[From Bill Powers (991002.0618 MDT)]
Bruce Abbott (various) --
As I understand your proposal, sequence control amounts to waiting for a
cue, and when it occurs, executing an associated behavior. When the
behavior is finished, either it produces another cue or another one is
presented by an outside agency. So the model is one in which there is a
series of stimuli and responses, each response leading to, causing, or
being followed by the next stimulus.
If that is how cued chaining is proposed to work, then I think we can
devise a simple experiment that will test this model. Chris Cherpas has
already described one experiment that partially challenges this idea. A
pigeon pecks a color key that cycles through a set of colors. When the next
color in a specific required sequence shows up, the pigeon pecks the other
key to get food. Of course since the colors show up in a fixed sequence,
this can also be interpreted as a sequence of, for example, one peck, two
pecks, two pecks on the color key. Thus what could be learned is that
appearance of a given color should be followed by zero, one, or two pecks
on the color-change key, which would fit the chained-response theory.
However, this experiment could be modified in a simple way so the colors on
the color key cycle at random instead of in a fixed sequence. Then there
would be no particular number of pecks required to produce any given color
in the required sequence of colors. The pigeon would simply keep pecking
(an unpredictable number of times) until the next required color showed up,
then peck on the other key to get food. There would be no way to associate
one cue with the number of pecks that will produce the next cue in the
sequence.
Even this variant, however, still allows the cued-chaining interpretation
to fit, since there is a fixed sequence of acts between pecking the color
control key and pecking the food-producing key. To rule it out, we need
remove all regular sequences of actions from the experiment and see if
sequence control still occurs.
Suppose the computer screen shows a scattering of red, blue, and green
solid disks of color. The task is to learn the sequence of colors that is
reliably followed by the printed phrase "you win." With the mouse, the
participant clicks on colors in sequence until "you win" appears.
On each trial, the locations of the solid-colored disks are shuffled at
random. Thus following a click on any given color during a given trial,
there is no way to predict which way the mouse would have to move to get
the pointer to a disk of the required next color. There is no way in which
clicking on a given color can cue any particular movement-response. There
is no sequence of movements that will produce the right sequence of
perceptions.
If, under this last variant, the participant can still learn to produce a
required sequence of clicks on colors, we will have shown that the only
_repeatable_ aspect of this sequence is the sequence of perceptions, not
the sequence of actions. So cued chaining can't explain the result.
Naturally, if the locations of the colors are the same on every trial, the
cued-chaining explanation will appear to work, since the same movement
follows clicking on a given color (if there are just three color disks on
the screen). The controlled-perception model also will work; an experiment
done under undisturbed conditions simply can't distinguish between the two
proposals. But as soon as the disk locations are made unpredictably
variable, the cued-chaining explanation is ruled out -- provided the
subject can still do the task.
Best,
Bill P.