Korea mandated xinqi for kimchi in 2021. Google Translate still says paocai in 2026

What happened, and what had already happened before it

2021. South Korea's Ministry of Culture, Sports and Tourism revised its official translation guidelines to designate xinqi (新奇) as the Chinese-language term for kimchi, and Seoul's city government registered it in its official foreign-language dictionary that September. The explicit purpose, per the Korea JoongAng Daily, was to draw a clear line between Korean kimchi and the Chinese pickled-vegetable dish paocai (泡菜), after a 2020-2021 dispute over whether a Chinese regional pickling standard could be applied to kimchi.

13 March 2026. A South Korean lawmaker reported, per fnnews.com, that Seoul's own Chinese-language tourism materials still rendered kimchi stew as "Paocai Tang" and listed the Kimchi Museum as the "Paocai Museum," five years after the city had registered xinqi as the official term. A city public relations official reportedly acknowledged the error and said it would be corrected.

27 to 28 April 2026. Nine Korean outlets, including Yonhap News and Maeil Business Newspaper, reported that Google Translate rendered the Hangeul for Dokdo into Japanese as Takeshima, the name Japan uses for the same disputed islets, and rendered kimchi into Chinese as Paochai rather than xinqi. Professor Seo Kyung-duk of Sungshin Women's University, who has separately flagged similar naming issues on Flightradar24, Naver's AI search summary, and Baidu's map labels, surfaced the Google Translate result with a screenshot. We found no on-record Google response to the April 2026 reports.

Historical precedent, for scale. This is not Google's first Dokdo dispute. In October 2012, Google changed its English-language map label for the islets to "Liancourt Rocks" and said the change followed "our long-standing global policy," not a government request, which Seoul protested at the time. The 2026 Translate incident is a different product, a different error type, and fourteen years later, but the same underlying sensitivity: a geopolitically contested name, resolved differently depending on which audience a system is serving.

The two terms, what they should be, and what actually appeared

The policy term lost even inside its own jurisdiction. The most striking row in Table 1 is not Google Translate's; it is Seoul's own tourism materials in March 2026, five years after the city itself registered xinqi as the official term, still defaulting to paocai. If the policy's author cannot keep its own output consistent with its own rule, a third-party translation product failing the same way one month later is not really a separate incident, it is the same gap surfacing twice.

No source confirms either error has been fixed. We found no Google statement responding to the April 2026 reports and no follow-up confirming whether the Dokdo or kimchi renderings were corrected afterward. Seoul's city PR officer acknowledged the March error and said it would be fixed; we found no later source confirming that it was.

Why a five-year-old policy still loses

This is a judgment, not a measured statistic: we did not obtain Google Translate's training data or a frequency count of "paocai" versus "xinqi" in any corpus, so treat the mechanism below as the most likely explanation given the pattern, not a confirmed fact.

This is the general shape of why cultural and political nuance fails in machine translation: the error is not usually a fluency failure, the sentences read fine, it is a term-selection failure, where the system (or the human writing to a deadline) reaches for whatever term dominates its exposure, and a newly mandated or culturally preferred term rarely dominates anything for years after it is introduced. A translation memory or glossary override solves this at the workflow level; a general-purpose MT engine with no domain glossary applied will not solve it on its own, and neither, evidently, will five years and a government mandate.

What this means for anyone localizing culturally sensitive content

A style guide is not self-enforcing. Seoul's own materials failing its own 2021 policy in 2026 is the clearest evidence in this article that writing the rule down is not sufficient. If a government's own commissioned translation work does not apply its own terminology, a brand's localization vendor will not apply it either unless the glossary is wired into the actual QA step, checked per delivery, not filed once and assumed to persist.

Treat named, contested or policy-sensitive terms as a glossary problem, not a model problem. Switching to a better or newer MT model will not fix a term-selection failure like this one, because the issue is which term the system defaults to, not how fluently it writes around it. A terminology override list, checked at output time against a maintained glossary, is the fix that generalizes.

Re-test known-sensitive terms after every model or vendor change. Nothing in the record suggests Google's April 2026 behavior was new; it may simply be when someone checked and posted it. If you ship a product that translates geopolitically or culturally contested terms (place names, ethnic or national dish names, disputed territories), build a standing test list and run it after every model update, not only when a user complains.

Check it yourself

# 1. Try the two terms in Google Translate yourself, Korean source language
#    - "허혜" (kimchi) -> Chinese: check whether the output is 新奇 (xinqi) or 泡菜 (paocai)
#    - "딩도" (Dokdo) -> Japanese: check whether the output transliterates Dokdo
#      or renders it as 竜島 (Takeshima)
#    Results drift across model updates, so date-stamp whatever you observe.

# 2. The original Korean-language reports, for anyone who reads Korean,
#    dated 27-28 April 2026:
#    https://www.mk.co.kr/news/society/12029062
#    https://www.yna.co.kr/view/AKR20260427098900371
#    https://www.ytn.co.kr/_ln/0103_202604280826232368

# 3. The March 2026 Seoul tourism-materials report:
#    https://en.fnnews.com/news/202603131747205803

# 4. The 2021 kimchi-naming policy background:
#    https://www.koreajoongangdaily.com/korea/seoul-pushes-xinqi-as-chinese-translation-of-kimchi/10426902

What would prove this wrong

The claim under test is that a government-mandated translation term, absent an explicit, maintained override applied at the point of output, will keep losing to the entrenched term in both machine translation and human-produced government materials, for years after the mandate. This is wrong if, by 1 April 2027, repeating the Table 1 test (kimchi into Chinese, Dokdo into Japanese, on Google Translate) returns the officially mandated term consistently across at least five repeated queries, and a check of Seoul's public-facing Chinese tourism materials shows the kimchi stew and museum names corrected. If both checks pass, the specific products examined here fixed the gap; if only one does, the finding holds for the other.

A second prediction, marked as judgment: we expect Professor Seo Kyung-duk or an equivalent watchdog to flag at least one more AI product (search summary, maps, or translation) making the same Dokdo or kimchi substitution before 1 April 2027, based on the recurring pattern already observed across Flightradar24, Naver and Baidu. If no further incident surfaces in that window, that reading was wrong.

FAQ

Why do AI translations fail on cultural nuance?

Because machine translation systems default to the statistically dominant term in their training data, and a government or community's officially mandated term is rarely the dominant one. Korea's Ministry of Culture, Sports and Tourism designated xinqi as the official Chinese translation for kimchi in 2021, specifically to distinguish it from the Chinese dish paocai. On 27-28 April 2026, Google Translate was reported translating kimchi into Chinese as Paochai anyway, and Dokdo into Japanese as Takeshima, the terms the policy exists to avoid. A mandated term can lose to an entrenched one for years after the mandate, because policy does not change what already dominates the corpus.

What exactly did Google Translate get wrong in April 2026?

Per reporting from nine Korean outlets on 27 and 28 April 2026 (Maeil Business Newspaper, Yonhap News, Wow TV, YTN, MBN, Asiae, Yonhap News TV and Kukmin Ilbo), Google Translate rendered the Korean word for the Dokdo islets into Japanese as Takeshima, the name Japan uses for the same territory, and rendered kimchi into Chinese as Paochai, the generic Chinese term for pickled vegetables that Korea's government had tried to retire in Chinese-language use since 2021. Professor Seo Kyung-duk of Sungshin Women's University raised the issue publicly with a screenshot.

Is this the first time Korea's kimchi translation policy has been ignored?

No. In March 2026, a South Korean lawmaker reported that Seoul's own Chinese-language tourism materials, produced by or for the city government that created the xinqi policy, still rendered kimchi stew as Paocai Tang and labeled the Kimchi Museum as the Paocai Museum. A city public relations officer acknowledged the error and said it would be corrected. The same mistranslation the policy was designed to prevent was present in the policy's own home jurisdiction's materials one month before Google Translate's April error was reported.

Sources

1. OECD.AI Incidents Monitor, Google Translate Mistranslates Korean Cultural Terms, Causing Controversy, incident dated 27 April 2026, aggregating nine Korean outlets.
2. Yonhap News, original report, 27 April 2026; Maeil Business Newspaper, original report, 27 April 2026.
3. fnnews.com, report on Seoul's Chinese-language tourism materials, 13 March 2026.
4. Korea JoongAng Daily, Seoul pushes xinqi as Chinese translation of kimchi, on the 2021 policy and its 2021 Seoul dictionary registration.
5. Hurriyet Daily News, Google drawn into South Korea-Japan island dispute, on the October 2012 Google Maps "Liancourt Rocks" episode, for historical context.

Related BLOMEGA guides: Consented AI training data providers.