The Pipelining Strategy
The Pipelining Strategy
The Pipelining Strategy is a schedule for taking in new Chinese characters. A character has three parts: its sound, its meaning, and its written form. The schedule spreads those three parts across three study sessions. That spread is the stagger. The learner learns the spoken word first, in both directions. Then the learner attaches the written form to a spoken word already known.
Binding a character to a spoken word the learner already owns costs less than learning form, sound, and meaning as three new things at once. The three sessions are arranged so that the learner owns the spoken word, both ways, before the character appears on a card. By then the spoken pair is overlearned.
Core takeaways
- A new written form is cheaper to learn when the spoken word is already known. One new link onto a known pair costs less than three new items at once.
- Each character takes three sessions: the spoken word in one direction, the same pair reversed, then the written form.
- A new cohort of 3 to 5 characters starts every session. Cohorts overlap, so in steady state one cohort finishes every session.
- The schedule covers intake only. After the form session, a character becomes an ordinary review card under Spaced Interleaved Retrieval.
- The numbers 3 to 5 new characters and 20 items per session are house defaults, not measured findings.
- The schedule is working when, after a few weeks, one cohort still exits each session and form cards feel like attaching a shape to a known word.
Why the usual flashcard costs more
A native child learning to write already knows the spoken word. Because the sound and meaning are known, plain repetition is enough to reach literacy. Analysis of a character’s internal structure is a tool for foreign learners. A native reader does not need it.
The default foreign-learner flashcard gives up the advantage of knowing the spoken word first. The front reads “self; from” and the back reads 自. Form, meaning, and sound all arrive in memory in one session. At the same time the learner is also working on grammar, tones, and vocabulary. Three Attributes of a Character describes the three parts this schedule separates. Rules of Effective Memorization gives the general rule the schedule follows: one new link at a time.
The three sessions
Session 1 is the spoken word in one direction. A card holds a pronunciation on one side and its meaning on the other. No written form appears anywhere on the card.
Session 2 is the same pair in the reverse direction. The forward link already exists, so the backward link costs less than learning the pair from nothing. How much less has not been measured.
Session 3 attaches the form. The front reads “dào: way, path” and the back reads 道. The shape attaches to a pair the learner already holds, instead of arriving as a third new item.
Either direction can go first. Meaning→pronunciation or pronunciation→meaning can be session 1. The fixed part is the order: one direction, then its reverse, then the form.
One session a day is the usual pace. The schedule requires the three-step order. It does not require one session a day.
Overlapping cohorts
A cohort is a small set of characters that move through the three sessions together. A cohort holds 3 to 5 characters. The 3 to 5 figure is a house default, not a measured finding. Each cohort spends three sessions in the rotation. A new cohort enters every session, so in steady state one cohort finishes every session.
The overlap keeps the number of finishing cohorts steady. Session 2 holds cohort B new plus cohort A reversed. Session 3 holds cohort C new plus cohort B reversed plus cohort A forms.
A worked rotation:
- A = 道 那 但
- B = 前 很 性
- C = 情 物 等
A common mistake is to read the schedule as five characters per three sessions. That rate is slower than one cohort per session. The slower rate comes from sequential batching: one cohort finishes all three sessions before the next one starts. Starting a new cohort every session avoids the slowdown. Sequential batching drops intake to one finished cohort every three sessions and gains nothing.
Each pairing of a pronunciation and a meaning gets its own card. 道 dào and 道 dǎo get separate cards. 那 nà and 那 nèi get separate cards too. So a cohort of three characters can produce five session-1 cards. Merging the readings onto one card is an acceptable personal variant.
What the stagger saves and what it costs
By the form session, two of the three parts are already known. Over thousands of characters, that saving per character adds up.
The cost is three sessions of attention per cohort before the form is on a card at all. The benefit is that the form attaches to a pair the learner has overlearned.
Bundled cards hold form, sound, and meaning while all three are unfamiliar. They triple the new material per repetition. The word triple describes the cost structure. It has not been measured. Bundled cards are the default that the stagger replaces.
Copying characters by rote, the way native learners do, fails without the spoken word already in place.
Session size and host words
A session holds 20 items. Of those, 3 to 5 are new characters. Both numbers are house defaults, not measured findings. The rest of the session is common words built from the new characters or from characters already known. A split of 3 new characters and 17 words is typical, and also a house default.
Each new character gets one or two high-frequency words that contain the character. Those host words put the written shape into a real sentence. They replace study of the character on its own.
Delaying the form session
The wait before the form session can be longer than one session. Three sessions in a row is the compact default. The form session can also wait a month or two after the spoken pair is overlearned. The cheap attachment still works then. The month or two is a house default, not a measured finding. The longest workable delay is not known.
Review after intake
The schedule covers intake only. After the form session it no longer applies to the character. The cohort leaves the rotation, and its characters continue only as ordinary review cards.
Spaced Interleaved Retrieval covers everything after that. SIR is retrieval spread over time and mixed across topics and forms. Spacing, interleaving, and ongoing reconstruction belong to SIR. The stagger never sets review timing.
A study session is the intake cards for the active cohorts plus every review SIR has scheduled. Intake cards never take the place of due reviews.
What the schedule does not cover
The schedule does not choose which senses of a character get cards. Reading sample glosses as a curriculum of meanings is a separate decision. Meaning Trees and Original Meanings owns it.
The schedule does not retrieve missing forms. The IME Method is production recall after intake. It is a separate method.
The schedule does not space reviews. That belongs to Spaced Interleaved Retrieval.
Signs it is working or failing
Early doubt is expected. Run the stagger for a few weeks before judging it.
If form-session cards still feel like three unfamiliar parts, the spoken pair was never overlearned. Then new cohorts wait, and the existing cohorts keep reversing for longer.
The schedule is working when, after a few weeks, one cohort still exits each session. Form-session cards then feel like attaching a shape to a word already held, not like a third new item.
How to practice this
- Make a card with the pronunciation on one side and the meaning on the other, with no character on it. Study it in one direction in the first session. Notice that you are learning one link, with no shape to hold.
- In the next session, study the same pair in the reverse direction. Notice that the reverse takes less effort than the pair took the first time.
- In the third session, add a card: pronunciation and meaning on the front, the character on the back. Notice whether the shape attaches to a word you already hold, or feels like a third new item.
- Start a new cohort of 3 to 5 characters every session, so three cohorts are active at once. Notice that one cohort finishes every session once the rotation is running.
- Keep new characters to 3 to 5 inside a 20-item session. Fill the rest with common words that contain the new characters. Notice that each character then appears inside a real word rather than alone.
- If form-session cards still feel like three unfamiliar parts, stop adding cohorts and keep reversing the existing pairs. After a few weeks, notice whether one cohort still exits each session.
Related pages
- How Chinese Characters Work: the overview page for the group of character pages this schedule completes.
- Three Attributes of a Character: the three parts this schedule spreads across sessions.
- Spaced Interleaved Retrieval: review after a cohort leaves intake.
- Meaning Trees and Original Meanings: which senses get cards.
- The IME Method: production recall after intake.
Open questions
One card per pairing versus merged cards for a character with two readings: which holds in practice?
How far the form session can be deferred after the spoken pair is overlearned, past the house “month or two.”
How a 20-item intake budget fits with a large SIR backlog on the same day.
Sources
Share, David L. “Phonological recoding and self-teaching: sine qua non of reading acquisition.” Cognition 55 (1995): 151–218. Phonological-primary, orthographic-secondary: a new written form attaches more cheaply to a word already held by sound.
L2 character work showing oral associations developing before form associations, for example the open-access study at PMC10313221. That study supports spoken-first as the cheaper order rather than a house preference.
Paired-associate and incremental vocabulary research (ERIC ED550749 and similar studies): a new link onto an existing pair costs less than three new nodes at once. Backward recall after a forward link is the step this schedule uses in session 2.