|Australian Journal of Educational Technology
1991, 7(1), 1-18.
The schism between operant behavioural and cognitive psychological views is examined with the aim of showing the potentials for their convergence when an instructional perspective is taken. As Lee (1988) points out, part of the problem lies in operant psychologists' use of stimulus-response language when they are really talking about condition-action sequences or means-to-ends. The inaccurate presentation of the operant position by cognitive psychologists is also part of the problem, and for the most part, neither reads the other's literature. With better communication, these problems could be overcome, although underlying philosophies of science might still differ.
An overtised-operant view of instruction on cognitive processes provides a further basis for closing the schism. Building from the behavioural processes of discrimination learning, chaining, verbal learning, etc., more complex cognitive structures can be described in terms of the more elemental structures from which they are built. Engelmann and Carnine's (1982) theory of instruction provides a key (through overt teaching strategies) in bridging the gap between basic operant principles and higher cognitive structures.
Operant psychology based interpretations of the changes that occur from being a novice to being an expert also are discussed to demonstrate additional commonalities between behavioural and cognitive positions.
Historically, instructional design grew out of research on behaviour change and the study of higher cognitive processes, such as concept learning and problem solving. In looking at this research, the first thing to impress one is the major schism between those professing an operant behavioural (or radical behavioural, if you prefer) approach, such as Skinner (1953,1957, 1968, 1977), Keller (1968), and Goldiamond and Dyrud (1966), and those professing a cognitive information-processing approach, such as, Gagne and Briggs (1974, 1979), Glaser (1987), and J. R. Anderson (1990). There clearly are polarised "good guys" and "bad guys" here, and which-are-which depends on your bias.
It is the thesis of this paper that the miscommunication between behaviourists and cognitivists need not exist when dealing with applications of psychology to instructional design, although it might go on forever when it comes to issues of a philosophy of science and research methods. I see three potential bases for improving communications and showing parallels in positions.
The first basis comes from the suggestion by Lee (1988) that operant psychologists give up their stimulus-response language and instead talk about means-to-reaching-ends-in-context. Instead of saying "The red light was a discriminative stimulus for a braking response, because in the past this response had been negatively reinforced by avoiding an accident (or ticket)," one might say "In the context of street intersection with heavy traffic, the braking action was a means of preventing an adverse consequence (end)." The language Lee proposes makes clear that operant psychology is concerned with human actions (operants) not with specific responses or specific acts. By use of such language, which is still technically correct within operant theory, the possibility of misinterpretation by cognitive psychologists is reduced.
A second basis for integration is in the demonstration (contrary to the views of many cognitive psychologists) of how basic concepts and procedures from operant psychology can be shown to produce higher order cognitive processes. Engelmann and Carnine's (1982) Theory of Instruction is at the heart of this demonstration, as is Gagne's work (1968, 1977).
A third basis for integration can be found in an analysis from an operant psychology perspective of the cognitive psychologists' work on the differences between novices and experts. In developing this paper, these three bases for bringing integration of thinking, and perhaps understanding of viewpoints, will be expanded with the hope of opening up communications between opposing camps.
Stimulus ---> Response ---> ConsequenceConsequences are often considered the major focus of operant paradigms and therefore will be discussed first. Consequences are defined by their functional effect on behaviour. Those that increase behaviour in the future under the given stimulus conditions are called reinforcers. Those that weaken behaviour are called punishers. It takes investigation to determine what, in fact, functions as an effective consequence. However, from knowledge of humans (ie, their similar genetic and learning histories) good guesses about what might serve as effective consequences are not too difficult to make.
Despite the common opinion that consequences are the primary (and to some "only") variable considered by operant psychologists, antecedent stimuli play a critical role in the learning actions that might be called intelligent. Preceding stimuli come to set the occasion for operant responses (ie, cue them) because of the effects of prior consequences. When a particular class of responses are reinforced (and responses not in that class are not reinforced) in the presence of a class of stimuli (but not others), the procedure is called differential reinforcement. The preceding stimuli that come to function to cue particular actions are called discriminative stimuli (or Sds). The important point for cognitive psychologists to take note of is that through differential reinforcement Sds can become what is common to a set stimulus examples; that is they can be the essential features of concepts. For example, a variety of steering wheels can control common steering responses; or a range of red colours can control the verbal response "red."
The operant response, by definition, is what has a common effect on the environment; that is, an operant response is described by the common effect of a class of responses. For example, in a Skinner box, it makes no difference whether a rat presses the lever with a front foot, a hind foot, or her nose. The lever-pressing action is defined by a common effect, depressing the lever 0.5 centimetre or so. For language conventions, the common effect is what teachers (or other social agents) will accept as correct. Thus, in every real sense operant psychology is describing a model for intelligent human behaviour, and the experimental work by operant psychologists on concept learning, prompting, chaining, shaping, problem solving, stimulus equivalence, the matching law, rule-governed behaviour, and so forth, have very real implications for instructional design. In other words, the simple Sd--->R--->C model is quite robust and, properly understood, can contribute to the understanding of lots of human learning usually restricted to domain of cognitive psychology. In view of all of this, Lee (1988) recommends that operant psychologists stop talking about discriminative stimuli and operant responses and talk about conditions, actions, and their contingent consequences. Actions are the means to certain ends under specified conditions - and this is how cognitive Psychologists talk about human action. The only real difference between operant and cognitive views is in the specification of causes. Cognitivists place causes in the mind, behaviourist place them in the environment and/or the learning history of the organism. The instructional designer must place the causes of learning where they are controllable by the teacher or teaching machine, ie, in the "environment" called the instructional program.
Instructional designers focus on teaching people when to do what. Consequences (feedback on correctness, corrections, praise, etc.) are still important, but are not emphasised in a discussion of design strategies. Instead the focus is on the an Sd--->R relationships (when to do what), which define the tasks which are the focus of instructional analysis. As such, Sd--->R units can be seen as the basic building blocks (component tasks) in behavioural chains, in procedures, and in more complex forms, such as problem-solving routines and strategies. These building blocks also can be found in the production tables of artificial intelligence programs. For example, Newell, Shaw, and Simon's (1957) AI program called the General Problem Solver deals with route-to-goal problems. One class of problems deals with how to get from one place to another in the United States. A database of bus, train, and air routes is used, along with maps, a distance reduction rule, and a production table to help decide when to do what. The distance reduction rule is: "Set up subproblems that maximally reduce the uncovered distances." The production table takes this form:
Newell, et al's basic artificial intelligence program was also effective in solving other route-to-goal problems with other databases and production tables. The important point here is recognition of the sameness of the concepts being used by cognitive and operant psychologists. By identifying these samenesses in route-to-goal problems, for example, strategies for teaching similar problem-solving procedures become apparent. For route-to-goal problems, problem solving requires learning how and when to reference databases (or fact systems, as they will be called later), how to discriminate relevant features of the problem that determine use of the procedure table, and how to apply the distance reduction rule in using the procedure table.(2)
Condition (Sd) ----------> Action (R) If over 1000 miles Use a plane If between 101 and 1000 Use a train or bus If between 1 and 100 Use a car or taxi If under 1 mile Walk
They begin with a macro-description of three kinds of analyses that are required to specify efficient conditions for the design of instruction for the teaching of cognitive skills - the analysis of behaviour, the analysis of knowledge, and the analysis of the communications used in teaching. These analyses are illustrated in Figure 1.
Figure 1: Three analyses necessary for the design of cognitive learning structures.
(Reproduced with permission of the authors from Engelmann and Carnine (1982)
Theory of Instruction. Copyrighted 1982 by S. E. Engelmann.)
The goals in the analysis of cognitive knowledge forms are two fold. The first goal is to identify types of knowledge forms (eg, concepts, rules) that can be taught with a common strategy. Some design strategies will be illustrate later. The identification of successful strategies for different types of knowledge provides the basis for a theory of instruction, ie, given a type of knowledge, a given design strategy is provided by the theory. The second goal is to find samenesses across pieces of knowledge that provide the basis for general-case instruction (Becker, 1986). These samenesses provide the basis for concepts, logical rules, empirical principles, and problem-solving routines. Success with this goal contributes to efficient instruction in that a basis for teaching generalisations is provided. Bruner (1966) called this goal of knowledge analysis economy of representation, and used the example of the logical rule called the Pythagorean theorem. This rule summarises properties common to all right triangles.
In general-case learning, one can teach through some examples, and the student can do any of the possible class members if the proper examples have been chosen. Non-general-case learning involves specific facts ("The yellow pencil is on the table") and fact systems (eg, a mapping of geological eras by the depth of layers found in a river gorge).
To better understand the hierarchical structure in Table 2, consider the problem-solving routine for fractions presented in Figure 2. Examine it and identify the rules and concepts that should be taught (component skills) before the full routine is presented.
Figure 2: A routine for picturing a faction. Teacher wording is in lower case letters, expected student response is in capital letters. The circles are the wholes to be divided into parts and filled in.The two transformation rules ("The bottom number tells us how many parts in each whole" and "The top number tells us how many parts we have") are joining forms (statements formed by joining together two or more concepts). The concepts that might have to be taught beforehand include (among others) fraction, equal, parts, and whole. Note that generalisable procedures can be derived from transformation rules and principles. For example, for the rule, "The bottom number tells us how many parts in each whole," the procedure is "Make as many parts in each whole as the bottom number tells you." For the principle, "If you heat matter, it expands," the procedure is, "To expand matter, heat it." Problem-solving routines involve sets of concepts, rules, and principles, and the procedures derived from them. From an operant psychology perspective, these activities involve primarily discrimination learning (concepts and rules), verbal response learning, and at times other motor operations under the control of verbal cues (eg, making lines to represent a number, counting from-a-number-to-a-number, writing answers).(3)
Basic concepts. Basic concepts are concepts that are best taught by examples, rather than definitions or synonyms (which are fact statements). Basic concept learning can be viewed as a multiple discrimination problem (Becker, 1986). Examples of concepts have to be discriminated from non-examples. The discriminations are between essential features positive examples (Sds symbolised as S+ features), and essential features of negative examples (Sds symbolised as S- features), If positive examples are red (S+), negative examples are other colours (S- ). If you know an object is blue, you know it's not-red. Within positive examples and within negative examples, it is necessary to discriminate essential features from non-essential features. For example, if S+ is red, non-essential features might include shape, size, texture, location, etc. (For more detail, see Becker, 1986).
To teach basic, single-dimension (or single-feature) concepts, such as right angle, smooth, or orange with examples requires a set of positive examples which all share the essential S+ features, and a set of negative examples, each of which do not share all S+ features. In initial teaching, Engelmann recommends that non-essential features be kept the same for the first 11 or so examples. A common setup, as illustrated in Figure 3, is used to focus learner attention on essential difference between positive and negative examples. In switching from positive to negative examples (or vice versa) keep everything the same except an essential feature (the minimum difference principle). Such pairs of examples clearly show concepts boundaries. If this is done consistently, the students learns that examples that are more different than the negative examples shown are also negative examples. In presenting positive examples for concepts with a range, sample the full range of possibilities with 3 or 4 examples to show the range of sameness (the sameness principle). The student will learn that other examples that fall within the range shown are also positive examples. Learner acquisition is then tested with 5 or 6 examples; and finally the irrelevant features are varied to expand the range of application.
These principles lead to an 11-step prescription for teaching any noncomparative, single-dimension basic concept. Consider the steps illustrated in Figure 3 to teach the concept flam before going on.
Figure 3: Initial steps for teaching a single-dimension non-comparative concept.
The concept taught in Figure 3 is, of course, over. Note that the setup does not change throughout the initial teaching examples. The sequence starts with two negative examples, then in going from example 2 to example 3, and from example 5 to example 6, minimally different positives and negatives are used to focus attention on the S+ versus S- features. Examples 3 to 5 show the range of S+. By starting with negatives, the learner is less likely to form a wrong hypothesis, since the minimally different positive will focus attention on the critical concept feature. Testing begins with example 6 and, finally, after example 11, the setup is varied to teach which features of examples are irrelevant and thereby extend the range of application of the concept being learned. This 11-step sequence should work efficiently for any non-comparative single-dimension concept, and thus provides a component for a theory of instruction.
Returning to Table 2, single-dimension comparative concepts come next. These can be taught by a 12-step sequence that follows that shown in Figure 3 except that it begins with a starting example. To illustrate, the starting example might be: "Look at this ruler in my hand." Then, the 11 steps used to teach non-comparatives are followed except that each new example is compared to the one before it. For example step 1 following the starting example might be: "It didn't get steeper" (it got less steep). Then step 2 might be: "It didn't get steeper" (example 2 is the same as 1). Step 3 might be: "It got sleeper" (example 3 is slightly steeper than 2), and so on. The same pattern of examples shown in Figure 3 is used, after a starting example, and each judgment is a comparison. A prescription for teaching basic comparative concepts is thus provided. For many single dimension concepts, such as steep, near, high, etc., where the meaning is relative to the context, the comparative form provides a clearer basis for initial instruction than the non-comparative form.
The next type of concept in Engelmann's classification is nouns. A different strategy is required to teach noun concepts. Because nouns have many features, there are no precise minimum differences between examples and non-examples (eg, a horse and a cow, or a chair and couch). Also, because there are so many possible negative examples for noun concepts, it is safer for the designer to constrain negative examples to those known to be in the student's repertoire. Starting the sequence with positive examples will likely focus attention on the essential features more quickly. Since the names of the negatives have already been taught, they are used in identifying examples. The sequence might start with 3 or 4 positive examples and then negatives interspersed with positives. For example, in teaching truck, you might present a semitrailer, a van, and a pickup as positive examples; and then start testing ("What's this?") with a train (S ), a bus (S-), a flatbed truck (S+), a car (S-), a van (S+), etc.
As noted earlier, concepts can also be taught through fact statements. For example: "Above is another word for over"; "It's a vehicle if it can take you places"; "A chair is furniture that can hold one person in a sitting position and support the back." Whether concepts are taught by statements or examples depends on the concept and the learner's skills.
Because of space limitations, only key aspects of the strategies for teaching the major joining forms and complex forms listed in Table 2 will be presented. For more details, the reader is referred to Engelmann and Carnine (1982) or Becker (1986).
Joining forms. Joining forms join together two or more concepts. Logical rules (transformation rules) are found in the rules of grammar (eg, "More than one X are Xs"), the rules underlying mathematical systems (eg, "The third place from the right of a number tells how many 100s"), propositional systems, and other logical structures, such as class inclusion ("Boys and girls are children"). Only positive examples are needed to practice application of transformation rules. The student is taught to use the rule to complete examples and to explain what was done using the rule.
Fact statements name something and then say something about it ("Birds have feathers," "Matter will expand if you heat it," "The pencil is on that table"). Facts join concepts observed to go together. Principles are generalisable fact statements. They are taught by having the student use the principle to predict what will happen and then explain why it happened. Sequencing of examples for some symbolic facts follows a transformation sequence. For other facts, sequencing follows the steps for single-dimension concepts, except that two questions are used in the teacher wording (predict and explain) rather than one. For the principle "Ice melts if the temperature is above 32 degrees Fahrenheit" (with distilled water at sea level), the first few examples might go like this:
In cognitive theory, interrelated pieces of declarative knowledge (concepts, principles, patterned sequences of events, etc.) lead to structures called schemes (Anderson, 1990). These are much like Engelmann's fact systems. A primary difference is their source. Most cognitive psychologists try to learn about schemes mainly by trying to find out how they are organised in peoples' minds. Engelmann would logically analyse the knowledge to be taught for its structure, and when it involves facts, attempt to organise it into a fact system that can be directly taught through a visual and verbal presentation. Engelmann's practice of initially building broad structures to which more detailed knowledge can be added is in keeping with the cognitivists' descriptions of how new information is more readily learned when it can be fitted into a schema already in place (Anderson, 1990).
The design of problem-solving routines (algorithms) must vary for each problem set, but follows four general rules:
In designing problem-solving routines, Engelmann plans for the transfer of component skills from one routine to another. This can provide a great savings in instructional time to accomplish a given objective. For example, nearly all of the steps in an addition routine (using an equation format and a counting strategy) transfer to an algebra addition format, a format for addition the fast way, and subtraction formats.
Cognitive psychologists' parallels to Engelmann's problem-solving routines can be found in Scandura's Structural Learning Theory (1983), and Landa's Algo-Heuristic Theory of Instruction (1983). As Reigeluth (1983) points out in his introduction to Landa's chapter in his book, "In spite of its cognitive orientation, Landa's theory is highly compatible with behaviourally oriented theories. The latter focus more on observable procedures (ie, overt behaviours), whereas Landa focuses more on unobservable procedures, (ie, cognitive processes). Nevertheless, ... [they] require essentially similar methods of instruction" (p. 165).
There are many other important considerations in designing problem solving routines. For example, it is more efficient to teach all component concepts, rules, and procedures by themselves before teaching the full routine. In early formats, all steps in the routine should require overt responses. Without this requirement, there is no basis for precise corrections. Later teaching formats chunk steps to speed up the routines so they eventually are covertised. In Anderson's (1990) terms, the steps are complied and tied to procedures. From an instructional design viewpoint, this is one basis for describing the differences between the novice and the expert.
Developing automaticity of basics. The expert has the component skills needed for solving problems automatically available. This allows a focus on critical aspects of the problem. To the operant psychologist, automaticity can been seen as the result of the frequent repetition of a chain of behaviour. The control of each component in the chain by external Sds drops out and is replaced, presumably, by response-produced stimuli (the internal stimuli produced by actions). Instead of typing a word one letter at a time by referring to a keyboard chart, the word is typed as a unit. Instead of sounding out each letter in a word phonetically, the whole word (or phrase) is read. New functional units of behaviour are formed where all the steps are fused and seem to occur automatically. In reading, a major goal should be to make decoding automatic through practice so that the learner can focus attention on comprehension of what is being read (LaBerge & Samuels, 1974).
Compilation and Proceduralisation. Steps in routines become chunked (compilation) and relevant declarative knowledge (ie, concepts and principles) are tied to problem-solving procedures (Anderson, 1990). As noted earlier, in designing problem-solving routines, Engelmann first keeps all steps overt so errors can be corrected and feedback given. Concepts, rules, and principles are made explicit in the routines (and taught separately before being use in a routine). Then overt steps in the routines are gradually combined through changes in formats until students perform on their own. Steps fuse together and become internalised as external prompts are faded. In Anderson's model of the expert, declarative knowledge is tied to procedural knowledge where it is explicitly needed. In Engelmann's scheme, concepts and rules become part of problem-solving routines through direct program design.
Glaser (1987) expresses a parallel set of ideas from a cognitive point of view focusing on rules and principles rather than procedures: "The knowledge underlying problem-solving skill is represented as a set of if-then goal-oriented production rules. The tutor monitors whether or not a student has carried out each rule correctly, and it responds to any errors or missing rules. The learning theory involved assumes a knowledge compilation process, in which, as experience is acquired in a domain, sequences of rules collapse into larger macro rules. This enables the tutor to adjust the grain size of instruction as learning proceeds" (Glaser, 1987, p.xv).
Becker, W. C. (1986). Applied Psychology for Teachers -A Behavioral Cognitive Approach. New York: Macmillan. (Originally published by Science Research Associates.)
Brophy, J. (1984). Designing curriculum and instruction. Contemporary Psychology, 29, 622-24.
Bruner, J. S. (1966). Toward a Theory of Instruction. Cambridge, MA: Belnap Press.
Chomsky, N. (1959). Verbal Behavior, by B. F. Skinner. Language, 35, 26-58.
Engelmann, S., & Carnine, D. W. (1982). Theory of Instruction: Principles and Applications. New York: Irvington. (Being republished in early 1991 by the Association for Direct Instruction, PO Box 10252, Eugene, Or.)
Engelmann, S., Davis, K. & Davis, G. (1981). Your World of Facts, Level 1. Chicago: Science Research Associates.
Engelmann, S., Davis, K. & Davis, G. (1982). Your World of Facts, Level 2. Chicago: Science Research Associates.
Gagne, R. M. (1965, 1970, 1977). The Conditions of Learning. New York: Holt, Rinehart and Winston.
Gagne, R. M. (1968). Learning hierarchies. Educational Psychologist, 6, 1-9.
Gagne, R. M. & Briggs, L. J. (1974,1979). Principles of Instructional Design. New York: Holt, Rinehart and Winston.
Glaser, R. (1987). In R. Glaser (Ed.), Advances in Instructional Psychology (Vol. 3). Hillsdale, NJ: Lawrence Erlbaum Assoc.
Goldiamond, I. & Dyrud, J. E. (1966). Reading as operant behavior. In J. Money & G. Shiffman (Eds.), The Disabled Reader, 93-115. Baltimore: John Hopkins.
Keller, F. S. (1968). "Good-bye teacher... " Journal of Applied Behavior Analysis, 1. 79-89.
Landa, L. N. (1983). The algo-heuristic theory of instruction. In C. M. Reigeluth (Ed.), Instructional-Design Theories and Model: An Overview of their Current Status. Hillsdale, NJ: Lawrence Erlbaum Assoc., 163-180.
LaBerge, D. & Samuels, S. J. (1974). Toward a theory of automatic information processing in reading. Cognitive Psychology, 6, 293-323.
Lee, V. L. (1988). Beyond Behaviorism. Hillsdale, NJ: Lawrence Erlbaum Assoc.
MacCorquodale, K. (1969). B. F. Skinner's Verbal Behavior: A retrospective appreciation. Journal of the Experimental Analysis of Behavior, 12, 831-841.
Newell, A., Shaw, J. C. & Simon, H. (1957). Preliminary Description of a General Problem Solving Program-I (GPS-I). Report CIP Working Paper 7, Carnegie Institute of Technology, Pittsburgh, Pa.
Reigeluth, C. M. (1983). Instructional-Design Theories and Models: An Overview of their Current Status. Hillsdale, NJ: Lawrence Erlbaum Assoc.
Scandura, J. M. (1983). Instructional strategies based on structural learning theory. In C. M. Reigeluth (Ed.), Instructional-Design Theories and Models: An Overview of their Current Status. Hillsdale, NJ: Lawrence Erlbaum Assoc., 213-246.
Skinner, B. F. (1953). Science and Human Behavior. New York: Macmillan.
Skinner, B. F. (1957). Verbal Behavior. New York: Appleton-Century-Crofts.
Skinner B. F. (1968). The Technology of Teaching. New York: Appleton-Century-Crofts.
Skinner, B. F. (1977). Why I not a cognitive psychologist. Behaviorism, 5, 1-10.
Watson J. (1930). Behaviorism. New York: Norton.
|Author: Dr. Becker got his PhD in Psychology at Stanford in 1955. He worked 15 years at the University of Illinois, before moving to Oregon. His early research focused on how parent behaviour influenced personality development in children. He then moved into schools where his research has been largely with younger children, first in showing teachers how to deal with behaviour problems and then with better instruction. From 1968 to 1978, he worked with Siegfried Engelmann and Doug Carnine within project Follow Through, a federal program for economically disadvantaged 5 to 9 year olds. It was an attempt to Follow Through on gains possibly made in Head Start programs. This instructional model was called the Direct Instruction Model. The model was most effective of nine comparison models in bringing disadvantaged children up to national norms in reading, mathematics, language, and spelling. During the past eight years he has been teaching a graduate course on Instructional Psychology.
Please cite as: Becker, W. C. (1991). Toward an integration of behavioural and cognitive psychologies through instructional technology. Australian Journal of Educational Technology, 7(1), 1-18. http://www.ascilite.org.au/ajet/ajet7/becker.html