Beijing puts artificial intelligence at the centre of Chinese learning
Nearly 5,000 participants from more than 160 countries and regions gathered in Beijing for the 2025 World Chinese Language Conference, where new tools promised to change how teachers prepare lessons and how students practise Chinese. The announcements included a knowledge graph containing 380,000 language nodes and about one million semantic relationships, an adaptive assessment product called HSK GO, and a smart classroom supported by a virtual teacher named Xiao Yu.
Contents
- Beijing puts artificial intelligence at the centre of Chinese learning
- How a knowledge graph connects the language
- Assessment that adjusts to each learner
- Lesson preparation and a virtual classroom teacher
- Pronunciation practice already has a working model
- What research says about sustained use and confidence
- Teachers need preparation as well as software
- Apps extend practice beyond the classroom
- A Saudi pilot combines student practice and teacher tools
- International partnerships connect language with practical skills
- Technical limits and changing language careers
- Schools can begin with a specific learning problem
- What to Know
The conference, held under the theme "Innovation Leads, AI Empowers: Learning Chinese Without Borders," presented artificial intelligence as a practical classroom resource moving beyond pilot projects. Its central ambition was to make Chinese learning more accessible, better matched to individual ability, and more closely connected to teaching needs.
Chinese Vice Premier Ding Xuexiang said China would continue supporting international Chinese language education and expand partnerships connecting language study with practical skills. He called for smart campuses, smarter classrooms and wider sharing of digital resources to make learning more convenient and efficient.
Interest in Chinese is growing through trade, study and cultural exchange. Schools seeking to meet that demand face difficulties hiring qualified teachers, developing curricula and providing enough speaking and listening practice. The technologies introduced in Beijing address different parts of that problem: organising teaching content, measuring ability, preparing materials and providing practice.
The position as of November 15, 2025, also includes a wider collection of classroom applications, consumer learning services and research on student motivation. Together, they show that software performance, teacher preparation and the learning environment all matter when a new tool moves into everyday use.
How a knowledge graph connects the language
The conference's knowledge graph brings together international Chinese education standards through 380,000 language nodes and about one million semantic relationships. A node represents an item in the system, while a relationship connects it to another item. In plain language, the graph is a map of how parts of Chinese fit together.
Characters, words, grammar points, topics and cultural references can be linked across texts and curriculum standards. When a learner has difficulty with a grammar pattern or a pairing of tones, software can use those connections to locate related examples and exercises aimed at that difficulty.
The companion education corpus collects large volumes of language data for multimodal searches and detailed analysis. A corpus is a collection of language material that can be searched and studied. Multimodal means working across different forms of information, including text, audio and, in some cases, images or video.
A teacher preparing a lesson about food, for example, could retrieve dialogues suited to a particular level and containing appropriate tone patterns. The same search could find menus with Chinese characters and pinyin, the system used to show pronunciation, along with authentic clips matching students' proficiency.
Assessment that adjusts to each learner
HSK GO uses AI to tailor assessment paths to individual learners. HSK, short for Hanyu Shuiping Kaoshi, is the standard Chinese proficiency test. Adaptive assessment changes the difficulty of questions as a student answers, helping build a more accurate picture of current ability.
The system connects assessment with curriculum standards and previous performance. It can suggest the next tasks, identify material needing review and estimate progress towards particular HSK bands. The assessment therefore becomes a guide to further practice as well as a measure of what someone already knows.
For teachers, information about performance can reveal where a class is making progress and where students are getting stuck. For learners, a pace matched to their ability and regular feedback can help build confidence. These functions address a practical classroom challenge: students following the same course can still need different kinds of help.
Lesson preparation and a virtual classroom teacher
An AI cloud platform introduced at the conference was trained with data from 3,000 Confucius Institutes. Teachers select textbook parameters, and the platform can generate lesson plans and slide presentations in seconds. Its intended benefit is less time spent on routine preparation and more time available for feedback, conversation practice and culture.
Xiao Yu, the virtual teacher in the smart classroom model, supports voice synthesis, speech evaluation and intelligent assessment. Voice synthesis produces spoken audio, while speech evaluation assesses a learner's spoken response. The combination offers instruction and opportunities to practise within the same classroom setup.
The model is designed for schools lacking qualified Chinese teachers. Local staff would supervise students and maintain community engagement, while the virtual system supplies consistent instruction and practice. Human supervision remains part of the proposed arrangement.
Across these tools, the educational aim is to combine precise practice and content delivery with a teacher's understanding of context and motivation. Culture, humour, appropriate language choices and genuine communication continue to depend on human interaction. Automated material generation also gives teachers more room to concentrate on those parts of a lesson.
Pronunciation practice already has a working model
Chinese Testing International, the organisation behind HSK, has built pronunciation training using Azure AI Speech. Its Learn Pronunciation to HSK application gives immediate feedback on individual sounds, tone contours and the rhythm of a whole sentence. Tone contours describe how the pitch of a spoken tone changes.
The application takes learners through repeated practice until they reach a set threshold. Its content includes words, phrases, dialogues, tongue twisters and classic poems, giving students experience with everyday language and literary styles rather than a single type of speaking exercise.
Chinese Testing International has established more than a thousand test centres in over 160 countries. It reports that AI speech scoring has made assessment more consistent, saved time and supported a complete cycle of teaching, practice, testing and assessment.
Immediate pronunciation guidance can help students work on tones and finals, the ending sounds within syllables. Teachers can then use assessment information to focus instruction where it is needed. The application illustrates how speech technology can connect independent practice with the broader work of a language programme.
What research says about sustained use and confidence
A study of learners at Confucius Institutes in Southeast Asia examined why students keep using AI chatbots for Mandarin practice. The research combined frameworks concerned with technology acceptance and motivation, identifying several factors associated with continued engagement.
Performance expectancy means believing that a chatbot actually helps someone learn. Effort expectancy concerns how easy the tool is to use. Reliable networks and technical support provide the conditions needed for regular practice, while encouragement from teachers and peers can strengthen motivation and satisfaction.
The practical lessons are specific: match chatbot tasks to course goals, keep interfaces simple, provide technical help and adapt prompts to a learner's level and interests. A conversation exercise becomes more useful when it supports the material a student is studying and is easy enough to begin without repeated assistance.
Research involving university students learning English in China examined the emotional effects of AI feedback. Corrective and motivational responses encouraged reflection on learning and supported creativity in writing and speaking. Familiarity with AI tools and the way feedback was delivered affected anxiety.
Carefully designed feedback also increased learners' resilience when they encountered difficulties. These findings are relevant to Chinese instruction, where learning tones, characters and appropriate language choices can feel intimidating. Timely guidance and encouragement can help students continue practising through those challenges.
Teachers need preparation as well as software
Educators working with AI in language instruction continue to place the relationship between teacher and learner at the centre of the classroom. AI can provide adaptive practice for many students, while a human guide helps explore cultural nuance, language choices suited to different situations and values.
Experienced teachers also shape the emotional atmosphere of a class, affecting students' persistence and progress. Their work includes helping learners remain motivated, understand context and build relationships through language.
Recent work on an AI readiness scale for teachers of Chinese as a foreign language identifies three areas: personal assets, beliefs about value and cost, and contextual resources. Personal assets include technological knowledge for teaching and willingness to try new tools. Value and cost beliefs include expected benefits and concerns such as the time required.
Contextual resources cover institutional support and the conditions enabling teachers to use technology. Together, these factors influence satisfaction with AI and the intention to continue using it. Training, peer support and practical guides therefore form part of adoption, alongside the software itself.
Apps extend practice beyond the classroom
Consumer applications offer learners more opportunities to practise outside scheduled lessons. Conversation platforms allow users to speak with AI that adjusts to their level and topic, then provides immediate feedback. Some services use voice models built from native speakers to create a stronger sense of a real conversation.
Reviews describe strengths including flexible role plays, feedback reports and vocabulary carried across different practice modes. They also identify problems: occasional mishearing of tones, voices that sound robotic and limited game elements. Learners may need considerable personal motivation to maintain a study streak.
Applications built around tutoring combine structured curricula with short sessions. One popular service offers eight levels across 170 topics, with hundreds of lessons lasting ten minutes. Its embedded AI acts as a personal tutor, explaining errors, suggesting better phrasing and responding to grammar and word choice in real time.
These tools cover pinyin, character structure, essential HSK vocabulary and everyday expressions. Native speaking teachers design the applications, while AI expands access to coaching and adjusts to learner answers.
Talkmate's International Chinese Education AI Agent connects language practice with cultural exploration. It presents AI avatars of more than 100 cultural figures and supports conversations in six major languages, with plans to expand. Advanced learners can discuss poetry, history and philosophy, while others can begin with modules comparing familiar subjects with Chinese contexts.
The platform aims to make practice personal and engaging while preparing learners for HSK examinations through more than 5,000 guided lessons. Its cultural conversations extend the learning experience beyond vocabulary lists and pronunciation drills.
A Saudi pilot combines student practice and teacher tools
HanLink underwent a pilot lasting four weeks with 500 students in Saudi Arabia. The pilot reported homework accuracy of 80 percent, an average oral score of 75 out of 100 and regular independent practice of 15 minutes a day.
The product includes pronunciation correction in real time, analysis of character writing, more than 180 conversation scenarios and continuous AI tutoring. Teacher tools provide automated grading and dashboards that track progress across a class and for individual students.
A company statement about a related product launch described its ambition to connect AI learning with cultural exchange and support students in Saudi Arabia:
HanLink marks a transformative leap in language education. By integrating AI with cultural exchange, we are proud to support Saudi Arabia's local students.
The company plans to expand the programme after the pilot. The reported figures describe students' performance and practice during that period; the programme combines tools for learners with information teachers can use to follow their progress.
International partnerships connect language with practical skills
Ding Xuexiang encouraged international Chinese education programmes to connect with vocational and professional training. The goal is to help learners use Chinese in workplace settings, linking language study to practical tasks and skills.
He also called for more partnerships between Chinese and foreign schools, more student exchanges and wider access to competitions such as Chinese Bridge. Governments, universities and organisations were urged to share digital resources and develop smart classrooms connecting learners at different locations.
International guests from Chile, Thailand, Egypt and Uzbekistan described Chinese as a bridge for relationships in trade, culture and education. They welcomed AI's potential to widen access and improve equity while affirming the value of direct dialogue between people.
Machine translation can support communication, but personal exchanges retain a role of their own. The conference's policy agenda connected digital access with school partnerships, exchanges and opportunities to use Chinese with other people, rather than limiting cooperation to the distribution of software.
Technical limits and changing language careers
Speech recognition can misjudge tones when recordings contain noise or when an accent is unfamiliar. AI voices can sound flat, weakening the feeling of a real exchange. Poorly designed assessment systems may also direct students towards material that is too easy or too difficult.
Schools need clear policies for student recordings and performance data, including where information is stored and processed and who can access it. Data privacy and security require transparent handling, especially when speaking practice produces recordings that can be retained by a service.
Internet connections, microphones and quiet places to practise are unevenly available. Without teacher training and technical support, tools can remain unused or create more work. Budgets need to cover subscriptions, equipment and training hours, as well as the technology being introduced.
AI is also changing language employment. Translation tasks previously handled by junior staff are increasingly performed by AI with high accuracy at low cost. Chinese universities have already reduced some foreign language majors, and graduates report fewer jobs focused directly on language in trade, translation and basic tutoring.
Demand is shifting towards language combined with law, business, data or area studies. Human expertise remains valuable in work requiring creativity, judgement and collaboration with technology. Programmes connecting Chinese with these subjects fit the conference's wider focus on professional skills.
Schools can begin with a specific learning problem
A programme introducing AI can start by identifying the task it needs to improve: beginner tone accuracy, consistent speaking assessment across classes or faster preparation of reading materials suited to different levels. Tool selection can then follow the curriculum, whether that means practice aligned with HSK or content matching a school's own syllabus.
A small pilot allows staff to review performance, engagement and satisfaction before expanding use. Simple interfaces help students start quickly, while teacher workshops can cover technical setup and classroom practice. Chatbot conversations work best when linked to lesson objectives.
Clear explanations and encouraging feedback provide emotional support, helping reduce anxiety and build resilience. Programmes also need to document data practices and secure consent, particularly for voice recordings. Sharing successes and difficulties across departments can help improve implementation and support new users.
The conference presented a broad toolkit, from connected language standards to virtual instruction. Its practical use will depend on how those tools fit lessons, how teachers are supported and whether students can access regular, purposeful practice.
What to Know
- The Beijing conference drew nearly 5,000 participants from more than 160 countries and regions.
- New tools included a knowledge graph with 380,000 nodes, an education corpus, HSK GO, an AI lesson platform and the virtual teacher Xiao Yu.
- Ding Xuexiang called for digital resource sharing, school partnerships and links between Chinese learning and professional training.
- Chinese Testing International uses Azure AI Speech for pronunciation practice and reports more consistent scoring and time savings.
- Research connects continued chatbot use with usefulness, ease of use, technical support and encouragement from teachers and peers.
- Teachers remain central to cultural understanding, motivation and appropriate communication.
- Talkmate combines cultural avatars with language practice; HanLink's Saudi pilot involved 500 students over four weeks.
- Speech recognition errors, data protection, uneven access and training costs remain practical constraints.
- AI is changing translation work, increasing the value of language skills combined with other professional knowledge.
- Targeted pilots, curriculum alignment, teacher training and clear consent procedures provide a practical starting point.






