Chapter 7 · A Handful of Slots

Ask most people to memorize ten unfamiliar digits after hearing them once, and they collapse in the middle. Yet the same ten digits, broken into the familiar shape of a phone number, are far easier. What you can lay in working memory at once is a handful at most, and beyond that handful, taking in something new pushes out something old by that much. Working memory's first limit is this narrowness. Exactly how many it can hold differs from person to person, so I will not pin it to a number, but that the number is small enough to count on your fingers is clear. Nearly everything we do in our heads meets a bottleneck here.

The Same Handful, Different Loads

Being narrow does not mean the information you can handle is small. The unit working memory holds is the representation, and as seen earlier, the amount of information one representation holds is not fixed. Even within the same handful of slots, how much is carried diverges enormously depending on whether you lay a shallow representation or a deeply composed one onto it.

Ten unfamiliar digits are ten separate, unrelated representations and soon exceed a handful, but in the familiar shape of a number they bind into a couple of chunks and fit within a handful. The same information taking up fewer slots is because it is already bound into one unit. In the language of the network seen in Chapter 3, one deeply composed representation loads the whole hierarchy beneath it, compressed into a single slot. So binding lower representations more deeply is itself the path to widening working memory's load. This is why an expert handles far more than a novice with the same narrow working memory: not because the number of slots is greater, but because a deeper composite representation is laid onto each slot. The narrowness is identical; what differs is what is laid on that narrow place.

Cost That Swells with Depth

There is a scene in reading where this difference in load decides the success or failure of comprehension. To build the meaning of a sentence is to stack representations upward—from letters to words, from words to phrases, from phrases to the meaning of a sentence. To bind an upper-layer representation, its material—the lower-layer representations—must float in working memory, one unit each.

Here a split arises. For a reader whose lower-layer materials are sufficiently bound by repetition, that material is retrieved and rises straight up as one-slot units. See a word and its meaning surfaces as one lump; a familiar phrase becomes one unit whole. Then only a few materials for the topmost binding need float in working memory, and building the sentence's meaning finishes cheaply. But for a reader whose lower-layer materials are not bound, things differ. The meaning of even a single word must be built on the spot by inference from materials further down, and if those lower materials too are not in place, you go down one more layer and build. Because working memory is narrow and cannot spread all this material out at once, building proceeds bottom-up, one at a time. You build one lower representation and fold it into a unit, hold onto it while you build the next, and bind the units thus gathered at the upper layer.

The trouble is that a unit built earlier dims while the next is being built. What is in working memory leaks out within seconds unless held, and while your hands are on building the next material, the one built earlier, unheld, is pushed out of place. If what is built earlier leaks away like this before all the materials needed for the upper binding have gathered, you must build it up again. But rebuilding takes the same working memory, and meanwhile yet another thing is pushed out. The more and deeper the materials to build, the more frequent this backtracking, until the moment when all the materials needed for the topmost binding float at once never comes. Then the topmost representation, the sentence's meaning, never closes. Having read all the letters yet with no understanding standing—this is exactly that. Not from laziness, nor from a lack of focus, but because in the narrow working memory there was too much material to build, and what was built earlier collapsed while the later was being built.

Where Background Knowledge Decides Comprehension

This explains why the same text is easy for one person and impossible for another. The two readers' working memories are equally narrow; the number of slots does not differ. What differs is whether the lower-layer material is already bound into one unit. For a reader with background knowledge, each material needs only one slot, leaving room to spare in working memory, and only the topmost binding remains, so comprehension is relatively easy. For a reader without background knowledge, the same text becomes a matter of building up each material one by one; working memory fills with that building, what was built earlier collapses, and the topmost binding fails to close. So deeply composing something does not merely increase what you know; it secures room in working memory on which to build something higher. The more you have, the more places there are to lay something new.

The narrowness of working memory makes this large a difference. But narrowness is only one of working memory's limits. Apart from how many you can hold at once, how many operations you can run at once on top of what you hold is another limit. And the three operations seen earlier do not each run on their own but share one household. The next is the story of that household.