6.4 Rule Transfer to New Symbols and Scenes

From recognition to transfer

Recognizing a pattern in one example is not enough. Rule transfer means using that pattern on a new example. If a symbol marks "inside" for a circle in a box, can you apply it to a triangle in a bowl? If a label marks plural for three keys, can you apply it to three stars? Transfer is the step that turns visual-symbolic recognition into genuine language aptitude, and it is the most heavily rewarded skill on DLAB-style items.

A memorized association is brittle. If you memorize that a red circle above a square is "pa," you may fail when the item shows a blue triangle above a star. If the true rule is above, the new object and color should not matter. The work is to separate the rule from the original surface.

Consider an original practice-style set: a small cup is "mi cup," a small key is "mi key," and a large key is "ra key." Shown a large cup, the best label is "ra cup," because "mi" marks small and "ra" marks large, and the size marker transfers across object identity. A choice giving "mi cup" preserves the original object word but ignores the changed size property, so it is a trap.

Transfer test questions

Transfer applies to symbols too

Transfer is not limited to objects. A drill might use a dot above a line to mean "before" and a dot below a line to mean "after." A new symbol, say a star above a line, may still mean "before" if the relation is unchanged. The star shape is irrelevant. This is abstraction: the rule is position relative to the line, not the exact object on it.

Beware over-transfer. If a rule is supported only for color, do not extend it to size. If "mi" marks small in two examples, it does not automatically mark young, quiet, or weak. Artificial systems can define any mapping they like, so stay close to the evidence. Over-transfer is especially costly on harder language-track items where one wrong inference cascades through a string of related questions.

Reading the answer choices for transfer items

In multiple-choice format, transfer items routinely include choices that preserve the wrong feature:

• One choice keeps the original object but uses the wrong relation.
• Another keeps the original color but the wrong number.
• Another reverses a relation (above becomes below).

The correct answer preserves the feature actually tracked by the rule and updates the surface to match the new scene. Use the contested features in the choices to figure out which feature the item is testing, then apply the rule. The DLAB measures potential to learn a foreign language, not prior knowledge, so every transfer item is engineered to reward flexible application over rote recall. For study, end every drill with a new item: ask what the system would do with a fresh object, answer it, and state the rule in one sentence. If you cannot state it, you memorized rather than transferred.

Transfer across structure, not just content

The most demanding transfer items change the structure of the scene, not only its content. Suppose the examples teach that a marker before the object means "the object acts" while the same marker after the object means "the object is acted upon," a position-driven active-passive contrast. A new item might introduce a brand-new object you have never seen labeled. You cannot rely on memory at all; you must apply the positional rule to the new word. This is structural transfer, and it is the closest visual analogue to learning genuine grammar, where the same case ending attaches to thousands of unseen nouns.

A second structural case is nesting. If a dot inside a circle inside a box uses two stacked relation markers, a new scene with a triangle inside a square inside a box should reuse the same nesting pattern with new object words. Candidates who learned the relation marker in isolation but never saw it nested often misplace it. Always ask whether the rule composes, that is, whether two simple rules can stack to describe a more complex scene.

A transfer checklist you can run in seconds

Before committing to a transfer answer, run three quick checks: does my chosen label preserve every feature the examples proved meaningful; does it update every feature that changed in the new scene; and does its marker order match the examples. If all three pass, commit and move on. If any fails, the choice is a distractor built to catch partial transfer. Because the test gives you under a minute per item across 126 questions, this checklist must become reflexive, not deliberate, which only happens through repetition on original drills rather than by reading about it.