Traditional Chinese output, in Taiwan wording
Traditional Chinese and Taiwan Chinese are not the same thing. A translation can be in correct Traditional characters and still read as foreign to a Taiwanese buyer, because the words are Mainland-standard: 軟件 where Taiwan writes 軟體, 屏幕 where Taiwan writes 螢幕. The parts of that gap a rule can decide — a fixed vocabulary list, spacing between Chinese and Latin, and metric units — Quravin applies as a fixed step on every translation into Taiwan Traditional Chinese, so they come out the same way every time instead of depending on how the model happened to sample that run.
Traditional characters are not the same as Taiwan wording
The writing system and the vocabulary are two separate questions. Convert Simplified to Traditional and you have settled the characters, not the words: 軟件 and 屏幕 are valid Traditional and still read as Mainland to someone in Taiwan, who would write 軟體 and 螢幕. On a product listing or a support reply that is meant to sound local, the wording is what gives it away. This page is about the wording — the part of it a rule can decide — not about character conversion.
Mainland-style in, Taiwan wording out
The table below is captured output, not a mock-up. Each row is a real source text run through Quravin's translation into Taiwan Traditional Chinese, with the result pasted back exactly as it returned. The note on each row says which rule produced the change.
| 入力 | Quravin の出力 | 実証する内容 |
|---|---|---|
| 这款软件的屏幕分辨率很高，净重500g，型号XR-500G，支持USB接口，默认网络设置可用。 | 這款軟體的螢幕解析度很高，淨重500公克，型號 XR-500G，支援 USB 連接埠，預設網路設定可用。 | Mainland-standard Chinese in, Taiwan wording out. Terms on the fixed list are rewritten (軟件→軟體, 屏幕→螢幕, 網絡→網路); a half-width space is added between Chinese and USB; 500g becomes 500公克; and the model number XR-500G is left untouched. |
| Net weight 500g. USB port. Model XR-500G. The software runs on the network. | 淨重500公克。USB 連接埠。型號 XR-500G。軟體在網路上執行。 | English in, Taiwan conventions out. 500g becomes 500公克, a space is placed between USB and the Chinese that follows, the model number XR-500G is preserved, and the wording lands in Taiwan form (軟體, 網路). |
A fixed list of Mainland terms, rewritten every time
A short, high-confidence list of Mainland terms is rewritten to its Taiwan form on every run — 軟件 to 軟體, 屏幕 to 螢幕, and a handful more. It is a fixed list, chosen so that each entry is safe to swap on its own without disturbing a longer word it might sit inside; it is not an attempt at complete Taiwan vocabulary, and a Mainland term that is not on the list is left to the model. A term you have put in your glossary is left as you wrote it, because your wording comes first.
A space between Chinese and Latin
Where a Chinese character runs straight into a Latin letter or a digit, a half-width space goes in: 支援USB becomes 支援 USB, 容量128GB becomes 容量 128GB. The space is added only at that Chinese-to-Latin boundary, so the inside of a model number or a URL is never touched, and a space that is already there is left alone.
Metric units as Chinese words
A metric unit written in Latin after a number is turned into its Chinese word: g becomes 公克, mm becomes 公釐, ml becomes 毫升, cm becomes 公分. A number and its Chinese measure word are treated as one token, with no space between them, so 500 公克 is closed up to 500公克. Ratings that stay Latin by convention — V, W, GB, %, °C — are left as they are, and a code such as XR-500G is never read as a measurement.
The same input gives the same output
Because these rules are applied as a fixed step rather than left to the model's sampling, the same text produces the same result on every run — including when the answer is served from cache. A listing corrected last month reads the same when it is regenerated this month. That repeatability is the reason for doing this with rules instead of a prompt.
Your approved translations are never rewritten
These rules act on machine output, not on wording a person has signed off. When a translation has been approved as an override, it is returned exactly as approved, byte for byte, with none of these rules applied on top. An approval you make to fit your own house style therefore holds, even where it differs from the default Taiwan rule.
What this does not do
These rules cover the mechanically decidable part and stop there. Phrasing and tone are not guaranteed; they stay with the model. Mainland terms outside the fixed list are left to the model too. Spacing is added only between Chinese and Latin, never inside a Latin run: 容量128GB is spaced to 容量 128GB, and the unit is not pushed away from its number. A price written with a thousands separator only closes up on the right: 售價 1,000 元 becomes 售價 1,000元, and the space on its left is kept. Where wording has to be locked, put it in your glossary; where it has to be permanent, approve it as an override. Engine default, glossary, override: three layers, each firmer than the last.
よくある質問
Does this convert Simplified Chinese to Traditional?
No. This step works on wording, not on the writing system. It assumes the output is already Traditional and adjusts the parts of the wording a rule can decide: vocabulary on a fixed list, spacing, and units. Choosing Traditional Chinese as the target language is what produces Traditional characters.
Which Mainland terms get corrected?
A fixed, high-confidence list — the terms that are safe to swap on their own without disturbing a longer word. 軟件 and 屏幕 are on it. It is deliberately short rather than complete; a Mainland term that is not on the list is left to the model, and you can pin any wording you need with a glossary entry or an approved override.
Will it touch my model numbers or SKUs?
No. The spacing rule acts only at a boundary between a Chinese character and Latin text, so a code with no Chinese inside it — XR-500G, AB-12-500 — is never split. The unit rule skips anything following letters or a hyphen, so a 500G inside a model number is not read as a weight. The rules are deliberately cautious here: a broken product code costs more than a missing space.
Can I make it always use a specific term?
Yes, in two ways. Put the term in your glossary and it is carried into the run with your wording and left untouched by these rules. Approve a translation as an override and that exact text comes back on every later run, byte for byte, with no rule applied on top. Use the glossary to steer wording, and an override to lock it.