Ask how Taiwan romanises names on passports and you will be told it uses Wade-Giles, or 通用拼音, or that it has moved to 漢語拼音 like the mainland. The regulation that governs it names none of them.
Article 14 of 護照條例施行細則, the enforcement rules under the Passport Act, is the operative text. It runs to eight numbered paragraphs and specifies two things about the spelling of a name: what script it must be in, and what it must be derived from. Which scheme performs the derivation is left open.
What the regulation actually fixes
The script: 護照外文姓名應以拉丁字母記載 — the foreign-language name shall be recorded in Latin letters, and anything not already in Latin letters shall be converted.
The source: for a first-time applicant with no existing foreign-language name, the rule is 以中文姓名之國家語言讀音逐字音譯為拉丁字母 — transliterated character by character into Latin letters from the national-language reading of the Chinese name.
That is the whole of it. A pronunciation to start from and an alphabet to arrive in. No table, no named system, no appendix.
⚠️ We also read the parent Act itself looking for the missing piece, and found no article dealing with foreign-language names, transliteration or romanisation anywhere in it. That is not the same as proving none exists, but the enforcement rules are where such a provision would live, and they do not contain one.
So where do Wade-Giles and 通用拼音 come from?
From the ministry's own conversion tool, which offers several schemes and lets the applicant pick. That is a service, not a legal instrument. It exists because the regulation requires a transliteration and declines to say how, so somebody had to supply the how.
The received account gets the causation backwards. Taiwan is not using a different system from the mainland. It is not mandating one, and the plurality people observe is the visible consequence of that silence rather than a policy choice between schemes.
The comparison that makes the point
We looked at Macau's rules recently and found the opposite arrangement: a statutory romanisation table, running to hundreds of pages, that fixes the spelling of each character so an official can look it up. Two jurisdictions, both dealing with Chinese names in a Latin alphabet, and one of them legislated a table while the other legislated a starting point.
Neither approach is obviously wrong. The table produces consistency at the cost of never being able to spell a name the way its owner does; the open rule produces the reverse. The received summaries describe both as "the romanisation system used in X", a label that fits Macau and not Taiwan.
What the applicant may actually choose
The regulation constrains the choice in three ways that matter more in practice than the scheme does.
- Existing documents take priority. If you already hold a Latin-alphabet name on a foreign or domestic official identity document, a hospital birth certificate, or a certificate from a properly registered school, that spelling may be adopted ahead of any transliteration.
- Renewals reuse the old spelling. Changing it requires a listed ground — that the existing spelling does not match the national-language reading, which may be used once only; or that it differs from the spelling used by a parent, child or sibling; or that a documented customary spelling exists.
- The whole thing must fit. Surname first, then given name, separated by a comma, and no more than 39 letters including the spaces.
The sibling clause often surprises people. A family transliterating the same surname through different schemes, or in different decades, ends up with passports that do not obviously belong together. The rule solves that by letting family members align, rather than by mandating one scheme for everyone.
Indigenous names are the one place the rule is mandatory
The regulation is permissive nearly everywhere and directive in one place. Where a member of a Taiwanese indigenous people has registered a traditional name written in that people's own script, the rule uses 應 — shall — and requires that script to be the foreign-language name. Where a Chinese name and a traditional name appear together, or for other minority groups and naturalised citizens, the wording relaxes to 得: it may be used.
The regulation declines to name a romanisation system for the majority case and mandates the script for this one. That is a deliberate distinction, not an oversight.
Which dialects count
"National-language reading" carries a definition from elsewhere. The National Languages Development Act defines national languages as 臺灣各固有族群使用之自然語言及臺灣手語 — the natural languages used by Taiwan's long-established ethnic groups, together with Taiwan Sign Language.
Read against the passport rule, that has a consequence people do not expect: Taiwanese Hokkien and Hakka readings are national-language readings, so a name transliterated from them is derived exactly as the rule requires. A Cantonese reading is not, because Cantonese is not the language of one of those groups, and it would have to arrive through the existing-document route instead.
⚠️ That is our reading of a definition applied to another instrument, not a published ruling on Cantonese. Treat it as an argument rather than a legal finding. The definition is quoted above so you can judge it.
What to do with this
If you are applying, the practical order is: check what spelling your existing documents already carry, because that route outranks transliteration and avoids the one-time-change rule later. If you have none, you are choosing a scheme rather than obeying one, and the choice you should optimise is family consistency and the 39-character limit rather than fidelity to any particular romanisation.
If you are building a tool to validate or generate these names, there is no canonical answer to check against. Our passport romanizer implements the mainland rule, which is a genuine rule with an issuing authority behind it; there is no equivalent to implement for Taiwan, and a tool that claims one is describing a ministry's convenience tool as though it were law.
Where this comes from, and what will date it
Article 14 was read in full from Taiwan's official laws database rather than from a ministry explainer page, and the definition of national languages from the National Languages Development Act in the same database. Both are quoted above in the original so the reasoning can be checked rather than taken.
Two limits. We did not open the amendment history to establish when the rule came to say "national-language reading" rather than something narrower, so we cannot say whether the wording that admits Hokkien and Hakka is recent or long-standing — that would change how surprising the dialect consequence is, though not whether it follows. And the claim that the parent Act contains no romanisation provision is a negative result from our own reading, stated as such above.
This dates if the enforcement rules are amended, which happens periodically and is published in the same database. A change to the schemes offered by the ministry's conversion tool would not date it, which makes the point: the tool is not the rule.
