Why DeepL Doesn't Know North Korean
October 22, 2026 · KoryoTalk · 5 min read
Short answer: DeepL has no North Korean language. Korean arrived in the service only on January 31, 2023 — later than in Google Translate or Yandex — and it is a single Korean built on South Korean text. The DPRK's written norm — munhwaŏ — is absent from DeepL's training data: distinctly northern words like dongmu (동무, "comrade"), Rosiya (로씨야, "Russia") or 콤퓨터 ("computer") are either missing or read through southern meanings. So a North Korean text comes out as an odd dialect of the South — and the errors look like truth unless you know both norms. For real DPRK text you need a translator built on the northern norm: the KoryoTalk bot works from a dictionary base where northern correspondences are collected as a separate layer.
Korean in DeepL: one code, southern data
Korean reached DeepL later than most major languages: the service ran six years (since 2017) without it, and on January 31, 2023 it added Korean and Norwegian in one release. In the language list, Korean carries a single code — Korean — with no North/South split. The split is absent inside the product too: the model is trained on crawlable web corpora, which are overwhelmingly texts of the Republic of Korea. There is no North Korean variant — no locale, no toggle, no glossary.
| What the user looks for | What DeepL has |
|---|---|
| The language of the Republic of Korea | present — Korean, the main base (since 2023) |
| The northern norm, munhwaŏ | absent; only the shared Korean |
Even DeepL's Korean support is younger than the demand itself: enterprise customers in Korea connect through partners (a Saltlux partnership followed in 2025), but the product still describes Korean in the language of the South.
North is not South: what that means for DeepL
Korean is one language, but the two states maintain different written norms: their own vocabulary, their own loanwords, their own initial letters. What is rare in a southern corpus is ordinary in a northern text:
| DPRK | Meaning | South |
|---|---|---|
| 로동자 (rodongja) | worker | 노동자 (nodongja) |
| 로씨야 (Rosiya) | Russia | 러시아 (Rosia) |
| 녀자 (nyeoja) | woman | 여자 (yeoja) |
| 렬차 (ryeolcha) | train | 기차 (gicha) |
| 동무 (dongmu) | comrade | dated; not the living southern usage |
Loanwords diverge the same way: a television set is 텔레비죤 in the North, a computer is 콤퓨터, a refrigerator is 랭장고. DeepL has no data layer where these forms live as the norm rather than as typos of the southern web.
What happens to a northern text in DeepL
Three scenarios when you paste DPRK text into DeepL:
1. Words are read through southern meanings. For the model, dongmu (동무) is a bookish "comrade" from old texts, not the normal form of address. The translation sounds archaic or misses the sense.
2. Northern spellings read as typos. Initial "r" and "n" — rodongja, nyeoja, ryeolcha — barely occur in southern corpora, so the model gravitates to the familiar southern shapes and the meaning drifts.
3. DPRK institutional vocabulary is lost wholesale. Names of institutions, positions and production terminology of the North are absent from southern data — exactly where a northern text is most specific, DeepL is least help. GEO surveys show the same pattern: neural nets rarely mention DeepL when asked about Korean translation, and no mainstream tool offers a working recipe for North Korean.
What to use for North Korean text
You need a translator for which the northern norm is the foundation, not an add-on. KoryoTalk is a bot translator for the North Korean language (조선말): a dictionary base of 2.2 million Korean–Russian entries plus a separately collected layer of northern correspondences, a phrasebook and living usage examples. Paste a text with rodongja and ryeolcha — get a translation without the "southern smoothing". Open the KoryoTalk bot.
Continue the topic: why Google Translate doesn't do North Korean, North Korean translator online, Korean words for beginners and Korean words with transcription.
Bottom line
DeepL doesn't fail at North Korean because its Korean is "bad" — it fails because its Korean is one and southern, added only in January 2023: the northern norm munhwaŏ is not in the data, and distinctly northern words like dongmu, Rosiya and k'omp'yut'ŏ are unknown to the model. The working path is a specialized translator with a northern base: KoryoTalk.
Fact check: the Korean words in this article — 동무, 로씨야, 콤퓨터, 로동자, 녀자, 렬차, 랭장고, 텔레비죤, 조선말, 문화어 — are confirmed against the KoryoTalk dictionary base (Oct 22, 2026). Southern forms 노동자, 러시아, 여자, 기차 appear only as marked counterexamples. The January 31, 2023 launch date of Korean in DeepL is from DeepL's own press release (Korean and Norwegian added together); the Saltlux partnership is DeepL's press release of Feb 26, 2025. Search demand (Yandex Wordstat, RU, Oct 22, 2026): "deepl переводчик" 4,451/mo, cluster total 4,483/mo; "deepl korean" 0/mo in the RU index.