The big idea
Teaching every skill directly would take forever. Procedures for emergent relations teach a small set of relations and produce others that were never taught. Teach a learner that the spoken word “seven” goes with the numeral 7, and that the numeral 7 goes with a picture of seven dots, and the learner may match the dots to the spoken word with no further teaching. That untaught performance is the payoff.
B.21 covers the processes, including stimulus equivalence and its properties. This guide covers the procedures: how to arrange teaching so new relations are likely to emerge, and how to test that they did.
Breaking it down
Equivalence-based instruction
Equivalence-based instruction (EBI) uses match-to-sample training to teach a few conditional discriminations (G.6) among stimuli, then tests for the relations that were not trained. Stimuli show stimulus equivalence when tests show reflexivity (A matches A), symmetry (after A-B training, B-A) and transitivity (after A-B and B-C training, A-C), including combined tests such as C-A.
A typical EBI program:
- Choose stimulus sets, such as A (spoken state names), B (state outlines) and C (printed capital cities).
- Train the baseline relations with feedback until mastery.
- Thin reinforcement on baseline trials so the learner is used to trials without feedback.
- Test the untrained relations without feedback, mixed with baseline trials.
- If tests fail, check that the baseline still holds, retrain it, or change the training structure.
Training structures
The training structure is the arrangement of the trained relations. In each pair below, the first stimulus is the sample and the second is the comparison.
| Structure | Trained relations | What stays the same |
|---|---|---|
| Linear series | A-B, B-C, C-D | Nothing: each set links to the next |
| One-to-many (sample-as-node) | A-B, A-C, A-D | One set is the sample in every relation |
| Many-to-one (comparison-as-node) | B-A, C-A, D-A | One set is the comparison in every relation |
Many studies find that one-to-many and many-to-one produce equivalence more readily than linear series, though results vary with learners and procedures. With four stimulus sets, training 3 relations can produce 9 untrained ones, which is where the efficiency comes from.
Testing for emergence
Testing is what makes a relation emergent. Untrained relations must be presented without reinforcement or feedback, or the test becomes training. After one-to-many training (A-B, A-C), typical tests are symmetry (B-A, C-A) and equivalence (B-C, C-B). Where relevant, also test whether the relations carry over to new response types, such as saying the capital when shown the state outline.
Matrix training and recombinative generalization
Matrix training arranges two or more sets of components on a grid, such as animals on one axis and actions on the other. Teaching only some combinations, often along the diagonal, can produce recombinative generalization: correct responding to untaught combinations of components learned in other combinations.
| Animal | jumping | sleeping | eating |
|---|---|---|---|
| horse | taught | probe | probe |
| pig | probe | taught | probe |
| duck | probe | probe | taught |
Three combinations are taught and six are tested. Diagonal teaching alone may not produce recombination, especially when the learner does not yet know the separate components. If so, teach overlapping combinations that share components. Matrix training is used for language (noun-verb, adjective-object), play actions, following directions, and early reading and spelling.
Multiple exemplar instruction and naming
Multiple exemplar instruction (MEI) teaches with many examples, or rotates one stimulus across several response types: pointing to it (listener), naming it (tact) and matching it. Rotating listener and speaker responses this way, across many stimulus sets, can establish a generalized repertoire in which learning one response type produces the other without direct teaching.
Bidirectional naming, often just called naming, is that repertoire. After hearing an object’s name, or after being taught one response to it, the learner can both select the object when the name is said and say the name when the object is shown. Teams check for it with probes: new stimuli are named for the learner without reinforcement, then both listener and tact responses are tested.
Seeing it in practice
- University. Dr. Mensah’s course software teaches students to match brain-region names to diagrams and diagrams to functions, then tests without feedback whether they can match names to functions.
- Clinic. Fatima, 5, learns “horse jumping,” “pig sleeping” and “duck eating” with toy figures. On probes she correctly acts out “pig jumping” and “duck sleeping,” which were never taught.
- School. A teacher uses a one-to-many structure: spoken coin names (A) to coins (B) and spoken coin names to printed values (C). The student then matches coins to values without being taught.
- Home. A parent rotates “Show me the giraffe,” “What’s this?” and “Find the same” with each new animal. After several sets, Leo names animals he has only been taught to point to.
Common mix-ups
- Training vs. testing. A relation is emergent only if it was never reinforced. Giving feedback on test trials turns the test into training.
- Sample vs. comparison. In A-B training, A is the sample presented first, and the B stimuli are the comparisons the learner chooses from.
- One-to-many vs. many-to-one. One-to-many keeps the sample set constant (sample-as-node). Many-to-one keeps the comparison set constant (comparison-as-node).
- Recombinative generalization vs. equivalence. Recombination produces new combinations of known parts. Equivalence produces new relations among stimuli in a class.
- MEI vs. sufficient exemplars. Both use many examples (G.15). MEI for naming also rotates the learner across response types.
Exam tip: On a training-structure question, find what stays the same. One sample set in every trained relation means one-to-many. One comparison set in every relation means many-to-one. Each set linked only to the next means linear series.
Quick recap
- EBI teaches a few conditional discriminations and tests for symmetry, transitivity and equivalence without feedback.
- Linear series, one-to-many and many-to-one are the three training structures, and linear series tends to be the least effective.
- Matrix training teaches a subset of combinations and tests the rest for recombinative generalization.
- MEI rotates listener and speaker responses to build bidirectional naming.
- Emergent relations make teaching efficient and generative.
Further reading
- Cooper, J. O., Heron, T. E., & Heward, W. L. (2020). Applied behavior analysis (3rd ed.), chapters 19 and 20.
- Sidman, M. (1994). Equivalence relations and behavior: A research story. Authors Cooperative.
Key terms
- Stimulus equivalence
- The emergence of untaught relations among stimuli after some relations have been taught, shown by reflexivity, symmetry and transitivity. Stimuli that pass these tests form an equivalence class and are interchangeable even when they share no physical features.
- Emergent relation (Derived relation)
- A relation between stimuli, or between stimuli and responses, that appears without direct teaching or reinforcement as a result of other relations that were taught.
- Equivalence-based instruction (EBI)
- Teaching that trains a small number of conditional discriminations, usually with matching-to-sample, and then tests without feedback for the untrained relations among the stimuli, so that more relations are learned than were taught.
- Training structure
- The arrangement of the relations taught in equivalence-based instruction. In a linear series each stimulus set is linked to the next (A-B, B-C); in one-to-many, or sample-as-node, one set is the sample for every other set (A-B, A-C); in many-to-one, or comparison-as-node, every other set is matched to one comparison set (B-A, C-A).
- Matrix training
- Arranging two or more sets of components, such as objects and actions, on a grid, teaching only some of the combinations, and then testing the untaught combinations for recombinative generalization.
- Recombinative generalization
- The production of new, correct combinations of components from responses that were taught separately. Matrix training arranges teaching to promote it.
- Multiple exemplar training (Multiple exemplar instruction, MEI, Teaching sufficient exemplars)
- Teaching with several varied examples of the stimuli, settings or response forms involved in a skill, or rotating one stimulus across different response types, so that responding extends to examples that were never taught.
- Naming (Bidirectional naming, BiN)
- A two-way relation between speaker and listener behavior. After learning to tact an object, a person can also select it as a listener when hearing its name, and the reverse, without separate teaching.
Check yourself
4 questions on G.19. Choose an answer to see whether you're right and why.
Show the answer
Answer: A. The same stimulus set, country names, is the sample in both trained relations, with flags and capitals as comparisons, which is one-to-many. Many-to-one would use flags and capitals as samples and names as the shared comparison. A linear series would link A to B and then B to C.
Show the answer
Answer: C. Matching flags to capitals (B-C) was never trained and requires combining symmetry and transitivity, so it tests equivalence, and running it without feedback keeps it a test. Flags to names (B-A) tests only symmetry, and identity matching tests reflexivity. Name-to-flag trials with praise repeat a trained relation.
Show the answer
Answer: D. Correct, untaught combinations of components (colors and objects) that were learned in other combinations show recombinative generalization, the goal of matrix training. Equivalence class formation concerns untrained relations among stimuli in a class, such as matching a picture to a printed word, not new combinations of parts.
Show the answer
Answer: A. Rotating listener and speaker responses across many stimulus sets is multiple exemplar instruction, and the result described, a tact emerging from listener training alone, is naming. One-to-many training arranges match-to-sample relations among stimulus sets; it does not rotate the learner across response types.
Adds a step to your study record. Saved in this browser only.