← All Research

Document Translation for the Korean Market in 2026: What Actually Breaks

By Linnk Research Team | August 2026 | 11 min read

Korean sits in an awkward position in most translation tooling: well enough supported that output looks plausible, not well enough supported that the specifically Korean failures get caught. This is a field report on what breaks and how to check for it.

The format problem comes first

Before any translation question, there is a file question. HWP — the format of Hancom Office's 한글 (hangeul) word processor — is the standard document format for Korean government agencies, universities, and a large share of Korean enterprise. Public filings, academic forms, official notices, and internal templates arrive as .hwp or .hwpx.

Almost no international translation tool opens it. The practical workflow is therefore conversion first — usually to PDF or DOCX — and conversion is lossy in the ways that matter: complex tables flatten, embedded objects drop, and the numbered-clause structures that Korean official documents rely on can lose their nesting.

If you work with Korean institutional documents at any volume, your pipeline's real constraint is not translation quality. It is what happens in that first conversion step, and it is worth testing deliberately rather than discovering downstream.

Contraction, and the line-breaking problem underneath it

Korean, like Japanese, generally contracts relative to English. The same layout consequences follow: under-filled text boxes, tables sized for the wrong language, slides that read as unfinished.

The sharper Korean-specific issue is line breaking. Hangul is written in syllable blocks, and Korean does use spaces between words (unlike Japanese), but the word-spacing rules — 띄어쓰기 (ttuieossugi) — are genuinely difficult and frequently applied inconsistently even by native writers. Naive line-breaking algorithms that treat hangul as breakable at any syllable will split a word across two lines. The result is not unreadable, but it is visibly wrong in the way a misspelling is visibly wrong, and it appears most often in exactly the fixed-width contexts where it is hardest to fix: table cells, slide text boxes, form fields.

Machine translation also inherits spacing errors from its source. A Korean source document with inconsistent spacing produces segmentation errors, and segmentation errors in Korean become mistranslations with no visual signal.

Hanja, and where it still lives

Modern Korean is written almost entirely in hangul, which leads many pipelines to assume hanja (Chinese characters) can be ignored. They cannot, in three places:

Legal and administrative documents retain hanja for terms of art, and in older documents for names and dates.

Academic and medical writing uses hanja parenthetically to disambiguate homophones — Korean has many, and the hanja gloss is often the only thing distinguishing two identical-looking terms.

Older scanned material — anything pre-1990s — mixes hanja far more heavily.

For OCR this is a mixed-script recognition problem, and engines tuned for hangul alone misread the hanja or drop it. For translation it is a disambiguation opportunity that most pipelines throw away: the hanja gloss is precisely the signal needed to pick the right sense of an ambiguous term.

Register, again, and differently

Korean encodes politeness and social relationship grammatically through speech levels — the 해요체 / 합니다체 distinction and the honorific infix 시. Like Japanese keigo, this is not stylistic garnish; it is a required choice that the source language does not encode.

Business documents use 합니다체. Marketing copy often uses 해요체. Getting this backwards produces a document that is fluent, accurate, and tonally wrong — a formal contract that reads conversational, or a friendly product page that reads like a legal notice.

The other recurring error is titles. Korean business address is title-based (부장님, 대표님), and translating these into English-style "Mr./Ms." — or dropping them — reads as a breach of basic courtesy in the documents where courtesy is most load-bearing.

Images are the biggest Korean use case

Judged by search demand, the largest single Korean translation need is not documents at all: it is images. Screenshots, photographs of signage and menus, product labels, captured chat threads, and social posts.

That is a different pipeline from document translation, and it has its own failure set: text rendered over photographs, decorative fonts, low-resolution screenshots, and the layout problem of putting translated text back into an image where the original text was sized for hangul.

Treating this as "document translation with an OCR step in front" produces poor results, because the output is an image, not a document.

Getting to a usable result

For digital documents, use a format-preserving document translator rather than extracting text and translating it separately — DocTranslator covers the general case, and for Korean output specifically the localized product is sulsul.ai, built around the Korean format spokes (PDF, 논문/academic papers, 자막/subtitles, PPT, Excel) rather than treating Korean as one target among many.

For scans and photographs there is an OCR stage first, which is where hanja and mixed-script problems live. Scanned.to handles scanned-document translation with layout preserved; ScanRead.ai is the lighter option when you want the text out and nothing else.

For HWP files, convert first and inspect the conversion before translating. That step, not the translation, is where the document is most likely to lose structure.

The review checklist:

• Check the speech level on the first paragraph. Business documents should be in 합니다체 unless there is a reason otherwise.
• Look for words broken across lines in tables and text boxes.
• Verify that numbered clauses kept their nesting through any format conversion.
• On mixed-script sources, confirm hanja terms were read rather than dropped.
• Check titles and honorifics on any document addressed to a named person.

<!-- linnk:faq -->

Frequently Asked Questions

Can I translate an HWP file directly?

Rarely. HWP is Hancom Office's format and few international tools read it. The usual path is converting to PDF or DOCX first — and that conversion is the lossy step, so inspect the converted file for table and clause-structure damage before translating it.

Why is my translated Korean text breaking mid-word?

Hangul is written in syllable blocks, and line-breaking algorithms that treat any syllable boundary as breakable will split words. It shows up most in fixed-width contexts — table cells, slide text boxes — and needs either a Korean-aware line-breaker or manual adjustment.

Does hanja still matter for Korean translation?

Yes, in legal, medical, academic, and older documents. Hanja appears there to disambiguate homophones, and it is often the only signal distinguishing two identical-looking Korean terms. OCR tuned only for hangul tends to misread or drop it.

What is the difference between 해요체 and 합니다체 for documents?

Both are polite, but 합니다체 is the formal register used in business and official documents, while 해요체 is softer and more conversational. Using the wrong one produces a document that is accurate and tonally inappropriate for its purpose.

<!-- /linnk:faq -->