China’s debate over AI language is about more than translation. In August 2026, the Communist Party newspaper People’s Daily urged wider use of standard Chinese equivalents instead of English terms such as “agent” and “large language model”. The argument connects terminology with China’s ability to define technology, develop domestic theories and influence global technology governance, according to South China Morning Post reporting from 2026.
That does not amount to a complete ban on English. The commentary supports a dual-track approach: Chinese terms should lead domestic communication, while English equivalents can remain useful for international research, standards work and cross-border collaboration. Why does that distinction matter? A ban restricts vocabulary; standardisation tries to give institutions a stable vocabulary of their own.
What People’s Daily Is Asking For
The proposal starts with vocabulary. Chinese state media has encouraged writers, educators and technology companies to use established Chinese equivalents for AI concepts rather than importing English jargon unchanged. Examples include zhinengti for “agent”, tishici for “prompt” and shengchengshi rengong zhineng for generative artificial intelligence. These examples appeared in 2026 reporting and commentary about the terminology campaign.
People’s Daily described excessive dependence on foreign terminology as “cognitive dependency”. The concern reaches beyond whether ordinary readers understand a word. If a society borrows the language of a field, who gets to define its categories, risks and acceptable uses?
People’s Daily’s 2026 argument treats language as part of technology governance, not as a cosmetic translation exercise.
The message followed China’s official designation of ciyuan as a standard translation for “token” at the China Development Forum in March 2026, according to the South China Morning Post’s 2026 account. The term combines ci, meaning “word”, with yuan, a character associated with a basic unit or element. In an AI model, a token can represent a word, part of a word or another piece of input processed by the system.
Why English AI Jargon Took Root
English became AI’s working language through research, software and investment networks. Chinese engineers often encounter terms such as “token”, “prompt” and “agent” in technical papers, programming libraries, benchmarks, product documentation and international developer communities. The vocabulary can therefore move from a research environment into a Chinese product meeting without waiting for an official translation.
Borrowed vocabulary also travels quickly. A short English term can move from a research paper to a product presentation, investor briefing or social-media post. That speed supports collaboration, but it can make technical ideas feel distant from people who do not work in AI.
China has localised major technologies before. The expression dian nao, or “electric brain”, became a Chinese term for computer. The example supports the argument that terminology can make an unfamiliar system easier to discuss rather than merely replacing one label with another, as the 2026 commentary noted.
Here lies the practical tension: English preserves a bridge to international research, while Chinese descriptive terms may explain a system’s function more directly to domestic users. Which label helps a buyer understand what the system actually does?
What “AI Discourse Power” Means
AI discourse power means the ability to influence how an emerging field gets described, interpreted and governed. Names do not determine how a model works. They can, however, make particular distinctions visible and allow certain assumptions to enter public debate.
Consider “hallucination”. In English AI discussions, the word usually describes an output that sounds plausible but contains unsupported or false information. A Chinese equivalent might emphasise fabrication, error or unreliable generation. Each choice directs attention toward a different risk: deception, technical failure or the need for verification.
The effect reaches several audiences:
- Journalists decide whether to describe an AI failure as deception, error or unreliable generation.
- Educators choose the concept students must learn before they use an AI system.
- Regulators define which behaviour requires testing, disclosure or human review.
- Companies map product features to procurement rules, standards and liability language.
- Users form expectations about whether a system acts independently, follows instructions or simply produces statistical predictions.
The same issue appears with “agent”. The English word can suggest autonomy, initiative or software acting on a user’s behalf. A Chinese expression may foreground intelligence, task-taking or a system’s functional role. The label changes which feature receives attention first.
A translation can preserve technical meaning while changing the public question: “Can it act?” becomes “What task does it perform, and under whose authority?”
Why Standards Make Terminology Political
Standards depend on definitions that let researchers, vendors, regulators and buyers describe the same capability. If two jurisdictions use different terms for similar systems, bilingual teams must map one vocabulary onto another before comparing safety requirements, procurement specifications or disclosure duties.
That mapping creates friction, but it can also reveal genuine differences. One framework may focus on autonomy; another may focus on human control, accountability or social effects. A single English label can hide those distinctions.
| Term used internationally | Chinese term or framing | Operational question | Why the framing matters |
|---|---|---|---|
| Token | Ciyuan（词元） | What unit does the model process? | Highlights a basic unit used in language processing and gives Chinese documentation a shared reference point. |
| Agent | Zhinengti（智能体） | What tasks can the system perform, and what authority does it have? | Can foreground an intelligent entity or a task-performing system. |
| Prompt | Tishici（提示词） | What instruction or cue does the user give the model? | Emphasises the input that guides generation. |
| Generative AI | Shengchengshi rengong zhineng（生成式人工智能） | What content can the system generate? | Describes artificial intelligence that generates text, images, audio, code or other outputs. |
| Hallucination | Chinese error- or fabrication-focused framing | How should unsupported output be detected and reported? | Moves attention from a human-like metaphor toward verification, reliability and remediation. |
| Compute power | Suanli（算力） | What infrastructure supports training or inference? | Connects model capability with chips, data centres, energy and access to computing resources. |
Terminology also affects interoperability. The European Union’s AI regulatory framework applies defined categories to risk management, transparency and provider duties. The EU AI Act entered into force in 2024, with obligations phased in across 2025 and 2026, so a term can carry operational consequences once it appears in a contract, technical file or compliance assessment.
When a term enters a standard or law, its translation stops being merely linguistic. It starts directing tests, records, responsibilities and remedies.
The Case for Cognitive Independence
China’s argument reflects a wider concern about intellectual dependence. If major AI concepts arrive through English-language research and products, Chinese institutions may appear to follow categories created elsewhere. Developing a native vocabulary signals that Chinese researchers and policymakers can frame the field on their own terms, a purpose connected to “discourse power” in 2026 state-media commentary.
The case goes further than national pride. A locally understood term can help schools explain AI, help officials communicate safeguards and help companies write clearer documentation for domestic users. The value depends on whether the term becomes precise and widely used, not simply whether it sounds Chinese.
Language alone cannot produce stronger models, safer deployment or better research. It can support those outcomes by giving institutions a shared vocabulary for discussing them. A term such as zhinengti becomes more useful when a company defines its permissions, tool access, logging requirements and human-approval points instead of treating the label as a marketing badge.
A Longer Chinese Translation Tradition
Chinese technical language has often combined translation with interpretation. A literal loanword preserves contact with an international field, while a descriptive term explains what the technology does. The contrast between an imported label and a phrase such as dian nao illustrates two different translation strategies: preserve the original sound, or build a meaningful local description.
That choice becomes especially visible with AI because the field changes faster than dictionaries and standards can keep pace. New terms appear alongside model updates, product categories and safety debates.
China’s current push therefore sits between two goals. Domestic users need terms that communicate clearly in Chinese. Researchers and companies still need English equivalents to collaborate across borders, compare papers and participate in international standards work. The dual-track approach attempts to serve both audiences rather than forcing one vocabulary into every setting.
What Changes for AI Companies
Companies operating in China should treat terminology as a product and compliance issue. Interfaces, training materials, contracts and safety reports may need consistent Chinese terminology alongside English technical references. A translation spreadsheet alone will not solve the problem if the product team, legal team and engineers attach different meanings to the same word.
- Create a bilingual glossary. Record the Chinese term, English equivalent, definition, permitted use, prohibited ambiguity and document owner for model features, safety failures, data practices and human-oversight controls.
- Map each term to measurable behaviour. For an “agent”, define the tools it can call, the actions it can take, the data it can access, the limits on its authority and the point at which a person must intervene.
- Separate translation from classification. Ask whether a Chinese term describes a function, a risk category, a legal duty or a marketing concept. These categories should not be treated as interchangeable.
- Check regulatory language. Match product claims with the terminology used in Chinese standards and applicable rules. Record the source and date for every controlled definition.
- Keep international equivalents visible. Engineers, auditors and suppliers need accurate cross-border mappings rather than isolated Chinese labels.
- Test the glossary with users. Give the same feature description to a developer, buyer, compliance reviewer and ordinary user. Where their interpretations diverge, revise the definition or add an example.
For a Bangkok-based team serving Chinese clients, this approach prevents a common mistake: translating a pitch deck while leaving the product, risk register and user documentation inconsistent. Terminology governance works best as a shared operational document.
China’s Chinese AI Terms vs an English Ban
The 2026 reporting points to standardisation, not a blanket prohibition. The People’s Daily commentary supports a dual-track model: English terms can remain useful in international research, while Chinese terms take priority in domestic communication, according to the South China Morning Post’s 2026 report.
| Approach | Domestic communication | International work | Likely effect |
|---|---|---|---|
| Blanket English ban | Chinese terms would be mandatory | Cross-border terminology could become harder to recognise | Maximum linguistic separation, with greater interoperability risk |
| Chinese-first standardisation | Preferred Chinese terms lead public, educational and official use | English equivalents remain mapped and available | Greater domestic consistency without severing international links |
| Uncontrolled borrowing | Teams choose terms ad hoc | English jargon remains easy to share | Fast communication, but more ambiguity for users, buyers and regulators |
That distinction matters. A ban would restrict vocabulary. Standardisation tries to create a preferred vocabulary and a stable mapping between languages. The second approach can strengthen domestic communication without cutting Chinese researchers off from global technical work.
NDTV’s 2026 coverage of China’s AI terminology campaign frames the policy as part of a broader effort involving artificial intelligence, robotics and 5G. The political symbolism matters, but so do the quieter consequences: how a product is documented, how a risk is classified and how a regulator compares one system with another.
The difference between “replace” and “standardise” determines whether readers see this as censorship, localisation or institutional capacity-building.
Frequently Asked Questions About China’s AI Terms
Why does China want Chinese AI terms?
China wants greater control over how AI concepts are explained and governed. Officials and state media connect standard Chinese terminology with “discourse power”, domestic theory-building and reduced dependence on foreign vocabulary, according to 2026 reporting.
Is China banning English AI words?
Available 2026 reporting describes a preferred Chinese vocabulary rather than a complete ban. English equivalents remain useful for international research and collaboration, while Chinese terms are encouraged for domestic communication.
What does “discourse power” mean in AI?
It means influence over the concepts and language used to discuss technology. That influence can shape journalism, education, standards, regulation and public understanding. In practice, the question becomes: which properties of an AI system receive a name, a measurement and a rule?
Why does the translation of “token” matter?
“Token” describes a basic unit processed by many language models. Establishing ciyuan as a standard Chinese term in March 2026 gives Chinese institutions a shared reference point for technical education, documentation and policy discussion, as reported by the South China Morning Post in 2026.
Why does “agent” create translation problems?
“Agent” can refer to several levels of capability. It may describe a model that plans, a software workflow that calls tools, or a system that takes actions for a user. A useful Chinese-English glossary should therefore define the system’s observable actions rather than rely on either label alone.
Should international AI teams use both English and Chinese terms?
Yes, bilingual mapping reduces ambiguity. Teams should define the Chinese term, retain the recognised English equivalent where needed and connect both labels to observable system behaviour. A glossary entry might read: “智能体 / agent: software system that may plan and call approved tools; no external action without human approval.”
Further Reading: AI Language, Standards and Governance
The next question is not which language wins. Ask which vocabulary lets users, engineers and regulators describe a system accurately across borders. Readers comparing China’s terminology strategy with European regulation can continue with the European Commission’s framework for AI rules, while the South China Morning Post’s reporting provides a detailed account of the 2026 People’s Daily commentary.
For practical work, build a bilingual glossary around capabilities rather than fashionable labels: input processing, retrieval, generation, tool use, autonomy, human approval and failure reporting. Add a source date, an example and a testable boundary to every entry; that method keeps the language precise as new AI terms arrive.