Skinner’s insight, and it is genuinely the useful one for teaching, is that language is not a vocabulary list. The same word can be several different behaviors depending on what occasions it and what it produces — and the ones that produce different things have to be taught separately.
The tact is naming what is present. A dog walks past, the child says “dog”, and what follows is social: you look, you agree, you say something back.
Why the same word is two skills
A child who says “cookie” because they want one is doing something controlled by their own state — they are hungry, the cookie is the point, and the reinforcer that maintains it is getting the cookie.
A child who says “cookie” while looking at a picture of one in a book is doing something controlled by the picture. Nothing is being obtained. What maintains it is that someone responded.
Two different sources of control, two different consequences, two different behaviors. That is why a child can request a dozen items fluently and still not name any of them, and why the reverse also happens. Neither is a failure of generalization — they were never the same skill.
The mistake that erases the difference
Teach tacts and then hand over the item each time the child names it, and the arrangement you have built is a request. The child learns that saying the word produces the thing, which is exactly what a mand is.
Reinforce a tact the way a person would: look, respond, say something back, be interested. If the child clearly wants the item, that is fine — but then you are working on the request, and it is worth knowing which one you are doing.
Where to teach them
The obvious place is a table with pictures, and that works. It is also the place where a tact most easily becomes a rote answer to a rote prompt — “what is it?” repeated forty times produces a child who names things when asked that exact question, in that chair.
The better half of the work happens during the day, where the things already are. Name what you both see and leave room. Ask less than you think you should: “what is it?” is a prompt, and a tact that only occurs after that prompt has not yet come under the control of the thing itself.
What the record should show
If a program is teaching both requests and labels, the record has to keep them apart — the same word counted in one column tells you nothing about which one is growing.
Interlaza records free-operant responses with the operant kept separate, along with whether the response was independent, prompted, or a partial approximation. That last one matters: a partial word is submitted as not correct, so it can never satisfy a mastery criterion, but it is reported separately rather than being filed as an error. A child moving from silence to an approximation is doing something, and a record that shows only right and wrong cannot see it.
Further reading
- Stimulus equivalence — how naming and matching start to feed each other.
- ABA therapy activities — where naming fits among the other things worth practicing.