Professional Documents
Culture Documents
Sidman, 1982
Sidman, 1982
NUMBER
(JANUARY)
A SEARCH FOR SYMMETRY IN THE CONDITIONAL DISCRIMINATIONS OF RHESUS MONKEYS, BABOONS, AND CHILDREN MURRAY SIDMAN, RICKI RAUZIN, RONALD LAZAR, SHARON CUNNINGHAM, WILLIAM TAILBY, AND PHILIP CARRIGAN
NORTHEASTERN UNIVERSITY
Procedures for generating arbitrary matching-to-sample performances may generate only conditional discriminations. Rational grounds for this distinction are proposed, based on the properties that any equivalence relation must possess. Empirical tests are described for determining whether subjects trained on conditional discriminations are also engaged in true matching to sample. A series of studies then leads to the conclusion that proof of true matching to sample by monkeys, pigeons, or baboons is yet to be provided. Whether the absence of such proof reflects experiential factors or species-defined limitations is not presently clear. Key words: conditional discrimination, matching to sample, stimulus equivalence, stimulus classes, stimulus control, key press, monkeys, baboons, children
A commonly used experimental arrangement presents a subject with two discriminative stimuli simultaneously, perhaps a vertical and a horizontal line, while a third stimulus, perhaps a green or red hue, determines the positive or negative function of each line. Proce-
durally, the relation of the hues to the lines is that of conditionality, or "if ... then . . ." (for example, if green, then vertical; if red, then horizontal). The line discriminations are conditional; hues are the conditional stimuli. Assumptions that extend the relations between stimuli beyond the procedurally defined conditionality reveal themselves when we call This research was supported by Grants MH30063, the behavior generated by the procedure HD05124, HD04147, and NE-G-00-3-0011O Preparation of the report was also supported in part by Grant "matching to sample." An important special RRO7143, Department of Health and Human Services. case is identity matching. The vertical disMuch of the data and systematic background were pre- criminative stimulus (comparison) may be possented in an address to A.P.A. Division 25 in Septem- itive when the conditional stimulus (sample) ber, 1976. Murray Sidman was involved throughout the experiments, which spanned a period of several years, is also vertical, with the discrimination reversand each of the other authors was indispensable to the ing when the sample changes to horizontal. conduct and conceptualization of particular studies. Experimenters often assume that for their subSpecial thanks are owed to Steve Brown, Gary Dubois, jects, too, these stimuli bear a relation to each Wendy Mackay, and Karen Oeschger for technical asbut of samesistance; to F. Garth Fletcher for apparatus design, con- other not only of conditionality struction, and maintenance; and to Betsy Constantine ness, or identity. This need not be so. Carter for suggesting a conceptual link betveen stimulus and Eckerman (1975) have presented data equivalence and the mathematically defined equivalence which suggest that "identity matching" can be relation. We are also grateful to Joseph V. Brady and a misnomer. For pigeons, a red sample and his associates and staff, who provided facilities and a congenial atmosphere for the conduct of Experiment 5 comparison may be as different from each during Murray Sidman's sabbatical year at the Behav- other as a red sample and a vertical compariioral Biology Laboratories, Department of Psychiatry son. For experimenters working with pigeons, and Behavioral Sciences, Johns Hopkins University then, matching to sample may be a case of Medical School; we particularly thank Alan Harris and mistaken identity. Robert Hienz for their generosity with equipment, The distinction between conditional disspace, and time; Frank Grollman for apparatus confor Sheckler Deborah and and matching holds also when all Bowers and Chris crimination struction; their competent day-to-day conduct of the experiment the stimuli are different-to subjects and exand maintenance of the baboons' feeding and health perimenter. By adding the qualifiers, "nonregimen. Reprints may be obtained from Murray Sidman, Department of Psychology, 282 Nightingale Hall, identity," "arbitrary," or "symbolic" to the Northeastern University, Boston, Massachusetts 02115. term, "matching to sample," investigators re23
24
Table 1 veal their assumption that the procedure generates another stimulus relation besides conThe Equivalence Relation ditionality. Although the assumed relation Equivalence Relations must be: cannot be identity, it can be equivalence. 1. Reflexive: aRa Matching to sample, therefore, sometimes re2. Symmetric: If aRb, then bRa fers not simply to the identity relation between 3. Transitive: If aRb and bRc, then aRc a conditional and a correlated positive stimulus, but to the more general equivalence relation. (Identity itself can be shown to be an must possess all three listed properties-reflexivity, symmetry, and transistivity-in order to equivalence relation.) An example may suggest why many of us fulfill the definition of an equivalence relation. The reflexive property requires the relation have overlooked the possibility that the conditional-discrimination procedure may generate R to hold between each stimulus and itself. nothing more than conditional relations, as- If R were the relation "larger than," reflexivsuming instead that the procedure also gen- ity could not be satisfied; a stimulus cannot erates equivalence relations. Suppose the con- be larger than itself. If R is the conditional ditional stimuli, hues, control discriminations relation, reflexivity requires "if a, then a," "if not just among simple tilted lines but among b, then b," etc. to hold true. Successful idenlines of greater complexity-the printed words tity matching is a necessary consequence of red, green, and their like. By calling a subject's reflexivity, which therefore conveys the notion performance "matching," we imply that each of sameness. The identity-matching procedure, however, color and its correlated word comprise a class not definitively test reflexivity unless we does of equivalent stimuli, permitting us to say, have correctly identified the stimuli that conor colors," for the "The words are symbols "The colors are the meanings of the words," trol the subject's behavior. If, for example, and to include such tasks in tests of simple control by the lines depended on their locareading comprehension. Equivalence is a tion on the left or the right key (given the tempting assumption, transforming a condi- usual arrangement in which the conditional tional discrimination into a semantic process stimulus is on the center key, and the others and supporting the notion that arbitrary are on the outer keys), the subject might not matching to sample is a linguistic perfor- match vertical to vertical, and horizontal to mance. If the assumption were valid, it would horizontal, but rather, vertical/center to verpermit us to adopt conditional discrimination tical/left and vertical/right, and horizontal/ as a procedural model for studying aspects of center to horizontal/left and horizontal/right. The experimenter's presumption of identical language even in nonhuman animals. stimuli would then be incorrect, and a perforthat looked like identity matching mance The Equivalence Relation would not be; to conclude that one was testWe cannot tell by looking at a subject's con- ing the relation R for reflexivity would be inditional-discrimination performance whether valid. A definitive indicator of true identity or not it involves true matching to sample. matching, and therefore of reflexivity, is genAlthough conditionality is directly inferable eralized identity matching. Successful generfrom current behavior, additional tests are nec- alized matching confirms reflexivity for each essary to validate an inference of equivalence. relation that is tested. Symmetry requires the relation R to hold Suppose conditional stimulus a controls a between two different stimuli. bidirectionally The which b prois positive. in discrimination cedurally defined relation R between stimuli If sample a bears a conditional relation to coma and b is the conditional relation aRb. If aRb parison b (aRb), the relation must also hold also satisfies the properties of an equivalence when b is the sample and a is the comparison relation, the stimuli will be members of a class (bRa); symmetry requires sample and compariof equivalent elements, and we can call the son to be interchangeable. A subject who has subject's performance "matching to sample." learned to select vertical when the sample is Table 1 summarizes the mathematical defini- green should, given the opportunity, select tion of the equivalence relation. A relation R green when the sample is vertical.
25
strate to our satisfaction that monkeys' conditional discriminations can properly be called matching to sample. The major purpose of this paper is to describe some experiments that failed to reveal equivalence relations in the
conditional-discrimination performance of rhesus monkeys-a failure that was confirmed with baboons. The experiments to be described represented our first approach to this problem; we asked whether the animals' performances satisfied the symmetric property of an equivalence relation. In the course of these experiments, we also obtained some data on transitivity. Although Hogan and Zentall (1977). Holmes (1979), and Rodewald (1974) have reported failures to find symmetric relations between the stimuli in pigeons' conditional discriminations, several variables that might have contributed to the absence of symmetry still require investigation. The present experiments, in addition to testing for symmetry with other species of subjects, will provide some of the information needed for a more definitive judgment. GENERAL METHODS The general subject, apparatus, and procedure specifications will apply to all experiments unless specific exceptions are noted.
Subjects In all but two experiments the subjects were adult rhesus monkeys, Macaca mulatta. Before becoming experimental subjects, and later, when nearing the end of "vacations" from experimental regimens, the animals were fed, at fixed times, enough to maintain their weights as close as possible to free-feeding levels. With any remaining food being removed one hour later, the animals learned to take their full ration. An animal that did not obtain its full ration in an experimental session received the remainder one-half hour after returning to its home cage. Up to Experiment 4, the monkeys underwent five experimental sessions, each no more than two hours long, in successive twoweek periods; one week, sessions took place on Monday, Wednesday, and Friday, and the monkey received its daily ration in the home cage on Sunday; the next week, sessions took place on Tuesday and Thursday, with a home-cage feeding on Saturday. During Experiment 4 the animals had a 1.5-hr session every weekday.
26
Both schedules maintained sufficient deprivation to ensure several hundred trials per session and at the same time kept the animals' weights close to free-feeding levels. In the event of excessive pausing or high error rates late in training sessions, the daily ration was reduced slightly until the performance became more stable. Apparatus The animals' experimental environment, a sound-attenuating, fan-ventilated chamber, was in the same room as the programming and recording equipment. The portion of the chamber that held the monkey was 63 cm high, 61 cm wide, and 59 cm deep. One stainless-steel inner wall, opposite the access door, contained a five-key touch-sensitive stimulus-response matrix, a trial-initiation key that operated a limit switch, two pellet dispenser trays, and a houselight. Figure 1 diagrams the keys and feeder tray. The trial-initiation key, 3.2 cm in diameter, was mounted on the vertical midline, its center 16 cm above the grating floor. The five stimulus-response keys were each 3.0 cm in diameter. Four were located at the corners of a 9-cm square (measured from the centers of the keys), with the fifth in the square's center. The matrix, on a circular black Plexiglas surround, was mounted with the center of the middle key at the vertical midline, 35.5 cm above the floor. Pellet trays on each side of the matrix, 22 cm above the floor and 19 cm to the
left and right of the vertical midline, received Noyes l-g banana pellets. A 60-W Sylvania Lumiline houselight, protected by an aluminum grill, was mounted above the key matrix, its top 55 cm above the floor. An IEE projector (Grason-Stadler #A5092A, Stimulus Pattern 156) behind each key of the matrix provided the stimuli, red or green patches filling the key, or a vertical or horizontal white line, .3 by 1.8 cm, on a black background. The trial-initiation key could be illuminated from the rear by two clear 7.5-W bulbs. A punched-tape reader programmed sequences of stimuli and trial types; solid-state equipment programmed the procedures. Events of interest were recorded on electromechanical and electronic counters, with temporal relations and other details being preserved by an Esterline 20-pen event recorder.
03
Fig. 1. Diagram (not to scale) of keys and feeder trays in the monkey chamber. The trial-initiation key (TI) was located below the matrix of stimulus-response keys (1-5). Sample stimuli always appeared on Key 3, and comparison stirnuli on Keys 1, 2, 4, and 5 (two per trial). If the correct stimulus was on Key 1 or 4, the pellet was delivered to the left tray (FL); if the correct stimulus was on Key 2 or 5, the pellet was delivered to the right tray (FR).
Procedures Five minutes after an animal entered the experimental chamber the houselight signaled the start of a session and remained on until the session ended. The animal stayed in the darkened chamber for an additional five minutes before being taken back to its home cage. Preliminary training. Animals first received magazine training, with pellets delivered unpredictably to either the left or the right tray. An autoshaping procedure (Brown & Jenkins, 1968; Sidman & Fletcher, 1968) then taught them to press the keys. All stimuli to be used later appeared on each key in a mixed sequence, one stimulus and one key per trial. If the stimulus appeared on the center key, the pellet was delivered alternately to the left or the right tray; on other trials the pellet dropped into the tray on the same side as the lit key. Once the animal reliably pressed the lit key, and only the lit key, within eight seconds, autoshaping contingencies ceased. Each trial in the next stage began with a stimulus on the center key. By pressing the center key, the animal produced a stimulus on one of the four outer keys of the matrix but no pellet. The center key stayed on. Pressing the outer key then produced a pellet and a five-second intertrial interval. Only one comparison key had a stimulus on it, the stimulus that was to be correct for the sample during conditional-discrimination training. In this and all subsequent training and testing procedures, pressing a dark key during a trial had
27
activate
recorder
the animal to press the illuminated trial-initiation key to produce a sample; when the sample came on, the trialinitiation key became dark. Pressing the center key then produced a comparison, and pressing that key produced a pellet and the intertrial interval. The animals then learned to press the trial-initiation key five times to produce a sample and thereby initiate a trial. The trial-initiation key permitted animals to pace their own trials. The final step introduced one incorrect stimulus along with the correct comparison after the animal pressed the sample key. This marked the start of conditional-discrimination training. The sample remained on Key 3 (Figure 1). Comparisons, only two per trial, appeared on outer Keys 1, 2, 4, and 5; on any conditional-discrimination trial, two outer keys had stimuli projected on them and the other two were dark. Each combination of stimuli and keys occurred equally often. With only two kinds of trials, as in the green/red identity-matching procedure, no more than three of the same type could occur consecutively. With more than two kinds of trials, all sample-comparison combinations had to occur before any could repeat. Also, all four comparison keys had to be scheduled as correct before any could be correct again. With those exceptions, all trial types and correct keys were equally probable on successive trials. Tape loops programmed sequences of 192 to 640 trials, depending on the number of trial types. Each session began with the trial that would have come up next at the end of an animal's previous session. Session durations and trials per session varied from animal to animal and in different stages of the experiments. Fewer trials had to be scheduled in sessions when a food pellet followed each correct trial than when leaner reinforcement schedules prevailed (see below). When an animal pressed the correct comparison key a pellet dropped into the tray on the same side of the matrix as the correct key, all keys became dark, and the intertrial interval began. Except for the pellet, the same consequences followed an incorrect response. The procedure was noncorrection; errors did not cause trials to be repeated. Intertrial intervals lasted at least five seconds, but could not
terminate sooner than five seconds after the animal pressed any dark key. Conditional-discrimination training. Monkeys R46 and R47 first had line-line training (line sample and line comparisons). The animal received a pellet if it pressed the vertical comparison when the sample was also vertical, and the horizontal comparison when the sample was horizontal. Because Monkeys R46 and R47 had great difficulty learning the line-line task, they were taught to respond differentially to each sample line. Five responses (FR) were required to vertical samples, and a pair of responses spaced two seconds apart (DRL) was required to horizontal samples before the comparison stimuli would be presented. The animals' accuracy quickly rose above 90%, confirming the facilitative effect of sample schedules reported by Cohen, Looney, Brady, and Aucella (1976), who used pigeons. Both monkeys then learned hue-hue (hue sample and hue comparisons) without differential sample schedules, pressing the green comparison when the sample was also green, and red when the sample was red. Before starting the present experiments, Monkey R44 had learned successive reversals of green/red and vertical/horizontal simultaneous discriminations. Perhaps because of its history, Monkey R44 learned the line-line task without differential sample schedules. All animals finally had sessions with line-line and hue-hue trials mixed. Before starting the experiments to be reported here, they underwent tests with simultaneous- and delayedmatching procedures in which sample schedules, where used, were removed and reinstated. Symmetry testing. In symmetry tests, new trial types were inserted as probes into a baseline of previously learned conditional discriminations, with all types of trials occurring equally often. Some experiments provided reinforcement on all correct trials, baseline and probes. Others provided reinforcement only on correct baseline trials, never on probes. In preparation for these extinction tests, before probes were actually inserted, the reinforcement probability on correct baseline trials was gradually reduced from 1.00 to .2. This reduction usually took place in steps of .1, and at least 90% of the trials had to be correct before a change. Occasional decreasing accuracy trends were halted by increasing the reinforcement probability and then using a stricter
28
criterion while bringing the probability down again. At the final level, .2, at least 90% of each baseline trial type had to be correct in each successive 25-trial block during two consecutive sessions. Then, since no probes would be reinforced, the reinforcement probability on baseline trials during the test was increased sufficiently to maintain the overall probability at .2. The output of a probability gate (BRS/ LVE PP-201) was modified to permit no more than nine consecutive correct baseline trials to go unreinforced.
EXPERIMENT 1 In asking whether monkeys' conditional discriminations satisfied the symmetric property of an equivalence relation, we at first simply added a few control features to the basic procedures used in the pigeon experiments (Hogan & Zentall, 1977; Holmes, 1979; Rodewald, 1974). Table 2 summarizes the conditional discriminations. The stimuli in each numbered line-a sample and two comparisons-define a trial type. Each pair of trial types, numbered 1 to 4, is a conditional discrimination with two components, A and B. The line-hue trial types, 3A and 3B, were of particular concern. Having learned this conditional discrimination, was an animal also doing arbitrary matching to sample? Was the relation between each line sample and its correct hue comparison just "if-then," or was it also equivalence? Had vertical and green be-
discriminations.
Although test procedures differed for each monkey (see below), the general training and testing sequence went as follows: animals first had line-line and hue-hue training, trial types IA, lB, 2A, and 2B. This gave them experience with lines and hues as samples and comparisons. They then had line-hue training; 3A and 3B were interspersed among the four previously learned trial types. With vertical samples, animals learned to select green, and with horizontal samples, to select red. Finally, with all six trial types mixed together as a baseline,
ble 2 Conditional discriminations and sample schedules (where used) in Experiments 1, 3, and 5. Where neither FR nor DRL is indicated, a single sample response sufficed to produce the comparisons. (Monkey R44 never had sample schedules, and in Experiment 3, only Subjects J. G. and A. B. had sample schedules.) Trial Types IA, IB (line-line), 2A, and 2B (hue-hue) were to provide discriminative prerequisites for the symmetry test. The starred Trial Types 3A and 3B (line-hue) constituted the critical conditional discrimination that was to be tested for symmetry. Trial Types 4A and 4B (hue-line) constituted the symmetry test. Symmetrytest sessions included all eight trial types.
VERTICAL HORIZONTAL
GREEN RED VERTICAL HORIZONTAL GREEN RED
FR DRL
FR DRL
29
Training and Testing Procedures Sample schedules. Table 2 also shows sample schedules, which, as noted earlier, were introduced because of difficulty in training Monkeys R46 and R47 on the line-line conditional
100-
R-46 .-,---
R-44
_---
90-
80 u w
70-
O
.--
6050-
z
w
. . t .
E!
4030-
20*
10VIH G
GR
V'H
0-
qGRjG
GR
VGV
VH VH GR G -R GR GR VH V
|
I-BASELINE-ITEST-I
BASE
LINE-kESTA
Fig. 2. Performances by Monkeys R46 and R44 during the symmetry-probe session of Experiment 1. Bars depict the percentage of correct responises for each trial type. Stimuli for each trial type are identified at the bottom of the bars, and a line connects each sample to its correct comparison. The bars, from left to right, correspond to trial types that are listed, from top to bottom, in Table 2. Asterisks denote the line-hue trial types tested for symmetry, and the two righthand bars for each animal depict the hue-line symnmetry probes. Probes were unreinforced. Each bar represents approximately 95 trials for Monkey R46, and 65 trials for Monkey R44. Only Monkey R46 had line-sample schedules.
discrimination (IA and 1 B). This difficulty persisted for conditional discrimination 3, in which lines again were samples, so schedules were used also in teaching them the line-hue task. Cohen, Brady, and Lowry (1981) and Urcuioli and Honig (1980) have suggested that schedule-generated behavior may actually take over the sample function from stimuli projected on the keys, in a manner reminiscent of coding responses (Lawrence, 1963; Schoenfeld & Cumming, 1963). Before Monkey R47's first symmetry test, therefore, attempts were made to remove the schedule requirements-by turning on the comparisons as soon as the animal pressed the sample key just once. Monkey R47 was unable to maintain both the line-line and the line-hue performances without schedules, but with schedules for line-line only, the linehue performance eventually met criterion. After the schedules had been used during original acquisition, therefore, they were maintained as shown in Table 2 for Monkey R46 only; Monkey R47 kept the schedules in trial types IA and 1B, but not in 3A or 3B. Monkey R44 never had differential sample schedules. In the course of attempts to remove the sample schedules, Monkey R47's comparison keys were permanently reduced from four to two. Samples still appeared on Key 3, but comparisons only on Keys 4 and 5 (Figure 1); Keys 1 and 2 remained dark and without function. Symmetry tests. After Monkeys R44 and R46 had learned the line-hue conditional discrimination, the reinforcement probability for all six trial types was reduced. Trial types 4A and 4B were then inserted into the sparsely reinforced baseline as unreinforced probes for symmetry. Monkey R47 had two symmetry tests, both of which provided reinforcement on all correct trials, baseline and probe. On probe trials, responses indicative of symmetry were arbitrarily designated as correct.
RESULTS AND DISCUSSION Figure 2 contains the probe-session results for Monkeys R44 and R46. Individual scores for each trial type make it possible to detect patterns of stimulus control that the customary accuracy measure might obscure (Sidman, 1980). Since each nonoverlapping pair of bars represents a conditional discrimination, averaging the pair would yield the usual accuracy
30
score. The first three pairs of bars for each somewhat diminished by the poorly mainsubject show baseline performances within the tained baseline, but insofar as the baseline test session, and the final two bars the sym- deterioration can be attributed to the probemetry probes. trial reinforcement, the test does bolster the All baselines were above 90%, correct. Line- evidence against symmetry. Hogan and Zentall line performances (the first two bars for each (1977) used a similar rationale for the design animal) were nearly perfect, as were hue-hue of some of their symmetry tests. In line-hue trial types 3A and 3B the ani(the second pair of bars). Bars with asterisks show the maintenance of high line-hue scores mals had apparently learned only a condi-vertical samples matched to green compari- tional relation between each sample and its correct comparison and gave no evidence for sons, and horizontal samples to red comparithe formation of equivalence relations. They sons. The final two bars for each animal represent the hue-line probes for symmetry. were unable to match the stimuli when the (Probe-trial responses indicative of symmetry original samples became comparisons, and the were arbitrarily designated as correct. Subseoriginal comparisons became samples, and proOne practice.) this follow will figures quent vided no justification for calling their perforanimal slightly preferred the vertical compari- mance matching to sample. son, and the other the horizontal; average scores on symmetry trials were 55 and 52% for EXPERIMENT 2 Monkeys R46 and R44, respectively. The aniIs it sufficient to say that the conditional remals did not reverse the sample-comparison roles of lines and hues; they gave no evidence lation is, "If the sample is vertical, then green is the correct comparison," or must one also of symmetry. Monkey R47's first hue-line probes averaged specify the incorrect comparison? Would it only 56% correct, but line-hue trial type 3A be more correct to say that the conditional in the baseline also dropped to 76%. After an- relation is, "If the sample is vertical and the other criterion session of baseline trials alone, comparisons are green and red, then green is summarized in the left side of Figure 3, the the correct comparison"? Farthing and Opuda (1974) presented pianimal received a second test. The right side of Figure 3 shows an average of only 49% cor- geon data indicating that the incorrect comparison is irrelevant to the relation between rect on the probes. Correct probe trials did not become more frequent as the sessions pro- sample and correct comparison. Nevertheless, gressed. Again, there was baseline disruption, the problem requires specific resolution in the most severe this time in the hue-hue trials. present context. If a complete definition of Even in its second session with reinforcement the conditional relation had to include incorrect stimuli, then a valid symmetry test would on correct probe trials, this animal performed no better on symmetry probes than did others require sample and correct comparison to be interchangeable only when the original incorin their extinction tests. Probing in extinction, however, did not af- rect comparison was also present. Experiment fect the other animals' baselines, even though 1, however, not only interchanged samples and they received only infrequent reinforcement. correct comparisons but also changed the inThe left side of Figure 3 shows that all trial correct comparisons from hues to lines (Tatypes were above 90% during the session be- ble 2). In Experiment 2, therefore, the symfore the probes; this typical stability indicates metry test substituted samples and correct that the baseline deterioration during the next comparisons for each other, whereas incorrect session came about because of the reinforce- comparisons remained unchanged. Table 3 summarizes the conditional discrimment on correct probe trials and was not a paradoxical effect of reinforcement on the base- inations. Trial types 1 and 2 were the same as line trials. It seems reasonable to assume that in Experiment 1. Asterisks mark the relations reinforcement on probe trials would not have to be tested for symmetry: in 3A, the sample disrupted the baseline if the conditional rela- was vertical and the correct comparison green, tions being tested were symmetric. The sig- as in Experiment 1, but the incorrect comparinificance of the probe performance itself is son was now horizontal instead of red; in 4B
31
R-47
100908070w
0 0
w
0
H50-
cr4030-
2010V HG R VH
V HG R VH G R GGR GRVH VH VH VHGu%UR
0-
VHVHGRGRGRGR
-.
I
line
BASELINE
I I- BASELINE- -TEST-I
Fig. 3. Monkey R47's performances during Experiment 1's symmetry test (right side) and the preceding basesessioni (Icft side). Bars depict the percenitage of correct responises for each trial type (corresponding to 'Table 2). Stiitiuli for each trial type are identified at the bottomn of the bars, withl a line connectinig eachl samiiple to its correct comiiparison. Asterisks deniote the line-hlue trial types tested for syllulnetry, aind the twvo righthand(I bars depict the hlue-linle symmiiiietry probes. All correct trials, including probes, were reinforced. Each bar represents 35 to 40 trials. Only line-line trials had differential sample schedules.
32
Table 3 Conditional discriminations in Experiment 2. Trial Types IA, lB, 2A, and 2B are the same as in Experiment 1. The starred Trial Types 3A and 4B maintain the same line samples and correct hue comparisons as the starred Trial Types in Table 2, but now the incorrect comparisons are lines instead of hues. Trial Types 3B and 4A were included in order to maintain conditionality in the line-hue relations. Trial Types 5A and 5B, the symmetry test, now maintain the same incorrect comparisons as Trial Types 3A and 4B. Symmetry test sessions included all 10 trial types.
SAMPLE
These, also, will not be reported here. The baseline for Experiment 2 was then reinstated in preparation for the test to be reported next.
Procedures The sequence of training and test procedures was similar to Experiment 1. After lineline and hue-hue training, trial types 3A, 3B, 4A, and 4B, with mixed line and hue comparisons, were interspersed among lA, lB, 2A, and 2B. Symmetry probes (5A and 5B) were introduced into this baseline of eight different trial types, with reinforcement on all correct trials.
RESULTS AND DISCUSSION
IA VERTICAL lB HORIZONTAL 2A GREEN 2B RED *3A VERTICAL 3B HORIZONTAL 4A VERTICAL *4B HORIZONTAL 5A GREEN 5B RED
VERTICAL HORIZONTAL GREEN RED GREEN HORIZONTAL VERTICAL RED VERTICAL HORIZONTAL
HORIZONTAL VERTICAL RED GREEN HORIZONTAL GREEN RED VERTICAL HORIZONTAL VERTICAL
the sample was horizontal and the correct comparison red, but the incorrect comparison was now vertical instead of green. The symmetry test, 5A and 5B, again had former correct comparisons, green and red, as samples, and former samples, vertical and horizontal, as correct comparisons. The incorrect comparisons, horizontal and vertical, remained unchanged from trial types 3A and 4B. What was the function of trial types 3B and 4A? Without 3B, animals might have learned to respond correctly on 3A even while ignoring the sample simply by selecting green whenever a green and a line comparison were present together. Inclusion of 3B required control by the sample if the animals were consistently to choose correctly between green and horizontal. Similarly, trial type 4A made control by the sample necessary for correct choices between red and vertical.
METHOD
The right side of Figure 4 presents percentages of correct trials of each type during the probe session. Again, asterisks denote baseline trial types that were tested for symmetry. The animal performed no better on the symmetry probes than it did during Experiment 1. (These results were also consistent with the probes in the unreported delayed-matching tests.) Reinforcement on correct probe trials, however, disrupted the baseline. Each line-line trial type dropped to 89%, trial type 3B (HHG) dropped to 59%0, and 4A (V-VR) to 83%,. Although 3A and 4B remained high, the decreases in 3B and 4A indicated considerable change in conditional discriminations 3 and 4 during the test. Probing in extinction had never affected this animal's baseline. Because of the disruption, the left side of Figure 4 shows the preceding baseline session. As with Monkey R47 in Experiment 1, the baseline disruption appeared to be a consequence of reinforcing the symmetry probes and provided additional evidence against symmetry. Baseline performances most severely affected by the reinforcement on correct probe trials involved identical sample and correct comparison lines, but failures to replicate this relation obscured
Subjects Monkey R44 had its first symmetry test in this experiment. The probes, carried out in extinction, will not be described here. Its second test, also in extinction, was reported in Experiment 1. Then came two delayed-matching tests (zero delay), the first in extinction and the next with reinforcement on all correct trials.
its significance. Even with incorrect comparisons held constant, a sample and its correct comparison were not substitutable for each other. The revised test was no more successful than the test in Experiment 1 in demonstrating symmetry. The conditional discriminations did not generate classes of equivalent stimuli, and the term matching to sample did not appear applicable.
33
R-44
10090 80-
-0
w 600 - 50z
t w
40302010v VH
/
0-
G R v H V l GR GR GH GH VR
N VR
H
v H G
VH VH
R GR
V
I
GH
H v H G R GH VR VR VH VH
BASELINE
I I -BASELINE
I-TEST-
Fig. 4. Performances by Monkey R44 during Experiment 2's symmetry test (right side) and the preceding baseline session (left side). Bars depict the percentage of correct responses for each trial type (corresponding to Table 3). Stimuli for each trial type are identified at the bottom ot the bars, and a line connects each sample to its correct comparison. Asterisks denote the line-hue trial types (with lines as incorrect comparisons) tested for symmiietry, and the two bars at the extreme right depict the hue-line symmetry probes. All correct trials, including probes, were reinforced. Each bar represents approximately 35 trials.
EXPERIMENT 3 A major justification for stressing the importance of symmetry has been the demonstration that the conditional-discrimination procedure does establish equivalence relations when the subjects are retarded teenagers (Sidman, 1971; Sidman & Cresson, 1973; Sidman, Cresson, &c Willson-Morris, 1974; Spradlin, Cotter, & Bax-
ley, 1973), normal children (Lazar & Kotlarchyk, Note 2; Sidman & Tailby, 1982), and normal adults (Lazar, 1977). The human subjects, however, did not learn the same conditional discriminations as the monkeys, nor were the methods of testing for equivalence relations the same. Is it possible that the techniques used with monkeys would have failed to produce symmetry even with humans? Ex-
34
periment 3 attempted to answer this question by repeating Experiment 1 reasonably closely with children.
METHOD
Subjects Six normal children, five boys and one girl, participated. Their ages (years-months) at the time of their symmetry tests were: 4-8 (C.H.), 4-8 (J.G.), 4-11 (P.L), 5-2 (A.B.), 5-2 (J.L.), and 5-9 (K.O.). Each child, recruited by newspaper advertisements, came several days a week with a parent who was paid after each session. The number of sessions, each 15 to 30 minutes long, varied because the children required different amounts of baseline training.
Apparatus and General Procedures The child sat before a stimulus-response matrix of nine translucent keys onto which stimuli were projected from the rear. The keys, each 7.3 cm in diameter, were arranged in a circle of eight, with the ninth in the center (Figure 5). The display diameter was 32.7 cm; the center-to-center distance was 9.5 cm between adjacent keys on the perimeter, and 12.7 cm between the center key and each of the others. Sample stimuli always appeared in the center, and comparison stimuli for a given trial appeared only on two of the four outer keys that formed the corners of a square (numbered 1, 2, 4, and 5 in Figure 5). A Kodak Carousel projector presented the stimuli, red or green patches filling the key, or a vertical or horizontal black line, .2 by 7.3 cm, on a white background. Solenoid-operated shutters
switch signaled the programming and recording equipment. A response was not defined until the key had been released for .2 sec; this permitted stimultaneous pressure on more than one key to be defined as incorrect. Unless a reduced reinforcement probability or an extinction probe precluded reinforcement, correct trials were followed by the sound of chimes, the delivery of a penny into a receptacle below and to the left of the key matrix, and a 1.5-sec intertrial interval. The children kept all pennies they received. Only the intertrial interval followed incorrect responses. Training and Testing Procedures Training. Three children (C.H., P.L., and K.O.) had already taken part in an experiment on delayed hue-hue matching and needed no preliminary training for the present study. With the three experimentally naive subjects (J.L., A.B., and J.G.), the delivery of three pennies sufficed to accomplish magazine training. The experimenter then provided nonvocal instruction by pressing the keys appropriately for several trials in which a sample and only a correct comparison stimulus (see below) appeared. Each trial began with a stimulus on the center key. Pressing the center key produced a stimulus on one of the four outer keys while the center-key stimulus remained on. Pressing the outer key then produced chimes, penny, and a 1.5-sec intertrial interval. The experimenter then ceased all demonstration and the children themselves pressed the keys, with instructions provided only by the reinforcement contingencies. Finally, both an incorrect and a correct comparison came on when the subject pressed the sample key. This marked the start of conditional-discrimination training, which followed the pattern of Experiment 1, as summarized in Table 2. Three children (P.L., A.B., and J.G.) learned hue-hue first, and the others learned line-line first. All four trial types (1A, lB, 2A, and 2B) were then mixed together. The children then learned the line-hue task that was to be tested for symmetry. Subject J.G. experienced great difficulty with the line-line conditional discrimination, and Subject A.B. learned line-line but then seemed unable to master line-hue. The sample sched-
000
4
r.
Fig. 5. Diagram (not to scale) of the key matrix in Experiment 3. Samples always appeared on Key 3, and comparisons on Keys 1, 2, 4, and 5 (two per trial). The other keys were never used.
35
P.L.
8060)
4020-
cr
0-
RVHGIR |VIHIGNE
I-BASELINE-I-TEST-I J.L.
IlII 4
I VIHIGINREV-HGIRI
I-BASELINE-1-TEST-1
A.B.
1002
ir80a.
V HG RV
HG
en in1|TSIBSLNE VH G R V H G R
V
iiietry tests in Experiment 3. Bars depict the percentage of correct responses for each trial type (corresponding to Table 2). Stimuli for each trial type are identified at the bottom of the bars, with a line connecting each sample to its correct comparison. Asterisks denote the linehue trial types tested for symmetry, and the two righthand bars for each subject depict the hue-line symmetry probes. Symmetry probes were unreinforced. These subjects had no differential sample schedules during test sessions.
reverse the sample and comparison roles of lines and hues. Her average score of only 17% on probe trials, considerably lower than the 50% "chance" level, might have suggested that the training actually generated strong "antisymmetry." The score, however, was deceptive; on one-third of the probe trials Subject K.O. did not select the antisymmetric comparison, but instead, pressed both comparisons simul-
taneously. Figure 7-B summarizes Subject K.O.'s second symmetry test, with the zero-delay matching procedure. During probe trials, which this time averaged 50% correct, she strongly preferred the vertical comparison. Although Hogan and Zentall (1977) and Gray (1966), who tested pigeons with zero delay, adduced statistical evidence for symmetry, their conclusion was not confirmed by Subject K.O. (or by the data, not presented, of Monkey R44). The zero-delay procedure did not suffice to transform the conditional relation into an equivalence relation. In Subject J.G.'s first probe session (Figure 8-A), he averaged only 56% on probe trials. In his second test (Figure 8-B), after the sample schedules had been removed, his probetrial performance did not improve, but because of his decreased accuracy (78%) on line-hue baseline trials, he was retrained and tested once more. The third test (Figure 8-C)
36
K.O.
100 806040c., 0
z
J.G.
100-
A
*
VIH|GIR
VHVHGRIGR
*l
200100-
V/HVH
V HIV'iIcRIGIGRI GR
G-Fl
B
80 6040200o0-
GR GR
G RVH
I-TEST-I
1--BASELINE--|1-TE ST-I
ix:
0
0
|-BASELINE
B
*
w-
o
w
8060200V
80-
40-
HIG RR
HIG 7R
6040200 -
VH VHjGR GR
VH GR GR GR GR
BASELINE-I
G R
I\I GHIVH
a.
TEST-I
during
sym-
-TEST-I
C
100-
metry tests in Experiment 3. Bars depict the percentage of correct responses for each trial type (corresponding to Table 2). Stimuli for each trial type are identified at
the bottom of the bars with a line connecting each sample to its correct comparison. Asterisks denote the linehue trial types tested for symmetry, and the two righthand bars in each panel depict the hue-line symmetry probes. Symmetry probes were unreinforced.
806040-
200
V HIG IR
I
VH
VW|R GR dR
V
H GR
~'H
IV'H
adequate baseline but no improvement in the probes. With or without sample schedules, Subject J.G. gave no evidence of symmetry. Unlike the three monkeys, four children showed symmetry in the relation between lines and hues. To the extent that Experiment 3 replicated the essential features of Experiment 1, one might conclude that the absence of symmetry in the monkeys' performances was not attributable to some procedural artifact. Two children, however, failed to show symmetry under conditions like those of Experiment 1. The possibility remained, therefore, that some aspect of the procedure was still responsible for the absence of symmetry in the performances of Subjects K.O. and J.G., and the monkeys. A reappraisal of the training and test procedures, and their rationale, led to changes in the baseline conditional discriminations, and the two children who had been exceptions joined the other four in yielding positive results in subsequent symmetry tests. Although the absence of controls for repeated testing clouded the relevance of these changes
a more
shows
BASELINE
Fig. 8. Performances by Subject J.G. during symmetry tests in Experiment 3. Bars depict the percentage of correct responses for each trial type (corresponding to Table 2). Stimuli for each trial type are identified at the bottom of the bars, with a line connecting each sample to its correct comparison. Asterisks denote the line-hue trial types tested for symmetry, and the two righthand bars in each panel depict the hue-line symmetry probes. Line-sample schedules were required only in Panel A.
I-TEST-I
(Spradlin, Cotter, & Baxley, 1973), the mere possibility that the new baselines might have been responsible for symmetry mandated similar training and test procedures for the monkeys.
EXPERIMENT 4 Among the prerequisites for the symmetry test were a successive discrimination between green and red samples, and simultaneous discriminations between vertical and horizontal comparisons. Experiment 3 brought about a reevaluation of the assumption that hue-hue and line-line training provided subjects with
this prerequisite repertoire.
37
Consider first the successive green-red discrimination (4A and 4B, Figure 2). Probe trials presented either a green or a red sample, and the subject never viewed them together-the hallmark of a successive discrimination. Like the probes, trial types 2A and 2B also presented a separate green or red sample. Unlike the probes, once the comparisons had come on, green and red stimuli were present simultaneously. Although one was a sample and the other a comparison, a subject could view them together. At probe-trial choice points, either a green or a red stimulus was present, but not both; at hue-hue choice points, the subject could compare green with red directly. For this reason, the hue-hue task may not have required the subject to learn the particular successive discrimination that the probes demanded. To help ensure that the monkeys learned a truly successive green-red discrimination in preparation for the symmetry probes, Experiment 4 changed trial types 2A and 2B from hue-hue to hue-form. Samples remained green and red, but the comparisons were now X and + (Table 4). Because of the change in the comparisons, green and red were no longer present together. Subjects now had to discriminate temporally separate instances of green and red stimuli, the same kind of discrimination that the probe trials required. Probe trials also required simultaneous discriminations between vertical and horizontal comparisons. Although trial types IA and lB (Table 2) were supposed to teach these prerequisites, the discriminations in the line-line trials did not have to be the same as those in the probe trials. The probes required discriminations between lines that were located only on comparison keys, but a subject could learn the line-line performance by discriminating between the line on the sample key and the line on a comparison key. This discrimination would not be available on probe trials, and the subject might fail to show symmetry simply because it had never learned the necessary
Table 4 Conditional discriminations in Experiment 4. Trial Types IA and lB, with X and + as samples, were to establish the simultaneous comparison-line discriminations, and Trial Types 2A and 2B, with X and + as comparisons, were to establish the successive sanmple-hue discriminations required for the symnmetry test. The starred Trial Types 3A and 3B (line-hue) constituted the critical conditional discrimination that was to be tested for symmetry. Trial Types 4A and 4B constituted the symmetry test. Test sessions included all eight trial types.
SAMPLE
IA x lB + 2.A GREEN 2B RED #3A VERTICAL *3B HORIZONTAL 4A GREEN 4B RED
COMPARISONS INCORRECT CORRECT
NTERTICAL
HORIZONTAL x + GREEN RED VERTICAL HORIZONTAL
HORIZONTAL
V'ERTICAL
+
x RED GREEN
HORIZONTAL VERTICAL
simultaneous line discriminations that were required during the probe trials; they had to discriminate between lines that were located only on comparison keys.
METHOD
Subjects Monkeys R44 and R46 were subjects. Although they underwent different training procedures, they yielded similar probe results, and only Monkey R46's experiment will be described.
Procedures The animal started with the hue-form conditional discrimination (Table 4, 2A and 2B) and learned it rapidly, but then had enormous difficulty with form-line (IA and 1B). Several procedural changes were instituted, both singly and in combination: trial types IA and lB were scheduled both with and without other baseline trial types; the sample was permitted to remain present for as long as two seconds after the animal had selected a comparison; and nine responses to the X and + samples were required before the comparison stimuli could appear. None of these had any clear beneficial effect, and all were terminated. A "yes-no" procedure, which Constantine (Note 1) had already analyzed extensively, was then used to teach Monkey R46 the successive discrimination between X and + samples.
discriminative repertoire. To give the monkeys discrimination training more like that required for line comparisons in the probes, Experiment 4 changed trial types IA and lB from line-line to form-line. The comparisons remained vertical and horizontal lines, but the samples were X and + (Table 4). Subjects now had to learn the same
38
In the "yes-no" procedure, trials began as usual with a sample presented on Key 3, but when the animal pressed the sample only one comparison stimulus appeared on Key 4 or Key 5. Sometimes the comparison was positive in relation to the current sample (Figure 9, A and C) and sometimes negative (Figure 9, B and D). The "yes" response, correct for positive and incorrect for negative trials, was to press the comparison stimulus within three seconds of its presentation; the "no" response, correct for negative and incorrect for positive trials, was to refrain from pressing the comparison for three seconds after its presentation. Either a correct "yes" (press within the threesecond limited-hold period) or a correct "no" (three seconds without pressing) produced a food pellet; an incorrect press or wait ended the trial without a pellet. The yes-no procedure was introduced in three steps, starting with trial types A and B (Figure 9), then adding the C-types, and finally the D-types. These steps also included temporary variations in the durations of the limitedhold and the waiting periods, and in the proportion of trials of each type. given during a session. A correction procedure was also used for a few sessions. When the performance met the usual criteria, form-line training with the yes-no procedure was complete; the animal successfully learned the intended sample discrimination.
(RESPOND)
A
POSITIVE TRIALS
NEGATIVE TRIALS
-(WAIT)
0 00 ceoe @ G0
00 0
On probe trials in both sessions (Figure 10), the animal almost always selected the vertical (incorrect) comparison when the sample was red. A vertical (correct) preference in the presence of the green sample was less pronounced during the extinction test but increased markedly when a pellet was delivered on all correct trials. Baseline error patterns differed during the two tests, with trial type 3B showing some disruption in the extinction test, and trial type lB slightly below criterion in the test with reinforcement. In neither test did the baseline errors seem severe enough to affect probe trials. Even the new baseline discriminations, therefore, which more explicitly required the kinds of stimulus control that the probe trials called for if symmetry were to emerge, did not bring out symmetry in the line-hue conditional dis-
crimination.
The animals' low scores on hue-line probes also revealed the absence of transitivity (Table 1) in the baseline conditional discriminations. If the stimulus relations in conditional discriminations 1 and 2 were transitive, then, even without line-hue training, hue-form and form-line should have permitted the emergence of hue-line relations in the probe trials.
Fig. 9. Representations of stimulus displays in the "yes-no" procedure after a subject had responded to the sample and brought on a comparison stimulus. Samples (X or +) are depicted in the center, and comparisons (vertical or horizontal lines) on either side. Shading indicates a blank key.
39
procedures. Experimental sessions took place every weekday and on occasional weekends.
Apparatus During experimental sessions a cage, like that used in the monkey studies but with dimensions appropriate to the larger animals, held the baboon within a sound-attenuated fan-ventilated chamber. A portion of one chamber wall, accessible to the animal through an opening in the cage, contained a horizontal row of three stimulus-response keys, each operating a limit switch when pressed. The keys were 3.2 cm in diameter, with a center-to-center distance of 4.4 cm between adjacent keys. Samples always appeared on the center key, and comparisons on the outer keys. Line and hue stimuli were the same as those in the monkey experiments. A pellet tray, located immediately below the center key, received Noyes l-g banana pellets. Except for illumination provided by the projected stimuli, the chamber was always dark. There was no trial-initiation key.
90-
8070w
EXT
O
z
w
C-)
60-
t t
4030-
cc.
W
a.
20-
IvJ Iv&
10FX
+IGIRIVIHIGIR
IX
GIR
VIHSVHIX+
0-
Fig. 10. Monkey R46's performances during its two symmetry tests in Experiment 4. Bars depict the percentage of correct responses for each trial type (corresponding to Table 4). Stimuli for each trial type are identified at the bottom of the bars, with a line connecting each sample to its correct comparison. Asterisks denote the line-hue trial types tested for symmetry, and the two righthand bars for each sessioni depict the hueline symmetry probes. In the session shown at the left,
symmetry probes were unreinforced (EXT); in the session shown at the right, all correct trials were reinforced (CRF). Each of the lefthand set of bars represents 65 trials; each righthand bar, 23 trials.
Training
The baboons had preliminary training like the monkeys, but had no trial-initiation key. Conditional discriminations were the same as those in Table 2, but the baboons never had sample schedules. Subject Bab-Sim learned line-line first. Huehue trials were then mixed with line-line. Adding line-hue trials then disrupted the lineline performance, so the animal was trained on line-hue alone before mixing all six trial types again. Reinforcement probability was then gradually reduced, as with the monkeys, and Subject Bab-Sim's first test had no reinforcement on probe trials. A second test on the next day gave reinforcement on all correct trials. Since some baseline disruption occurred during the first two tests, the animal had three more baseline sessions with 100%; reinforcement, followed by a third probe like the second (reinforcement on all correct trials). Reinforcement on all eight trial types then continued for 20 more sessions. Subject Bab-Win started with both line-line and hue-hue trials, then had line-hue alone, and finally had all six trial types mixed together. During this final baseline phase, the intertrial interval was increased to 10 sec and
EXPERIMENT 5 Whether the different outcomes in the experiments with monkeys and children arose from species-related or experiential factors, or both, is not at all clear at present (see General Discussion). Because tests with other species might help resolve the question, Experiment 5 repeated Experiments 1 and 3 reasonably closely with baboons as subjects.
METHOD Subjects Two baboons (Papio anubis) were subjects. At the time of their symmetry tests, Subject Bab-Sim was adolescent and Subject Bab-Win was adult. Bab-Sim was experimentally naive; Bab-Win had been confined in a restraint chair during a study that involved hue discriminations but had never been exposed to the chamber or the stimuli used in the present experiment, or to conditional-discrimination
40
remained at that duration. Subject Bab-Win had a single test, with reinforcement on all correct trials.
RESULTS AND DISCUSSION
Probe trials during Subject Bab-Sim's first test in extinction (left section of Figure 11) averaged 46% correct, with the animal manifesting a slight preference for the horizontal comparison on both hue-line (probe) and lineline (baseline) trials. In the second test, with reinforcement on all correct trials (center section of Figure 11), probes averaged 44% and the horizontal preference grew larger in both sets of trial types. Criterion baseline performances were then reestablished, in the hope of ruling out loss of the discrimination be-
tween vertical and horizontal comparisons as the source of the animal's difficulty with the probes. During the third test, with all correct trials again reinforced, probes still averaged only 48% (right section of Figure 26). Although the animal's comparison preference had now shifted strongly to vertical, the score of 88% on trial type lB did not indicate sufficient baseline disruption to implicate a generalized loss of the line-line discrimination as the source of the animal's difficulty with the probes. It seems reasonable to conclude that baseline disruption was not responsible for the probe-trial performances but rather, that unspecified control generated in probe trials was responsible for the baseline disruptions. For example, a comparison-line preference that de-
BAB-SIM
a ^^
IlVU-
T
90-
EXT
8070w
CRF
CRF
60-
C.)
w a.
50- +
403020-
+ + + + + + + + + + + + + + + + + ++ + + + + +
*I*
10-
*I*
H G \V/ V VHVHIGR
V
*1*
V
G R VH VH GR GR GR GR VH VH
V H G R
\
\ I
GR GR
H G R \ GR VH VH
H \V
G R
\
H G
\
0-
41
or 90J
100 90
V.%
*G-GR
I
oR-RG
I I
I I
80 [
70
LUi C) gOOF
*V-VH
a
a
oH-HV
I A
9o
0
F 100
LL 90
*V-GR
I
oH-RG
I
a
z
0-
80
WL 70 G-VH 60 50
40
~~~~~~~~I
R-HV
30 20
10
"q
'
a. .
I
.
BAB-SIM
11
13
15
17
19
21
CONSECUTIVE SESSIONS
Fig. 12. Baboon Bab-Sim's performances during its third symmetry test (Session 1) and the subsequent 20 training sessions. All correct baseline and probe trials were reinforced. Hue-hue (G-GR and R-RG), line-line (V-VH and H-HV), and line-hue (V-GR and H-RG) trials appear, respectively, in the topmost, second, and third panels; probe trials (G-VH and R-HV) are in the bottom panel. Each session had 300 to 360 trials.
veloped during probe trials might have caused the animal occasionally to ignore the sample and, instead, to select the preferred comparison even during baseline trials IA and lB. Subject Bab-Sim's failure to reverse the sample-comparison roles of lines and hues was confirmed during subsequent training sessions,
with reinforcement on all correct trials (Figure 12). Throughout these sessions the animal maintained scores of 90%0 or better on hue-hue and line-hue trials (topmost and third panels, respectively). The line-line baseline and the hue-line probes (second and bottom panels, respectively) continued to fluctuate uneven
til Session 13, with the probe scores always lower and more variable than the baselines. Generally, the comparison-line preference during line-line trials reflected the probe-trial preference but in reduced magnitude. Subject Bab-Win's symmetry test, with reinforcement, showed the animal maintaining an excellent baseline while averaging only 49% correct on probes (Figure 13). The baboons, therefore, confirmed the monkeys' data both without and with reinforcement. They gave no evidence that their conditional-discrimination training had established symmetric relations between samples and correct comparisons.
42
BAB-WIN
100
-
90-
8070w
600
z
w
50- + + + + + + + +
40-
cr
w a.
30-
20-
*1
10
-
H G R V H G R I l VH VH UR GR GR GR VH VH
V
0-
BASELI
NE-Il-TEST-I
Fig. 13. Subject Bab-Win's performance during its symmetry tests in Experiment 5. Bars depict the percenttage of correct trials for each trial type (correspondinig to Table 2, but with no sample schedules). Stimuli for each trial type are identified at the bottom of the bars, with a line connecting each sample to its correct comnparison. Asterisks denote the line-hue trial types tested for symmetry, and the two righthand bars depict the hue-line symmetry probes. All correct baseline and probe trials were reinforced. Each bar represents 48 trials.
GENERAL DISCUSSION Procedures that have become more or less standard (e.g., Cumming 8c Berryman, 1965) generated symmetry in the conditional discriminations of 5-year-old children, but not of monkeys or baboons. The failures confirmed and extended findings with pigeons (Hogan &c Zentall, 1977; Holmes, 1979; Rodewald, 1974). Experiment 4 also proved transitivity to be absent in monkeys' conditional discriminations.
Because symmetry and transitivity are necessary properties of equivalence relations (Table 1), the absence of either was sufficient to disprove the inference that each sample and its arbitrarily corresponding comparison stimulus had formed a class of equivalent stimuli. The monkeys', baboons', and pigeons' conditional discriminations could not legitimately be called matching to sample. By extension, any previous interpretations of conditional discriminations (if ... then relations) as matching to sample (equivalence relations) must be viewed skeptically, particularly, perhaps, if the subjects were not humans. It is of course impossible to prove by failures alone that conditional-discrimination procedures are incapable of establishing symmetric relations for any organism. Such a conclusion would demand an understanding, which we currently lack, of the prerequisites for symmetry, and an explanation of the organism's inability to meet one or more of the requirements. On the other hand, the data have made it clear that symmetry cannot be taken for granted. This demonstration by means of empirically specifiable behavioral techniques is perhaps the most significant contribution of the present experiments. If failures to demonstrate symmetry are to be attributed to experiential rather than to genetically related variables, proof that the subjects lacked one or more of the necessary experiences must be provided. Whether or not subsequent investigation succeeds in identifying critical gaps in the subjects' behavioral histories, shifting the burden of proof and thereby provoking the search may prove constructive. Perhaps the most relevant experience to provide would be additionial symmetry tests, with initial test failures being followed by explicit reinforcement of the desired performance, as was done with Subject Bab-Sim (Figure 12). One might argue that pigeons, monkeys, and baboons, unlike humans, do not ordinarily experience enough exemplars for them to appreciate the concept of symmetry. Demonstrations of symmetry with normal 5-year-olds who had not yet started school (Experiment 3), and of equivalence relations with severely retarded, institutionalized, environmentally restricted subjects, who had little language and no constructive academic experience (Sidman, 1971, unpublished data; Sidman & Cresson, 1973),
43
of the controlling stimuli in the conditional discriminations may be the most fundamental factor underlying the absence of symmetry. The number of testable differences between the experimenter's specifications and the actual controlling stimuli is limited only by one's ingenuity. One cannot rule out the possibility that continued attempts to sharpen a monkey's training history will succeed in achieving congruence between the experimenter's and the subject's definitions of the controlling stimuli, thereby permitting a valid and successful test for symmetry. Equally difficult to rule out, however, is the possibility that species-related factors will prevent the development of such congruence. It may, for example, not be possible for some species to abstract certain stimulus features from the environment, or to form certain functional classes, or even to separate stimulus features from stimulus functions. Such species differences need not be absolute, but may depend on the ethological validity of the stimuli and functions being examined. The present experiments might have yielded positive evidence of symmetry if they had used monkey odors and monkey sounds as stimuli instead of lines and color patches, or if the functional classes had been "edible" and "safe," or "inedible" and "dangerous" rather than "sample" and "comparison." In that event, the species differences would lie in the arbitrariness or in the variety of the equivalence relations that could develop. When the conditional-discrimination procedure does generate true matching to sample, the formation of stimulus classes defines a semantic correspondence between each sample and its matching comparison stimulus. The procedure therefore has features that recommend it as a model for the study of complex behavior, and the discovery that children, chimpanzees, monkeys, and pigeons could be taught conditional discriminations fostered the hope that such relatively simple organisms could be used to develop experimental and theoretical models even for language (e.g., Davenport & Rogers, 1970; French, 1965; Nissen, Blum, 8c Blum, 1948; Riesen & Nissen, 1942; Weinstein, 1941, 1945). If nonhuman species fail to develop true matching performances or prove limited in comparison to humans, then studies with nonhuman subjects take on a somewhat different significance. Rather than accept the nonhuman perfor-
44
mance as a model, homologous to the human, a thorough analysis of failures of the procedure to generate true matching offers the opportunity to specify the missing components and thereby to identify elements that are unique to the normal human performance. Then, attempts to build into the subjects' repertoires those components that would make nonhuman and human matching-to-sample behavior homologous might yield practical methods applicable to humans with developmental
deficiencies.
REFERENCE NOTES
1. Constantine, B. An experimental analysis of stimulus control in simple conditional discriminations: A methodological study. Unpublished doctoral dissertation, Northeastern University, 1981. 2. Lazar, R. M., & Kotlarchyk, B. J. The acquisition of sequence-class membership through stimulus equivalences. Paper presented at the meeting of the Eastern Psychological Association, Hartford, April 1980. 3. Mackay, W. E. Stimulus class formation with an oddity procedure. Unpublished master's thesis, Northeastern University, 1979.
REFERENCES
Brown, P. L., & Jenkins, H. M. Auto-shaping of the pigeon's key peck. Journal of the Experimental Analysis of Behavior, 1968, 11, 1-8. Carter, D. E., & Eckerman, D. A. Symbolic matching by pigeons: Rate of learning complex discriminations predicted from simple discriminations. Science, 1975, 187, 662-664. Cohen, L. R., Brady, J., & Lowry, M. The role of differential responding in matching-to-sample and delayed matching performance. In M. L. Commonis & J. A. Nevin (Eds.), Quantitative analyses of behavior (Vol. 1). Discriminative properties of reinforcement schedules. Cambridge: Ballinger Publishing Co., 1981. Cohen, L. R., Looney, T. A., Brady, J. H., & Aucella, A. F. Differential sample response schedules in the acquisition of conditional discriminations by pigeons. Journal of the Experimental Analysis of Behavior, 1976, 26, 301-314. Cumming, W. W., & Berryman, R. The complex discriminated operant: Studies of matching-to-sample and related problems. In D. I. Mostofsky (Ed.), Stimulus generalization. Stanford: Stanford University Press, 1965. Davenport, R. K., & Rogers, C. M. Intermodal equivalence of stimuli in apes. Science, 1970, 168, 279-280. Farthing, G. W., & Opuda, M. J. Transfer of matching-to-sample in pigeons. Journal of the Experimental Analysis of Behavior, 1974, 21, 199-213. French, G. M. Associative problems. In A. M. Schrier, H. F. Harlow, & F. Stollnitz (Eds.), Behavior of nonhuman primates. New York: Academic Press, 1965.