Intersection of Technology and Pedagogy (pp. 225–242). Springer.

Ren, X., & Wang, R. (2025). Integrating human-AI collaboration into translation education: A comprehensive protocol for assessment, diagnosis, and strategy development. PLoS One, 20(12), e0338089. DOI: 10.1371/journal.pone.0338089.

Slator. (2025). Slator 2025 Language Industry Market Report.

Sun, R. (2024). Evaluating the translation accuracy of ChatGPT and DeepL through the lens of implied subjects. Arab World English Journal for Translation & Literary Studies, 8(4).

The Markup. (2025, April 2). Are AI models advanced enough to translate literature? The debate is roiling publishing.

Woesler, M. (this volume). Appropriateness Theory: An integrative ethical framework for translation studies.

Vaswani, A., et al. (2017). Attention is all you need. In Advances in Neural Information Processing Systems, 30.

Weaver, W. (1949). Translation. Reprinted in W. N. Locke & A. D. Booth (Eds.), Machine Translation of Languages (1955). MIT Press.

Woesler, M. (this volume). Ethical frameworks for AI in higher education: Between European regulation and Chinese innovation.

Woesler, M. (this volume). Learning a foreign language with and without AI: An empirical comparative study.

Zhang, H. (2025, April). Quality Comparison of MT in Translating Chinese Music Historical Texts and Inspiration. In Learning Technologies and Systems: 23rd International Conference on Web-Based Learning, ICWL 2024 and 9th International Symposium on Emerging Technologies for Education, SETE 2024 Shanghai, China, November 26–28, 2024 Revised Selected Papers (Vol. 15589, p. 193). Springer Nature.

Zhang, J., & Doherty, S. (2025). Investigating novice translation students’ AI literacy in translation education. The Interpreter and Translator Trainer. Taylor & Francis.

Zhang, W., Li, A. W., & Wu, C. (2025). University students’ perceptions of using generative AI in translation practices. Instructional Science. Springer.