You can't teach every relation a learner will ever need. Generative teaching arranges a few trained relations so that many more show up untaught. This lesson covers how to plan for that and how to test whether it happened.
Why it matters
Teaching time is limited. If you teach a learner to match a spoken word to a picture and the picture to a printed word, several more relations may show up for free, such as matching the printed word back to the picture or picking the printed word when they hear it spoken. Programs built this way can reach more targets in less time and support more flexible language.
This task is about design and evaluation: choosing an efficient training arrangement, choosing the probe that answers your question, and spotting a probe procedure that undermines the conclusion.
Key ideas
- An emergent relation is one the learner shows without having been taught it directly. It is evidence that the trained relations formed a larger network.
- Stimulus equivalence is tested with three properties. After training A→B and B→C with matching-to-sample:
- Reflexivity: matching a stimulus to itself (A→A).
- Symmetry: reversing a trained relation (B→A after A→B was trained).
- Transitivity: linking two trained relations through the shared stimulus (A→C).
- Combined test (equivalence): C→A, which requires both symmetry and transitivity. Passing it is strong evidence the stimuli form an equivalence class.
- Choose a training structure. Each one shares a "node" stimulus:
- Linear series: A→B, then B→C.
- One-to-many (sample-as-node): A→B and A→C.
- Many-to-one (comparison-as-node): B→A and C→A.
- The combined test depends on the structure. In a linear series it is C→A; in one-to-many it is between the comparisons (B→C and C→B), which were never trained together.
- Stimulus functions can transfer through a class. If one member of an equivalence class gains a function, such as serving as an SD or a conditioned reinforcer, other members may take on that function without training. Relational frame theory describes similar effects (which it calls transformation of function) for relations beyond sameness, such as opposite or more-than.
- Matrix training builds recombinative generalization. Arrange components (for example, actions × objects) in a grid. Teach a subset, often the diagonal so each component appears at least once, then probe the untrained combinations.
- Multiple exemplar instruction supports generative repertoires. Rotating across examples and across listener and speaker responses can help untaught responses emerge, such as tacting an item learned only in listener training. Results vary across learners, so always test.
- Test before you teach the rest. Run probes for untrained relations with no differential consequences. If a relation doesn't emerge, first check that the trained relations stay accurate when intermixed; probes mean little until they do. If they hold, teach the missing relation directly.
Definitions
- Emergent relation: an untrained stimulus relation that appears after other relations were trained.
- Generative performance: novel, untaught responding that results from how earlier instruction was arranged.
- Stimulus equivalence: accurate responding to untrained, unreinforced stimulus relations (reflexivity, symmetry, and transitivity) after some related relations were trained, so the stimuli become interchangeable.
- Equivalence class: the group of stimuli that are related through equivalence.
- Transfer of function: a function trained with one class member appearing in other class members without direct training.
- Matrix training: teaching selected combinations from a grid of components and probing the rest.
- Recombinative generalization: correct responding to new combinations of previously learned components.
Examples from a session
Generative teaching in three settings
- Equivalence (school): Mateo, age 10, is taught to match the printed name of a coin, such as "dime" (A), to a picture of the coin (B), and the picture (B) to its printed value, "10¢" (C). In probes, he selects "dime" when shown "10¢" (C→A), a relation nobody taught him.
- Matrix training (clinic): Hana, age 4, works on action-object phrases with 3 actions (push, roll, drop) and 3 objects (ball, car, block). Her RBT teaches only push ball, roll car, and drop block. Hana then says "roll ball" and "drop car" in probes.
- Transfer of function (adult day program): Walt, age 58, has formed a class of a coffee-cup icon, the printed word "break," and a green card. Staff teach him that the posted icon means he can go to the break room. Later, when the printed word is posted instead, he heads to the break room on his own.
When you write the probe plan, first thin reinforcement on the trained baseline trials so the test phase does not bring a sudden loss of reinforcement. Then intersperse probe trials among already-mastered trials, keep reinforcement on the mastered trials only, and give no prompts or corrections on probes.
Non-example: An RBT praises Ana, age 9, for each correct C→A probe answer. Her 100% does not show emergence, because the reinforced relation is now a trained one.
Common mistakes / exam traps
- Mixing up symmetry and transitivity. Symmetry reverses one trained pair (B→A). Transitivity links two pairs through a shared stimulus (A→C).
- Calling C→A transitivity. C→A is the combined test, which needs both symmetry and transitivity.
- Calling emergence "stimulus generalization." Stimulus generalization rests on shared physical features. Members of an equivalence class are usually arbitrary and look nothing alike.
- Reinforcing or correcting probe trials. Once a relation is reinforced, it's trained, not emergent. Probe without feedback.
- Teaching a full matrix row. Teaching push ball, push car, and push block never exposes "roll" or "drop." Spread training across rows and columns.
- Assuming emergence. Some learners don't show it. Test, then teach what's missing.
Exam trap
Before you name the relation, write out what was trained and what was tested. Reversed pair = symmetry. Chained pair = transitivity. Reversed and chained = combined (equivalence) test.
Exam trap
Transfer of function vs. stimulus generalization: if the new stimulus shares physical features with the trained one, think generalization. If it looks nothing alike and is linked only by training history, think transfer through an equivalence class, and check that the class formed before expecting transfer.
Quick check
Kofi, age 7, was taught to select a picture when he hears its spoken name, and to select a printed word when shown the picture. To test reflexivity, what would his BCBA probe?