An earlier version of this paper was presented at the ABA meetings in Boston, May 2004. This is still a DRAFT .....
THE BEHAVIOR OF THE LISTENER:
Dichotomies real and false.
Paul K. Brandon
Psychology Department
Minnesota State Univeristy
Mankato, MN 56001
This paper was occasioned by Charlie Catania's tutorial on verbal behavior at last year's ABA. His comments about higher order response classes generated some questions, and suggested an application of some data that we collected back in 1987.The central topic of this presentation is the behavior of the listener.
Verbal behavior is behaviorally defined as operants whose reinforcement is mediated by a change in behavior of a listener. Often the additional specification is added that the listener and speaker can exchange roles, or as Ferster put it, share a common intraverbal repertoire. This makes the behavior of the listener an integral part of the analysis of verbal behavior.
It must be emphasized that terms such as verbal and rule governed are descriptions of functions; not category labels. Thus they are not necessarily exclusive. Beware of pigeon holes!
The behavior of the listener is characterized as rule governed, or as Shimoff and Catania (1998, chapter in Handbook of Research Methods in Human Operant Behavior, Lattal and Perone, eds.) prefer: verbally governed behavior.
A case can be made that the term 'verbally governed' be used for the broader class of behaviors under the control of verbal discriminative stimuli, while the term 'rule governed' is restricted to that subclass mainted by generic reinforcement not specific to the behavior (see the discussion below of higher order response classes).A good discussion of the conditions necessary for acquiring these behavior is found in Greer and Keohane (2006) <http://www.slp-aba.com/SLP-ABA-1-2.pdf>
They address the question of the acquisition through multiple exemplar training of higher order emergent functions such as naming and mand/tact and tact/mand functional transfer.I'll address some issues involving the analysis of this class of behavior.
The labels "Rule governed behavior" and "Rule Following" are used with the (tacit) assumption that they apply to a unitary function.
I would posit that in fact there are two functional relationships involved, which can be distinguished on the basis of a molar vs molecular level of analysis.
The first use of the term is applied on the molecular level, and is the most common text book definition: behavior acquired through the control of a verbal discriminative stimulus that specifies a contingency (Skinner, Operant Analysis of Problem Solving, in Contingencies of Reinforcement, 1969). This is the original definition given by B. F. Skinner, a most molecular analyst. Many of the common-behavior examples that we give of rule following are of this type.
Parenthetically, this statement is limited as a definition.
First, this definition is structural rather than functional in nature. It says that a rule is a verbal operant that takes the form if a certain behavior occurs, then a certain consequence will follow. This is a good guide to the kind of statements that we label rules, but does not clearly state the functional relationships maintaining them.Second, it is not clear that rules can be classified as discriminative stimuli, since they are not usually present when the listener's behavior occurs and is reinforced (although some kinds of instruction might meet this requirement).
One might draw a distinction between direct stimulus control (meeting the classical definition of a discriminative stimulus) and indirect stimulus control.
This latter category has been variously labeled function altering stimulu (after Schlinger and Blakely, 1987) where rules change the way in which other future stimuli control behavior,
a conditional discrimination, where the control exerted by one stimulus is conditional on the value of some other stimulus dimension, and
joint stimulus control (Lowenkron, 1998) where two different stimuli, one of them more remote in time, jointly control a particular response topography.To return to our discussion:
Skinner draws a distinction between two methods of acquiring a behavior (instruction and shaping); not methods of maintaining it.
The second usage derives from a more molar analysis, and refers to behavior that is maintained by rules rather than by contingencies. Joyce and Chase (JEAB 1990) allude to this distinction.
On the molecular level, the main function of rule following is to switch the control of behavior from one contingency to another. This might be related to a changeover response in a concurrent schedule, or, when there is a significant delay in the emission of the response, a function altering stimulus.
Thus, this is a transitory effect. Behavior is not maintained by the rule; rather control is shifted from one contingency to another. There is still a clear and specific relationship between the behavior and the reinforcer maintaining it.
It is true that a history of rule following (being under the stimulus control of verbal discriminative stimuli) is necessary, but this is true for stimulus control in general.
molecular molar transitional stable specific contingency generic contingency One important aspect of behavioral transitions is that the history of consequences is much more important that the specific direct consequence. Thus the nature of the specific direct reinforcement contingency is less important here than in steady state behavior.
The point is that the effect of the occurrence of a discriminative stimulus is not due to the reinforcement that will occur, but to the past history of reinforcement of responding in the presence of that or similar stimuli.
In the case of transitional rule following the changed behavior will be maintained by its natural consequences. If this happens frequently enought (even though intermittently) then rule following will be repeated.
Of course, as Andronis (Response to Chase; Dialogs on VB) has pointed out, in this situation truly rule governed behavior can occur only once: once the behavior cas contacted the contingency its control has shifted to that contingency).
On the molar level, on the other hand, we are looking at the rule as the primary source of control over the behavior in a stable state. There is no specific relationship between the behavior controlled by the rule and a particular source of reinforcement.
Rather, rule following is being maintained by the generic (primarily social) reinforcement contingencies specific to rule following as a class of behaviors in its own right. This is the kind of functional relationship that Hayes refers to as pliance.
There is no clear relationship between behavior and the reinforcement that maintains it.
Most experimental studies of rule governed behavior appear to belong in this category. Given the relatively weak reinforcers contingent upon performance in these experimental models it is not surprising the the generically reinforced rule following predominates. This is an endemic problem in human subject research: the most powerful reinforcer is usually the one causing the subject to participate in the experiment.
This, of course, is not a dichotomy.
Like the molar/molecular distinction itself it designates ends of a continuum. There will necessarily be intermediate situations whose functional category is ambiguous.
This is similar too, though definitely not the same as, Zettle and Hayes' (1982) distinction between pliance (rule following controlled by social contingencies) and tracking (rule following controlled by a relationship between the verbal instructions and environmental events). The difference is that while Zettle and Hayes are making a distinction based on types of contingencies, there is little indication of different levels of analysis.
In addition, the pliance/tracking distinction has the problem that while pliance is defined in terms of consequences, tracking is defined primarily in terms of antecedents. This asymetry is also true of Skinner's mand/tact distinction; not surprising since Zettle and Hayes set up pliance and tracking as listener's analog of the mand and tact for the speaker.
This asymetry may be more of a problem in the behavior of the listener, since it introduces a more basic difference in the controlling reinforcement contingencies. Basic mands and tacts both produce (and are maintained by) immediate changes in the behavior in the listener. Pliance, however, is maintained by much more diffuse contingencies.
Of course, these social contingencies may still be natural (Hayes, 1989). This does not imply that they are thus delayed or generic.
This takes us to the concept of the higher order response class, which has its own ambguities.
Following the logic of sets, a higher order response class might be defined as one made of elements which are themselves responses. It's not this simple however (Higher-order class of behavior: a class that includes within it other classes that can themselves function as operant classes (as when generalized imitation includes all component imitations that could be separately reinforced as subclasses). A higher-order class is sometimes called a generalized class, in the sense that contingencies arranged for some subclasses within it generalized to all the others. Generalized matching and verbally governed behavior are examples of higher-order classes. Catania, L&P 1998).
The applied child compliance procedures which use compliance to a high probability command to increase the likelihood of responding to a lower probability command are a good example. The High-P/Low-P command procedure is usually explained as an instance of Behavioral Momemtum. This analysis has problems because it is questionable whether discrete trial procedures fit the model of Behavioral Momentum Theory (Nevin, 1996; Houlihan and Brandon, 1996), which emphasises rate of a given reinforcer.
However, this procedure can be analyzed in terms of the verbal governance of higher order response classes (Shimoff and Catania, Catania, 1998).
If complying with a request is looked at as a higher order response class containing many subclasses of compying to specific requests, then increasing the likelihood of any of these discriminated operants should increase the likelihood of all of them.
If we define responses classes functionally, then if an element is a member of a higher order class and under the control of the contingencies that define that class, it cannot be an independent response as well unless we can show that it is simultaneously under the control of contingencies operating at both levels: that of the elementary response and that of the higher order response class. Rule governance vs. contingency shaping implies that we are talking about differences in the acquisition of behavior. In this context we may in fact be able to support a dichotomy in how behavior is acquired (prompting vs shaping).
It is when we shift to talking about the maintenance of behavior that the distinction begins to get fuzzy. Joyce and Chase (JEAB 1990) have also raised this issue.
Clearly, when a response is acquired, we can see that it's control might shift from one function to another. We might prompt a behavior (stimulus control due to history of reinforcement for following prompts) which then comes into contact with more direct contingencies and is maintained by them. I believe that this is the essence of Skinner's original conceptualization.
At any given time, there is only one source of control over the response.
If a response is defined topographically there is no problem in assigning it simultaneously to two classes. On the other hand, to say that a behavior in a stable state is a member of both a lower and higher order operant class, one must demonstrate simultaneous control by both sets of contingencies. Otherwise, we are simply aggregating responses acquired separately into a new, differently defined response class.
In the case of rule following, if we are to define rule following as a higher order of response class whose elements are themselves functionally defined response classes, we must demonstrate that the behavior controlled by the rules (verbal discriminative stimuli) is also controlled by a more specific contingency.
Once we show this, we have shifted the argument from a false dichotomy of rule governed VERSUS contingency shaped behavior to one of describing the different contingencies involved in the control of a given behavior (Catania also makes this point). Now we are describing the relative contributions of two different contingencies to the control of a multiply determined behavior. Rules ARE contingencies, not an alternative to them.
Saying the rule governed behavior is (relatively) insensitive to immediate contingencies is therefore a tautology; this is part of the definition of rule governed behavior and thus not an explanation of it.
THE DATA
The data I'm presenting is based on a Masters dissertatation completed by Mark Smasal in 1987!
This is a systematic replication of Hayes, Brownstein, et al (1986).
One should note that while they state that they use a multiple schedule of reinforcement, they are in fact using a mixed schedule (no correlated discriminative stimulus) with an instructional stimulus (lights labeled "go fast" or "go slow" imposed on the schedule.
The basic experimental setup involves a human subject pushing buttons.
The subject is seated before a console containing a 5 X 5 matrix of lights.
Button pushes move the lighted location from the upper left to the lower right of the matrix according to some reinforcement schedule. A point is earned when the path through the matrix is completed. There is also a discriminative stimulus light on the console.
In the original study the light matrix was discrete bulbs; in our replication we used a matrix on a computer (Apple II) screen.
<show example from Mac version>
To examine the relative effects of instructions and immediate reinforcement contingencies, a multiple schedule is used which requires a high rate of response (FR contingency) in one component and a low rate of response (DRL) in the other.
Verbal instructions (minimal, "Go fast", "Go slow", "Sometimes go fast, sometimes go slow, depending upon the stimulus light") are given which may or may not be consistent with the actual contingencies programmed into the response task (reinforcement schedule).
In Hayes, et. al. 1986, the data were analyzed in terms of overall response rates under given experimental conditions.
This analysis showed that when rules (verbal instructions) and contingencies (immediate outcomes of responses) were in conflict, response rates were better predicted by rules than contingencies.
This is described as "insensitivity to reinforcement contingencies".
To be fair, Hayes (1989) does quality this and refers to 'relative insensitivity' to immediate contingencies.
However, this mode of analysis is a blunt instrument which dictates that the conclusion will be in the form of a dichotomy.
We applied a more sensitive and detailed measure: the IRT (Interresponse Time distribution) analysis. Blough (JEAB 1963) has demonstrated that this measure can reveal separate functional classes of responses in FR responding. Joyce and Chase (JEAB 1990) did apply an IRT analysis to rule governed behavior, but not in the context of multiple schedules.
This analysis breaks down response rates (average time between responses) into its components. In this case we measured the number of responses which occurred less than one second apart, one to two seconds apart, and so on up to a bin which contained all responses more than 16 seconds apart). One can see that many different IRT distributions could produce the same overall rates.
In particular, this analysis lets us pick up sub peaks in a multi modal distribution which would be washed out if we just used the average IRT (response rate).
Our data show the effects of both the immediate schedule contingencies and the instruction contingencies.
While the strongest effects are attributable to the instructions (not surprising given the relatively weak reinforcers that are specific to earning points), all of the data for all subjects showed subpeaks attributable to the immediate contingencies of button pushing.
This demonstrates simultaneous functional control by verbal discriminative stimuli and immediate reinforcement contingencies and lets us say that button pushing was simultaneously a member of two response classes:
The lower order class of pushing a button for points, and
The higher order response class of rule following for more delayed and diffuse social reinforcement.
Speculation:
It is conclusion that seeing schedule effects in human behavior (such as FI scallops) comparable to classic animal data (but my experience ....) is very much the exception.
Is it possible that we would see more similarity looking at human data on a more molar level? (Exam passing behavior of classes).
To conclude, behavior analysts define operants functionally on the basis of their situation and outcome. ALL operants are by definition contingency governed.
We can, however, make some distinctions between the types of contingencies which control behaviors on a given level of analysis.
In particular, we can distinguish between the control of behavior by the immediate outcomes of responding and the stimulus control exerted by membership of the operant in the more generic response class of rule following.
If we characterize these operants as lower and higher order classes of responses, then we can analyze the relative contributions of these two classes of contingency to the current behavior of the individual.
Hayes, Brownstein, Zettle, Rosenfarb and Korn (1986).Rule-governed behavior an sensitivity to changing consequences of responding.JEAB, 45(3), 237-256.
top | psych home page | MSUS,M home page | CSBS home page webmaster: paul.brandon@mnsu.edu