Traditionally, and almost certainly as a result of the fact that early research in this area was done with animals, performance in matching tasks has been ascribed to the development of unmediated stimulus selection. But over the years, in studying these performances with humans, I have encountered much evidence suggesting that many complex matching performances are indeed mediated, and by a variety of verbal operants.
Clearly, in both of these examples, the correct comparison stimulus is unique because it is the only comparison that allows the subject to make one hand sign that is both a self-echoic with respect to the sample tact, and also itself a tact for that comparison. An accurate hand sign of any other comparison would require the subject to change from the rehearsed hand sign. As you can see in the two examples, only the comparison identical to the sample allows one hand sign to be jointly under both tact and self-echoic control. I call this unique form of stimulus control joint control. It is unique because it can only occur with one comparison on each trial. But as you can see by looking at the two examples, the event of joint control is common to both cases where a comparison is encountered that evokes the rehearsed hand sign. And it is this joint control event that evokes the selection response. Probably the most important question these considerations raise is with regard to their relevance to the description of ordinary behavior, when it is not accompanied by overt mediating responses.
Thus the correct comparison is unique -- it alone evokes joint tact
self-echoic control. Now, having laid this out, the question naturally
arises as to the evidence for this sort of control. And so I would like
to report today on some experiments we have concluded examining the performance
illustrated in the last slide: the selection of complex, multi-dimensional
stimuli in response to spoken phrases, or perhaps less formally, the selection
of complex objects in response to their descriptions.
The goal of these experiments was to look for evidence for the operation of each of the two legs of joint control. And so the first experiment looked at the role of tact behavior, and the second looked at the role of the self-echoic.
The first stimulus marked with an asterisk, would be described as King-Bus-Point
because it is purple, has the internal shape named bus, and has the border
named Check. Six stimuli were constructed for a training set. The top of
the figure illustrates some of them. In Phase 1 of training, subjects were
trained in the successive comparison procedure I described. They were trained
to select each of the six stimuli when given a spoken description as a
sample. Learning here was very slow. It took over 400 trials for the children
to learn to select the six stimuli in response to their descriptions.
When they finally did learn this, the children were immediately given the naming test listed here as Step 2. Here, they were shown one stimulus at a time, and simply asked give its description. No child could fully describe more than two of the stimuli, and most could only give a complete description, for one stimulus. This, even though they had just spent 400+ trials learning to select these stimuli in response to the descriptions, they could not give the descriptions from looking at the stimuli. No symmetry was apparent here.
In Step 3, subjects were given a test for generalization. New stimuli were constructed from novel combinations of the same features, as shown in the bottom of the figure, and subjects were required to select each novel stimulus when given its spoken description. But we found that subjects could not do this. There was no generalization from initial training to these novel combinations. Apparently teaching comparison selection with the training set was, in and of itself not an antecedent adequate to produce generalized selection. So the question we addressed next was what was missing from the subjects repertoire that would produce generalized selection to the novel combinations. In light of our earlier findings illustrating the efficacy of joint control in producing generalized matching, this was the route we chose to explore. We proceeded to bring selection behavior under joint control.
We suspected that this was somehow due to the fact that the stimuli and their names were familiar, and that subjects could repeat the names as memorized intraverbals rather than by actually generating the names by tacting each element of the training-set stimuli. And so in the next phase, the tact-training procedure was repeated, but with some novel, unfamiliar combinations of color, border and shape. Here, the subjects could not just repeat the names they had heard over 400 times in training, but rather were forced to literally construct the names for the stimuli by tacting each element.
Apparently this was the key. Because when given the generalization test
in Step 9, the subjects immediately performed virtually without error.
It seems that the generalized selection of novel combinations of color,
shape and border could only occur after subjects learned to actually emit
appropriate tacts for each of the elements of the novel stimuli. It thus
appears that generalization depended on subjects learning to accurately
tact novel comparison stimuli.
From a larger perspective, we see here that at least one verbal relation, the tact leg of joint control, was necessary for generalized matching.
As to the second verbal relation that can comprise joint control, the self-echoic, that was explored in a second experiment.
Thus, subjects here could apply tacts they had learned pre experimentally for these stimuli. And given this pre experimental history, the generalization seen in Step 3 was dramatic, but of course, not unexpected. Thus, after learning to select combinations with names like Blue-Boy-Box and Purple Clown-Circle, there was immediate generalized selection to stimuli comprised of novel combinations of these same names.
The next phase got to the heart of the matter: the role of self-echoic rehearsal was explored. This phase took advantage of the fact that with successive presentations, the comparison stimuli only appeared one at a time. Thus, on some trials the correct comparison appeared immediately after the sample description was pronounced. But on other trials 2, 3, 4, or even 5 comparisons might intervene before the subject saw the correct comparison.
To prevent subjects from rehearsing the sample description while waiting
the correct comparison to appear, the subjects were now required to read
numbers from a constantly changing display that appeared on the screen
just above the comparisons. Thus, as before, subjects would hear the sample
spoken and then see the comparisons one at a time, and as before, they
could select a comparison by pressing on it, or see the next comparison
by pressing the black square. but now, while seeking the correct comparison,
they had to read the numbers appearing on the screen. Presumably, this
impeded the rehearsal of the sample description.
As soon as this data was collected, these same subjects were then given matching to sample training, but with the 6 stimuli with unfamiliar names like King-Bus Point. Once they learned to select the 6 training set stimuli in response to their descriptions, we examined the effect of preventing rehearsal on matching accuracy.
As we see, there isn't much difference with comparisons in the first three positions, but after that, with the familiar names like blue-boy-circle, the accuracy of a correct selection declined as a function of the number of comparisons intervening between the sample and the correct comparison -- This trend was found to be statistically significant. That is, subjects made fewer correct selections when 5 or 6 comparisons intervened between the sample and the correct comparison, than when only 1 or 2 intervened. This strongly implies that some self-echoic rehearsal process mediated selection here.
Interestingly, with the unfamiliar names like King-Bus-Check, the gradient
is flat! With these stimuli, subjects maintained their selection accuracy
regardless of the number of comparisons intervening between the spoken
sample and the correct comparison. Thus, they were just as accurate selecting
comparisons that appeared in the 5th or 6th position as in the first or
second. Apparently, whatever the subjects were doing with these stimuli,
it did not involve vocal rehearsal of the stimulus names. Here then we
have evidence for a direct, unmediated form of stimulus control.
|TYPE OF SAMPLE|
|Acquisition of initial
|Very rapid||Very slow|
|Naming Test performance||Errorless - Subjects named all training set stimuli correctly||Poor - Subjects named no more than 2 out of 6 correctly|
|Generalization to novel stimuli||Immediately after initial training||None - until trained to tact stimulus elements.|
|Effect of rehearsal prevention on selection accuracy||Accuracy deteriorated with increasing distance between samples and comparisons||Accuracy was maintained|
|Initial form of stimulus control||Joint Control||Conditional discrimination|
On the other hand, with the familiar names, shown in the middle column, initial acquisition was very rapid, and in the naming test subjects could name all of the stimuli, thereby showing that tacts were available. Furthermore, generalization here was immediate, and with rehearsal prevention, there were declining gradients of accuracy -- thereby illustrating a dependency of selection on self-echoic rehearsal. And so, while it is indeed possible for subjects to learn a matching to sample task as a simple, unmediated conditional discrimination, it also appears that subjects may learn additional verbal relations whose interaction mediates generalized relational responding.