Traditional Chinese Output in Mac Dictation, Not Simplified

For Traditional Chinese output in Ducduc, choose Settings → Cleanup → Chinese script → Always Traditional. If you use Apple’s own Dictation and only need to convert existing text, select it and use the Mac’s Chinese conversion service in a supported app. Neither method changes what the recognizer heard or guarantees Taiwanese or Hong Kong wording.

This guide is published by the maker of Ducduc. It separates three settings that are often confused: recognition language, written characters and regional vocabulary. The conversion example is illustrative, not a recording or an accuracy test.

If you use Apple Dictation, try the built-in conversion

Apple documents a way to convert selected text in some apps:

1. Select the Chinese text after dictating.
2. Choose Convert Text to Traditional Chinese in the Input menu, or Control-click and choose Services → Convert Text to Traditional Chinese.
3. If the service is missing, use Services → Services Settings from the app menu and enable the Chinese conversion service under Text.

See Apple’s conversion instructions. If your app does not expose the command, try a compatible text editor and paste the converted text back after checking it.

For ongoing system Dictation, choose the language and region you speak under System Settings → Keyboard → Dictation → Languages. The Dictation language is separate from a Pinyin, Zhuyin or Cangjie keyboard input source. Apple’s Dictation guide describes the speech setting. Adding a Traditional Chinese keyboard is not, by itself, a dependable output-script policy for a third-party dictation app.

Set Ducduc’s output script

1. In Settings → Dictation → Recognition language, use Chinese (with mixed English) for mainly Mandarin speech with English terms.
2. In Settings → Cleanup → Chinese script, choose Always Traditional.
3. Dictate a short sentence containing a character that visibly differs between scripts, such as 学/學 or 发/發, and an English term you use.
4. Compare the final text with the raw transcript in Dictation History. Inspect the final text, not just the live recognition preview.

Ducduc’s Chinese recognition choice requests zh-CN, the mainland Mandarin locale. Its raw transcript can therefore be Simplified even when your writing preference is Traditional. Choosing Traditional changes the final written output; it does not select a Taiwan or Hong Kong recognition locale.

The alternatives are Auto (follow speech) and Always Simplified. Auto retains the recognizer’s script; it does not infer your regional writing preference from your accent.

What a successful conversion looks like

Illustrative raw transcript:

这个 feature 先不要发布。

Illustrative Traditional output:

這個 feature 先不要發布。

Here 这个 and 发布 change script. feature remains English, and “do not publish yet” remains a restriction. The example shows what to inspect; it does not claim that a recording produced either line.

Test a short sentence as well as a longer one. In the reviewed implementation, the final script transform applies even when a short dictation bypasses AI cleanup, cleanup fails, or the built-in quota is exhausted. The raw transcript in History can remain Simplified, which is useful evidence rather than a failed lock.

Is this just an instruction to the cleanup model?

The cleanup model receives a script preference, and the Mac also applies a deterministic character transform to normal dictation output. The final step covers built-in cleanup, BYOK and local cleanup. The built-in service additionally normalizes script server-side.

This describes the normal dictation path. Translation and selected-text voice editing have their own request behavior; their output is not a test of an ordinary dictation setting.

Diagnose the remaining problem

What you see	Likely next check
Raw transcript is Simplified; final text is Traditional	The intended separation between recognition and output
Final text is still Simplified	Selected script, installed build and which operation you used
Traditional characters, wrong word	Recognition or cleanup; use the dictionary guide
Traditional text, unwanted regional term	A vocabulary preference, not merely script conversion
English is translated	Translation setting or cleanup behavior; use the bilingual guide
No recognized text	Speech assets, language and permissions

Script conversion cannot guarantee every preferred variant, proper name or regional expression. For example, converting the characters in 软件 does not establish a Taiwan vocabulary policy that chooses 軟體 rather than 軟件. Review names and ambiguous variants. Add a narrow exact replacement only when you deliberately want that wording in the chosen app.

An already misheard word remains wrong after conversion. Changing the system display language, installing a different keyboard or selecting a cleanup model cannot by itself establish better recognition of that word.

Does conversion require cloud processing?

The Mac’s final character transform runs locally and adds no cloud request. Recognition uses Apple’s on-device speech path. Built-in or cloud BYOK cleanup still sends text to its configured service; choosing Traditional does not change that boundary. See the mode-by-mode data flow.

For same-Mac recognition and cleanup, download the required speech resources and a local Ollama model, use a loopback endpoint, and disable Ollama cloud features. Then select Ducduc Offline mode using the local setup guide. That mode controls Ducduc’s cleanup route, not Ollama’s network access. A .local address can refer to another computer on the LAN.

Try Ducduc with a short mixed-language sentence first. If final output remains Simplified, send feedback with a redacted raw/final pair, selected script, operation, app and build number. Those details distinguish a conversion issue from a recognition or version mismatch.