People’s Daily, the Chinese Communist Party’s official newspaper, has urged researchers, media organisations, and technology companies to use Chinese equivalents for common English AI terms. The commentary focused on vocabulary such as “agent” and “LLM”, according to reporting by the South China Morning Post in 2025.
The proposal is advisory rather than a technical ban. Its significance comes from the institution behind it: state media is treating AI vocabulary as part of a wider question about public understanding, national standards, and China’s influence over how emerging technologies are described.
People’s Daily Calls for Chinese Equivalents to English AI Terms
The commentary challenges the routine use of English abbreviations in Chinese AI discussion. “LLM” refers to a large language model, while “agent” commonly describes a system that can plan or carry out tasks with limited step-by-step instruction. Chinese technical teams may understand both terms immediately, but general readers may not.
| English term | Meaning in AI discussions | Chinese-language direction |
|---|---|---|
| LLM | Large language model | Use a Chinese equivalent instead of the abbreviation where clarity matters |
| Agent | A system that can plan or execute tasks | Describe the system’s role in Chinese |
| Hallucination | Incorrect or fabricated model output | Prefer a clear Chinese description for public communication |
| Prompt engineering | Designing instructions for an AI model | Translate the concept in documentation and education |
The argument has two parts. First, Chinese terminology can make technical developments easier for non-specialists to follow. Second, a consistent local vocabulary gives regulators, teachers, journalists, and companies a shared set of terms for discussing AI.
The dispute is not simply about replacing a few abbreviations. It asks who gets to name the systems, risks, and capabilities that will shape the next stage of AI policy.
Analysis of the 2025 People’s Daily commentary
Why English AI Terms Have Become Common in China
English entered Chinese AI work through research papers, software documentation, programming libraries, and international developer communities. Technical teams often keep the original term because it matches an API, a code variable, a paper title, or a feature name. Translation can introduce a second label without removing the first one.
Consider “prompt”. A developer may use the English word in code, discuss the same input as a Chinese-language instruction with a product manager, and describe it with a third phrase in customer documentation. The three labels can refer to the same object while serving different audiences.
- Research: papers and benchmark discussions often retain original English terminology.
- Engineering: library names, APIs, and code make direct translation awkward.
- Product work: marketing teams may use familiar global labels to signal compatibility.
- Public communication: English-heavy explanations can make a product seem harder to understand.
That overlap explains why a language campaign faces friction. An official glossary can standardise a term in policy documents, but engineers still need to recognise the English wording used in source code and international research. Should a company choose the official Chinese term, the English term, or both? In practice, the answer may depend on the reader.
The “Discourse Power” Argument Behind the Push
The commentary also reflects a debate about “discourse power”: the ability to define how a technology is understood and governed. Vocabulary does more than label an object. It can frame the problem, suggest which risks deserve attention, and establish the assumptions that guide regulation.
Terms such as “alignment”, “frontier model”, and “hallucination” emerged within English-language AI research and policy discussions. Chinese researchers can use those terms productively while still questioning whether their definitions fit local institutions, users, and regulatory goals.
Standards work makes the issue more concrete. When governments and companies negotiate technical requirements, terminology affects what a rule covers. A translation can clarify a concept for local readers, narrow its meaning, or introduce a different emphasis. The wording eventually appears in procurement documents, safety procedures, educational materials, and product specifications.
China’s push therefore resembles a bid for conceptual control as well as linguistic clarity. A domestic vocabulary lets institutions describe AI without depending entirely on definitions created elsewhere. It also gives Chinese policymakers a way to present local interpretations in international standards and bilingual research.
Concerns Over “Cognitive Dependency” on Foreign Jargon
The strongest version of the argument says that imported jargon can make a research community follow another field’s assumptions. If every new capability arrives with an English label, local researchers may spend more time adopting the existing frame than building a new one.
That concern should not be confused with a claim that translation automatically produces better research. A new Chinese term cannot replace experiments, evaluation methods, or reliable model documentation. It can, however, make a concept easier to teach, debate, and apply across institutions that do not share the same technical background.
The policy question becomes practical when a Chinese term has several possible meanings. “Agent” might describe a software workflow, an autonomous model, or a product feature that calls external tools. A broad translation could improve accessibility while losing technical precision. A narrow translation could help engineers while confusing the public.
For companies, the answer may be controlled bilingual terminology. Internal documentation can pair the official Chinese expression with the English term used in code and international research. Search systems can index both labels. Training materials can explain the relationship rather than forcing employees to choose one vocabulary in every setting.
Localization Is Not New: The “Dian Nao” Precedent
Chinese technology language already contains examples of concepts that became ordinary through native terminology. The word 电脑, pronounced diànnǎo, literally combines the ideas of electricity and the brain to mean “computer”. Its familiar form shows how a technical category can become part of everyday language rather than remaining a foreign label.
AI vocabulary follows a similar path. 人工智能, or réngōng zhìnéng, expresses artificial intelligence through Chinese characters. 大语言模型 describes a large language model, while 智能体 can refer to an intelligent agent. These expressions give readers a semantic clue that an acronym such as LLM does not provide on its own.
The comparison has limits. Older computing terms settled over long periods, while AI products and research methods change quickly. A term that seems clear today may become too broad after models gain new capabilities. Language policy can encourage consistency, but users will still test each expression in classrooms, software interfaces, technical papers, and daily conversation.
What the Push Means for AI Companies
Companies operating in China should watch terminology in four areas:
- Interfaces: feature names should remain understandable without hiding the technical function.
- Documentation: Chinese help pages may need official terms alongside internationally recognised equivalents.
- Training: staff need glossaries that connect Chinese policy language to English research vocabulary.
- Compliance: product and safety documents may increasingly be reviewed through standardised Chinese definitions.
A terminology change does not require a complete rewrite of a product. Teams can start by mapping each English term to its Chinese equivalent, noting where the meanings diverge, and testing both versions with technical and non-technical users. Which label helps a developer debug a system? Which one helps a customer understand its limits? Those answers may differ.
International teams also need a translation layer for contracts, standards submissions, research partnerships, and safety evaluations. A shared glossary reduces the risk that “agent”, “open source”, or “hallucination” carries one meaning in English and another in Chinese. The exercise can expose disagreements early, before they become product or compliance problems.
A Language Policy, Not an English Ban
People’s Daily’s commentary does not remove English from AI research. Chinese engineers still need to read international papers, use software built around English names, and communicate with overseas partners. The near-term change is more likely to appear in public-facing language, official documents, education, and domestic product communication.
The debate will turn on adoption. If Chinese equivalents remain confined to editorials, technical teams may continue using familiar English shorthand. If regulators, universities, platforms, and major vendors adopt the same terms, the vocabulary could move from guidance into normal professional practice.
That makes the next stage worth watching: not whether Chinese AI workers stop saying “LLM” overnight, but whether official translations begin shaping search indexes, model cards, safety frameworks, and the labels users see inside AI products.
Further reading
For a deeper technical perspective, compare bilingual glossaries across model documentation, safety standards, and developer APIs; the differences often reveal where translation ends and policy interpretation begins.