Multiple Exemplar Instruction (MEI): Training for Generalization

Generalization and Generativity Are Not the Same Outcome

A learner answers every trial right at the table. Then nothing carries over. The word never shows up in a new way, in a new place. Dr. Jessica Osos built this talk around that gap. The goal of language work is not a perfect score in session. The goal is language that is flexible and shows up without a prompt.

Two words get swapped in clinic talk, and they should not be. Generalization (the same skill used in new places) means variation in responding. It spreads across stimuli, responses, settings, and contexts. The skill itself stays the same. A child learns to tact (name) one dog. Soon they name many dogs. That is generalization at work.

Generativity (new language nobody taught) is a different thing. New verbal relations appear that were never trained directly. LaFrance and Tarbox draw this line in their conceptual analysis, and Osos leaned on it throughout the talk. Programs stall right at this line. Learners pick up operants one by one. The novel responding never comes.

generalization is necessary, but not sufficient, right?

From the talk — Dr. Jessica Osos

Why Verbal Operants Start Out Separate

Skinner defined verbal behavior by function, not by form. The same word can be many different behaviors. A tact (naming a thing you can see) is one. A mand (asking for what you want) is another. An echoic (repeating what you just heard) copies a model. A listener response (finding the item when someone names it) is different again. An intraverbal (words that answer other words) has no item in view at all.

Each one has its own antecedent and its own consequence. Early in development they are functionally independent (each skill learned on its own). That phrase sounds abstract, but the clinic version is blunt. You teach dog as a tact and you get a tact. You then teach it again as a listener response. Then again as a mand. Then again as an intraverbal.

Functional interdependence (one operant carries to the rest) is the flip side. Some autistic children build it on their own. Many do not, and for them it has to be taught. Osos framed the cost in minutes. Early intervention time is the scarcest thing you have. Teaching one word four separate ways is a poor use of it.

If you teach it as a tact, they only acquire it as a tact.

From the talk — Dr. Jessica Osos

MET and MEI Are Two Different Procedures

Multiple exemplar training (MET) is the one most clinicians know. The operant stays fixed and the stimuli change. You teach the tact dog with five or ten different photos. The mechanism is broader stimulus control (what the child learns to notice). It is a sound, well studied way to program for generalization.

Osos pushed hard on example selection. Vary the pictures as widely as you can while they all still show the target. One slide showed a dog swimming, well outside the usual context. Another showed a small dog with poofy hair, and another a dog of a different color. That spread teaches the child what makes a dog a dog: four legs, ears, fur, a long mouth and nose. Your generalization only reaches as far as your examples do.

Multiple exemplar instruction (MEI) flips the design. The stimulus set stays the same. The operants rotate, fast, one trial after the next. Same picture of a dog: name it, find it in an array, then say what it says. MET strengthens responding inside one operant. MEI builds a relationship between operants.

MET is a great strategy, but not one that's specifically programming for generativity.

From the talk — Dr. Jessica Osos

Bidirectional Naming Is the Cusp MEI Targets

The most studied outcome in MEI work is bidirectional naming (listener and speaker skills feed each other). Teach the word as a listener response and the tact shows up. Teach the tact and the listener response shows up. It is often treated as a developmental milestone. Osos called it the first step of functional interdependence.

That status is not loose language. A behavioral cusp (a change that unlocks new learning) changes what a learner can learn next. After the cusp, words can come from ordinary contact with the world, not only from planned trials. MEI has been shown to produce this naming reliably, which suggests it builds the interdependence the cusp requires.

The empirical base is wider than naming alone. Across studies, MEI has been tied to untaught tacts, intraverbals, and listener responses. Those effects have been replicated across learners, settings, languages, and targets. Teaching the same operants one at a time, in blocks, rarely does this. The rotation is doing the work.

Greer and colleagues in their article in 2005 really described bi-directional naming as a behavioral cusp because its emergence dramatically alters how learners can acquire language.

From the talk — Dr. Jessica Osos

Why MEI Works: Arrangement, Not Trial Count

A common misread is that MEI simply adds practice. Rotate fast, run more trials, get more learning. Osos rejected that. The change is structural. It is not about the gap between trials. It is about mixing operant types inside one seated block. That keeps the learner contacting new functional relations for the same target.

MEI is also not only for first words. LaFrance and Tarbox describe its use with complex repertoires. That list includes autoclitic frames (small word changes that shift meaning), spelling and dictation relations, and higher order intraverbals. So MEI may work as a general engine for relational learning across development, not a beginner tool.

Relational frame theory (RFT, a theory of how learned word links work) offers a second read on the same effect. From that view, MEI strengthens the contextual cues that support relational responding. Stimulus functions then transform with no direct teaching. Stimulus equivalence (treating related items as the same) is the simplest case. Teach A equals B, then B equals C, and A equals C arrives free. Verbal behavior names the contingencies. RFT accounts for the derived relations. MEI arranges the conditions for both.

It's a different instruction arrangement altogether.

From the talk — Dr. Jessica Osos

What a Ten Minute MEI Session Looks Like

Picture three items on the table: a dog, a cup, a ball. One picture of each. The pictures do not change for the whole block. Only the demand does. Point to the cup and ask what it is. Next trial, ask the learner to find the dog. Next, ask what you bounce, and have them touch the ball. Then back to the dog for a tact.

Twelve trials gets you four per target with the operants mixed. In this example the block covers tacting and two kinds of listener responding: by name and by function. Pick the operants to match the learner. For a young child working on naming, two operants are enough. The whole block runs five to ten minutes.

Osos was blunt about improvising. Even running the demo alone, with the plan in mind, the wrong trial almost slipped out. So do not hand a technician a concept. Write the twelve trials, randomize the order ahead of time, and build a data sheet that matches. One or two MEI sessions inside a three hour ABA (applied behavior analysis) session is plenty. Spreading the same targets across the whole session is not MEI at all.

the stimulus stays constant while the operant changes, right?

From the talk — Dr. Jessica Osos

Probe Sets and Teaching Sets: Knowing When to Stop

First, prove the learner needs MEI. Osos described starting with a probe set of three common items: a pencil, a teddy bear, and a fork. Teach all three as tacts with plain discrete trial teaching (one clear trial at a time), rotating the targets. Take them to mastery. Then lay them out and ask the learner to touch each one. If the listener response is not there, functional independence is confirmed.

Now pick three new targets that matter to the child, such as a plate, a toy car, and a crayon. These become the MEI teaching set. Run the rapid rotation on them, mixing both the targets and the operants, until they hit mastery. The probe items stay out of this set on purpose.

Then test with a fresh set: banana, train, doll. Teach those as tacts only, to mastery. Probe the listener responses. If they emerge, that is your evidence. Osos would celebrate it and test a couple more sets before stopping. If they do not, teach the listener responses directly and run another MEI set with three new targets.

Expect to burn through targets fast. Keep two running lists, one for probes and one for teaching, and keep every item socially meaningful. This is still ABA session time. Osos would usually stop once three or four probe sets transferred cleanly.

you don't have to use this strategy forever you just have to use it until that multiple exemplar or that functional interdependence emerges

From the talk — Dr. Jessica Osos

Where MEI Goes Wrong in Practice

The top error is running MET and calling it MEI. The stimuli rotate and the operants never do. Nothing generative follows, because nothing was arranged for it. The second error is stopping too early. MEI gets tried for a week, looks messy, and gets dropped. Emergence needs repeated contact with the same targets across several functions.

Mands need extra care. A mand runs on a motivating operation (what makes an item worth having right now). Rapid rotation makes a real establishing operation hard to hold. Without one, you risk teaching a child to request an item they do not want. That is faulty stimulus control, not functional language. Probing the mand later is often the safer move.

Prerequisites matter too. Learners need some listener and echoic skill so they can contact reinforcement during the block. Do not wait for perfect mastery before you start, but do check that the basics are there. And keep the trial format intact. Every MEI trial still follows a normal teaching trial. Gain attention and present the SD (the cue to respond). Prompt if needed, then wait for the response. Finish with reinforcement or error correction.

One last framing worth keeping. MEI changes how instruction is organized. It does not replace careful target selection, good prompting, or direct teaching.

MEI should be viewed as a powerful supplement, not a shortcut.

From the talk — Dr. Jessica Osos