Skip to content

Converge the Korean README on one speech level, and fix what that exposed - #753

Merged
MongLong0214 merged 3 commits into
mainfrom
readme-ko-humanize
Aug 18, 2026
Merged

Converge the Korean README on one speech level, and fix what that exposed#753
MongLong0214 merged 3 commits into
mainfrom
readme-ko-humanize

Conversation

@MongLong0214

Copy link
Copy Markdown
Owner

Ran README.ko.md through a Korean AI-tell analysis. The honest headline is that most of what it looks for is absent — the document was translated by hand, not by machine — so this targets five named patterns and leaves everything else alone.

risk_band = low · risk_score = 2
double passive 0 · by-passive 0 · have/make literalism 0 · inanimate subject 0
hype vocabulary 0 · hedging 0 · sentence-initial conjunctions 0

A blanket rewrite would have been a no-op or harmful. 42 of 546 lines changed.

What was wrong, in order of weight

The register was mixed three ways — 한다체 97, 합쇼체 64, 해요체 2 — and flipped inside a single bullet list, in the section that lists what this project cannot do:

설치하기 전에 읽어보세요.                        해요체
- …라고 말합니다 / …적습니다 / …않았습니다        합쇼체
- …구현되지 않았다 / …없다 / …이다                한다체   ← same list

That is the section where being believed matters most.

It is now 한다체 throughout, with six exceptions that follow one rule: address the reader and it stays polite, describe the product and it does not. The hero paragraph, the CTA, the contents, two headings and the "read this before installing" line are all direct address. The headings are also anchor targets — changing them breaks the contents links and the parity with ja/zh.

Four other patterns treated, one deliberately not:

  • Connective commas (the trace of English verb-ing, and): 10 → 0
  • Left-branching modifier stacks: three unwound, including the highest-scoring one, where three clauses had to be held before reaching the noun they modified
  • Em-dash appositives followed by a Korean particle: five resolved. Sentence-final dashes stay — those are the author's rhythm, not a translation artifact
  • A가 아니라 B: left alone. It is what walks the document's claims back; the repetition is the point

The verification pass caught three defects I introduced

And one of them was not a style problem.

Rewriting the requireSignedDirective sentence changed a security claim. Unwinding the modifier stack turned an adnominal into a new requirement clause and added a summary assertion (두 가지가 필요하다) the original does not make. Restored.

Restoring it settled an ambiguity the Korean had all along. Its surface reading put the signature in the allowlist alongside the fingerprint. README.md:260-263:

commitlore.requireSignedDirective=true also requires Git's verified signature status from this verifier's trust store and an exact %GF fingerprint in repository-local commitlore.trustedSigner

Two requirements; only the second is in the allowlist. The sentence was mistranslated before this branch touched it, and unwinding the stack is what surfaced it.

Splitting a sentence moved a subject. 읽을 수 있다 went from the reader to the hook. Split again to restore the impersonal reading.

Removing a comma collapsed two promises into one. 제거하지 않으며, 남기는 것을 명시한다제거하지 않고 남기는 것을 명시한다 reads as naming only what is kept. That comma was a syntactic boundary, not decoration — comma removal only works where the following clause is short and the subject does not change.

Preserved

Code blocks, tables, links, anchors, HTML, version pins, every measured figure, issue numbers, protocol terms, headings, and the field-report quote (someone else's words). Heading count is identical to main; the numbers gate and conformance suite both pass.

Known limits

Terminology is still doubled in places (path/경로, host/호스트, wiring/배선). That is translation-consistency work and has to move ja and zh at the same time, so it is not here.

One connective comma survives behind bold markers, where splitting would make the bold span cross a sentence boundary.

…osed

A diagnosis pass over README.ko.md found the register mixed three ways -- 한다체 97, 합쇼체 64, 해요체 2 -- and flipping inside a single bullet list, in the section that lists what this project cannot do. That is the section where being believed matters most.

The register is now 한다체 throughout, with six deliberate exceptions that follow one rule: address the reader and it stays polite, describe the product and it does not. The hero paragraph, the CTA, the table of contents, two headings, and the line that says to read the limits before installing are all direct address. The headings are also anchor targets, so changing them would break the contents links and the parity with the other three translations.

Four other patterns were treated and one deliberately was not. The connective-comma habit -- the trace of English `verb-ing, and` -- went from ten instances to zero. Three left-branching modifier stacks were unwound, including the one the analysis scored highest: three clauses had to be held in mind before reaching the noun they modified. Five em-dash appositives that were followed by a Korean particle are now parentheses or separate sentences, while the sentence-final dashes stay -- those are the author's rhythm, not a translation artifact. The `A가 아니라 B` figure was left alone: it is what walks the document's claims back, and its repetition is the point rather than a tic.

A verification pass then caught three defects this introduced, and one of them was not a style problem.

Rewriting the `requireSignedDirective` sentence turned an adnominal modifier into a new requirement clause and added a summary assertion the original does not make -- in a sentence about a security boundary. Restoring the original wording also settled an ambiguity the Korean had all along: its surface reading put the signature in the allowlist alongside the fingerprint, and the English says only the fingerprint goes there. The sentence was mistranslated before this branch touched it.

Splitting one sentence moved the subject of "읽을 수 있다" from the reader to the hook. And removing a comma after a connective collapsed two promises into one: "제거하지 않고 남기는 것을 명시한다" reads as naming only what is kept. That comma was a syntactic boundary, not decoration -- comma removal only works where the following clause is short and the subject does not change.

Limit: terminology is still doubled in places (path/경로, host/호스트, wiring/배선), and that is translation-consistency work that has to move ja and zh at the same time; one connective comma survives behind bold markers where splitting would make the span cross a sentence
Blast: local
Undo: easy
Certainty: firm
Record-Id: r-kohumanize
Provenance: authored
Verified: the corrected sentence was checked against README.md:260-263, which requires "Git's verified signature status from this verifier's trust store and an exact %GF fingerprint in repository-local commitlore.trustedSigner" -- two requirements, only the second in the allowlist; heading count, version pins and every measured figure are unchanged, and the numbers gate and conformance suite both pass
CommitLore-Version: 2.0.0
@github-actions

github-actions Bot commented Aug 18, 2026

Copy link
Copy Markdown

CommitLore — record lint

Trailers: clean — 3 commits in origin/main..aed6ad071fcbcf7ef0c4e6164f4a33cfb8d084a9
Active constraints: not read — commitlore: the index is incomplete: the build stopped after its time budget with 652 commit(s) or note(s) unread — records in them are missing from this answer. fix: commitlore init (or commitlore index) to finish the index (1 changed path)

Trailer violations fail this check. Active constraints are informational — they are what the repository already decided, not a verdict on this PR.

@MongLong0214
MongLong0214 merged commit 8e9cb09 into main Aug 18, 2026
12 checks passed
@MongLong0214
MongLong0214 deleted the readme-ko-humanize branch August 18, 2026 03:35
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant