The IME Method

The IME Method is a recall protocol, a pinyin keyboard’s candidate window run in the head. Typing Chinese is recognition: feed a pronunciation to an input method editor and pick the form that looks right. Writing is production, the harder direction; the method closes the gap by crossing a sound-part list with a meaning-part list until one stored pairing looks familiar.

The loop

A form goes missing mid-stroke. Three lists get built. Sound components that could carry the syllable. Meaning components that fit the sense — especially the original sense, the meaning the character was invented to write. Then the two lists are crossed until one pairing feels familiar.

It works because memories interconnect. A temporarily forgotten form remains reachable through neighbours still held — sound and meaning — by walking associative links. Reciting the alphabet to surface a name is the everyday analogue. Anything genuinely learned remains stored, waiting on the right cue. The prerequisite is character knowledge organized by functional components. A few dozen characters learned that way already support the method on those characters.

The sound list is harvested first. If none surface, sound components are pulled from other characters with that pronunciation, and the search runs across related initials: {b, p, f} behave as one group. Sound Components supplies the sound-side candidates. Sound Series is the related-initial map.

The meaning list is built from original senses. 頁 qualifies as a head-component only because it originally meant “head.” Meaning Trees and Original Meanings is where those original senses live. Form Components is the depicting role behind them. Meaning Components supplies the sense-side candidates.

Cross-match continues until recognition fires. The learned combination feels different from invented ones. This is recognition, not invention. When the components are right and the order is not, every plausible layout is written and recognition arbitrates.

Heard tones cannot be trusted at face value. Mandarin sandhi moves third tone, 一, and 不; the dictionary’s citation tone may differ from what was heard. The syllable is searched across tones.

Any fragment can start the walk. Components recall characters; characters recall components. The loop is practised during ordinary recall review, so it is already warm when a form stalls. Surface vs Deep Structure is the functional decomposition the lists draw on.

Two reconstructions

頓 in 停頓 had gone unwritten for more than a year. The sound list produced 屯. The meaning list produced 頁, still readable as a head. 屯 plus 頁 is 頓. Recognition fired on a stored form, not on a new invention. (享 as a lookalike in the dun neighbourhood is a short extra; it is not the pairing that fired.)

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 fired. Sandhi plus a related-initial group is what made the heard syllable usable.

What the loop is

A blocked production problem becomes a self-generated multiple-choice question. The price is the functional-component organization the loop assumes, plus the seconds of listing. The benefit in the same breath: a missing form that still has neighbours is reachable.

The same loop flips. A known form whose reading or meaning has gone dark is not a different method. It is the same lists, asked in reverse.

Reverse gear, and the lookup diet

The flipped question: which part most likely gives sound, which meaning. It works best on the one-sound-plus-one-meaning build most characters use.

Role tendencies prune the lists. 各 almost always plays sound. 穴 always plays meaning — 空, 穿, 窗.

Swing components still resolve. 立 is meaning in 站 and 端, sound in 拉 and 粒, both in 位. The walk is reactivating a stored character, never decoding a cold one.

Unlearned characters get a constrained guess, then a lookup. 盆 in 臉盆: container 皿 plus sound 分 suggests fen / ben / pen; lookup returns pén; {b, p, f} licenses 分. An unlearned form stays uncertain until that lookup. A confirmed guess exposes the systematic connection — meaningfulness, association, organization — the general rules on Rules of Effective Memorization.

Circling is a different job. It is a lookup diet, not the production protocol. Constant lookup means never actually reading. An unknown is marked at most once per page; lookup waits until a count — three is a workable start; the definition is written in the margin of a book already owned; character and page are logged on the last page. If lookups still crowd reading, the threshold rises to four or five and the material drops to only slightly above level. On a screen, tap-lookup costs nearly nothing, so the threshold has to be self-imposed.

Where the strings do not reach

Storage bounds retrieval. Never-learned characters get the guess-and-verify route, not string-plucking.

Only durable knowledge can cue. Mnemonic stories and folk etymologies decay. Sounds and meanings of words actually read and written persist.

Typing tempts surrender. IME typing gets cited as proof handwriting should be skipped. Production is harder, and a working model of how characters encode sound and meaning makes it less hard. Handwriting is not a requirement of this page. Recognition-first is a live alternative. The protocol still helps in reverse on a known form.

Component training spots sound components natives miss — sound change hides relations. Natives stay more correct about most things, on massive experience. There is no superiority story here.

The method cannot retrieve what was never stored. It does not replace intake, and it does not replace review. Circling on a screen is still unsolved. If every recovery is a cold invention that fails lookup, the functional organization is not in place yet — the component pages come first. What “working” looks like: a character unwritten for a long time, whose sound and meaning are still held, becomes a short candidate list rather than a blank.

The same choice a keyboard offers is now runnable without one, in both directions, on what is actually stored. Circling is how reading stays reading.

Links into the knowledge base

• How Chinese Characters Work — the hub: the cluster this protocol runs on.
• Surface vs Deep Structure — functional decomposition the lists draw on.
• Sound Components — sound-side candidates.
• Meaning Components — meaning-side candidates.
• Form Components — depicting role behind original senses.
• Sound Series — related-initial groups ({b, p, f}).
• Meaning Trees and Original Meanings — original senses as retrieval cues.
• Rules of Effective Memorization — meaningfulness, association, organization.

Open Questions

What circling threshold holds when 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 treats that gap as the thing the in-head window closes.

Taft & Zhu 1997: component knowledge aids recall. The lists this protocol builds are those components, generated then recognized.

Standard Mandarin tone sandhi (third tone, 一, 不) — why a heard syllable is searched across tones before the sound list is closed.