People’s Daily, the official newspaper of the Central Committee of the Communist Party of China, has urged Chinese media, researchers and technology companies to favour Chinese equivalents over widely used English AI terms. A 2025 report by the South China Morning Post described the commentary and the debate it prompted in China’s technology sector.
The discussion focuses on words such as “large language model,” “LLM,” “agent,” “hallucination” and “prompt engineering.” These expressions remain familiar to engineers worldwide, but Chinese state media is presenting their replacement as more than a translation exercise. The campaign connects language with public understanding, technical education and China’s ability to define AI on its own terms.
What People’s Daily Is Asking For
The commentary calls for standardised Chinese expressions instead of unnecessary reliance on English abbreviations and jargon. It does not mean that Chinese developers will stop reading English research, using English-language code libraries or communicating The immediate target is not to replace English AI terms everywhere, but the language used in public explanations, media coverage, policy documents and domestic technology products. technology products.
That distinction matters. A software engineer may still write “LLM” in code or refer to a model’s API documentation in English. A government document, product interface or news report may increasingly use a Chinese term alongside—or instead of—that abbreviation.
| English term | Common Chinese equivalent | Why the wording matters |
|---|---|---|
| Large language model / LLM | 大语言模型 | Provides a standard descriptive label for a model class. |
| AI agent | 智能体 | Frames an agent as a capable software entity rather than a borrowed label. |
| Artificial intelligence | 人工智能 | Uses an established Chinese term already embedded in policy and education. |
| Large model | 大模型 | Offers a shorter expression common in Chinese technology discussions. |
The dispute is not about whether Chinese engineers understand AI. It is about who gets to name the concepts that shape public discussion and policy.
Why English AI Jargon Took Root
English became the working language of much of modern AI research. Academic papers, software documentation, benchmark descriptions and open-source projects frequently appear in English first. Chinese engineers therefore encounter terms such as “fine-tuning,” “embedding,” “transformer,” “token” and “reinforcement learning from human feedback” in the material they use to build and test systems.
Technical teams also prefer words that connect directly to code. “Prompt” can describe the text passed to a model. “Token” matches the way many systems measure input and output. “Fine-tuning” appears in libraries, tutorials and configuration files. Even when a Chinese translation exists, the English term may remain faster to recognise inside a mixed-language engineering environment.
The vocabulary then moves beyond laboratories. Product marketing, technology journalism, university courses and online tutorials reuse the language of the engineers who build AI systems. Consumers may discuss whether a chatbot “hallucinates,” whether a model is “open source,” or whether an application has an “agent” feature. The terms travel with the products and become familiar before formal Chinese alternatives gain wider use.
- Research: English remains common in papers, documentation and international conferences.
- Engineering: English labels often match code, APIs and software libraries.
- Media: Short abbreviations make technical stories easier to write and recognise.
- Marketing: Familiar global terms can signal access to the wider AI ecosystem.
AI discourse power: the ability to influence how an emerging field is described, interpreted and governed. “discourse power”: the ability to influence how an emerging field is described, interpreted and governed. Names do not determine technology by themselves, but they can shape which distinctions become visible and which assumptions enter public debate.
Consider the word “hallucination.” In AI discussions, it describes an output that sounds plausible but contains unsupported or false information. A Chinese equivalent can emphasise a different aspect of the problem, such as fabrication, error or unreliable generation. The choice affects how journalists explain the risk, how educators teach it and how policymakers describe potential safeguards.
The same question applies to “agent.” The term can suggest autonomy, initiative or software that acts on a user’s behalf. A Chinese expression may foreground the system’s role, intelligence or task-taking capacity. Neither language automatically produces a better definition. The wording determines which features receive attention.
Standards work makes this influence more concrete. Technical standards depend on definitions that allow researchers, vendors, regulators and buyers to discuss the same capability. If terminology differs across jurisdictions, bilingual teams must map one vocabulary onto another before they can compare requirements. That creates friction, but it can also preserve different approaches to safety, accountability and system design.
The Case for Cognitive Independence
The push also reflects concern about what officials and commentators may describe as dependence on foreign technical language. If every new AI concept enters Chinese discussion in English, the argument goes, researchers may inherit not just a label but the assumptions attached to the field in English-speaking countries.
That claim should be treated as a policy argument rather than a proven limit on Chinese research. Chinese engineers can understand a concept without translating its name, just as scientists in many countries use English terminology while contributing original work. Still, a stable local vocabulary can help schools, public agencies and non-specialist users discuss complex systems without relying on unexplained abbreviations.
Translation can also expose gaps in a field. When practitioners search for a precise Chinese expression, they must decide what the term actually means. Is an “AI agent” defined by autonomy, tool use, planning, memory or a combination of those capabilities? Does “alignment” refer to human preferences, policy compliance, model behaviour or all three? Naming forces those questions into the open.
A Longer Chinese Translation Tradition
China has used localised technical vocabulary for decades. The familiar word 电脑 (diànnǎo) means “electric brain” and serves as the everyday term for “computer.” 人工智能 (rén gōng zhì néng) is the established expression for artificial intelligence. 大语言模型 means “large language model,” while 智能体 is commonly used for an AI agent.
These terms make unfamiliar technologies readable through Chinese characters with recognisable meanings. That can support teaching, official communication and search. It also gives institutions a consistent vocabulary for legislation, procurement documents and product guidance.
The precedent does not guarantee that every proposed AI translation will succeed. Some terms spread because they are short, technically precise or already embedded in global software. Others fail because users find them awkward. Everyday practice, not a single editorial, will decide which alternatives survive.
What Changes for AI Companies
For companies serving Chinese users, terminology can affect several layers of a product:
- User interfaces: Feature names may need Chinese labels that explain capability rather than repeat English jargon.
- Documentation: Help centres and developer guides may need consistent Chinese definitions and English cross-references.
- Training: Sales, support and compliance teams may require glossaries for emerging AI concepts.
- Marketing: International terms may remain useful, but domestic campaigns could place Chinese wording first.
- Cross-border work: Teams will need controlled vocabularies that map Chinese policy language to English research and commercial terms.
The likely result is not a clean replacement of English with Chinese. Technical work rarely changes that way. A bilingual model may emerge instead: Chinese terms for public-facing communication and official documents, with English abbreviations retained in code, research exchanges and international business.
A Signal, Not Yet a Complete Ban
The People’s Daily commentary functions as state-media guidance, not necessarily as a technical regulation that imposes penalties for using “LLM” or “agent.” The available reporting describes an appeal for standardisation and wider adoption. Companies should therefore distinguish between a preferred vocabulary and an enforceable requirement.
Even without a formal rule, state-media language can influence how officials, schools, publishers and technology firms frame new products. The impact will become clearer if Chinese regulators, standards organisations or major platforms adopt the suggested terminology in later documents.
For international observers, the episode offers a useful test of AI localisation. When a Chinese technical term appears in a policy document or product announcement, does it translate directly into an English concept? Or does it carry a different view of autonomy, safety or responsibility? Comparing the glossaries—not just the models—reveals how different technology systems are being built around different linguistic assumptions.
Further reading
- South China Morning Post’s report on Chinese AI terminology provides the source context for the People’s Daily commentary and the reaction it generated.