Zhuyin (注音) is the set of phonetic symbols Taiwan's Ministry of Education dictionaries use for readings, and on a computer each symbol sits on an ordinary key. Leave the input method in English mode and the keys print letters instead, so 你好 comes out as su3cl3, the garble searched for as 注音亂碼. The symbols come back exactly once you know the layout. We compared the two that Microsoft and Apple both describe, which Microsoft lists as Standard and Eten (ET, 倚天). Besides the space bar they share 39 keys, and 37 of those mean a different symbol on each. So a 5 or a 6 in the garble means the standard layout, and an apostrophe or an equals sign means ET.
Each key in English mode prints its letter every time
A Zhuyin layout gives each key one symbol. The open-source input library libchewing publishes these tables; its standard layout file (version 0.12.0, April 2026) calls it the Dai Chien (大千) layout and "the most commonly used one". That is the project's description; we found no survey of what typists in Taiwan use.
On that table s is ㄋ, u is ㄧ, c is ㄏ, l is ㄠ and 3 is the third-tone mark, so su3cl3 reads ㄋㄧˇ ㄏㄠˇ. Nothing is guessed, because on both layouts each key carries one symbol. We typed all 1,337 distinct toned single-character readings we extracted from the Ministry's 《重編國語辭典修訂本》 on both, and every key string but one read back to its own reading; ˙ㄑ, typed 71 on ET, stays a number.
On Windows, Microsoft's page for its Traditional Chinese IME pairs the Shift key with "Switch between English and Chinese mode", and lists four layouts: Standard, HanYu Pinyin, Eten and Taiwan Pinyin. The symbols themselves are in our Zhuyin and pinyin guide.
A syllable always ends on a tone key or the space bar
An input method holds a syllable's symbols until a tone key closes it. The standard layout puts the second, third and fourth tones on 6, 3 and 4 and the neutral-tone dot on 7; ET uses 2, 3 and 4, with the dot on 1. The first tone has no mark, and both layouts type it with the space bar. Apple's ET input method page puts it plainly: 「空白鍵用來輸入一聲」.
So su3cl3 splits at its two 3s, and in Apple's ET example f82te3,x- x92 (繁體中文) the space is the first tone of 中. The neutral dot is written before a syllable and typed after it, so 2k7 on the standard layout is ˙ㄉㄜ, the character 的.
Standard and ET layouts rarely give a shared key the same symbol
ET puts ㄅㄆㄇㄈ on b, p, m and f, as pinyin spells them; the standard layout puts them on 1, q, a and z, down the first column. Apple gives 植物學 as 56j4vm,6 on its 繁體注音 page and ,2x4cuw2 on its ET page. Both contain a comma, ㄝ in one and ㄓ in the other; only the 4 means the same.
Only the standard layout uses 5 and 6, and only ET uses the apostrophe and the equals sign. On ET, 1 and 2 end a syllable as tone keys, while on the standard layout they begin one as ㄅ and ㄉ. A missing tell proves nothing; su3cl3 has none.
Some strings make sense either way. Reading every key string back on the other layout, we found 158 of the 1,337 that also spell a dictionary reading there. The count is the same both ways, since each match pairs the same two readings. ET's xo3, for 我, is ㄌㄟˇ on the standard layout, a reading of 累.
Short ET text sometimes passes for standard text
Symmetric ambiguity still yields one-sided mistakes, because a decoder must break ties. Our Zhuyin keyboard decoder reads text both ways and keeps the layout under which more syllables have characters in its list, built from Unicode's Han database; a tie goes to standard. Of the 1,281 readings the list has characters for, 192 single syllables typed on ET (15.0%) were read as standard, and none typed on standard was read as ET. For two-syllable strings the ET figure fell to 36,031 of 1,640,961 ordered pairs (2.2%). The 12 standard misreads among all 1,337 readings are all missing from the list. Real text favours some readings, so these rates describe arithmetic rather than typing.
Until 21 October 2026 the page named the layout it picked but gave no hint that the other layout also fit. It also called the 56 dictionary readings missing from its list "not a Mandarin syllable", and turned a bare 2026 into "2昂6". The revised page offers the ET reading with a switch whenever that layout finds characters for every syllable, covering all 192. It tells "not a Zhuyin spelling" apart from "no characters in this page's list" and leaves a lone number alone. Paste xo3: it still reads 累, with 我 offered on ET. Paste gjo45l6xk7 for 睡ㄓㄠˊ了, its middle syllable marked as missing from the list.
Doubled-up keys on Hsu and ET26 almost never lose a syllable
Two other libchewing layouts fit Zhuyin onto the letter keys by stacking symbols. On Hsu, 14 of its 25 letter keys carry more than one. The libchewing 0.5.1 source settles a doubled key by position: once an initial or medial is in place, a key carrying both kinds of symbol means "the user wants to input the rhyme".
Typed through our port of those rules, with results checked against the dictionary readings, Hsu returned 1,333 of the 1,337 and ET26 1,334. Each has two key strings that stay ambiguous: on Hsu, aj is ㄘˋ or ㄟˋ, and ej is ㄧˋ or ㄝˋ. Our port assumes keys typed initial first, and the real library does more: version 0.12.0's Hsu file keeps an alternate-syllable table covering both pairs, and the 0.5.1 code checks each syllable against its dictionary, so it may recover some of Hsu's four misses.
Characters are seldom certain even when the symbols are
Getting ㄐㄧ back from the keys is exact; knowing the character is a ranking at best. In the list we built from Unicode's Han database, that syllable carries 66 traditional characters. Unicode's UAX #38 says the kHanyuPinlu frequency field draws in part on a Beijing frequency dictionary first published in 1986. Our homophones guide counts them across Mandarin.
The readings have limits too. The MOE dictionary calls itself 「歷史語言辭典」, a historical dictionary for language researchers, and its licensing page says it takes in 「現代及傳統音讀」, modern and traditional readings alike. It is released under CC BY-ND 3.0 Taiwan, and the Ministry's usage note bars changing a reading or converting it to simplified characters. Our count is credited as the licence asks: 中華民國教育部（Ministry of Education, R.O.C.）。《重編國語辭典修訂本》網址：http://dict.revised.moe.edu.tw/; the copy we used did not record its 版本編號. How Taiwan's readings differ from the mainland's is in our Taiwan readings guide. A count over a historical dictionary can say which syllables exist, but never which character the person who typed su3cl3 had in mind.
