Conditional Discrimination Learning and Conditional Reasoning

by Nonhuman Animals

©Roger K. Thomas 2005

 

Roger K. Thomas, Ph.D.

Professor Emeritus

Department of Psychology

The University of Georgia

Athens, GA 30602-3013 USA

 

Abstract

There is an extensive history of investigating "conditional discrimination learning" by nonhuman animals (hereafter "animals") using various paradigms, and often it is stated or implied that the animals had demonstrated conditional reasoning corresponding to forms such as, "if p, then q." However, this paper questions whether there has ever been a valid demonstration of conditional reasoning by nonhuman animals. Previous investigations have involved methods that either (a) confound conditional reasoning with the possibility of rote-learning or (b) confound the possibility of conditional reasoning with conjunctive reasoning. The only nonverbal procedure of which I am aware that might be used to show unequivocal conditional reasoning by an animal was developed for use with humans. However, that experiment appears impractical for nonhuman animals, and its author (Bourne, 1970) relied partly on the subjects' verbal explanations to confirm that they had reasoned conditionally.  It is hoped that one result of this critique will be to prevent future researchers from misinterpreting or misrepresenting, either inadvertently or intentionally, the results of conditional discrimination learning research.

Introduction

The literature addressing conditional discrimination and conditional reasoning by nonhuman animals (hereafter "animals") is too vast to attempt a full review, and such a review is not necessary for the critique that is to be presented here. Enough examples from the literature will be cited to (a) demonstrate the pervasiveness of conditional discrimination learning research and (b) to show that misinterpretations that attribute conditional reasoning to animals often occurs. To the best of my knowledge, no study using animals has provided a valid demonstration of conditional reasoning, and, in any case, two simple tests or questions that any researcher may apply to any study in question can determine whether that investigation provided a valid example. The two questions are: (1) Did the investigation confound the possibility of rote learning with the possibility of conditional reasoning?  (2) Did the investigation confound the possibility of conjunctive reasoning with the possibility of conditional reasoning?  If the answer to the either question is “yes,” then the investigation did not provide a valid example of conditional reasoning.

The following examples of the misattribution of conditional reasoning to animals were quoted from literature that suggested the potential relevance of conditional learning to the acquisition of language or its prerequisites in nonhuman primates.  Levy (1988) wrote:

...left-hemisphere superiority for true conditional discriminations, but hemispheric symmetry for simple associative learning, suggests a specialization for pre-propositional reasoning. It is not difficult to perceive how these abilities could become elaborated, differentiated, and integrated together to form the basis for the evolution of human language, including speech, phonetic analysis, and syntactical organization (p. 165; emphasis added).

In a related vein, Deacon (1988a,b) wrote:

....prefrontal...[cortical]...expansion is probably correlated with the evolution of language. This association is suggested by...[among other examples that Deacon cited]...the role of lateral prefrontal areas in conditional learning....The earliest development of symbolic communication... approximately two million years ago provided significant...reproductive advantage to those most facile with its acquisition and use and so established powerful selection for these abilities. Most significant of these were selection for enhanced learning of complex conditional associations and enhanced oral-vocal motor-sequencing skills (p. 407-408; emphasis added).

Deacon (1989) iterated this view in his article, "The neural circuitry underlying primate calls and human language;" see especially the section headed "Prefrontal cortex and combinatorial learning" (pp. 386-387) and, more recently, in The symbolic species (1997, pp. 260-265).

To support his view that the involvement of the "lateral prefrontal areas in conditional learning" is related to the evolution of language, Deacon (1988a,b; 1989) cited data reported by Dewson (1977) and Petrides (1982, 1987), and Levy relied on Dewson's data (1977, 1978).  Dewson's accounts of his experiments were not presented in terms of his monkeys using conditional rules, such as, "if p, then q."  However, Petrides wrote, "...the animal had to learn to select the appropriate...[response]...according to the conditional rule: If stimulus A is shown, select Response X...." (1987, p. 98)  As will be discussed below, Petrides' experiments are confounded both by the possibility for rote learning and for the use of conjunctive as opposed to conditional reasoning to support such an interpretation.

It is beyond the scope of the present article to address the merits of Deacon's and Levy's general arguments, but there are serious questions about the experimental evidence upon which they based significant aspects of these arguments, namely, the evidence that they cited for monkeys' use of conditional reasoning. Further, it is reasonable to ask whether any report of any animal's use of conditional reasoning has been valid. The present article will show why all previous studies purporting to show animals' use of conditional reasoning are likely to be found to be inconclusive.  The present article will also suggest how studies with valid experimental procedures might be conducted.

Before proceeding, some important qualifiers must be expressed. First, it is not being asserted that the animals in studies such as those of Dewson (1977, 1978) and Petrides (1982, 1987) did not use conditional reasoning but that the experimental procedures do not enable us to show conclusively that they did.  Second, Dewson’s and Petrides’ research does indicate that experimentally induced, left-anterior, neocortical lesions selectively affect monkeys' performances on the tasks used. While these lesion effects might suggest some validation of Dewson's and Petrides' research in the context of their use by Deacon and Levy, it is not sufficient validation to support their use as providing evidence for conditional reasoning.  The criticisms raised here regarding the experimental evidence for conditional reasoning should not be taken to demean the considerable value of Dewson's and Petrides' findings nor the merits of Deacon's and Levy's general speculations. 

Inconclusive "Conditional Discrimination" Learning Tasks

The procedures used by Dewson (e.g., 1977, 1978) and Petrides (1982, 1987) exemplify an extensive literature in animal learning that is based on a procedure known as "conditional learning" or "conditional discrimination learning" (e.g., French, 1965; Carter & Werner, 1978; Zentall, Hogan, & Edwards, 1984).  However, as will be shown, procedures with such names do not provide conclusive evidence for an animal's use of conditional reasoning.  It should be emphasized that use of procedures with these names, even when the animals learned successfully, does not mean that the authors also claimed that the animals showed conditional reasoning.  Many, perhaps most, do not make such claims, and it is only with those who do claim to have shown conditional reasoning by animals that this article questions.

The typical conditional learning task, including those used by Dewson (1977, 1978) and Petrides (1982, 1987), involves two successively-presented discriminanda, represented here as A and B, only one of which is used on a given trial and two simultaneously-presented discriminanda, say X and Y, which appear on every trial. A or B serves as an associative cue to select either X or Y. It is tempting to describe and conceptualize such tasks, as many investigators have done, as embodying conditionals such as: "If A, then X and if B, then Y."

Typically, the same four discriminanda are used throughout training or are presented repeatedly, as was done in Dewson's and Petrides' experiments.  Such repeated presentations make it likely that the few specific configurations afforded by the discriminanda can be learned by rote-memorization.  Of course, such configuration learning is confounded with the possibility that the animals used conditional reasoning, as French (1965) and others before him (e.g., Nissen,1953) have noted. Such confounding prevents Dewson's and Petrides' studies from providing conclusive evidence for animals' use of conditionals. However, even if specific configuration learning is precluded, there remains a fundamental problem with the experimental design used in most "conditional learning" studies. Before discussing that design problem, it may be useful first to show how the specific-configuration-learning confound can be avoided.

There are three basic ways to avoid specific configuration learning: (a) use exemplars of conceptual categories for the successive discriminanda as Riopelle and Copelan (1954) did, (b) use exemplars of conceptual categories for the simultaneous discriminanda as Thomas and Kerr (1976) did, and (c) use conceptual categories as exemplars for both the successive and simultaneous discriminanda as Burdyn and Thomas (1984) did.  I will describe now the research by Burdyn and Thomas to illustrate the use of conceptual categories as discriminanda and, later, to illustrate the design problem mentioned above.

Burdyn and Thomas (1984) used exemplars of the conceptual categories "same" and "different" as the simultaneous discriminanda; the exemplar of "same" was an identical pair of objects and the exemplar of "different" was a non-identical pair of objects.  New pairs of objects were used on each trial in the conceptual category phases of the testing which precluded the monkeys from memorizing specific discriminanda and reinforcement associations.  As the successive discriminanda, we used the conceptual categories "triangularity" and "heptagonality" which were represented by 120 discriminable exemplars of each to make it unlikely that the monkeys memorized specific triangles and heptagons.

An apparatus with three guillotine doors was used.  During most of the training, all three doors were raised and lowered at the same time. On a given trial, (a) either a triangle or a heptagon was presented as a result of raising the center door, (b) a pair of identical objects was presented as a result of raising one of the outer doors, and (c) a pair of non-identical objects was presented as a result of raising the other outer door; the left-right positions of the same and different pairs were determined quasi-randomly using the Fellows (1967) series. When a triangle was presented, the correct response was to displace the object-member of the same-pair (which covered a food well) that was closest to the center door in order to gain access to a bit of fruit beneath the object. When a heptagon was presented (as illustrated below), the correct response was to the object-member of the difference-pair (which covered a food well) that was closest to the center door.

Later during training, the center door was raised to expose either a triangle or a heptagon; then, it was closed to cover the triangle or heptagon before the outer doors were raised to expose the same and different pairs of objects. We systematically increased the interval between closing the center door and raising the outer doors and found that our best performing monkey could reach a stringent criterion of correct responding with a 16 sec. interval; the criterion was a minimum of 13 of 15 correct on 15 triangle-same trials and 13 of 15 correct on 15 heptagon-different trials in a 30_trials session. Therefore, when the cues were visually absent, triangularity and heptagonality symbolically represented same and different, respectively.

It is tempting to conceptualize these monkeys' successful performances as using conditionals which might be expressed as "if triangle, then same" and "if heptagon, then different." However, Burdyn and I realized that we could not say unequivocally that the task required the monkeys to use conditionals, because it was possible that they were forming conjunctive associations such as "triangle and same" and "heptagon and different." It is also possible that they were forming biconditional associations, such as, "triangle if and only if same" or "heptagon if and only if different." This general interpretational problem also affects Dewson's and Petrides' studies and, most likely, all other so-called "conditional learning" studies in animals. It should be noted also that most animal studies, including Dewson's and Petrides', have not used conceptual-category discriminanda which means their subjects might have rote-memorized the specific configurations associated with the discriminanda-reinforcement contingencies.

Conditional Versus Conjunctive Reasoning in Standard Logic

To distinguish whether an animal had used conditional or conjunctive reasoning will require tasks that differentiate logically between the two types of reasoning. That means, for example, that if one wants to show that animals can reason conditionally, the task must incorporate all the logical requirements of conditional reasoning. To explain this further, I will begin with a review of the truth tables for the conjunctive and the conditional.

Conjunctive                    

p     q     p&q                    

T    T        T                     

T    F        F                      

F    T        F                     

F    F        F                      

 

Conditional

 

p     q      p>q

 

T    T        T 

T    F        F

F    T        T

F    F        T

Of course, truth tables are abstractions. To render them concrete for the purpose of experimental research, p and q above may represent discriminanda or behavioral responses, as will be illustrated below using examples from Burdyn and Thomas (1983). Whether the relationship between p and q in column 3 is a true or false is axiomatically determined by the rules of standard logic.

As seen in the truth table above for the conjunctive, the only valid or true example of a conjunctive relationship is when p and q are both T. However, as may be seen for the conditional, three p-q relationships are deemed to be T. One of these is the same as the only p-q relationship that is T for the conjunctive. Therefore, for an experimenter to distinguish whether an animal has reasoned conjunctively or conditionally, the experiment must be designed to show that the animal also performs consistently with all p-q relationships of both the conjunctive and the conditional. It is easy to design an experiment that incorporates all p-q relationships for the conjunctive, but it is very difficult, perhaps impossible, to design a nonverbal experiment that is likely to be successful with nonhuman animals that can distinguish whether an anima had reasoned conditionally. Perhaps, the most troublesome requirement to include in an experimental test of the conditional is that represented in rows 3 and 4 of the truth table.

The following verbal example will illustrate the conditional: "If the boat sinks, then we swim." When both antecedent (boat sinks) and consequent (we swim) are true, the statement represents row 1 in the truth table. If the boat sinks and we do not swim (row 2; antecedent is true and consequent is false), we will not have manifested the conditional relationship as required by standard logic in row 2, and the result according to the conditional relationship will be false or invalid. However, if the boat does not sink (rows 3 and 4, antecedent false), there is no condition set to swim or not swim; that is, we can swim or not swim and still have a "true" manifestation of the conditional.

In mapping the p-q relationships on truth tables in typical animal "conditional learning" research, the design problem exemplified in most of that research including Dewson's (e.g., 1977, 1978), Petrides' (1982, 1987), and Burdyn and Thomas's (1984), is that two inextricably linked relationships are being investigated concurrently. We can map only one relationship at a time on a truth table, but the two relationships are not independent. The contingencies are restrained in a way that prevents us from knowing whether the separate relationships involve conjunctions or conditionals (or, for that matter, biconditionals. e.g., A if and only if B).

There is a further problem in the case of the conjunction and the conditional that is caused by the two interdependent relationships. The problem can be illustrated by citing virtually any animal conditional learning experiment. Because the Burdyn and Thomas (1983) experiment was more rigorous than any other study that I know about in terms of using conceptual categories as discriminanda rather than specific-configuration-learnable, discriminanda-reinforcement associations, it will be used here to illustrate the problem.

Recall that the Burdyn and Thomas experiment involved the triangularity-same and heptagonality-different relationships. For purposes of relating the research to the conditional truth table, substitute "triangularity" (t) for p and substitute "same" (identical pair of objects) or s for q. The most meaningful "translation" of T is to denote the presence (+) of an attribute such as triangularity which will be done symbolically as "+t" and F will be denoted symbolically by the absence of triangularity (-t). Note that the absence of a triangularity (-t) in this experiment was consistently associated with the presence of an exemplar of heptagonality (symbolically, +h), and the absence of same was consistently associated with the presence of a pair of non-identical or "different" objects or (+d). Thus, in this experiment, for example, -s and +d are equivalent and might be written as -s/+d; similar equivalency exists between -t and +h, thus, -t/+h. Note also that a separate and equivalent truth table is needed to describe the heptagonality-different relationship which was running concurrently with the triangularity-same relationship. Column 3 denotes the axiomatic outcome, T or F, according to standard logic. Column 4 denotes the "translated" outcome in terms of whether the monkeys should be reinforced for associating the "triangularity" and "same" relationship. Appropriate (i.e., "correct") responses to the triangularity-same relationship result in reinforcement, as denoted by "Yes" in column 4, and inappropriate responses are not reinforced, denoted by "No."

                             Conditional                                        Conditional

p(triangle)    q(same)       Outcome re: p>q       Outcome re: Reinforcement

T(+t)            T(+s)                                   T                                          Yes

T(+t)           F(-s/+d)                                F                                          No

F(-t/+h)       T(+s)                                    T                                          No

F(-t/+h)       F(-s/+d)                                T                                          Yes

It can be seen that there is incongruence between the truth-functional outcome and the reinforcement-contingency outcome in row 3. A comparable analysis for the conjunctive will show incongruence in row 4. Do we conclude that the subject has used neither a conditional nor a conjunctive relationship? No, that conclusion could not be justified, because the design in the typical conditional discrimination task constrains the outcomes. The absence of triangle in rows 3 and 4 is confounded with the presence of a heptagon, a meaningful alternative discriminandum, and the non-choice of "same" in row 4 is confounded with the choice of "different," also a meaningful alternative.

It may be noted that had the above been diagrammed in terms of the truth-table for the biconditional, the truth-functional and reinforcement-contingency outcomes would have been congruent for all four rows. Such consistency might imply that the biconditional relationship expressed as "triangularity if and only if same" had been demonstrated in our experiment. However, it is my view that the demonstration of the biconditional is also inconclusive owing to the compelling and confounded alternative (i.e., the heptagonality-difference relationship); that is, a demonstration of the biconditional should involve more open-ended alternatives to the discriminanda-responses involved in the focal relationship (i.e., triangularity-same). As it is, one still cannot say whether triangularity-same and heptagonality-different relationships were understood (reasoned) by the monkeys as conjunctive, conditional, or biconditional relationships.  It is reasonable to say that Burdyn and Thomas's monkeys used one of the three connectives.  Using empirical measures of the relative ease (or difficulty) with which humans learn to apply conjunctive, disjunctive, conditional, and biconditional reasoning in nonverbal concept learning tasks (see below) the conservative interpretation would be to say that Burdyn's and Thomas's monkeys reasoned conjunctively.

It appears, then, that traditional conditional learning experiments with animals are inherently flawed. Are there procedures available to the animal researcher to show whether animals can use conditional reasoning? It will be useful next to consider a nonverbal procedure that was used with humans.

Towards Valid Tests of Animals' Use of Conditionals

Bourne's (1970) article, "Knowing and Using Concepts," provides good examples to make several useful points. Bourne used problems constructed from three features of two dimensions, namely, three forms (circle, square, and triangle) and three colors as illustrated in in the Figure below (based on Bourne's Figure 1).  The figure shows how objects would be partitioned or classified as correct or incorrect choices according to the logical operation that was was used to define the relationship between the two focal attributes, red and square.

It will be useful next to illustrate how the positive and negative examples for the conditional based on red and square map onto the truth table. Note that a fundamental difference between the red-square example and the triangle-same example is that the focal attributes in the red-square problem are manifested in a single discriminandum. Thus, the translations are T and F for presence (+) or absence (-) of red and square and T or F in the third column represent whether the example is T or F according to the contingencies specified by the conditional.

Conditional

    p             q                               p>q

T (+R)     T (+Sq)                         T   (Red-square discriminandum is an example)

T (+R)     F  (-Sq)                          F   (Red but not square discriminanda are not examples)

F  (-R)     T  (+Sq)                         T   (Square but not red discriminanda are examples)

F  (-R      F   (-Sq)                          T   (Not red and not square discriminanda are examples.

The conditional requires that if it is red, then it must be square to be an example, but if it is not red, any discriminandum may be an example, including not-red squares.  If this was all that was done, an obvious alternative solution (see Figure above) to using the conditional rule is that the subject might merely learn by rote the two that are non-examples and/or the seven that are examples among the discriminanda.

Shifting to conceptual categories as attributes (e.g., triangularity rather than square) instead of two specific attributes does not fully eliminate the 2-7 rote-learning solution. For example, we might construct a set of discriminanda using three colors such as black, white, and gray and three form categories such as triangularity, pentagonality, and heptagonality. Let's say we used gray and pentagonality as the focal attributes. The subject only has to learn that gray-triangles and gray-heptagonals are not examples but that all other discriminanda are.  In brief, the subject might learn the 2- 7 division among examples and nonexamples regardless of whether it was based on specific discriminanda or conceptual classes of discriminanda, and it might learn this division independently of learning anything about applying conditional reasoning.

Bourne (1970) recognized this alternative, rote-learning solution.  In fact, he identified four solutions that were alternatives to using conjunctive (Cj), disjunctive (Dj), conditional (Cd), or biconditional (Bd) rules, and he predicted outcomes on tasks such as those illustrated in the above Figure according to these four alternative solutions. For example, generalizing from the abovementioned 2-7 division solution suggests that the order in terms of ease of learning would be that the conjunctive would be easiest with its 1-8 division of the discriminanda, followed by the conditional with its 2-7 division, followed equally by the disjunctive and biconditional with their 4-5 divisions.

Without discussing here the bases of the other three alternative solutions (which may be seen in Bourne, 1970), outcomes, expressed as order of ease in attaining a criterion performance, predicted from them together with Bourne's actual experimental outcome are diagramed below (following Bourne's Table 2); the aforementioned solution based on the division of examples and non-examples is number 3 among Bourne's alternative solutions.  Because the actual outcome differed from those predicted by the four alternative solutions, it may be tempting to suggest that Bourne's subjects were using the rules appropriately. However, Bourne did not make this suggestion, probably because he realized that there may be other, as yet unidentified, alternatives to using the rules.

Alternative Solution 1: Cj = Dj = Cd = Bd

Alternative Solution 2: Cj < Dj = Bd < Cd

Alternative Solution 3: Cj < Cd < Dj = Bd

Alternative Solution 4: Cj = Dj = Cd < Bd

Actual Exp. Outcome: Cj < Dj < Cd < Bd

Subsequently, in the same investigation, Bourne (1970) was able to determine through a series of transfer experiments that his subjects had, in fact, learned the rules. Some of the transfer experiments involved the experimenter and the subjects verbally discussing the applicable rule. It is unlikely that such verbal validation will be available to animal researchers, and it remains to be seen whether animals will show the kind of perfect or near-perfect transfer of training that is required otherwise to confirm that the subject used the rule. By "near-perfect," it is meant that there must be so few mistakes that the subject likely could not have memorized specific discriminanda and reinforcement relationships.

A minimum of four trials is necessary merely to present the minimal information to show which rule is operating, namely, one trial each to manifest each row contingency in a given truth table. After being trained on a succession of problems based on the same rule, Bourne's human subjects learned to use the four informational trials to attain thereafter perfect or near-perfect performances on new problems. Presumably, this could be done only if the subjects had inferred correctly and used the appropriate rule.

Recommendations for Future Research

Based on the foregoing, animal research on conditional reasoning can and must be improved as follows:

(1) Abandon traditional "conditional learning" procedures with their inherent confounding of meaningful alternative discriminanda.

(2) Include conceptual-category discriminanda to preclude rote memorization of discriminanda-reinforcement relationships.

(3) Use responses that allow the subject to affirm or negate exemplars.

(4a) If the investigator wishes to use Bourne's procedures, the relevant attributes must be incorporated into each discriminandum. Presumably, however, the relevant attributes or discriminanda could be physically separated, as most "real life" examples of conditionals involve spatially and temporally separate p and q components.

(4b) Animal experiments based on Bourne's procedure would involve reinforcing an animal's responses that correctly affirmed or negated single discriminanda in accordance with the applicable rule. A series of problems should be administered according to a single rule, until, following the administration of the four mandatory informational trials on new problems, the animal continued with perfect or near-perfect performances, or until it seemed likely that the animal would not be able to attain such performances. If perfect or near-perfect performances were seen on new problems, it should be reasonable to attribute the use of the conditional reasoning to the animal (or conjunctive, etc., depending upon which was being tested).

Standard Logic Versus Natural Logic

This paper would be incomplete without acknowledging that some scholars have tried to reconcile standard logic with what some refer to as natural or mental logic (e.g., Braine, 1978; Braine & O’Brien, 1998; Gentzen, 1964; Miller & Glucksberg, 1988). For example, natural logic is said to apply to cases of reasoning that reflect genuine, "if-then" conditional reasoning, yet which may fail to fulfill the requirements of the standard logic of the conditional.

It is beyond the scope of this article to do little more than acknowledge the issue of natural versus standard logic. Those interested may consult the references cited. However, two points should be made. The first is that although, for example, Braine (1978) and Gentzen (1964) have attempted to develop systematic methods for the analysis and study of the use of natural logic, my consideration of natural versus standard logic has not informed me how the methods associated with natural logic will enable us to design experiments to distinguish whether animals use conjunctive, conditional, or biconditional reasoning. Thus, it appears that the most conservative and justifiable approach is to continue to attempt to confirm animals' use of the logical operations based on methods that embody standard logic.

References

Bourne, L. E., Jr. (1970). Knowing and using concepts. Psychological Review, 77, 546-556.

Braine, M. D. S. (1978). On the relation between the natural logic of reasoning and standard logic. Psychological Review, 85, 1-21.

Braine, M. D. S., & O’Brien, D. P. (Eds.) (1998). Mental logic. Mahwah, NJ: Lawrence Erlbaum Associates.

Burdyn, L. E., Jr., & Thomas, R. K. (1984). Conditional discrimination with conceptual simultaneous and successive cues in the squirrel monkey (Saimiri sciureus). Journal of Comparative Psychology, 98, 405-413.

Carter, D. E., & Werner, T. J. (1978). Complex learning and information processing by pigeons: A critical analysis. Journal of the Experimental Analysis of Behavior, 29, 565-601.

Deacon, T. W. (1988a). Human brain evolution: I. Evolution of language circuits. In H. J. Jerison & I. Jerison (Eds.), Intelligence and evolutionary biology (pp. 363_381). New York: Springer-Verlag.

Deacon, T. W. (1988b). Human brain evolution: II. Embryology and brain allometry. In H. J. Jerison & I. Jerison (Eds.), Intelligence and evolutionary biology (pp. 383-415). New York: Springer_Verlag.

Deacon, T. W. (1989). The neural circuitry underlying primate calls and human language. Human Evolution, 4, 367-401.

Deacon, T. W. (1997). The symbolic species. New York: W.W. Norton, Co.

Dewson, J. H., III (1977). Preliminary evidence of hemispheric symmetry of auditory function in monkeys. In S. Harnad (Ed.), Lateralization in the nervous system (pp. 63-71). New York: Academic Press.

Dewson, J. H., III (1978). Some behavioural effects of removal of superior temporal cortex in the monkey. Recent Advances in Primatology: Volume One. Behavior, 1, 763-768.

French, G. M. (1965). Associative problems. In A. M. Schrier, H. F. Harlow, & F. Stollnitz (Eds.), Behavior of nonhuman primates (Vol. 1, pp. 167-209). New York: Academic Press.

Gentzen, G. (1964). Investigations into logical deduction. American Philosophical Quarterly, 1, 288-306.

Levy, J. (1988). The evolution of human cerebral asymmetry. In H. J. Jerison & I. Jerison (Eds.), Intelligence and evolutionarybiology (pp. 157-173). New York: Springer-Verlag.

Miller, G. A., & Glucksberg, S. (1988). Psycholinguistic aspects of pragmatics and semantics. In R. C. Atkinson, R. J. Hernn hhstein, G. Lindzey & R. D. Luce (Eds.) Stevens' handbook of experimental psychology: Vol. 2. Learning and cognition (pp. 417-472). New York: John Wiley & Sons.

Nissen, H. W. (1953). Sensory patterning versus central organization. The Journal of Psychology, 36, 271-287.

Petrides, M. (1982). Motor conditional associative-learning after selective prefrontal lesions in the monkey. Behavioural Brain Research, 5, 407-413.

Petrides, M. (1987). Conditional learning and the primate frontal cortex. In E. Perecman (Ed.), The frontal lobes revisited (91-108). New York: The IRBN Press.

Riopelle, A. J., & Copelan, E. J. (1954). Conditional reversal to sign. Journal of Experimental Psychology, 48, 143-145.

Thomas, R. K., & Kerr, R. S. (1976). Conditional discrimination in Saimiri sciureus. Animal Learning & Behavior, 4, 265-271.

Zentall, T. R., Hogan, D. E., & Edwards, C. A. (1984). Cognitive factors in conditional learning by pigeons. In H. L. Roitblat, T. G. Bever & H. S. Terrace (Eds.), Animal cognition. (389- 405). Hillsdale, N.J.: Lawrence Erlbaum Associates, Publishers.