There is a strawberry on the table. Below it, two options: another strawberry and a banana. The child touches the strawberry. Correct. Twenty trials later accuracy sits around 95% and the stage is marked as mastered.
What has been learned?
It is a fair question and it has an answer. The child may have learned to relate two things by a shared property — or may simply have learned that touching the strawberry makes a pleasant sound happen. Both produce the same 95%.
Telling them apart is not an academic nicety. It is the difference between a skill that holds up in a new situation and one that stays locked inside the screen it was trained on. INTERLAZA is built around that distinction.
Two contacts that look identical
Working in the interbehavioural tradition, the psychologist Emilio Ribes distinguishes forms of contact with stimuli that are indistinguishable at the table. The difference is not in what you see; it is in what the child has to do in order to be right.
Coupling is recognising and repeating. If the correct answer is always the same thing — always the strawberry, always the red circle — nothing needs to be compared. Identifying one object and repeating the gesture is enough.
“Recognising the same stimulus object does not constitute comparative, relational behaviour. It is simply a repetition pattern.”
Comparison is responding to a relation. Things change on every trial and the only constant is the criterion — “the same colour”. No object ever wins twice, so memorising one is useless: the sample has to be consulted each time.
Which yields a question any instructor can ask of any task:
Could the child be right without comparing? If the answer is yes, then however well they perform, they are not relating.
Something here usually surprises people: identity matching is not an easier version of comparison. It is a different thing altogether. Neither are fixed arbitrary pairings — if “when red appears, touch yellow” never varies, these remain constant occurrence contingencies “with no permutations”. Two memorised pairs, not a relation grasped.
None of which makes those tasks worthless — quite the opposite. Ribes is explicit that coupling contacts “do not disappear, but are necessary as the absolute constants from which relational contingencies emerge”. They are the floor. A child who cannot yet tell two cards apart cannot compare anything. What matters is knowing which ground you are standing on at any given moment.
Four design decisions
INTERLAZA does not leave this to chance. Four concrete decisions, all of them visible in how trials are generated.
1. Roles rotate
Every object appears as sample, as correct option and as distractor across a session. The generator does not choose at random: it favours whichever has appeared least.
This sounds like a detail. It is not. Shuffling positions is not enough. Varying whether the answer sits left or right prevents a side bias, but it does nothing about an object bias. And pure randomness only distributes well in the long run — a stage is a few dozen trials, and across that stretch it can quite easily leave one object sitting in the same post from beginning to end.
2. Values vary
When a stage is prepared, concepts are spread across several values of the dimension being worked on. If eight of ten tagged foods were red, “red” would stop being one value among others and become a fixed datum — and the child could climb to 100% without ever attending to the relation.
3. The relation holds while everything else changes
This is the heart of it. Objects and their properties change from trial to trial; the only constant is the criterion. Success becomes reachable only by comparing.
4. Every stage closes with a no-feedback probe
And here is the decision that sets us apart.
If accuracy during training proves nothing — and it does not — something else is needed:
“Accurate performance is not sufficient to identify the type of functional contact developed.”
Ribes accepts one indicator only: a transfer probe, under two conditions that are uncomfortable and informative for exactly that reason.
The varying attribute changes — what Varela and Ribes call the modality. If the child trained on “the same colour”, the probe asks for “the same size”, with objects they have not seen. If the relation was grasped — “the same in some respect” — it carries. If colours were memorised, it does not.
And they are not told whether they were right. No confirmation, no sound, no green. The moment feedback appears we stop measuring and go back to training.
In the early stages — identity, word→picture, sound→picture — the probe works differently, because no dimension is trained there: what changes is the exemplar. The answer image is presented in its other version (realistic photograph to pictogram). A correspondence bound to the concept survives the picture being redrawn; one bound to that particular image does not.
No probe scores, and none gates progression. It is not an exam. It is an instrument.
What to expect from a probe
Better to say it before it happens: performance drops. A child at 90-100% during training may land around 60% on the probe.
That fall is not a failure of the child or of the programme. It is the measurement. It reports how much of the earlier accuracy rested on repeating. Without the probe, 95% would read as understanding and nobody would know otherwise.
And the idea is not ours. The method comes from Julio Varela and Carmen Quintana, whose Competence Transfer Matrix (1995) does exactly this: it decomposes any situation into four factors — dimension, relation, modality and instance — and asks what happens when some are held and one is varied. Fifteen distinct types of transfer fall out of that.
Both of our probes are cases of theirs. Changing only the exemplar, holding relation and modality, is their case 1: extra-instantial transfer — what we use going from photograph to pictogram. Changing the attribute that varies is their case 2: extra-modal transfer, colour → size.
It is also worth not confusing this with generalisation. Varela notes that behavioural writing routinely treats the two as synonyms, and that this blurs them: in this tradition transfer is the simultaneous generalisation of stimulus and response (Ribes, 1990). Classical generalisation is a one-term affair — a response spreads to new stimuli because they resemble the trained ones. Transfer requires both to detach. Calling it “generalisation” names one term where there are two.
Sidman’s equivalence tests belong to the same family of reasoning: what matters is not what the learner does with what was trained, but what they do with what was never trained.
What changes when you read a session
There is a practical consequence that goes beyond software and applies as much at the table as on a screen.
When a child gets something wrong, the usual reading is “they don’t have it”. But very often what they have done is respond to a criterion other than the one we had in mind: position, the most salient object, a partial feature. From their point of view nothing failed: they were consistent with something else.
Recording “incorrect” throws that away. Recording “responded by position”, or “matched by shape when we asked for colour”, keeps it — and points at what to do next.
Because the criterion is not in the cards. Nothing about two cards makes matching by shape wrong and matching by colour right. That criterion is held by the instructor, the family, the community that agreed what each word names. “Correct” and “error” are not properties of the child’s response: they are relations between what the child does and a criterion somebody is holding.
Designing a teaching platform is, in large part, deciding how carefully that criterion gets held.
INTERLAZA draws on the interbehavioural tradition (Kantor; Ribes and López), Varela and Quintana’s Competence Transfer Matrix (1995), Sidman’s stimulus equivalence, and Relational Frame Theory. Full references are on our science page.