Summary
In the Korean (ko) docs, "X-to-Y" terms (text-to-image, image-to-image, image-to-video,
text-to-model, …) are written in many inconsistent forms. Before fixing it I'd like to check
the preferred direction, since it's a wording choice rather than a clear bug.
The inconsistency
For the single concept text-to-image, the ko docs currently mix at least:
| count |
form |
| 131 |
텍스트-to-이미지 |
| 21 |
텍스트-to-image |
| 12 |
텍스트 to 이미지 |
| 4 |
텍스트-to-Image |
| 3 |
텍스트-to이미지 |
Across all X-to-Y concepts there are ~330 hybrid occurrences in ~44 files. The same pattern
shows up in other concepts too — e.g. image-to-image appears as both 이미지 to 이미지 (20)
and 이미지-to-이미지 (16); image-to-3D as 이미지-to-3D (13, Korean spliced with "3D"); also
텍스트-to-비디오, 이미지-to-비디오, 텍스트-to-모델, etc. Most splice an English "to" between
Korean words, which reads awkwardly in Korean.
This is mostly ko-specific: in ja/zh the same concepts almost never appear as this hybrid —
they keep the English form (text-to-image, Text to Image) or localize fully (e.g. ja
テキストから画像生成). The hybrid is essentially a ko-only artifact of the auto-translation.
Direction
Two reasonable options for ko:
- Keep English (
text-to-image, Text to Image) — matches what ja/zh mostly do, and
T2I is already in preserve_terms.
- Localize fully to natural Korean (e.g.
텍스트 기반 이미지 생성).
I think (1) fits best given the ja/zh precedent and the existing T2I rule, and I have a
ko-only branch ready that restores the English-source form. Mainly I wanted to confirm this
is a change you'd want, and that "keep English" is consistent with how the project wants
locale docs to treat these terms — happy to go with (2) instead if you'd prefer.
(Not proposing a glossary change: to check whether a future re-translation would just bring
the hybrid back, I took the fixed files (hybrids already restored to the English form) and
re-translated them with --force (pnpm translate -- --force <file>) using both
deepseek-v4-pro and qwen-mt-plus. Neither reintroduces the hybrid, so fixing the content
is enough on its own — no preserve_terms/glossary change needed.)
Summary
In the Korean (ko) docs, "X-to-Y" terms (text-to-image, image-to-image, image-to-video, text-to-model, …) are written in many inconsistent forms. Before fixing it I'd like to check the preferred direction, since it's a wording choice rather than a clear bug.
The inconsistency
For the single concept text-to-image, the ko docs currently mix at least:
Across all X-to-Y concepts there are ~330 hybrid occurrences in ~44 files. The same pattern shows up in other concepts too — e.g. image-to-image appears as both
이미지 to 이미지(20) and이미지-to-이미지(16); image-to-3D as이미지-to-3D(13, Korean spliced with "3D"); also텍스트-to-비디오,이미지-to-비디오,텍스트-to-모델, etc. Most splice an English "to" between Korean words, which reads awkwardly in Korean.This is mostly ko-specific: in ja/zh the same concepts almost never appear as this hybrid — they keep the English form (
text-to-image,Text to Image) or localize fully (e.g. jaテキストから画像生成). The hybrid is essentially a ko-only artifact of the auto-translation.Direction
Two reasonable options for ko:
text-to-image,Text to Image) — matches what ja/zh mostly do, andT2Iis already inpreserve_terms.텍스트 기반 이미지 생성).I think (1) fits best given the ja/zh precedent and the existing
T2Irule, and I have a ko-only branch ready that restores the English-source form. Mainly I wanted to confirm this is a change you'd want, and that "keep English" is consistent with how the project wants locale docs to treat these terms — happy to go with (2) instead if you'd prefer.(Not proposing a glossary change: to check whether a future re-translation would just bring the hybrid back, I took the fixed files (hybrids already restored to the English form) and re-translated them with
--force(pnpm translate -- --force <file>) using bothdeepseek-v4-proandqwen-mt-plus. Neither reintroduces the hybrid, so fixing the content is enough on its own — nopreserve_terms/glossary change needed.)