In Search of Cumulative-Hierarchical Learning

Behavior Analysis Program
Department of Psychology
Western Michigan University

Yes, we can say, CRF is really FR 1 (we can say, continuous reinforcement is really fixed ratio reinforcement where the ratio of response per reinforcer is 1 to 1). But if such labeling is not a reduction to absurdity, it’s at least a reduction to triviality. Continuous reinforcement shares none of the properties of fixed ratio reinforcement, such as the pause after reinforcement followed by the rapid acceleration of responding up to a hell-bent-for-leather, maximum rate. Continuous reinforcement does not capture the spirit, the essence, of fixed-ratio reinforcement. 

Yes, we can say, a pigeon’s food reinforced key peck is really a “mand” for food. But such labeling trivializes the concept of “verbal behavior.” The reinforced key peck, that simple, ubiquitous form of the operant shares none of the rich properties of verbal behavior, the properties that make verbal behavior so unique—verbal behavior’s participation in Hockett’s linguistic productivity, a participation that allows the verbal organism to make appropriate novel responses to novel stimuli, verbal behavior’s grammatical nature. And, it is a short distance from saying, the pigeon’s reinforced key peck is a mand, to saying, the non-verbal child’s head banging is her effort to COMMUNICATE her needs and wants to you, to communicate that she wants you to pay more attention to her and to stop making such difficult demands of her. It is with good reason that we classify pigeons as non-verbal organisms. 

And, maybe we can say, shaping, chaining, and mediation are really examples of cumulative-hierarchical learning, but none of those three behavioral processes capture the essence, the importance, of cumulative-hierarchical learning. I think equating them trivializes the important, but complex concept cumulative-hierarchical learning.

Why? 

The first step to answering this question, was a review of some of Staats (1977) earlier writing on what he then simply called hierarchical learning: “. . . behavioral skills a child should have before he should attempt a new learning task” (p.287). “. . . the child in many instances must acquire one repertoire of skills—if not several—before he is able to move on to the learning of a more advanced repertoire” (p. 287). “The child learns skills that determine how rapidly he will learn new skills” (p. 287). “The child is involved in a progression of learning that moves from the acquisition of basic repertoires of skills to the acquisition of more advanced skills based upon the earlier learning.” (p. 288). And I also reviewed Hixson’s (200?) quotation from (Rosales-Ruiz and Baer, 1997, p. 533): “A behavioral cusp…is any behavior change that brings the organism’s behavior into contact with new contingencies that have even more far-reaching consequences.” 

The next step was to use these quotes to construct a definition of cumulative-hierarchical learning: the learning of a skill or value that affects learning in a wide variety of other contexts (by value I meanreinforcers and aversive conditions). And it is this wider variety of contexts that makes cumulative-hierarchical learning such an exciting, theoretically profound, and utilitarian concept. The wider variety of contexts is Staats’ new learning task, more advanced repertoire, new skills, and more advanced skills; and more clearly, it is Rosales-Ruiz and Baer’s more far-reaching consequences.

For example, consider the autistic child’s behavior of looking at the trainer. This may be merely part of a simple behavioral chain: 

SD1 (the trainer says, “look at me,”)

R1 (the child looks at the trainer) 

That looking response always produces the same reinforcer/discriminative stimulus.

Sr/D2 (the trainer says, “do this”; and then touches his own nose) 

This second SD is always the occasion for reinforcement of the same response (the child’s touching her nose). 

In this case, to call looking at or attending to the trainer an example of cumulative-hierarchical learning seems to trivialize the concept. 

But, the child’s response of looking at the trainer could be a behavioral cusp in an example of cumulative-hierarchical learning: Suppose the skill of looking at the trainer affects the child’s performance or learning in a wider variety of contexts, such as imitating eye touching, hand clapping, toy playing, shoe tying, going to the lunch room, playing the piano, and drawing. In this case, learning the single behavioral cusp of looking at the trainer is clearly an example of cumulative-hierarchical learning because it allow the child to learn new, more advanced skills (e.g., shoe tying); the skill of looking at the trainer participates in a wider context than the simple nose-touching chain; the looking-at-the-trainer cusp has far-reaching consequences.

And the look-at-the-trainer cusp participates in even more cumulative-hierarchical learning when it produces SDs for a wider range of instructions than, “do this” (e.g., “put same with same,” “wash your hands,” and “put on your coat, so we can go to the playground”).

And, just as we trivialize cumulative-hierarchical learning by offering simple behavioral chaining as an example, I think we add to the trivialization by offering shaping and even simple response mediation as examples. Consider response mediation. A mediating response is a response that is correlated with an SD, with the result that the stimuli produced by that response will acquire the evocative function of the SD and evoke the response in the absence of that SD. Mediating responses may be learned in one situation (the presence of the original SDs). And they may result in the attainment of otherwise missed reinforcers in another situation (when the original SD is no longer present). But it is not clear to me that these mediating responses always participate in cumulative-hierarchical learning. 

As an example, the Hixson cited work by, Lowenkron (1998) on delayed matching to sample, where he established a set of responses that mediated between a corresponding set of sample stimuli and the correct selection of delayed comparison stimuli. This training in the use of these mediating responses facilitated the use of a novel set of mediating responses to mediate between a corresponding set of novel sample stimuli and the correct selection of delayed comparison stimuli. That the earlier mediation training facilitated or even eliminated the need for later mediation training is an excellent example of cumulative-hierarchical learning. But, as Hixson implied, by itself, the use of the original mediating responses with the original delayed-matching-to-sample stimuli is not an example of cumulative-hierarchical learning; we have cumulative-hierarchical learning only when the subsequent set of mediating responses facilitated performance with a novel set of delayed-matching-to-sample stimuli. 

The learning is hierarchical in that learning the response-mediation components facilitated learning the original delayed matching performance. And it is cumulative in that learning the response-mediation process for one set of delayed identity-matching stimuli, facilitated the learning of a novel set. 

Though cumulative-hierarchical learning may involve something like chaining, shaping, or response mediation, not all chaining, shaping, and response mediation are examples of cumulative-hierarchical learning.

References

Hixson, M. D. (200?) Response Mediation and Cumulative-Hierarchical Learning. The Analysis of Verbal Behavior, ?, ??

Hockett, C.F. (1960). Logical considerations in the study of animal communications. Animal Sounds and Communications, 7, 392-429
.
Lowenkron, B. (1998). Generalization of delayed identity matching in retarded children. Journal of the Experimental Analysis of Behavior, 50, 163-172. 

Rosales-Ruiz, J. and Baer, D.M. (1997). Behavioral cusps: A developmental and pragmatic concept for behavior analysis. Journal of Applied Behavior Analysis, 30, 533-544. 

Staats, A.W. (1977). Child learning, intelligence, and personality: Principles of a behavioral interaction approach (revised edition). Kalamazoo: Behaviordelia