How does China talk about the world? This site tracks the Ministry of Foreign Affairs' press conferences and the Ministry's use of confrontational language towards other countries (2002–2026).
Each country is coloured by how often it appeared in MFA spokesperson answers across the selected years. Drag the two handles to pick a single year or a span — counts are summed across the range — or click Whole register for the full 2002–2026 record. Colour reflects average mentions per year, so the scale stays comparable whatever span you choose.
Note: counts reflect how often a country was mentioned in MFA spokesperson answers, not questions. China is excluded (it appears in almost every answer). Taiwan and Hong Kong are shown separately. Xinjiang and Tibet appear in the data but cannot be rendered as separate map units. Data for 2002–2023 uses NLP entity extraction from the original corpus; 2024–2026 uses keyword matching on question and answer text and is broadly comparable but not strictly equivalent.
This section measures confrontational language in the spokesperson's own answers — instances where China itself issues a threat, makes a demand, urges or requests action of another party, or lodges a formal diplomatic representation. These are markers of China's diplomatic assertiveness, not of pressure on China. Use the tabs to switch between yearly counts, each code's share, confrontational exchanges vs the whole corpus, and seasonality by calendar month. (A separate category, attacks on Chinese nationals abroad, is shown only on the map further down.)
Note: a single exchange can be coded for multiple categories, so bar segments sum to more than the number of distinct exchanges. The dashed line shows the actual exchange count. 注意：单条问答可编入多个类别，因此各类别之和可能超过实际问答条数；虚线表示实际条数。
Click a category to see its definition. 点击各类别查看定义。
MFA press conference — confrontational acts in the spokesperson's spoken answers, by year (2003–2026). "Vs whole corpus" compares the confrontational subset to all ~35,000 exchanges.
MFA — domestic questions — the same MFA answers, but only the exchanges whose question came from a domestic Chinese outlet (mainland state and non-state media — CCTV, Xinhua, Global Times, People's Daily, China Daily, The Paper and the like; Chinese-language Hong Kong outlets such as Phoenix TV are included). It is the same coding of China's own answers, isolated to the friendly, often teed-up questioners. The questioner is only recorded from 2020, so this series begins there; the confrontation rate runs a little above the full record (7.4% vs 5.7%).
State media commentary — the same acts in combined People's Daily commentary columns (钟声, 和音, 国际论坛, 望海楼, 国纪平, plus the English-edition 钟声) and China Daily editorials — 706 confrontational pieces, 2007–2026 (charted from 2019, when the corpus becomes substantial — before then it is only a handful of pieces a year), coded by the validated Sonnet coder (κ ≈ 0.61). The mild "request" tier is excluded (matching the commentary analysis), so it shows five acts.
What "whole corpus" means here. For state media it is not the whole newspaper. It is the 5,082 opinion pieces we scraped from these specific slots — the People's Daily's dedicated foreign-affairs commentary columns (望海楼, 和音, 国际论坛, 钟声, 国纪平; the English-edition 钟声) and all China Daily editorials — so "vs whole corpus" is the confrontational share of those commentary pieces that year, not of everything the papers published. (For the MFA it is every press-conference Q&A exchange.)
Levels are not unit-comparable (spoken Q&A vs whole articles) and the state-media series is far smaller — read each on its own scale; the comparable signals are the shares, the mix of acts, and the timing.
Which countries are most often the subject of exchanges where China issues a threat, demand, urge, or formal representation — or, for "Attacked abroad," where Chinese nationals and interests were targeted overseas? Select a category in the panel to filter the map. Hover a country for a count breakdown.
Countries are coded from the question text (q_loc field) — they reflect which country the exchange was about, i.e. the target or subject of China's pushback (or, for "Attacked abroad," where the attack occurred), not necessarily who initiated the underlying dispute. The United States and Taiwan account for the largest share by a wide margin.
MFA press conference — the spokesperson's spoken answers (2003–2026, ~35,000 Q&A exchanges). A country is counted when it is the subject of an exchange in which China issues a threat, demand, urge or representation; target country from the q_loc field.
State media commentary — combined People's Daily foreign-affairs columns (钟声, 和音, 国际论坛, 望海楼, 国纪平, plus the English-edition 钟声) and China Daily editorials — the 706 pieces flagged confrontational. A country is counted when it is the subject of such a piece, on the same five acts (threat, active threat, demand, urge, representation; the mild "request" tier is excluded, as in the validated commentary coding, κ ≈ 0.61). Non-country targets (EU, NATO, G7) are omitted from the map.
The two are not unit-comparable in level — one counts spoken Q&A, the other whole articles — so read each map on its own scale; the comparable signal is the geography of who China pushes back against. "Request", "Attacked abroad" and "Protested" exist only for the MFA record.
The confrontation coding above is applied to the MFA spokesperson's spoken answers. Here the same six categories — threat, active threat, demand, urge, request, formal representation — are run over China's official commentary: the People's Daily's dedicated foreign-affairs columns — 钟声 (Zhong Sheng), 和音 (Heyin), 国际论坛 (International Forum), 望海楼 (Wanghailou) and 国纪平 (Guo Jiping), the Party paper's assertive "voice on international affairs" (its English-edition 钟声 shown alongside) — and China Daily's institutional editorials. This sets the diplomat's register beside the editorialist's. The corpora differ hugely in size (tens of thousands of Q&A vs a few hundred op-eds), so everything below is a share, never a raw count. Commentary is topic-agnostic — every 钟声/国纪平 piece.
What each share is a share of. The commentary percentages are not shares of the whole newspaper. Each is the confrontational share of the pieces we scraped from specific slots only — the People's Daily's dedicated foreign-affairs commentary columns (钟声, 和音, 国际论坛, 望海楼, 国纪平, and the English-edition 钟声) and all China Daily editorials — 5,082 pieces in total. So a source's n (shown on each card) is that column-set's size, and "32%" means 32% of those pieces confront, not 32% of everything the paper ran. The MFA figures are shares of press-conference Q&A (per exchange) or briefings (per day).
These shares come from the project's validated confrontation coder (Claude Sonnet, codebook v4 — Cohen's κ = 0.61 against a hand-coded gold standard, within the 0.6–0.8 band the project's other coders reach; the mild "Request" tier is excluded as it sits below the confrontation line). An independent audit confirmed the shares are computed correctly. Still, the gaps are not apples-to-apples — read them with four more things in mind:
1. Different units — both MFA bars are shown. "MFA — per exchange" counts each question-and-answer separately (~10 per briefing), so a spokesperson confrontational in one of ten questions still scores low (5%). "MFA — per briefing" counts each press conference as one event — confrontational on 32% of days, far closer to the article-level commentary. The commentary figures count whole articles, so the per-briefing bar is the unit-comparable one.
2. Chinese reads a little hotter than English. The same 钟声 column scores 32% in Chinese versus 25% in its English edition (pd_en) — a modest gap. (The old loose coder exaggerated this to 79% vs 38%; the validated coder narrows it.) Chinese phrasing (必须, 必将, 坚决) still trips the coder somewhat more readily than English, so compare the Chinese PD columns and the English China Daily with that in mind.
3. One column vs many. "People's Daily (all columns)" averages the combative 钟声 (32%) with much quieter columns (望海楼 9%, 国际论坛 3%, 和音 0%) — which is exactly why 钟声 is shown on its own.
4. "Confrontational" lumps mild and severe together. It means any of the five counted acts fired — from a polite urge to an active threat. The Confrontation mix tab below shows sources confront in different ways: the MFA and China Daily most often via threats, 钟声 via demands.
This chart shows each spokesperson's activity across the corpus; the cards below summarise key stats, including the share of their exchanges coded as confrontational.
Each bar shows the breakdown of confrontation-type exchanges (as a share of all confrontational exchanges) for that spokesperson. Hover for counts.
The MFA corpus records which organisations and countries are mentioned in each question, and which news outlet asked it. These charts reveal whose interests dominate the press conference agenda.
Bar length = total question mentions; colour = confrontation rate (darker = higher)
Bars show count; colour shows confrontation rate (darker = higher)
Note: 59% of exchanges have no outlet attribution (recorded as "–" in the original data — common for written questions or when the transcript omits the asker). Western wire services (AFP, Reuters, Bloomberg) prompt confrontational answers more often (9.1%) than Chinese state media (6.8%). Confrontation rate = share of an outlet's exchanges where China's answer was coded Threat, Demand, Urge, or Active threat (or the exchange involved an attack on Chinese nationals abroad).
Browse the flagged records from both sources — MFA press-conference exchanges flagged confrontational and state-media commentary pieces flagged confrontational (People's Daily columns + China Daily editorials). Choose a source, filter by year and confrontation type, or search the text. Each Chinese commentary piece links to a Google-Translate view of the original. Both update weekly.
This site draws on the Chinese Ministry of Foreign Affairs Press Conference Corpus — a structured dataset of MFA spokesperson Q&A sessions from October 2002 to the present, covering over 35,000 individual exchanges.
Geographic locations in spokesperson answers were extracted using named entity recognition and manually validated. Country mentions reflect the text of answers only; China itself is excluded from geographic counts as it appears in almost every response.
Podium vs commentariat. The "same confrontation, on the page" section applies the same confrontation categories to official commentary — the People's Daily's dedicated foreign-affairs commentary columns (钟声, 和音, 国际论坛, 望海楼, 国纪平 in the Chinese edition; the translated 钟声 / "Zhong Sheng" in the English edition; from the people.cn archive), and China Daily's institutional editorials (from chinadaily.com.cn) — coded by a language model on a codebook adapted from the MFA scheme (the six categories that transfer to written editorial voice: threat, active threat, demand, urge, request, representation). Foreign-affairs opinion in People's Daily is concentrated in these dedicated columns; the paper's other editorial genres are predominantly domestic. Because the commentary corpus is far smaller than the 35,000-exchange press-conference record, the two are compared only as shares, never raw counts. Commentary is topic-agnostic — every 钟声/国纪平 piece. The corpora update weekly alongside the press conferences.
Original data source:
Mochtak, M., & Turcsanyi, R. Q. (2021). Studying Chinese foreign policy narratives: Introducing the Ministry of Foreign Affairs press conferences corpus. Journal of Chinese Political Science, 26(4), 743–761.
Data has been updated and coded further by Jamie Gruffydd-Jones.
Spokesperson biographical data:
Villegas-Cruz, A. M. (2026). Who speaks for Beijing? Insights from a new dataset on China's diplomatic elite. Journal of Current Chinese Affairs. https://doi.org/10.1177/18681026261439122
Chinese Diplomatic Elite Biographical Dataset (CDBD). Education, regional origin, career postings, and Party-congress membership in the spokesperson cards are drawn from this dataset.
Status: Preliminary research. Data and classifications are subject to revision.
Related: State media analysis →