8.2 Manage Working Memory With Chunking

Chunk the rule, then apply it

Working memory is the small mental workspace you use to hold information while doing something with it. Classic estimates put its capacity at only about four chunks at once. On the DLAB that information might be a sound contrast, a word-order rule, a plural ending, a tense marker, or a symbol-to-meaning mapping. The challenge is not just noticing the rule — it is keeping it stable long enough to answer, often after the audio has already stopped.

Chunking means collapsing several details into one unit. Instead of holding "the subject comes first, the object second, and the verb last," you hold "SOV." Instead of "the ending -im signals more than one," you hold "im = plural." Each chunk frees capacity for the next piece of evidence.

Practice-style example, not official DLAB content:

A weak approach silently repeats every full sentence. A strong approach chunks: raku = driver, savi = guard, el = one, ir = many, order = SOV. Now "The guard moves many carts" assembles directly as savi mon-ir te. You held five tiny chunks instead of three long sentences.

Match the chunk to the DLAB skill

The form of the chunk changes with the item type, but the habit is identical:

• Sound items: "stress on syllable 2," "final consonant flips meaning," "long vowel = different word." Lock this on first listen because the clip will not repeat.
• Word-order items: "SOV," "adjective after noun," "negative goes last," "question particle first."
• Morphology items: "-na = not," "-ir = plural," "prefix ke- = actor."
• Visual-symbol items: "circle = person," "double line = plural," "shading = past."

Keep each chunk short enough to survive interruption. A label like "the suffix that appears at the end of the second word indicating more than one" is itself a memory load; "im = plural" is not.

Sequence and recovery

Use the order observe → contrast → label → apply. Do not label before you have compared at least two examples, or you may lock onto the wrong pattern (the overfitting trap covered in 8.4). When a chunk slips mid-item, do not reread everything — return to the single cleanest contrast pair (two sentences that differ by one element). One clean pair rebuilds the rule fast.

Fatigue makes working memory noisier, and on a 126-item test the systems blur late. Reset between items by mentally saying "new system" so a plural marker from item 40 does not contaminate item 41, where the same-looking ending might mean past tense or nothing.

Log memory failures separately from reasoning failures. "Forgot the plural marker" calls for chunking drills; "misidentified the plural marker" calls for better contrast analysis. Because the DLAB is an aptitude test built on constructed languages, chunking generalizable rule structure is a far better preparation target than memorizing fake vocabulary from a commercial drill — there is no fixed real language to learn, so flexible rule handling is the actual skill being measured.

The best chunk is accurate, brief, and disposable: accurate enough to fit the examples, brief enough to hold under the clock, and disposable the instant the next constructed system begins.

A worked chunking procedure

Walk through a full item the way you should under time pressure. Suppose you are given three examples and must build a fourth target sentence.

Target: The scientist reads many books.

Step one, observe: three sentences, each three words, with the middle word changing endings. Step two, contrast: polu versus mira swaps the actor, and -am versus -az swaps one versus many; dru and the order never change. Step three, label four tiny chunks: polu = teacher, mira = scientist, -am = one, -az = many, order fixed. Step four, apply: scientist + many books = mira vek-az dru. You held five chunks, never five sentences.

Notice what chunking protected you from: you did not waste capacity wondering whether dru carries meaning (it is constant, so it is just "book/object" and you set it aside), and you did not assume English structure. Constant elements are free — chunk them once and stop attending to them.

When chunks collide

The hardest working-memory failure on a long test is proactive interference: an earlier chunk intrudes on a later, similar-looking system. If item 30 used -az for plural and item 31 uses -az for past tense, your fingers want the old answer. Two defenses help. First, attach a system tag to each chunk in your head ("system-31: az = PAST"), so the label itself carries its scope. Second, when two systems feel confusable, re-derive from one fresh contrast pair in the current item before answering — never from memory of the previous one.

Chunking is not memorizing forever; it is holding the right rule for exactly one item and then letting it go.