Chinese, Japanese & Korean

Chinese and Japanese aren’t typed directly: you type a reading (pinyin, bopomofo, a radical code, romaji) and pick the character it means from a candidate bar. Korean is different: Hangul jamo assemble into syllable blocks on their own, so it types and suggests like any other language. This page covers both.

Chinese Pinyin: the candidate strip replaces the usual word-suggestion row while you're composing a reading.

Two different input flows

Conversion input methods (Pinyin, ClearFlow Pinyin, T9 Pinyin, Zhuyin, Cangjie, Cangjie Quick, Stroke for Chinese; Jyutping, Yale and Cantonese Pinyin for Cantonese; Romaji, Godan, Flick, Kana JIS and 50 keys for Japanese) hold what you type in a composing buffer and replace the entire suggestion strip with ranked character/word candidates. Nothing commits until you pick a candidate, press space, or otherwise flush the buffer.

Korean types the other way. HangulComposer composes jamo directly into syllable blocks (ㄱ + ㅏ + ㄴ → 간) with no reading step and no candidate list. There’s nothing to convert unless you want Hanja, which the Korean options can add to the strip. Five keymaps ship: the standard two-set (두벌식) grid, where one key is ㄱ whether it starts or ends a syllable and the composer decides by position, the two three-set (세벌식) grids, 390 and Final, whose keys say for themselves whether they are an initial or a final, and the two 12-key pads, 천지인 (Cheonjiin) and 나랏글 (Naratgeul). On a three-set grid a tense initial is typed by pressing the plain one twice (ᄀ ᄀ → ㄲ), and a vowel after a final never pulls that final into the next syllable. Syllables build the same way in the keyboard’s own text boxes, such as the emoji, GIF and clipboard searches, Find and replace, and the AI prompt: backspace there takes back the last jamo rather than the whole block. The same goes for the other keyboards that build letters from keys, Vietnamese Telex and VNI and Khipro. Korean typing gets the ordinary word-suggestion and autocorrect strip instead, backed by a real downloadable dictionary (see Downloadable dictionaries). If you’ve used a Chinese or Japanese IME before, don’t expect Korean to work the same way. It’s a direct phonetic keyboard.

Hanja

Korean can offer the Hanja for the word you are typing. Open Korean options on the language’s screen, download the Hanja (漢字) dictionary (about 7 MB, 300,000 entries) and turn on Offer Hanja. It’s off by default, since most Korean text has none.

The choices sit behind the first word on the strip. A tap writes one in place of the Hangul with no space after it, and nothing is learned from it. An ending stays in Hangul: 학교에 offers 學校에. With few slots on the strip you see only the first few, so turn on the scrolling suggestion strip to reach the rest.

Fix spacing

Selecting Korean text can offer Fix spacing, which corrects the spaces between words: 나는학교에간다 becomes 나는 학교에 간다. It runs on the phone with the Kiwi analyser, and nothing leaves the phone. It needs two downloads from Korean options: the Spacing correction library (about 18 MB) and the Spacing correction model (88 MB, about 100 MB unpacked), and the action stays out of the selection row until both are there. The first use in a while takes a few seconds, since the model opens and takes hundreds of megabytes of memory, and it’s closed again a minute after you stop. Only phones with a 64-bit processor can run it, and it isn’t in the F-Droid build.

천지인 (Cheonjiin) is the phone keypad, laid out the way Samsung’s is. Vowels are drawn from three strokes, ㅣ, ㆍ and ㅡ, in the order you would write them: ㅣ ㆍ is ㅏ, ㆍ ㅣ is ㅓ, ㆍ ㅡ is ㅗ, ㅡ ㆍ is ㅜ, a second dot gives the y-vowel (ㅣ ㆍ ㆍ is ㅑ), and a ㅣ after a vowel gives ㅐ, ㅔ, ㅚ, ㅟ and the rest. A dot that isn’t part of a vowel yet waits on screen as ㆍ or ‥, and a third dot goes back to one. Each consonant key holds two or three letters, picked by tapping it again: ㄱㅋ once is ㄱ, twice ㅋ, three times ㄲ. To type the same consonant twice in a row (the ㄱ ㄱ in 먹고), either pause for a second before the second tap or press Space between them. A space straight after a consonant tap only ends that letter and types nothing; press it again for a real space. The .,?! key cycles through its four marks the same way. Delete takes back one tap: a whole consonant, or one stroke of a vowel (ㅏ back to ㅣ).

나랏글 (Naratgeul) is the other phone keypad, laid out the way Samsung’s is. It has keys for six consonants, ㄱ ㄴ ㄹ ㅁ ㅅ ㅇ, and for the vowels ㅏㅓ, ㅗㅜ, ㅣ and ㅡ. Tap ㅏㅓ once for ㅏ and again for ㅓ, and ㅗㅜ the same way. Every other letter is built from these with two keys that change the last letter you typed. 획추가 adds a stroke and goes round: ㄱ ㅋ, ㄴ ㄷ ㅌ, ㅁ ㅂ ㅍ, ㅅ ㅈ ㅊ, ㅇ ㅎ, and ㅏ ㅑ, ㅓ ㅕ, ㅗ ㅛ, ㅜ ㅠ. 쌍자음 doubles a consonant and undoubles it: ㄱ ㄲ, ㄷ ㄸ, ㅂ ㅃ, ㅅ ㅆ, ㅈ ㅉ. On a vowel it gives the y-form, the same as 획추가: ㅏ ㅑ, ㅓ ㅕ, ㅗ ㅛ, ㅜ ㅠ, and ㅐ ㅒ, ㅔ ㅖ. Both work on a syllable you are still building, so 가 becomes 갸, 롬 becomes 롭 and then 롶, and 각 becomes 갂. A letter that can’t stay in the syllable moves out of it: 밥 doubled is 바ㅃ, and 낛 with a stroke is 낙ㅈ. ㅐ, ㅒ, ㅔ and ㅖ have no key: type the vowel and then ㅣ (ㅏ ㅣ is ㅐ), and 획추가 or 쌍자음 on ㅐ gives ㅒ. Once a word is finished with Space or a mark, the two keys do nothing to it. Both keys are also in the layout editor’s action picker, for a pad of your own.

10 키 (10 key) is the same pad in Gboard’s arrangement: ▶ ends a consonant’s cycle without waiting and without a space, so ㄱ, ▶, ㄱ is two ㄱ, and . and , sit either side of ㅇㅁ in place of the .,?! cycle.

천지인 플러스 (Cheonjiin Plus) builds vowels from the same three strokes as 천지인, and gives every consonant a key of its own, so most letters take one tap. Tap ㅋ ㅌ ㅍ ㅎ ㅊ again for the tense letter beside them: ㄲ, ㄸ, ㅃ, ㅆ, ㅉ.

베가 (Vega) is a phone keypad where every key holds a cycle of letters, vowels included. Tap ㄱㅋ for ㄱ, ㅋ, ㄲ and ㅏㅑ for ㅏ, ㅑ. ㅣㅡ runs ㅣ, ㅡ, ㅢ. As on 천지인, a pause or a Space ends a cycle, so ㄱ, Space, ㄱ is two of them, and ㅐ is ㅏ then ㅣ.

모아키 (Moakey) types a whole syllable in one stroke. Drag a consonant key towards a vowel: right for ㅏ, left for ㅓ, up for ㅗ, down for ㅜ, either upper corner for ㅣ, either lower corner for ㅡ. Bend the drag without lifting for the rest. Right and back is ㅐ, forward again ㅑ, back again ㅒ, and left works the same way for ㅔ, ㅕ, ㅖ. Up and back is ㅚ, forward again ㅛ, and up then right is ㅘ. Down and back is ㅟ, forward again ㅠ, and down then left is ㅝ. A lower corner then up is ㅢ. Tap a consonant to end the syllable before it, and one backspace takes back one stroke. 모아키 양손 (Moakey two-hand) adds ㅣ, ㅡ and ㆍ keys: ㅣ joins the vowel before it (ㅏ ㅣ is ㅐ) and ㆍ gives its y-form (ㅏ ㆍ is ㅑ, and ㆍ again is ㅏ).

나랏글 and 베가 also come centred, with the letters in the middle columns and punctuation down the left.

Chinese input schemes

Chinese carries ten typing layouts. Adding it from the language catalogue turns on Pinyin alone, and the other nine are a tap away on Chinese’s own screen:

Cangjie: radical keys build a shape code; the strip fills in once the code is complete.

Layout	Reading unit	Buffer commits
Pinyin	Romanized syllables (ni, hao)	Prefix: a multi-syllable buffer offers the whole phrase plus its leading syllables
ClearFlow Pinyin	The same syllables on the ClearFlow grid	Prefix
T9 Pinyin	Same syllables on a 9-key/T9 grid (ABC/DEF/GHI…)	Prefix
Zhuyin (Bopomofo)	ㄅㄆㄇㄈ… symbols, with four dedicated tone-mark keys (ˇˋˊ˙)	Prefix
12-key Zhuyin	The same symbols on ten keys of three or four each. The tone key only ends a syllable, since the dictionary has no tones	Prefix
Cangjie (倉頡)	Up to five radical keys, a–y, each printed as its radical glyph	Whole buffer: one code run is one character
Cangjie Quick (速成)	Just the first and last radical of a character	Whole buffer
Wubi 86 (五笔)	Up to four root components, on the letters a–y. Each key’s corner shows its stroke zone and root, and z stands in for one you aren’t sure of	Four keys at most: one code run is one character or phrase
Stroke (笔画)	Five stroke-class keys (一丨丿丶乙) plus a ＊ wildcard for an uncertain stroke	Whole buffer
12-key stroke	The same strokes on a phone keypad, with 通 as the wildcard	Whole buffer

Pinyin, T9 Pinyin, and Zhuyin are all prefix-commit: type nihao and the strip offers 你好 first, then leading-syllable candidates like 你 alone. Tapping one consumes only the part of the buffer it covers and re-converts the rest, so you don’t have to clear the buffer between characters. Cangjie, Cangjie Quick, and Stroke are the opposite: shape and stroke codes map to exactly one character each, so a commit always consumes everything you’ve typed.

Each letter on the Pinyin layout carries a digit or symbol on long-press, shown in its corner as on the English layout: the digits on the top row, then @ # ¥ % & - + ( ) across the middle row and * " ' : ; ! ? across the bottom one. With the number row on, the top row’s corners stay empty, since its digits are already above it. To give a key other alternates, open the layout in the layout editor and change that key’s Press and hold list.

Wubi’s letters are grouped by the first stroke of the roots on them: 一 on g f d s a, 丨 on h j k l m, 丿 on t r e w q, 丶 on y u i o p and 乙 on n b v c x. A code spells a phrase as well as a character: khlg offers 中国 and wyth offers 信息. An exact code comes first, then whatever starts with it, commonest first. A fifth key starts the next code, so wythkhlg offers 信息, and picking it leaves khlg converting. The table is Wubi 86 only; a Wubi 98 code spells something else. Like Cangjie and Stroke, Wubi does not learn from your picks.

Zhuyin, 12-key Zhuyin and T9 Pinyin don’t need their own dictionary. They reuse the Pinyin pack and translate each segmented syllable to its pinyin spelling. Cangjie and Cangjie Quick share one radical-code table. Stroke and Jyutping (below) each have their own pack.

Picking syllables on T9 Pinyin. Each key on the T9 pad stands for three or four letters, so the keys you press could spell several syllables, and the keyboard guesses which. The column down the left of the pad shows what the next syllable could be while you type. WXYZ GHI ABC MNO lists xiao, xian, zhao and zhan, then xia and zha, then xi, yi and zi: longest first, with the keyboard’s own guess at the top. Tap one and that syllable is settled. The strip then offers only characters read that way, and the column moves on to the next syllable, so a phrase the strip keeps getting wrong (优点真遗恨 for 有点小遗憾) can be pinned down one syllable at a time. The text you are typing shows a ' between syllables, so you can see how your keys were split. Delete takes back the last key you pressed, and once only settled syllables are left it unsettles the last one, back to its keys. While you aren’t typing a reading, the column lists punctuation (，。？！、：； and more) to tap instead. The column is in the layout editor’s action picker too, as “Pinyin syllable list”, for a pad of your own.

ClearFlow Pinyin is Pinyin with the letters where ClearFlow puts them: the commonest ones in the middle, and tall keys down both sides. It reads the same Pinyin dictionary and takes the same settings, so Jianpin, Fuzzy Pinyin, Double Pinyin and swiping all work on it. Shift’s tall key is ‘词 here, as on the Pinyin layout, and the top corners type ！ and ？. One thing stays QWERTY: a missed key is repaired from the keys beside it on a QWERTY board, so a slip on this grid is less often caught.

Cantonese: Jyutping, Yale and Cantonese Pinyin

Cantonese is a separate language entry (not a Chinese sub-option) with three layouts, one for each way of spelling it. All three are standard QWERTY plus a permanent digit row, all three are prefix-commit conversion schemes like Pinyin, and all three read the same 163k-entry dictionary. Adding Cantonese turns on Jyutping alone, and the other two are a tap away on its screen.

Layout	你好	粵語	水	How it differs from Jyutping
Jyutping (粵拼)	neihou	jyutjyu	seoi
Yale (耶魯)	neihou, neihhou	yutyu, yuhtyuh	seui	j and ch for z and c, y for j, eu for oe and eo, and a for an open aa (媽 is ma)
Cantonese Pinyin (教院拼音)	neihou	jytjy	soey	dz and ts for z and c, y for yu, oey for eoi, and oe for eo

Tones are optional on all three, and they narrow nothing yet. Jyutping and Yale write them as a digit from 1 to 6 after the syllable. Cantonese Pinyin also takes 7, 8 and 9 for the entering tones, so 粵語 can be jyt9jy5. Yale can mark a low tone with an h after the vowel instead, as in sihk and neih, and you can leave it out: neihou could be nei + hou or neih + ou, and the keyboard reads both and lets the words decide.

Yale and Cantonese Pinyin are Jyutping’s dictionary under another spelling, so they need no download of their own. What you pick on one of them moves up on the others, and your own Cantonese word lists work on all three. They always convert on the built-in engine, even with RIME picked for Cantonese, because RIME’s Cantonese schema spells Jyutping.

Double Pinyin

For Pinyin typists who want fewer keystrokes, Double Pinyin encodes every syllable as exactly two keys (one for the initial, one for the final), which the composer expands to full Pinyin before segmenting as usual. Five real schemes are available: Microsoft, Sogou (identical key table to Microsoft), Xiaohe (小鹤), Ziranma, and Pinyin++. Turning one on replaces the syllable-typing rules on the Pinyin and ClearFlow Pinyin layouts. It isn’t a separate keyboard, because the keys stay the same.

Your own scheme

Pick Custom in the scheme list, or open Custom Double Pinyin scheme below it, to set every key yourself:

WM Keyboard › Languages & layouts › Chinese › Custom Double Pinyin scheme

The first time you pick Custom, the editor opens on a copy of the scheme you were typing with (Xiaohe if you were on full Pinyin), so you change a working table rather than start from nothing. Start from a scheme in the menu swaps in another built-in table the same way. Nothing changes on the keyboard until you press Save, and saving also switches you to Custom. The editor always opens with three # comment lines that explain the format, and a scheme you paste or import without them gets them added on top. While the custom scheme has nothing but comments you type full Pinyin.

The scheme is plain text in the same format as fcitx’s sp.dat, so a scheme file written for fcitx works here unchanged, and one you export works in fcitx. The menu has Copy and Paste for sharing it as text, and Import and Export for files. One line per key:

方案名称=My scheme
[声母]
zh=v
ch=i
sh=u
[韵母]
iu=q
ing=k
uai=k
[零声母]
=aeo
ang=ah

• pinyin=K puts an initial or a final on key K. Several finals can share a key, and the one that makes a real syllable wins. The one-letter initials (b, p, m and so on) stay on their own keys unless a line moves them, so you only have to say where zh, ch and sh go.
• =keys lists the keys that start a syllable with no initial, followed by the final’s key, so oj is an when o is listed. A * among them is fcitx’s rule for schemes without such a key: a one-letter final is typed twice (aa), a two-letter one as itself (ai), and a longer one as its first letter and its own key (ah). With no = line, a, e and o are the lead keys, as in the built-in schemes.
• pinyin=KK, with two keys, spells one whole syllable and wins over everything else.
• # starts a comment, lines in square brackets are only headings, case doesn’t matter, and ü can be written v.

Under the text, a status line names the first line the scheme skips and why (tap it to select that line), or else every initial and final that no key types yet. Only the first scheme in a file is read: fcitx’s own sp.dat holds several in a row, each opened by its 方案名称= line, so delete the ones you don’t want or let the first one stand.

Jianpin (简拼): initials are enough

Nobody types 中国人民 as zhongguorenmin. Jianpin lets a syllable be its first letter alone, so zgrm reaches the same word, wm finds 我们 and bj finds 北京. It’s on by default, the way every desktop and phone Pinyin IME has it, and it’s a switch on the Chinese screen if you’d rather not.

The rule is one letter, one syllable, boundaries included: hd means h… plus d…, never “any reading with an h and a d in it”. That’s what makes mixed input work too. Spell out as many syllables as you like and abbreviate the rest: haod finds 好的, zguo finds 中国, beij finds 北京. A syllable you type in full is matched exactly and never widened, so hao still means 好 and nothing beginning with hao.

Three conventions borrowed from Sogou and Google Pinyin:

• z, c and s cover the retroflex initials too. zg finds 中国 (zhōng guó) as well as 祖国 (zǔ guó). A typed zh, ch or sh reads both ways: sh is s + h (上海, 生活) and also the initial sh… on its own (是, 说), and zhg still reaches 中国. Both are on the strip and rank by likelihood. An apostrophe is your own boundary: s'h is s + h only.
• The vowels a, e and o are both a syllable and an abbreviation. aq finds 安全. Typed on their own, the exact reading leads: a offers 啊 before 爱 and 安.
• A phrase beats a stitch. Where the dictionary has the word (我们, 中国), it comes first; single characters strung together to cover the same letters rank below it. So the strip leads with what the abbreviation was for, and the ranking the issue asks for falls out of that: an exact reading first, then a partly spelled one, then pure initials.

Everyday pairings

A lot of Chinese isn’t dictionary words. 我的 and 我不 are two single characters that turn up together constantly, and no dictionary lists either as an entry. Ranked one word at a time, wbxh came out as 亡八喜欢: two real words beat four, whatever the words are.

The Chinese word pairs pack fixes that. It counts which words follow which in a large body of everyday Chinese (film and TV subtitles, plus web and news text), so the keyboard can tell that 不 after 我 and 喜欢 after 不 are what people write. With it, wbxh leads with 我不喜欢, wbxz with 我不想再, wdmm with 我的妈妈, nine with 你呢 and hd with 好的. It comes with the Pinyin dictionary when you download that, and if you already had the dictionary, a chip on the strip offers it once: “Download Chinese word pairs for better candidates”. It’s about 11 MB. T9 Pinyin and Zhuyin use it too.

Without the pairs, from three abbreviated letters up and only where no single entry covers everything you typed, the strip carries a few stitched readings right after its first answer: whole-buffer ways of reading the letters as separate words, so 我的妈妈 for wdmm is still a tap away.

Learning steers the search as well as the finished list. The characters you keep choosing for a reading are the ones paths get built out of, so the longer you type, the more the readings come out in your own words.

Two things follow from abbreviations being real units:

• Candidates appear from the first key. n alone offers 你, 那, 能 and so on, and nih already leads with 你好, so a whole word is often on the strip before its second syllable is finished.
• A half-typed trailing syllable is guessed, not ignored. With nihaoz on the strip, the top candidate covers all six letters (你好在, say), where before it would have covered 你好 and left the z. Keep typing and the guess resolves; tap 你好 and only those five letters commit, the z stays.

Two readings of the same letters

Some letters split into syllables more than one way. xian is one syllable (先, 现) and also two (西安, Xi’an); nine is nin + e and also ni + ne (你呢); fangan is 方案. Pinyin marks the second kind with an apostrophe, which almost nobody types on a phone, so the keyboard reads both and the more likely reading wins: xian offers 先 and 现 first and 西安 a few places on, and nine leads with 你呢. Type the apostrophe and it’s settled: xi'an is two syllables only, so 先 is no longer offered.

Splits that need more abbreviations, or more syllables, than the letters require are left out. xian never reads as xia + n (下年), and nihao never as ni + ha + o.

A missed key

Where the letters you typed only make sense as a run of initials, the keyboard also tries the slips a finger makes: each letter as the key next to it, two letters the wrong way round, one letter too many and one too few. xiqn reads as xian (the a beside the q missed), qign as qing, zhng as zhong and zhonng as zhong, and the repair is offered alongside whatever the initials spell. It only looks where the letters need two or more bare initials to read at all, so ordinary pinyin and deliberate initials (wbxh) are never second-guessed. A repair is only offered while it reads nearly as well as the letters as typed, at least a tenth as likely, so a slip never pushes a real reading of your letters off the strip.

Two slips come from the keys around the letters rather than among them (since 0.6.6). A thumb aiming for the space bar can catch the row above it, so one of c, v, b, n or m alone between two syllables can be read as a missed space and left out: nihaobwo offers 你好我. With Jianpin off, that letter used to have no reading at all and stopped the conversion where it sat. With Jianpin on it is also an initial (b for 不), and both readings are offered. A letter that starts a real syllable, like the b of bu, is never left out. And with Fix number row slips on, a digit typed in the middle of a reading is the letter under it on the number row, so niha9 converts as nihao.

Other spellings

Some syllables have more than one spelling in use, and a few slips are so common that every Pinyin keyboard takes them. They all read as the syllable they stand for, with nothing to switch on:

• ü as u or v: lue and nue for lüe and nüe, jv, qv, xv and yv for ju, qu, xu and yu.
• The long forms that pinyin shortens: liou for liu, guei for gui, duen for dun.
• gn for ng (bagn is bang) and on for ong (zhon is zhong).

A spelling that is a syllable in its own right keeps its own meaning: lu is always lu.

Before you finish the syllable

The last syllable can be unfinished. zho at the end of what you typed also reads as zhong and zhou, nia as nian, niang and niao, so 中 and 年 are on the strip before you type the rest. This only applies to the last few letters, and only where they aren’t a syllable already.

Rare characters

Every character has a place in the ranking from how often it shows up in a large body of text, and for rare characters that count is unreliable: one odd file is enough to put a character nobody writes at the top of its reading. The keyboard also knows which characters China’s standard list of common characters (通用规范汉字表) includes and at which of its three levels. A character outside the list, or in its rarer levels, sinks below the everyday ones that share its reading. This list is part of the app, not a download.

Dates and numbers

WM Keyboard › Languages & layouts › (a CJK language) › (language) options › Date, time and numbers

Some letters type things that aren’t in any dictionary:

• rq (日期) offers today’s date: 2026年10月13日, 二〇二六年十月十三日, 2026-10-13 and 10月13日.
• sj (时间) offers the time: 14:05, 14点5分 and 下午2点5分.
• xq (星期) offers the weekday: 星期二, 周二 and 礼拜二.

These come right after the best candidate for the same letters, so sj still leads with 时间. Picking one teaches the keyboard nothing.

Type v and then a number, with a decimal point if you like, for the number in Chinese numerals: v1234.5 offers 一千二百三十四点五, the financial 壹仟贰佰叁拾肆点伍, 一二三四点五 digit by digit, and for whole numbers the amount 壹仟贰佰叁拾肆元整. No pinyin syllable starts with v, so this never changes what pinyin reads. All of this is on by default and one switch turns it off.

Narrowing the list by strokes

When the character you want is far down a long list, narrow it by how the character is written. Hold ‘词 and slide to the backtick, then type the first strokes with the letters fcitx and Sogou use: h for 横 (一), s for 竖 (丨), p for 撇 (丿), d or n for 点 (丶) and z for 折 (乙). yi then `z leaves 乙 and the other characters that start with 折; the composing text shows the strokes as you type them. Picking a candidate ends the filter, and only the letters that candidate covers are used up. It needs the Stroke dictionary download, which is where the stroke order comes from. Without it the filter narrows nothing.

The word separator key

On the Pinyin layout the key left of z is ‘词 (‘詞 with Traditional characters on), not Shift. It types the apostrophe that marks a syllable boundary, so xi'an and s'h take one tap each, the way Gboard’s Pinyin keyboard has it. Hold it and let go to switch between Simplified and Traditional characters, see Switching scripts from the keyboard; the backtick is the second entry in the same popup.

Swiping

You can swipe on the Pinyin layout, with gesture typing on. A swipe spells a reading, either the full pinyin (nihao) or the initials of a word (wm for 我们), and the reading goes into the text you’re composing as if you’d typed it, so the strip converts it the usual way and a second swipe carries the same sentence on. Nothing commits and no space follows. It needs the Pinyin dictionary, since that is where the readings come from, and it’s off under a Double Pinyin scheme.

The rest of it

Jianpin is a matching rule, not a second dictionary. Its index is the ~410-syllable inventory grouped by first letter, so a run of initials costs a handful of binary searches per syllable rather than a scan. It has no effect under a Double Pinyin scheme, where a syllable is always exactly two keys, and it’s independent of Fuzzy Pinyin: a syllable typed in full can still be matched fuzzily, an abbreviated one is expanded, and the two combine. Picks are learned under the reading the abbreviation resolved to, so choosing 我们 for wm also promotes it for women.

Fuzzy Pinyin and lazy Jyutping

Two independent toggles relax reading matching for imprecise typing:

• Fuzzy Pinyin treats commonly-confused sounds as equivalent, so a syllable typed loosely still finds its characters. It’s eleven switches rather than one: six initial groups (zh↔z, ch↔c, sh↔s, n↔l, r↔l, f↔h) and five nasal-ending groups (an↔ang, en↔eng, in↔ing, ian↔iang, uan↔uang), each with a row of its own and all on once the master switch is on. That way you can take the regional nasal endings without paying for n↔l on every syllable that starts with either. Switch one off and a Match every pair again row appears to put them all back.
• Lazy pronunciation (懶音), Cantonese-only, matches the sound mergers common in Hong Kong speech: n↔l, an initial ng- that can drop entirely (我 ngo5 → o5), gw→g and kw→k before a rounded final only (國 gwok3 → gok3, but never 誇 kwaa1 → kaa1), and the coda mergers toward the alveolar place (-ng/-n, -m/-n, -k/-t and -p/-t). It also takes z↔j and c↔ch on the Jyutping layout, which aren’t 懶音 at all but Yale-era romanization habit. The Yale and Cantonese Pinyin layouts leave those two out, since their own spelling already says which sound a key means. Every merger works in both directions, because hypercorrection is as common as the merger itself.

Fuzzy Pinyin is off by default and lazy pronunciation is on, and they change ranking rather than what you can type. A precisely-typed reading still works exactly as before. Japanese has a matching of its own for the flick pad, covered under Kana without marks.

Traditional characters and regional wording

A single toggle switches candidate output from Simplified to Traditional Han, applied as each composer builds its candidate list (not as a find-and-replace afterward, which would break prefix-commit matching). Many Simplified characters have more than one Traditional form, and only the word tells which: 发 is 發 in 發展 but 髮 in 頭髮, 后 is 後 in 後來 but stays 后 in 皇后, 干 is 幹 in 幹部 but 乾 in 乾淨. The keyboard converts word by word with the OpenCC phrase and character tables (release 1.4.2), using OpenCC’s own matching: the candidate is split into the longest phrases it contains, each phrase converts as a unit, and only a character that is in no phrase falls back to its most common form. A candidate the keyboard stitched from several words is split the same way, so 头发干净 comes out as 頭髮乾淨. A character typed on its own has no word to decide it, so Pinyin, T9 Pinyin, Zhuyin and Jyutping list each of its Traditional forms, the most common first: fa offers 發 and then 髮, gan offers 幹, 乾 and 干, hou offers 後 and 后. Cangjie and Stroke do not, because their codes describe the shape of one character.

The character tables ship with the app. The phrase table is the Traditional characters phrase table pack, about 1 MB, which you download from the Chinese or Cantonese options page like the other tables below. It is one download: the row on either page is the same file. Without it, Traditional output still works, but one character at a time: each character takes its most common form, so 头发 comes out as 頭發. Conversion itself never goes online. The tables load the first time you turn the toggle on, so for a moment after that the candidates can still show in Simplified.

With Traditional on, a second control appears: a region picker (Standard / Taiwan / Taiwan wording / Hong Kong). Taiwan follows OpenCC’s s2tw: Taiwan’s character preferences (為, 裡, 著), and the word stays the one you typed, so chuzuche gives 出租車 and jichengche gives 計程車. Taiwan wording follows OpenCC’s s2twp, which also swaps regional vocabulary: Taiwan’s 計程車 instead of the mainland’s 出租車 for “taxi,” for instance, even though you typed the reading of 出租車. Vocabulary is matched inside a candidate too, so 出租车光盘 becomes 計程車光碟. It is a separate choice because a swap can also change the meaning: 支持 becomes 支援 and 文件 becomes 檔案 in every sense of the word. Hong Kong follows OpenCC’s s2hk: the standard wording with Hong Kong’s character preferences (説, 衞, 温). The Hong Kong choices come straight from OpenCC’s tables and have not been reviewed by a Hong Kong writer yet; a wrong one is worth reporting. The region applies under the RIME engine too, whichever schema is in use.

Switching scripts from the keyboard

For a message in the other script you don’t need the settings app. Two switches flip Traditional characters from the keys, and each one changes the same setting as the row on the options page, so it stays the way you left it:

• Hold ‘词 on the Pinyin layout and let go. A toast says which script the keyboard now writes.
• The 繁 button. Turn on 繁 button under WM Keyboard › Languages & layouts › Chinese › Chinese options › 繁 button (also on the Cantonese page; off by default). A 繁 button then sits beside the candidates on every Chinese and Cantonese layout, lit while candidates come out Traditional. Like the other buttons beside the words, it shows while the candidate strip is up, not while the toolbar is.

A reading you are in the middle of typing switches with it: the candidates on the strip are drawn again in the new script, so you can type toufa, see 头发, switch and pick 頭髮. Any layout can carry the switch too: in the layout editor it’s the Simplified / Traditional Chinese action, and in a layout file it’s {"type": "han_script"}.

Both controls are on the Chinese and Cantonese options pages only, and they are one setting shared by the two languages. Japanese candidates never pass through this conversion, so the Japanese page does not show them.

Handwriting layouts

Chinese, Japanese and Korean each also ship a Handwriting layout, as do 296 other languages, for the character you can’t spell: a kanji whose reading you don’t know, a name, a hanzi you only recognise by shape. Enable it like any other layout of the language, and the keys above the bottom row become a writing area. Write one or several characters, pause, and the best reading goes into your text while the others wait in the suggestion strip for a tap to swap them in. The bottom row keeps ?123, the globe, 、 and 。 (Chinese: ，), space, delete and enter, so you can switch back to your typing layout with one tap. Characters written one after another are never separated by a space. It is the same on-device recognition as the Handwriting tool, so the language needs its model downloaded once; the writing area offers the download itself. The F-Droid build reads Chinese and Japanese here with zinnia, and has no Korean handwriting.

Japanese input

Japanese ships five typing layouts, all feeding the same composer, plus the handwriting one above:

• Romaji: type romanized Japanese on a standard QWERTY layout. Keystrokes turn into hiragana as you type, by the same romaji table Google Japanese Input uses: doubled consonants become っ, yōon like kya become きゃ, and the foreign-sound spellings work too (vo ゔぉ, thi てぃ, kwa くぁ, ye いぇ), so vo-karu offers ヴォーカル. nn and n' are always one ん, whatever comes after: konnya is こんや and kinnen きんえん, which means こんにちは takes three n’s, konnnichiha. A single n before a consonant is ん as well (shinbun しんぶん). The long-vowel mark ー has its own key at the end of the middle row, as on other Japanese keyboards, and a hyphen typed inside a word reads as ー too, so ka-do converts to カード. Between words a hyphen stays a hyphen. Each letter carries a digit or symbol on long-press, shown in its corner as on the English layout, with ¥ on d and % on f as Japanese keyboards place them.
• Flick: a 12-key pad with kana glyphs directly on the keys. Flick a key in a direction to reach its row-mates (あ flicks left/up/right/down to い/う/え/お).
• Kana JIS: each key emits a kana glyph directly, matching the physical JIS Japanese keyboard layout. Since 0.6.6 the ゛ and ゜ forms are a swipe away, so you don’t have to tap a second key: swipe down on か for が (on every kana that takes ゛, う for ゔ included) and up on は, ひ, ふ, へ or ほ for ぱ, ぴ, ぷ, ぺ, ぽ. Swipe up on ⇧ to run 小゛゜ on the kana you just typed, the same as the ゛ key does. The swipes aren’t drawn on the keys, which are too narrow for them. With the swipes in hand you can drop the ゛ and ゜ keys in the layout editor and give the space to something else.
• Godan (五段): romaji on five rows, with the vowels a i u e o down the left and the consonants beside them, the way Gboard draws it. Letters without a key of their own are flicks: k right for g and up for q, s for z and j, t for d and c, h for b, f and p, m up for l, y up for x, w up for v. Flick most keys down for a digit, n up for ん and r for 。？！、. Small kana are the usual x spellings, or tap 小゛゜, which turns ka into が and tsu into っ.
• 50 keys: a key for every kana in gojūon order, with あ at the top right and the rows running leftward as a kana chart does. Hold a key for its marked and small forms: か for が, は for ば and ぱ, つ for っ and づ. 小゛゜ does the same to the kana you just typed.

Japanese Flick: kana glyphs sit on the keys themselves, with directional flicks for the rest of the row.

The Flick pad is an ordinary layout, so you can change it in the layout editor. Each key’s sheet has a 3×3 pad with a cell for every flick direction, the four corners included, and a Shift field for each so the same pad can type katakana while Shift is on. The editor’s preview draws a key’s flicks around its edges, so you can see and move every kana on the pad, and the keyboard draws them on the keys the same way unless Show flick arms on keys is off. Any key of a Japanese layout can also be told to become 小゛゜ after a kana: while the last kana you typed has a small, ゛ or ゜ form, that key shows 小゛゜ and changes it, and once the word is done it goes back to its own job. Give that job to the globe or emoji key and the separate 小゛゜ key can go, which is how phone kana pads save the space. Either key cycles katakana as readily as hiragana, so a pad built on アイウエオ keys reaches ェ from エ and バ or パ from ハ.

All five read the same kana or romaji into JapaneseComposer, which converts kana readings to kanji/word candidates using a Mozc-derived dictionary. If nothing in the dictionary matches (or before you’ve downloaded the pack at all), the reading still commits as plain hiragana, katakana, or half-width katakana, so every Japanese layout works as a bare kana keyboard with zero setup. A phrase the dictionary lists only in kana still gets its kanji reading offered: またあした puts また明日 on the strip next to the plain kana, so you don’t have to convert また and 明日 one at a time.

Kana without marks

On Flick, Kana JIS and 50 keys, every small kana, ゛ and ゜ is an extra key or a hold. Match kana without marks lets you leave them off: type かつこう and the candidates still offer 学校, which the dictionary files under がっこう. Each plain kana is also read as the forms the 小゛゜ key would cycle it through (か as が, は as ば or ぱ, つ as っ or づ, や as ゃ, a vowel as its small form), anywhere in the word and as many times as the word needs, so しゆきよう reaches 授業.

It only ever adds marks. A が or a っ you typed was asked for by name and is never read back as か or つ. Each guessed mark costs the candidate a little, so between two words of about the same frequency the one spelled exactly as you typed it comes first, while a clearly more common marked word leads: かつこう opens with 学校 ahead of the given name 葛洪, and かき still opens with 書き, 描き and 下記 before 鍵 arrives. Nothing the exact spelling offered is removed, so expect a longer list rather than a different one. What’s left in the field while you compose is the kana you typed; the marks arrive with the candidate you pick. Picks are learned against the dictionary’s reading, so 学校 moves up whether you next type it with its marks or without.

Romaji is never widened. ga costs exactly what ka does on a QWERTY layout (or one flick on Godan), so kaki means かき and nothing else, which keeps the strip as tight as it was.

Swiping on Kana JIS

With glide typing on, you can swipe across the Kana JIS layout. A stroke spells a reading, which goes into the text you’re composing as if you’d tapped its kana, and the strip converts it the usual way. Nothing commits and no space follows, so a second swipe carries the same sentence on.

You don’t have to visit ゛ or ゜ on the way. A stroke through か, つ, こ and う reads as かつこう, and Match kana without marks finds 学校 in it. When the word you meant doesn’t come up, draw the mark: pass over ゛ straight after か and the stroke reads as がつこう. With Match kana without marks off, that’s the only way a swipe gets a mark. Small kana are always drawn on their full-size key (つ for っ, や for ゃ), and the candidates put the small form back.

The readings come from the Japanese dictionary, so swiping waits until it’s downloaded, and only readings common enough to be swiped are in the list: a rare name still has to be tapped. Conjugated forms the dictionary doesn’t list, such as わかった or いきました, don’t come out of a single stroke yet. 50 keys glides the same way, and so does a layout of your own with a key for most kana. 50 keys has no ゛ to pass over, so its swipes rely on Match kana without marks. Romaji, Godan and Flick don’t glide.

Punctuation

Every Japanese and Chinese layout carries its language’s punctuation, whichever input method you type with.

• Beside the space bar. The comma key types 、 in Japanese and ， in Chinese, and the full stop types 。. Hold the comma for the quotes: 「 」 『 』 in Japanese; “ ” 「 」 《 》 in Chinese, alongside 、 ！ ？.
• Flick. Flick や left for 「 and right for 」. Hold 、 for 『』, （）, ・ and 〜.
• ?123. The first symbols page swaps the ASCII quotes and the Spanish ¿ ¡ for the language’s own marks. Japanese gets ・ 「 」 、 ： ！ ？, Chinese gets 、 “ ” ： ； ！ ？, and the brackets become （ ）. Hold a quote for the nested pair (『』, ‘’) or the other convention.
• =\<. The second page has the rest of the bracket family as keys of their own. Japanese: 『』【】, holding 〈〉《》 and 〔〕. Chinese: the title marks 《》 (hold for 〈〉) and the corner brackets 「」 (hold for 『』 and 【】), so Mainland and Taiwan or Hong Kong quoting are both a tap away. It also has 〜 and … (Japanese) or —— and …… (Chinese). © ® ™ moved onto ¶‘s hold.
• Dashes. Hold - on the first symbols page for ー, 〜 and ―― (Japanese) or —— and …… (Chinese) as well as the usual dashes.

The , and . on the symbols page itself stay ASCII, because that is where numbers like 1,000.5 get typed. Their holds lead with the wide forms. A layout that writes its own symbols page keeps it exactly as written.

Full-width space

Japanese and Chinese text is spaced with the ideographic space (U+3000, 「　」), which is as wide as a character, rather than the narrow space between words in other languages. Turn on Full-width space in the language’s options and the space bar types it. It is set per language, off by default, so you can have it for Japanese and not for Chinese. A space pressed while a reading is still waiting to convert picks a candidate as it always has; only a press with nothing composing types a space.

The candidate bar

The expanded candidate grid: every ranked candidate, wrapped to fit, covering the key rows while it's open.

Whenever a conversion composer has a non-empty buffer, the usual word-suggestion strip is replaced by a horizontally-scrolling row of ranked candidates. They’re sized to their text and packed from the left rather than spaced evenly, so the best candidate stays in a consistent spot instead of jumping around as the list changes length. Each new reading opens the row back at its left edge, however far the last one was scrolled. A chevron at the end opens an expanded view, and it sits outside the scrolling area so it never scrolls off. That expanded view is a wrapping grid, and it covers the key rows while it’s open, the same way Sogou, Baidu, and QQ handle it. By then you’re choosing a character rather than typing one. Neither view pages. Both scroll.

Tapping a candidate (in the strip or the grid) resolves by its position in the ranked list, not by its text, because the same word can legitimately appear twice at different reading lengths (a Japanese reading might list 行 for both い and いき). Only the part of your typed buffer that the candidate covers gets deleted. The rest re-converts. Space is the fast path: it commits the top-ranked candidate for the leading syllables and moves on, so repeated space presses walk a whole sentence through without ever opening the strip. No path here adds a trailing space, since Chinese and Japanese don’t put spaces between words. Enter converts nothing: it confirms the reading exactly as typed and does nothing else, so no line break and no send. In Japanese that is the hiragana you see, which is how you keep a word in kana; in Chinese it is the letters you pressed, so with Double Pinyin on, nihc goes in as nihc and not as the nihao the composing line spells out. That makes Enter the quick way to drop a Latin word into a sentence without switching keyboards. Any other commit (punctuation, moving the cursor away, switching layout, or leaving the field) ends the reading too. In Japanese it keeps the kana as typed, the way Enter does, and only space or a candidate you pick converts. A candidate you highlighted with the arrow keys or a stepping space bar still goes in. In Chinese it flushes the whole buffer by repeatedly taking the top candidate until nothing’s left, so you’re never left with pinyin letters in the text.

Japanese is more forgiving about punctuation. Its punctuation (、。，．！？・…〜) typed after kana joins the reading instead of ending it, so なるといいです！ stays one reading that space can still convert, and the mark goes in with it. Brackets still end the reading.

Moving the cursor into the reading doesn’t end it either, in Japanese and in Chinese typed on letters (Pinyin with Jianpin or Double Pinyin, Jyutping, Yale and Cantonese Pinyin). A tap inside the underlined reading puts the cursor there: what you type goes in at the cursor, Delete takes out the letter or kana in front of it, and the candidates follow the corrected reading. So a j typed for an h early in mmbxjn is fixed where it is, without typing the rest again. In Japanese 小゛゜ changes the kana in front of the cursor. Tap at the end of the reading to carry on there. Converting starts from the reading’s beginning wherever the cursor is. Since 0.6.6 dragging along the space bar to move the cursor does the same: while a reading is open the cursor walks through it one letter or kana at a time and stops at either end, instead of ending the reading and losing its conversion. Confirm the reading first to move on through the rest of your text. With Mozc, this works until the arrow keys split the reading into segments; after that the space bar moves from segment to segment, the way the arrow keys do, and the candidates you picked for each one stay. On the 9-key pads, Zhuyin, Cangjie, Wubi and strokes a tap inside the reading still ends it.

In Pinyin the list runs from the whole buffer down to its first letter. Every reading of everything you typed comes first, then the best readings of shorter stretches from the start, longest first, then the single characters for the first syllable. wbxh offers 我不喜欢 and 我不想和, then 我不想 and 我必须, then 我不, then 我. Each one commits the letters it covers and leaves the rest to convert.

If a reading has no dictionary match at all, it always falls back to committing what you typed as-is (raw pinyin letters, kana, or a stroke code) rather than trapping you in an empty buffer.

Earlier, later, never

Hold a candidate in the strip or the grid and the strip turns into its menu: Earlier, Later and Never show, with ✕ to close it. Earlier counts it as picked, which puts it first for that reading. Later moves it one step down, behind the candidates it was level with, and also makes every sentence built with it a little less likely. Never show takes it off the list for good, along with every reading built through it, so blocking 亡八 also removes 亡八喜欢. The strip updates at once. It works on every conversion layout (Pinyin, T9 Pinyin, Zhuyin, Jyutping, Japanese and imported input methods), each in its own learning history. On an English word in the strip, Never show adds it to the suggestion blacklist instead. To undo the lot, clear the learned words below.

One character of a word

The hold menu on a candidate of two or more characters also shows its first and its last character. Tap one to type only that character, for the whole reading the word covered: hold 选择 and tap 选. This is the quick way to a character that is hard to find under its own reading (以词定字, fcitx’s [ and ]). Nothing is learned from it.

What comes next (联想)

WM Keyboard › Languages & layouts › (a CJK language) › (language) options › Next-word suggestions (联想)

Once a Chinese conversion is complete and nothing is composing, the strip offers the words that usually come next: after 我 it might offer 们, 的 and 是; after 喜欢, 你 and 吃. Tap one and it goes in as it is, with no space, and the strip offers what follows that. The words you have written after the same text before come first, then the common pairs from the word pairs pack, then the rest of a longer word the last character starts (无 offers 聊, from 无聊). It needs the word pairs pack, follows the strip’s own switches like any other suggestion, and is on by default. Start typing pinyin and the strip goes back to converting.

The suggestions come from the text in front of the caret, not only from your last tap. Move the caret into the middle of something you wrote, or backspace a character away, and the strip offers what usually follows the Chinese now behind the caret.

Backspace a word you just converted all the way out and the strip shows the other conversions of the reading that spelled it. Delete 他 and it offers 他, 她 and 它; tap the one you meant and it goes back in where the old one was, and the keyboard learns the pick the same way as one made while composing. This works for the conversions typed into the current field, and a word only half deleted gets the ordinary next-word suggestions instead, since part of it is still in the text.

English words and emoji

Pinyin, romaji and Jyutping are typed on the same letters as English, so with English turned on under the language’s Also suggest from, the strip offers English words for those letters too. Type hello and hello sits right behind 和. When the top candidate already covers everything you typed, as 很 does for hen, the English waits behind the first three candidates instead. Space still commits the top candidate, never the English. Tapping an English word puts it in exactly as shown, in place of the whole reading, with no space after it, and counts it as a word you use, the way typing it on an English layout does. Words the keyboard has learned from you on an English layout are offered here too, ahead of the word list, so a name or a brand you type is there when you start spelling it. Only words that start with what you typed are offered, never corrections, since the letters are a reading first. T9, Zhuyin, Cangjie and Stroke layouts don’t offer English, because their keys aren’t the alphabet.

Emoji come up the way they do on any other layout: the top candidate is looked up in the language’s emoji keywords, so xiao showing 笑 brings 😄 alongside, and an English word brings its own. They need the language’s emoji keyword pack, and they follow Emoji suggestions and the strip’s own switches, since these are suggestions rather than the input method. Tapping one with Keep word as the insert mode commits the candidates first and the emoji right after (笑😄); Replace word puts the emoji in place of the reading. A long press does the other one, as it does everywhere.

Picking with space, the arrows and Enter

WM Keyboard › Languages & layouts › (a CJK language) › (language) options › Space steps through candidates

Out of the box the space bar types the first candidate straight away. Turn on Space steps through candidates and it works the way Japanese input methods do instead: the first press highlights the first candidate, each press after it moves the highlight one along (wrapping back to the start at the end), and Enter types the highlighted one. Keep typing and the highlighted candidate goes in before the next reading starts; ⌫ drops the highlight and leaves you editing the reading again. The strip scrolls to keep the highlight in view, and it draws the same outline the keyboard uses for anything picked with the keys. The setting is per language, and off by default, so you can have it for Japanese and keep the quick space for Chinese.

On a physical keyboard the candidates come up even though the on-screen keys stay hidden: just the strip appears while a reading is waiting to convert, and goes again once it’s done. The arrow keys move the highlight whether or not the setting is on (→ or ↓ for the next candidate, ← or ↑ for the one before), and Space or Enter types it. Once a candidate is highlighted, 1 to 9 pick one by its place on the strip and Esc goes back to the reading. Before that the number keys are left alone, because Jyutping’s tones and the T9 pad type digits into the reading. Enter only stops confirming the reading as typed once you’ve moved the highlight, so it still works for dropping in a word as you typed it.

Where the reading shows, and switching language mid-reading

WM Keyboard › Languages & layouts › Chinese › Chinese options › Show the reading in the text field

The pinyin, zhuyin or Jyutping you type sits underlined in the text field until you pick a candidate, as it does on Gboard and Sogou. Turn Show the reading in the text field off and the field stays as it was: the reading shows at the start of the candidate strip instead, in front of the candidates, and only the candidate you pick goes into the field. That is how fcitx5 works. A long reading shows its end on the strip, where the next key lands. Move the cursor yourself, by tapping somewhere else in the text or selecting some of it, and the reading is dropped, as in fcitx5, so nothing you typed lands in a spot you didn’t mean. The setting is shared by Chinese and Cantonese, sits on both options pages, and is on by default. Japanese always shows its reading in the field.

WM Keyboard › Languages & layouts › Chinese › Chinese options › When you switch language mid-reading

Switch to another language or layout with a reading still waiting, and by default its best candidates go in, as if you had pressed space until it was all converted. When you switch language mid-reading offers two other choices. Type what you typed puts the reading in as letters (nihao, or the zhuyin itself), the way fcitx5 does. Discard it throws the reading away, and nothing goes in. Like the row above, it covers Chinese and Cantonese together.

The keyboard’s own boxes

The same thing happens in the keyboard’s own text fields: the dictionary tool’s search, the clipboard and emoji filters, the AI prompt, Find and replace, a plugin’s box. They compose and convert exactly as the app’s field does: romaji turns into kana as you type it, the 小゛゜ key cycles the last kana, the candidates appear in the strip above the keys, and a tap commits the part of the reading it covers and leaves the rest composing. Nothing reaches the app behind the panel, and a pick there is learned the same way a pick anywhere else is.

These boxes used to take the keys straight: ka stayed ka, there was no way to add a dakuten, and no kanji was ever offered, so a Japanese user could not search the dictionary tool for anything.

The candidate bar is not the ordinary suggestion strip wearing a different coat, and it isn’t switched off by the things that switch that strip off. Turning Suggestions off, or typing into a field that asks for a silent strip (search and filter boxes, address bars, email fields, password boxes), leaves the candidates exactly where they are. For a conversion method the candidate list is the input method, so hiding it wouldn’t quiet a guess, it would stop you typing the language at all. Password boxes convert too, and what they don’t do is remember the pick afterwards, which is the same rule the rest of the keyboard’s learning follows.

Learning from your picks

Every candidate you pick from the strip or the grid is remembered against the reading you typed, and so is the one a space press commits. Whatever goes in when you leave the buffer some other way (punctuation, moving the cursor) teaches the candidates nothing, since it’s only the order the list already had. A Pinyin or Jyutping reading that goes in as the letters you typed, because you pressed Enter or nothing in the dictionary converts it, is learned as a word in Latin letters instead, the way a word typed on an English layout is: a spelling no list knows joins your personal dictionary once you’ve used it a few times, and from then on the strip’s English words offer it for those letters. The next time the same reading comes up, your previously-chosen candidates move to the front (most-picked first) while everything else keeps the order the dictionary gave it. In Pinyin they move to the front of their own group: a 我 you picked for wo comes first among the single characters for wbxh, never ahead of 我不喜欢. It’s a reordering of your own history, not a blended relevance score, so a candidate you’ve never picked never gets promoted just because it’s common in general.

In Pinyin it learns three more things:

• Phrases you build. When you commit one buffer in two or more picks, say 我的 and then 妈妈 for wodemama, the whole thing is learned as one phrase, so next time it’s offered whole, and wdmm finds it too. Only short ones are kept, up to four syllables (or six single characters), and none where the word pairs already join every piece to the next.
• What you write after what. Each pick is remembered as following the word before it, whether that was the pick before or the Chinese text in front of the cursor. Those pairs help decide between readings (ta after 她说 leans to 她) and come first in what comes next.
• The text before the cursor. The first word of a new reading is ranked as following whatever Chinese is already written before it, using the word pairs pack and your own pairs. After a space, a punctuation mark or Latin text, the reading starts a new sentence.

Nothing you taught it stays forever if you stop using it. After every 3,000 picks or so, each count that hasn’t been used in that time is halved, and a count of one goes. A word you pick every day never fades, and one you picked a few times last year stops outranking it.

Learning is scoped separately per reading system (pinyin, T9 Pinyin, zhuyin, Cantonese, and Japanese kana readings each keep their own history, with the three Cantonese spellings sharing one), because the same word can be spelled several different ways depending on which Chinese scheme you’re using. Cangjie, Cangjie Quick, and Stroke don’t participate in learning at all: their radical and stroke codes map deterministically to one character, so there’s no ambiguity to learn from.

This history is governed by the same setting as ordinary word learning:

WM Keyboard › Privacy › Learning on this device › Learn from typing

Turning it off, using incognito mode, or typing into a field that opts out of learning all stop new picks from being recorded, exactly as they do for the regular word suggestion history.

Options

WM Keyboard › Languages & layouts › (a CJK language) › (language) options

Opening a Chinese, Cantonese, or Japanese entry from the Languages screen shows a dedicated options group (Korean gets a shorter one, covered under Hanja and Fix spacing):

Setting	Default	Applies to
Conversion dictionary download	Not downloaded	Chinese, Cantonese, Japanese (see below)
Traditional characters	Off	Chinese and Cantonese, shared between them
繁 button	Off	Same, a switch for the row above beside the candidates
Show the reading in the text field	On	Chinese and Cantonese, shared between them
When you switch language mid-reading	Type the first candidate	Same
Regional wording (Standard / Taiwan / Taiwan wording / Hong Kong)	Standard	Same, only shown once Traditional is on
Lazy pronunciation (懶音)	On	Cantonese only
Match kana without marks	On	Japanese only, and only kana typed on Flick, Kana JIS or 50 keys
Full-width space	Off	Chinese, Cantonese, and Japanese, each on its own
Space steps through candidates	Off	Chinese, Cantonese, and Japanese, each on its own
Jianpin (简拼)	On	Chinese only; ignored under a Double Pinyin scheme
Next-word suggestions (联想)	On	Chinese only
Date, time and numbers	On	Chinese only
Fuzzy Pinyin	Off	Chinese only
The eleven fuzzy pairs, one row each	All on	Chinese only, shown while Fuzzy Pinyin is on
Double Pinyin scheme	Off (full Pinyin)	Chinese only
Custom Double Pinyin scheme	Empty	Chinese only, used when the scheme above is Custom

To clear everything the keyboard has learned from your CJK picks:

WM Keyboard › Privacy › Your data › Learned words

CJK picks live in a file of their own, but the Delete the learned words button clears the lot in one action: your typed-word history, emoji usage, autocorrect statistics, the CJK pick history (with what you told the candidate menu) and the per-app language-mix record. There’s no way to clear the CJK half on its own.

Conversion dictionaries are downloads, not bundled data

None of the conversion dictionaries ship in the app. Until you download the matching pack, or import a list of your own, a Chinese or Cantonese layout still composes and shows the buffer, but the candidate list stays empty. Japanese is the exception: its plain hiragana, katakana and half-width katakana forms are appended to every ranking, so they’re on offer with no pack at all.

You don’t have to know that in advance. Adding Chinese, Cantonese or Japanese puts the packs its layouts need into the Download language data? prompt, sized with the rest (the setup wizard fetches them along with the word list). Switching on one of their layouts later, on the language’s page or under More layouts, asks Download the Chinese Pinyin dictionary? when that layout’s pack isn’t on the device yet. And if a layout whose pack is missing comes up on the keyboard anyway, a chip appears on the strip straight away, before you’ve typed anything, naming the pack you need: “Download the Chinese Pinyin dictionary to type characters”. Tapping it opens that language’s page, where the pack’s row is. The chip asks once per pack, and the ✕ dismisses it. It never appears in a password field.

Language	Pack	Approx. size	Coverage
Chinese	Pinyin (also covers Zhuyin & T9 Pinyin)	~2.6 MB	~121k entries, CC-CEDICT-derived
Chinese	Chinese word pairs (comes with Pinyin, for ranking whole readings)	~11 MB	~85k words, 700k pairs
Chinese	Stroke	~450 KB	Stroke-order table
Chinese	Cangjie (also covers Cangjie Quick)	~350 KB	~29k characters
Chinese	Wubi 86	~1.7 MB	~93k characters and phrases, from the Jidian table in ibus-table-chinese. Characters past U+FFFF are left out
Chinese, Cantonese	Traditional characters phrase table (optional, for Traditional output)	~1 MB	~50k phrases, OpenCC
Cantonese	Jyutping (also covers Yale & Cantonese Pinyin)	~3.4 MB	~163k entries
Japanese	Kana → kanji	~42 MB	~1.08M entries, Mozc-derived

When a new version of a pack comes out, its row offers the download again. Until you take it, the copy you already have keeps converting, so an app update never leaves you unable to type. The Pinyin pack got one with the October 2026 rescore, which ranks common characters like 齐 and 议 near the front of their reading instead of behind rare ones. Downloading it replaces the old copy rather than keeping both.

The shape-based tables have a second source. Import from FlorisBoard, in the same group, reads a FlorisBoard language pack (.flex) and fills the Cangjie, Stroke or Wubi pack from the SQLite table inside it, with no download at all. Only those three land, since Wubi 98, Zhengma and the rest have no composer here; the app names whatever else it found back to you. See Importing other keyboards.

Each download is checksum-verified before it’s used and, once on disk, needs no further network access. See Network policy for what the keyboard contacts and when, and Downloadable dictionaries for how the separate word-list system (which does have entries for plain Japanese and Korean word suggestions, just not for the conversion dictionaries here) works.

Your own word lists

WM Keyboard › Languages & layouts › (a CJK language) › Dictionaries › Custom dictionaries

A list you import for Chinese, Cantonese or Japanese goes into the candidates, next to the pack. It only needs a reading on each line, and the formats people already have all carry one:

• CC-CEDICT and CC-Canto (關係 关系 [guan1 xi4] /relation/). Chinese takes the Simplified form, the one the Pinyin pack lists, and Traditional characters turns it into the other; Cantonese takes the Traditional form and the Jyutping in braces.
• Rime dictionaries (.dict.yaml), such as Luna Pinyin: word, the spaced reading, then a count or a percentage. The YAML header is skipped.
• Mozc and Google Japanese Input user dictionaries: reading, word, part of speech. A katakana reading is read as hiragana.
• The packs’ own format, reading, word, frequency, one tab between each.
• Sogou’s .scel cell dictionaries (细胞词库), as Sogou publishes them: each word with its pinyin and Sogou’s count.

Tone numbers and tone marks are dropped and ü is spelt v, the same as in the packs. A line with no reading, such as a plain column of characters, is skipped, and the count on the Custom dictionaries screen is the number of lines that do have one.

A word your list adds appears under its reading. A word the pack already has keeps its place in the pack’s ranking, unless your list gives counts of its own (Rime’s do), in which case it can move up but never down. Your picks reorder both the same way. A list works on its own too: with no pack downloaded, the candidates come from your lists, and the missing-pack chip stops asking. Use only my word lists leaves the pack out once you have imported a list with something in it; with nothing imported it has no effect here, so it can’t leave the language unable to type a character.

A list goes live the next time you open a text field, and only the table for its own language is reloaded.

Dictionaries made for fcitx5, Rime or Sogou

Most Chinese dictionary projects on GitHub are built for the desktop input methods, and many ship two files per release: a .dict for fcitx5 and a .dict.yaml for Rime. Take the .dict.yaml. It is the Rime format from the list above, so it imports as it is: pick Chinese, open Custom dictionaries and import the file.

Sogou’s own .scel cell dictionaries import directly too, from Custom dictionaries or by opening the file with WM Keyboard; the import names them as a Chinese list.

fcitx5’s .dict is compiled, and the keyboard can’t read it: it is libime’s binary dictionary. Look for the Rime file or the .scel files in the same release, or convert the project’s source text instead.

Each file can be at most 32 MB. That rules out the whole-corpus builds, such as fcitx5-pinyin-sougou-dict, whose sougou.dict.yaml is over 400 MB: the import refuses it, and splitting it wouldn’t help, because the keyboard reads every list into memory when it builds the Chinese table. That project’s release also carries sougou_dict.tar.gz, the raw Sogou cell dictionaries it was built from. Unpack it and import the .scel files for the topics you actually type, such as computing, place names or medicine. A few thematic lists of a few megabytes each give most of the benefit and load fast.

The RIME engine

WM Keyboard › Languages & layouts › Chinese › Conversion engine › Engine

Chinese and Cantonese can convert on RIME instead of the keyboard’s own engine. RIME is the open-source engine behind Weasel on Windows, Squirrel on macOS, fcitx5-rime and ibus-rime on Linux, and Trime on Android. It reads RIME schemas: a schema says which keys spell which readings and which dictionary they look up, so anything written for those keyboards types here too, including schemas you find on GitHub or make yourself.

Engine on a language’s page picks Built-in or RIME, for that language alone, and Built-in is the default. The layouts don’t change: Pinyin is still the Pinyin board, Zhuyin still has its bopomofo keys, and the keyboard turns them into the keys RIME’s schemas expect (bopomofo onto the Dachen key map, Cangjie radicals onto their letters, strokes onto h s p n z). Only the conversion is RIME’s: the candidates, how far each one reaches into what you typed, and what it learns from your picks.

RIME needs three things before it takes over, and the group lists them as they arrive:

1. Its library, about 3 MB. The Google Play edition fetches it from Play, the GitHub edition from the project’s data repository, and the F-Droid edition from the same place once Download feature libraries is on under Privacy. Picking RIME asks for it, after the usual question on a metered connection.
2. A schema pack. Each row downloads one, together with RIME essentials (punctuation, symbols, a word-frequency table and the Simplified and Traditional tables every schema uses). Each pack is checked against a SHA-256 built into the app, like the conversion dictionaries.
3. Compiling. RIME turns schemas into fast lookup tables on the phone, which it calls deploying. The RIME status row starts it by itself as soon as a download finishes and shows each schema as it goes. Compile runs it again. Luna Pinyin takes seconds; Rime Ice, with its far bigger vocabulary, can take a minute or more and a few hundred megabytes of memory while it runs.

Until all three are in place, and while a compile is running, the language keeps typing on the built-in engine, so switching never leaves you unable to type.

Pack	Schemas	Layout it serves	Licence of the data
Luna Pinyin (朙月拼音)	luna_pinyin and its Simplified, Taiwan and sentence variants, luna_quanpin, stroke	Pinyin, Stroke	LGPL-3.0
Rime Ice (雾凇拼音)	rime_ice, Double Pinyin schemes, t9	Pinyin, T9 Pinyin	GPL-3.0
Bopomofo (注音)	bopomofo and its variants, terra_pinyin	Zhuyin	LGPL-3.0
Cangjie 5 (倉頡五代)	cangjie5, cangjie5_express	Cangjie	LGPL-3.0
Jyutping (粵拼)	jyut6ping3	Cantonese Jyutping	CC BY 4.0 and ODbL

Each pack carries its upstream licence in a licenses folder inside the zip. The Rime Ice pack leaves out its Tencent word-vector list, the largest table, which would take minutes to compile on a phone. RIME’s Lua plugin is not in this build either, so the Lua parts of Rime Ice (its date, calculator and correction scripts) are skipped and the rest of the schema works. It would make the library a third bigger and let any imported schema run its own code inside the keyboard, and the built-in engine has date, time and Chinese numeral shortcuts of its own.

Which schema a layout uses. Once something is compiled, a picker appears per kind of layout: Pinyin layouts, 9-key Pinyin layout, Zhuyin layout, Cangjie layout and Stroke layout for Chinese, Jyutping layout for Cantonese. Automatic takes the first downloaded pack that serves the layout, in the order of the table above. Cangjie Quick has no RIME schema here and stays on the built-in engine.

The 9-key pad. Rime Ice’s t9 schema serves the 9-key Pinyin layout, and the syllable column works as it does on the built-in engine: it lists what the next keys could spell, longest first, with the syllable RIME’s first candidate reads there at the top, and a tap settles that syllable for RIME too. The text you are typing shows RIME’s spelling of its first candidate, xi'an for 西安. Rime Ice is the only pack with a 9-key schema, so without it the 9-key layout stays on the built-in engine.

What comes next. RIME has no next-word suggestions of its own, so once a conversion lands, Chinese gets the keyboard’s own what comes next, whichever schema typed the text: your own habits first, then the Chinese word pairs pack, which you can download from the language’s dictionaries. Cantonese gets none, because the pairs are Mandarin.

Your own schemas. Import RIME schemas takes a zip of .schema.yaml and .dict.yaml files, the way most schema projects publish them on GitHub. The folder inside the zip doesn’t matter: the keyboard unpacks the files beside the first schema it finds. A file one of the packs already provides is kept as the pack has it, and so is RIME’s own default.yaml. An imported schema compiles with the packs and shows up in every picker. The same zip also installs through Import on the Storage screen, or by opening it with the app, which recognises it by the schema inside. Each import gets its own row with a delete button.

Traditional characters and Regional wording work here too: they switch on the matching filter inside the schema (Simplified, Traditional, Taiwan or Hong Kong forms), so a candidate is converted before the keyboard measures how much of your reading it covers. Cantonese stays in Traditional, as the built-in Jyutping dictionary is.

Words on the keys. On the Pinyin layout, candidates float over the key that carries on their reading, read from the spelling RIME shows for each one, so a schema that shows no spelling floats none. After a commit, the words from what comes next float over the key that starts them, which needs the Chinese dictionary pack as well.

Learning. RIME learns into its own user dictionary when you pick a candidate, and it learns whole phrases: picking 你好 and then 世界 for nihaoshijie teaches it 你好世界. Like the rest of the keyboard’s learning, it learns nothing in incognito or a password field. Delete RIME data removes the compiled schemas, your imported ones and everything RIME learned; the downloaded packs stay and compile again.

Before the first unlock. RIME’s schemas and its user dictionary sit in the app’s credential-protected storage, with your other learned words, so on a phone that has restarted and not been unlocked yet, Chinese types on the built-in engine. RIME takes over again after you unlock.

The Mozc engine

WM Keyboard › Languages & layouts › Japanese › Conversion engine › Japanese engine

Japanese can convert on Mozc instead of the keyboard’s own engine. Mozc is Google’s open-source Japanese engine, the one inside Google Japanese Input and Gboard’s Japanese keyboard. Where the built-in engine looks up words, Mozc converts whole sentences, so kyouhaiitenkidesune comes out as 今日はいい天気ですね in one pick. It also predicts as you type, splits a reading into its parts (文節) the way a desktop Japanese input method does, and learns from what you pick.

Japanese engine picks Built-in or Mozc, and Built-in is the default. Every Japanese layout converts through Mozc once it’s picked: Romaji, Godan, Flick (the 12-key pad), Kana JIS, 50 keys and any layout of your own that composes kana. The keys don’t change. The keyboard still turns romaji into kana itself, the 小゛゜ key still cycles the last kana, and only the conversion is Mozc’s.

Mozc needs two downloads before it takes over, and the group lists them as they arrive:

1. Its library, about 3 MB. The Google Play edition fetches it from Play, the GitHub edition from the project’s data repository, and the F-Droid edition from the same place once Download feature libraries is on under Privacy. Picking Mozc asks for it, after the usual question on a metered connection.
2. The Mozc dictionary, about 19 MB: Mozc’s own system dictionary, connection costs and segmenter, built from Mozc’s sources at the same release as the library. Picking Mozc starts this download too, and it is checked against a SHA-256 built into the app like the other conversion dictionaries. A phone that can’t download gets an import row under it, with a link to the file to fetch on another device. The same file installs offline through Import on the Storage screen.

Mozc status says where it stands. Until both are in place, and before the phone’s first unlock, Japanese keeps typing on the built-in engine, so switching never leaves you unable to type. The built-in kana dictionary above is not needed for Mozc and can stay or go.

The candidates. While you type, the strip shows Mozc’s predictions: whole-sentence conversions first, then words and phrases that cover only the start of the reading. Picking one of those commits that much and leaves the rest of the reading composing, with its own candidates already up. The plain hiragana, katakana and half-width katakana of the reading stay at the end of the list. The expanded grid pages further down the same list. After you pick, the strip offers what Mozc expects next.

Moving the boundaries. The arrow keys work on the reading while it’s composing, as they do on Gboard’s Japanese pad. ← and → split the reading into its parts if it isn’t yet, then narrow or widen the highlighted part by one kana. The field shows every part converted, with the highlighted one shaded, and the strip lists the candidates for that part alone. With selection on, the arrows move the highlight to the part before or after instead. Picking a candidate commits every part up to the highlighted one, each as it shows, and the rest stays split so the next part’s candidates come straight up. Backspace or another key goes back to typing the reading. The arrow row and the text-editing pad both have these arrows.

On a physical keyboard it works as on a desktop: Shift+←/→ resizes the highlighted part, and once the reading is split ←/→ move between parts while ↑/↓ step through the candidates.

Match kana without marks works here too, through Mozc’s own version of it, so かつこう still finds 学校.

Learning. Mozc learns from each pick: which conversion a reading takes, where you put the boundaries, and what you type after what, which is how its predictions get better. Like the rest of the keyboard’s learning, it learns nothing in incognito or a password field. What it learns is written to the app’s credential-protected storage a few seconds after you pick. Forget what Mozc learned clears it and keeps the dictionary.

Details & edge cases

• A Japanese word-list download exists but does nothing for typing. Japanese has its own entry in the ordinary downloadable-dictionary catalog (see Downloadable dictionaries), but because Japanese is a conversion input method, the regular suggestion engine never runs for it. That download currently has no effect on any of the three Japanese layouts. Korean’s word list, by contrast, is fully active, since Korean isn’t a conversion method.
• The Traditional/region toggles show on the Japanese screen but are inert there. JapaneseComposer never applies the Han-variant conversion, so leave those settings alone if you’re a Japanese-only typist. They were placed above the Chinese-specific options rather than gated per-language, but they have nothing to act on for Japanese.
• Cangjie Quick needs the same dictionary as full Cangjie. With only one key typed, Quick behaves like an ordinary prefix search, since the “last” radical isn’t known yet. The second key turns it into a first-and-last shape match, which usually narrows to a handful of candidates.
• The candidate depth differs by scheme. Pinyin, Zhuyin, T9 Pinyin, Jyutping, and Japanese all rank up to 100 deep in the expanded grid. Cangjie and Stroke cap at 24, because they look candidates up from a fixed code table rather than a ranked lattice.
• There’s no “lock” or “pin” on a candidate. Once you commit one, it only affects future ranking through the learning history above. There’s no separate pinning feature to hold a candidate in place.