What is wrong with “just Korean”

One Korean language exists in two literary norms: the Southern 표준어 (based on the Seoul dialect) and the northern 문화어 (based on Pyongyang speech, codified in 1966). Speakers understand each other, but after eighty years of separate development the norms diverged in pronunciation, lexicon and orthography — we unpacked this in “One language or two norms?”.

For a translator this is not theory. “Woman”: the northern norm writes 녀자, the Southern — 여자. “Worker”: 로동자 versus 노동자. “Russia”: 로씨야 in the North versus 러시아 in the South — a difference not of rule but of transliteration source: the DPRK renders Russian pronunciation, South Korea renders English.

The rule that breaks machine translation

The main systemic orthography difference is 두음법칙, the “initial-sound rule”: a Southern-norm rule that softens ㄹ and ㄴ at word onset. The northern norm does not apply it — hence 녀자, 로동자, 리론 (“theory”), 량심 (“conscience”), 령도자 (“leader”). The mechanics are in our 두음법칙 explainer.

The problem: neural networks trained on Southern text apply 두음법칙 automatically. You ask for “woman” — you get 여자. Formally Korean. Practically the Southern norm, which in a text for a North Korean audience reads as foreign.

False friends and myths

Northern lexicon attracts invented “northern words” — recycled around the internet as fake curiosities. Verifiable sources say otherwise: “ice cream” in the DPRK is 에스키모 (hard) or 아이스크림 (soft); “pizza” is 삐짜, a phonetic borrowing like the Southern 피자 but through a different path. “Taxi” is simply 택시, the same word as in the South. The 조선말대사전 dictionary and the Republic of Korea Ministry of Unification portal confirm each of these words with two independent sources. We do not reproduce the fakes even to debunk them: quoting a fake spreads it further.

Myths are dangerous precisely in translation: a Korean-studies specialist or a native speaker spots a fake in thirty seconds, and trust in the text — and in whoever commissioned it — is gone for good.

Loanwords: one object, two words

Another layer is consumer tech and media. “Television” in the North is 텔레비죤, in the South 텔레비전. “Radio” — 라지오 versus 라디오. Small on sight, systemic in substance: the northern norm transcribes foreign words by its own rules, and an “almost right” word reads as foreign.

How to check you actually got the northern norm

Three quick tests for any Russian-to-Korean translation:

  1. The 두음법칙 test: ask for “woman”, “worker”, “theory”. The Southern norm gives 여자, 노동자, 이론. The northern — 녀자, 로동자, 리론.
  2. The “Russia” test: 로씨야 is the northern spelling, 러시아 the Southern.
  3. The address test: “comrade” in a northern context is 동무; if a tool only offers 친구 (“friend”), it works in the Southern norm.

A service that fails all three translates into 표준어, whatever its homepage promises. That is exactly what the KoryoTalk bot is built against: a 2.2M-entry Korean-Russian dictionary base plus a separately collected layer of verified northern correspondences, so northern spellings like 로씨야, 녀자, 로동자 survive the trip.