Chapter 11 · Four Strands That Decide Understanding

We said the situation model is built by the inferential synthesis that fills gaps. Then where does the difference come from—reading the same text, one accomplishes that synthesis while another stays at the text base? There are four strands that govern which propositions remain in long-term memory to become material for the next synthesis, and whether the gap-filling inference happens or is blocked. The four look as if they run on their own, but in the end they gather into one question: while reading the text, does inferential synthesis happen, or is it skipped?

What Remains · Re-entry and the Primacy of Surface

Which propositions remain in long-term memory? Working memory is narrow and cannot hold every proposition while reading; a processed proposition is soon pushed out. But a pushed-out proposition does not vanish forever. If a following sentence points again to that proposition's member, the proposition is summoned back up into working memory. The principle that a member is a cue operates here too. If one proposition shares members repeatedly with several following sentences, it is summoned up repeatedly, and each time it is summoned it is processed once more, accumulating the probability of remaining in long-term memory.

That a text's central claim is remembered far better than the details is due to this accumulation of re-recall. The central claim shares members repeatedly with following sentences throughout the text, so it is summoned up again and again and its count of processings piles up, while a detail in one corner is processed once and, with no occasion to be summoned again, slips away. What matters here is that what remains is not set by a declaration that "this is important." Even if the body writes "this is the key point," if that representation does not share members with following sentences, it is not summoned up again and slips away as is. Conversely, even unmarked as important, a representation entangled with several sentences is summoned again and again and remains. What decides what remains in memory is not the declaration of importance but how often it is summoned again within the text's weave.

Gaps Call for Synthesis · The Paradox of Cohesion

Next is whether the gap-filling inference happens. Intuitively, a smooth text that spells out every connection always looks good. But it is not so. When there is a connection not spelled out in the text—a gap—the reader draws on long-term memory's background knowledge and fills that spot by inference, and that very inference is the synthesis that builds the situation model. A high-cohesion text that spoon-feeds every connection does away with this filling. With no gap to fill, the reader passes smoothly through the text base with no occasion to infer, and the synthesis that would build a situation model does not happen.

So for a reader with sufficient background knowledge, a gapped, low-cohesion text yields the better result—and that difference can be confirmed only on the inference-and-application test that measures the situation model. But this effect is tied to the reader. To a reader without the background knowledge to fill the gaps, a low-cohesion text is merely an empty hole with no material to fill it, and for such a reader a high-cohesion text that spells out every connection is better. So neither "a text that fills in every connection is good" nor "a text that leaves gaps is good" is true in itself. Whether the reader has the material to fill those gaps divides the answer.

Making Yourself Ask · Self-Explanation

There is also a way to force the gap-filling inference on oneself: while reading, asking oneself "why is that so" and "how does this connect to what came before." This question makes you draw causal knowledge from long-term memory and connect it to the text's propositions—that is, raising inferential synthesis with your own hands. That one who reads asking "why" in a string understands more deeply than one who merely reads the same text along is because those questions raised more of the synthesis that builds the situation model, and indeed understanding deepens in proportion to how many explanations were actually produced.

But the same condition attaches to this method too. Even if you ask "why," if the material to build the answer is not in long-term memory, the question ends in empty processing. There must be causal knowledge to draw on for the synthesis of connecting it to the text to happen; ask "why" a hundred times from nothing and there is nothing to fill with. Self-explanation works when there is material to retrieve.

If retrieval fails, there is no path ridden and so nothing firmed. Too easy, and retrieval strength is high, so riding the path firms it almost none; too hard and it fails, so the path is not ridden at all and again nothing is firmed. Reinforcement comes only from that passage in which fallen retrieval strength was nonetheless hauled up with effort and succeeded. So retrieval practice too is premised on success. In one analysis, when the success rate of recall fell below half and no answer was given either, the testing effect effectively vanished. But when the answer was checked right after recalling—that is, when feedback was paired—the effect revived greatly even at a low success rate. When, right after groping a path and failing in the end to find it, the correct path is checked, that groping effort is not wasted but becomes, if anything, deeper learning.

So the condition that makes difficulty desirable is clear: it must be a difficulty the learner can, with help if needed, finally succeed at. If you have them recall, check the answer right after, or ask at a level mostly likely to succeed. Have them recall, give no answer, and gauge no difficulty, and what was meant to help becomes wasted effort.

Does the Text Guide Synthesis? · Connectives and Form

The last strand is how much the text tells the direction of the inferential synthesis. Connectives that specify the actual relation between propositions—the contrastive "but" or the causal "because"—tell which relation to bind two propositions in and so aid synthesis. By contrast, additive connectives that specify no relation and only add things to process—"and" or "also"—the more they multiply, the more they lower understanding. Specifying no relation while only increasing the propositions to bind, they merely eat working memory's resource and do not guide synthesis. This is why a text strung along with "also" looks connected yet does not stay in the head. It is a list in the shape of connection.

The text's form works by the same logic. A narrative in which events are densely linked in a chain of cause is easier to synthesize into a situation model than an expository text that merely classifies and lists the same content. The narrative has the relations of "and so" and "because of that" laid densely, guiding the reader's synthesis at every step, while the listing expository text gives no relation to bind between items, so the reader must make each relation one by one.

Four Strands Gathering into One Question

The four strands gather, in the end, into one. The situation model is built in the reader's inferential synthesis that fills gaps, and whether that synthesis happens divides the depth of understanding. If the text spoon-feeds everything or fills itself with a relationless list, synthesis is skipped and only the text base remains; if the text leaves spots to fill and the reader has the material to fill them, synthesis happens and the situation model stands. What remains is set by how often the material of that synthesis is summoned again within the text's weave.