Why do some IMEs treat 「台灣」 and 「臺灣」 as two different words?

Question

Weirdo · Accepted Answer

Two different characters. 「台」 is U+53F0 and 「臺」 is U+81FA. Both read ㄊㄞˊ, but an IME dictionary stores exact strings, so 「台灣」 and 「臺灣」 come up as two candidates.

To a reader, 「台」 and 「臺」 look like variants of the same character. Software does not merge them. Unicode assigns 「台」 U+53F0 and 「臺」 U+81FA; Taiwan’s CNS 11643 maps them to 1-463A and 1-6A57. Cangjie codes differ too: 台 is IR, 臺 is GRBG. Unihan records them as semantic variants and lists 「台」 as the simplified form of 「臺」 — a relationship, not a single code point.

MOE’s Dictionary of Chinese Character Variants files 「臺」 as a zhengzi (正字, A03391). 「台」 is also a zhengzi in its own right (A00496), and, when read ㄊㄞˊ, a variant of 「臺」 (A03391-012). The Revised Mandarin Chinese Dictionary uses 「臺灣」 as the lemma; a headword search for 「台灣」 returns none. That is a usage standard. It does not collapse the two encodings.

An IME phrase table keys on code-point sequences, not on “the name of this place.” Zhuyin for both characters is ㄊㄞˊ, so a phonetic IME can list 「台」 and 「臺」 together — and the two-character strings 「台灣」 and 「臺灣」 as two entries. The open-source Rime schema Phah Tai-bun says so in its README: type tai uan and the candidates are 「臺灣 / 台灣」. Cangjie is the other way around: IR yields 台, GRBG yields 臺.

Search, comparison, and databases follow the same rule: unequal strings are two words. The IME is not making a political choice. Encoding never unified the variants.

Why do some IMEs treat 「台灣」 and 「臺灣」 as two different words?

Two code points, not one word

Dictionaries store strings, not meanings