The IME Method
The IME Method
The IME Method is a way to recall a Chinese character form that will not come while writing. The writer builds, from memory, the same candidate list a pinyin keyboard would show, without the keyboard.
Typing Chinese is recognition. The typist gives a pronunciation to an input method editor and picks the form that looks right from the candidates it shows. Writing a character by hand is production, and production is harder than recognition. The method makes production less hard by turning it into recognition. The writer lists sound components that could carry the syllable. The writer lists meaning components that fit the sense. The writer crosses the two lists until one stored pairing looks familiar.
Core takeaways
- When a form will not come, the writer lists sound components that could carry the syllable and meaning components that fit the sense, then crosses the lists until one pairing looks familiar.
- The method works because a learned character stays stored and can be reached through its sound and meaning components. This requires character knowledge organized by functional components.
- A heard syllable is searched across tones and across related initials such as {b, p, f}, because tone sandhi and sound change can hide the dictionary reading.
- The same lists run in reverse to recover the reading or meaning of a known form. A character that was never learned gets a constrained guess and then a lookup.
- Circling limits lookups during reading. The reader marks an unknown character at most once per page and looks it up only after a set count, starting at three.
- The method retrieves only what was stored. It does not replace intake or review.
How the loop runs
A form will not come while the writer is writing. The writer builds three lists. The first holds sound components that could carry the syllable. The second holds meaning components that fit the sense, especially the original sense, the meaning the character was first invented to write. The third holds the pairings made by crossing the first two. The writer goes through the pairings until one looks familiar.
The method works because memories are linked to each other. A form that cannot be recalled at the moment is still reachable through the sound and meaning components the writer still holds. Reciting the alphabet to recall a person’s name works the same way. Anything genuinely learned remains stored and can be reached with the right cue. The method requires character knowledge organized by functional components. A few dozen characters learned that way are enough to run the method on those characters.
Building the sound list
The sound list comes first. The writer lists sound components that could carry the syllable. If none come to mind, the writer takes sound components from other characters with that pronunciation. The search also covers related initials: {b, p, f} are searched as one group. Sound Components supplies the sound-side candidates. Sound Series maps the related initials.
A heard tone may not be the dictionary tone. Mandarin tone sandhi changes the third tone, 一, and 不. The citation tone in a dictionary can therefore differ from the tone that was heard. The writer searches the syllable across all tones.
Building the meaning list
The meaning list is built from original senses. 頁 counts as a head component only because it originally meant “head.” Meaning Trees and Original Meanings lists those original senses. Form Components describes the depicting role behind them. Meaning Components supplies the sense-side candidates.
Crossing the lists
The writer pairs items from the two lists until one pairing looks familiar. A pairing that was learned looks different from a pairing invented on the spot. The familiar pairing is a form already in memory. Sometimes the components are right and their order is wrong. Then the writer writes out every plausible layout and picks the one that looks familiar.
Any fragment can start the search. A component can bring back a character that contains it, and a character can bring back its components. The writer practises the loop during ordinary recall review, so the loop is already familiar when a form will not come. Surface vs Deep Structure describes the functional decomposition the lists draw on.
Two recovered characters
頓 in 停頓 had not been written for more than a year. The sound list produced 屯. The meaning list produced 頁, which still reads as a head. 屯 plus 頁 is 頓. The writer recognized a form already stored in memory. 享 came up as a lookalike while searching around the syllable dun. It is a short extra, and it is not the pairing that was recognized.
綁 was recovered with no memory of having studied it. A spoken phrase about closing a lunch box was heard as báng before 起來. The citation reading is bǎng. The sound list ran {b, p, f} against that syllable, across tones. The meaning list ran “bind / tie.” The stored form was recognized. Tone sandhi plus the related-initial group made the heard syllable usable.
What the loop costs and what it gives
The writer turns a blocked recall into a multiple-choice question they make themselves. The cost is the functional-component organization the loop needs, plus the seconds spent listing. The gain is that a missing form whose sound and meaning components are still stored is reachable.
The same loop runs in reverse. A known form whose reading or meaning cannot be recalled uses the same lists, asked the other way round.
Running the loop in reverse
The reverse question is which part of the form most likely gives the sound and which part gives the meaning. It works best on the one-sound-plus-one-meaning build that most characters use.
Knowing which role a component usually plays shortens the lists. 各 almost always gives sound. 穴 always gives meaning, as in 空, 穿, 窗.
Components that take either role still resolve. 立 gives meaning in 站 and 端, sound in 拉 and 粒, and both in 位. The search reactivates a stored character. It does not decode a character that was never learned.
A character that was never learned gets a constrained guess and then a lookup. 盆 in 臉盆: the container component 皿 plus the sound component 分 suggests fen, ben, or pen. The lookup returns pén. The {b, p, f} group allows 分 as the sound component. A form that was never learned stays uncertain until that lookup. A confirmed guess shows the systematic connection between the parts and the reading. Meaningfulness, association, and organization, the general rules on Rules of Effective Memorization, describe that connection.
Limiting lookups while reading
Circling is a separate practice for limiting lookups while reading. It is not part of the recall loop above. Constant lookup means the reader never actually reads. The reader marks an unknown character at most once per page. The reader looks it up only after it has been marked a set number of times; three is a workable starting threshold. The definition is written in the margin of a book the reader owns. The character and the page are logged on the last page. If lookups still crowd reading, the threshold rises to four or five and the reading material drops to only slightly above the reader’s level. On a screen, a tap lookup costs nearly nothing, so the threshold has to be self-imposed.
What the method cannot do
The method retrieves only what was stored. A character that was never learned gets the guess-and-lookup route described above, not the list search.
Only durable knowledge can serve as a cue. Mnemonic stories and folk etymologies fade. The sounds and meanings of words the reader has actually read and written last.
IME typing is sometimes cited as proof that handwriting can be skipped. Production is harder than recognition, and a working model of how characters encode sound and meaning makes production less hard. The method does not require handwriting. Recognition-first study is a live alternative. The loop still helps in reverse on a known form.
Component training finds sound components that native speakers miss, because sound change hides the relation between a component and the reading. Native speakers remain more correct about most things, on massive experience. Component training does not make a learner better than a native speaker.
The method cannot retrieve what was never stored. It does not replace intake or review. Circling on a screen is still unsolved. Every recovery may turn out to be a cold invention that fails lookup. In that case the functional organization is not in place yet. The component pages come first. When the method works, a character not written for a long time, whose sound and meaning are still held, becomes a short candidate list instead of a blank.
The choice a keyboard’s candidate window offers can be made without a keyboard, in both directions, on characters that are stored. Circling keeps the reader reading instead of looking characters up.
How to practice this
- List sound components that could carry the missing syllable, searching {b, p, f} as one group and across tones. If none come, take sound components from other characters with that pronunciation. Notice that a heard tone may differ from the dictionary tone.
- List meaning components that fit the sense, using each component’s original sense. Notice that 頁 counts as a head component because it originally meant “head.”
- Cross the two lists until one pairing looks familiar. Notice that a stored pairing looks different from one you invent. If the parts are right but the order is not, write every plausible layout and pick the familiar one.
- Run the same lists in reverse on a known form whose reading or meaning you cannot recall. Ask which part most likely gives the sound and which gives the meaning. Notice that 各 almost always gives sound and 穴 always gives meaning.
- For a character you never learned, make a constrained guess and then look it up. Notice that the guess stays uncertain until the lookup confirms it.
- While reading, mark an unknown character at most once per page and look it up only after three markings. Write the definition in the margin and log the character and page on the last page. If lookups still crowd reading, raise the threshold to four or five and read material only slightly above level.
Related pages
- How Chinese Characters Work: the hub page for the character cluster this method runs on.
- Surface vs Deep Structure: the functional decomposition the lists draw on.
- Sound Components: sound-side candidates.
- Meaning Components: meaning-side candidates.
- Form Components: the depicting role behind original senses.
Open questions
What circling threshold holds when a tap lookup costs nearly nothing?
What threshold is current, in practice, on paper?
Where does this loop sit inside Spaced Interleaved Retrieval as scheduled practice, rather than only as an emergency?
Sources
Recognition is easier than production: the CFL recognition/production gap, and keyboarding-versus-handwriting studies. The method aims to close that gap with the candidate list built in memory.
Taft & Zhu 1997: component knowledge aids recall. The lists this method builds are those components, generated and then recognized.
Standard Mandarin tone sandhi (third tone, 一, 不): why a heard syllable is searched across tones before the sound list is closed.