Behavioural Processes 112 (2015) 61–71
Contents lists available at ScienceDirect
Behavioural Processes journal homepage: www.elsevier.com/locate/behavproc
Selective attention and pigeons’ multiple necessary cues discrimination learning夽 Y. Teng a , O.V. Vyazovska b , E.A. Wasserman a,∗ a b
The University of Iowa, USA Institute for Problems of Cryobiology and Cryomedicine of the National Academy of Sciences of Ukraine, Ukraine
a r t i c l e
i n f o
Article history: Available online 20 August 2014 Keywords: Discrimination learning Selective attention Attentional tradeoff Dimensional stimulus control Vision Touchscreen pecking Pigeon
a b s t r a c t We deployed the Multiple Necessary Cues (MNC) discrimination task to see if pigeons can simultaneously attend to four different dimensions of complex visual stimuli. Specifically, we trained eight pigeons on a simultaneous discrimination to peck only 1 of 16 compound stimuli created from all possible combinations of two stimulus values from four separable visual dimensions: shape (circle/square), size (large/small), line orientation (horizontal/vertical), and brightness (dark/light). Some pigeons had CLHD (circle, large, horizontal, dark) as the positive stimulus (S+), whereas others had SSVL (square, small, vertical, light) as the S+. All eight pigeons acquired the MNC discrimination, suggesting that they had attended to all four dimensions. Learning rate was similar to all four dimensions, with learning along the orientation dimension being a bit faster than along the other three dimensions. The more dimensions along which the S−s differed from the S+, the faster was learning, suggesting an added benefit from increasing perceptual disparities between the S−s and the S+. Of particular note, evidence of attentional tradeoffs among the four dimensions was much weaker with the simultaneous task than with the successive task. We consider several reasons for this empirical disparity. © 2014 Elsevier B.V. All rights reserved.
Everyone knows what attention is. It is the taking possession by the mind, in clear and vivid form, of one out of what seem several simultaneously possible objects or trains of thought. Focalization, concentration, of consciousness are of its essence. It implies withdrawal from some things in order to deal effectively with others. These famous lines of William James (1890/1950, pp. 403–404) emphasize the phenomenology of selectively attending to some stimuli and ignoring others. However obvious James’s mentalistic proposal may seem to us, it is utterly useless to researchers who are interested in the developmental and comparative psychology of selective attention—babies and baboons do not spontaneously relate their states of mind to us. Any hope of gaining objective insight into the nature of selective attention requires behavioral study—even in the case of verbal adults.
夽 This article is part of a Special Issue entitled: Tribute to Tom Zentall. ∗ Corresponding author at: Department of Psychology, The University of Iowa, Iowa City, IA 52242, USA. Tel.: +1 319 335 2445; fax: +1 319 335 0191. E-mail address: [email protected] (E.A. Wasserman). URL: http://www2.psychology.uiowa.edu/Faculty/Wasserman/ (E.A. Wasserman). http://dx.doi.org/10.1016/j.beproc.2014.08.004 0376-6357/© 2014 Elsevier B.V. All rights reserved.
In this connection, Herbert Spencer Jennings later pioneered a behavioral approach to the study of psychological phenomena, like attention. For Jennings, attention is not a conscious mental state; rather, “at the basis of attention lies objectively the phenomenon that the organism may react to only one stimulus even though other stimuli are present which would, if acting alone, likewise produce a response (1906/1976, p. 330).” The organism can then be said to attend to the particular stimulus to which it responds. This equating of attention with the stimulus control of overt behavior was most carefully and explicitly stated by George Reynolds (1961) in regard to his well-known study of pigeons’ visual discrimination learning, in which redundant relevant cues (color and shape) were associated with reinforced or nonreinforced key pecking. Reynolds proposed that, “an organism attends to an aspect of the environment if independent variation or independent elimination of that aspect brings about variation in the organism’s behavior (p. 203).” Reynolds’ pioneering experiments showed “that a pigeon may attend to only one of several aspects of a discriminative stimulus. Every part of the environment that is present when a reinforced response occurs may not subsequently be an occasion for the emission of that response (p. 208).” Therefore, according to Reynolds, “attention refers to the controlling relation between a stimulus and
62
Y. Teng et al. / Behavioural Processes 112 (2015) 61–71
responding. An organism attends to a stimulus when its responding is under the control of that stimulus (p. 208).” Many different experimental methods have been deployed to study selective attention in animals (Riley and Leith, 1976; Sutherland and Mackintosh, 1971; Thomas, 1970; Zentall, 2012). However, none of these methods affords researchers the opportunity to study the dynamics of selective attention while discrimination learning is actually unfolding. We therefore developed the Multiple Necessary Cue (MNC) task, which allows the concurrent monitoring of stimulus control by several different physical dimensions as learning progresses. Since 1993, we have deployed the MNC task to study a variety of issues in discrimination learning, with selective attention emerging as one of our prime concerns (Chatlosh and Wasserman, 1993; Gottselig et al., 2001; Kirkpatrick-Steger and Wasserman, 1996; Kirkpatrick-Steger et al., 2000; Soto and Wasserman, 2010, 2011; Vyazovska et al., 2014; Wasserman et al., 2002). For instance, in the course of our studies of stimulus control by geons and their spatial relations in object discrimination (reviewed by Wasserman and Biederman, 2012), we discovered that the MNC task could yield very useful information. In one project, KirkpatrickSteger and Wasserman (1996) arranged a successive go/no go version of the MNC discrimination procedure to teach eight pigeons to peck just 1 of 16 pictorial stimuli, each of which displayed two abutting shapes. As one example of a positive stimulus (S+), a wedge would be located to the right of a cube. The remaining three locations of the wedge relative to the cube (left of, above, below) were negative stimuli (S−s); so too were all four locations (right, left, above, below) of three different shapes (cylinder, cone, handle). The birds rapidly learned this go/no go discrimination task—pecking the 1 S+ at a much higher rate than any of the 15 S−s—thereby documenting stimulus control by both geon identity and spatial relation. Most interestingly, across all of the pigeons, there was an inverse relation between stimulus control by component shape and component location: that correlation was large and statistically significant, −.84. This strong negative correlation suggests that the more stimulus control was acquired by one aspect of the line drawings, the less control was acquired by the other—a classic attentional tradeoff. Our most recent MNC project (Vyazovska et al., 2014) studied the behavior of pigeons given a successive go/no go discrimination task involving four different dimensions of integral visual stimuli. Specifically, we trained nine birds to peck only 1 of 16 compound discriminative stimuli created from all possible combinations of two stimulus values from four separable visual dimensions: shape (circle/square), size (large/small), line orientation (horizontal/vertical), and brightness (dark/light). Some of the pigeons were assigned CLHD (circle, large, horizontal, dark) as the S+, whereas others were assigned SSVL (square, small, vertical, light) as the S+. All of the pigeons acquired the MNC discrimination, indicating that they had attended to each of the four dimensions of the stimuli. In addition, the more dimensions along which the S−s differed from the S+, the faster was discrimination learning, suggesting an added benefit from increasing the number of perceptual disparities between the S−s and the S+. Finally, clear signs of attentional tradeoffs among the four dimensions arose during the course of discrimination learning, with marked upswings in discriminating one dimension accompanied by marked downswings in one or more other dimensions, particularly for pigeons taking longer to master the MNC discrimination. Such attentional tradeoffs are believed to be the result of two basic and logically related aspects of attention (Pashler, 1998): limited capacity and selectivity. If an animal’s attentional capacities are overloaded, then selectivity is a necessary consequence of limited capacity. The notion that paying more attention to some
discriminative stimuli causes the loss of attention to others has been called the “inverse hypothesis” (Thomas, 1970). The present project asked how replicable the results of the Vyazovska et al. (2014) study would be if the MNC task were given as a simultaneous discrimination rather than as a successive discrimination; indeed, all of our prior MNC investigations had used successive discrimination procedures. One might expect some disparity in performance because the simultaneous task allows the organism to directly compare the S+ with each of the S−s before deciding to which to respond, a possibility that is precluded in the successive discrimination, where only one stimulus at a time is presented and the organism must decide whether or not to respond to it. The inability to compare the S+ and S− compounds in the successive task might amplify any attentional and/or memory demands imposed by the MNC discrimination. 1. Method 1.1. Subjects We studied 8 feral pigeons kept at 85% of their free-feeding weights by controlled daily feedings. The pigeons had served in unrelated studies prior to the present investigation and therefore needed no further pretraining before participating. 1.2. Apparatus The experiment used four 36-cm × 36-cm × 41-cm operant conditioning chambers detailed by Gibson et al. (2004). The chambers were located in a dark room with continuous white noise. Each chamber was equipped with a 15-in. LCD monitor located behind an AccuTouch® resistive touchscreen (Elo TouchSystems, Fremont, CA). The portion of the screen that was viewable by the pigeons was 28.5 cm × 17 cm. Pecks to the touchscreen were processed by a serial controller board outside the chamber. A rotary dispenser delivered 45-mg pigeon food pellets through a vinyl tube into a Plexiglas cup located in the center of the rear wall opposite the touchscreen. Illumination during the experimental sessions was provided by a houselight mounted on the upper rear wall of the chamber. The pellet dispenser and houselight were controlled by a digital I/O interface board. Each chamber was controlled by an Apple® iMac® computer. The program that ran the experiment was developed in MatLab® . 1.3. Stimuli and experimental design We prepared a total of 16 different integral shape/size/ orientation/brightness compound visual stimuli created from two possible values along four dimensions (circle/square, large/small, horizontal line/vertical line, dark/light)—(paired S+ and S− stimuli are depicted in Fig. 1). The width/diameter of the large stimuli was 5.6 cm, whereas the width/diameter of the small stimuli was 3.8 cm. The RGB value of the dark stimuli was (110, 110, 110), whereas the RGB value of the light stimuli was (160, 160, 160). We presented two compound stimuli on every training trial: an S+ and an S-, with the left-right positions of the two stimuli randomized across trials. The centers of the stimuli were placed 14.5 cm apart from each other, 6.5 cm from the edge, 5.0 cm from the top, and 9.0 cm from the bottom of the touchscreen area that was available to the pigeons. The discriminative stimuli were presented in the center of the touchscreen frame on a blue field (RGB values were 0, 0, 255) which filled the entire LCD display. The effective pecking area containing each discriminative stimulus (large or small) was 6.4 cm × 6.4 cm in order to equate the opportunity to record pecks from stimuli of both large and small sizes.
Y. Teng et al. / Behavioural Processes 112 (2015) 61–71
63
remained on the screen for 1 s, whereas the unselected stimulus disappeared. Differential food reinforcement was used to promote discrimination learning. If the birds made a correct response, then food was delivered (two or three 45-mg pellets) and an 8-s intertrial interval ensued, during which the screen blackened and the pigeons awaited the next trial. If the choice response was incorrect, then no food was delivered and the screen blackened; a 1-s timeout period ensued, followed by a correction trial. One or more correction trials were given until the correct response was made. Only the first choice response of a trial was recorded and used in data analysis. The training sessions lasted 40 days. 2. Results 2.1. Discrimination learning All eight pigeons learned the four-dimensional two-alternative forced-choice MNC discrimination task, although the speed of learning varied among the birds. Seven of the eight pigeons responded correctly to the S+ more than 85% of the time in the presence of each of the 15 different S−s in at least one (usually many more than one) session; bird 88B responded correctly to the S+ at least 75% of the time in the presence of each of the 15 different S−s in a single session. We deemed each bird to have first met these learning criteria when those performance levels were initially attained. The mean number of days to reach criterion was 13.5. Table 1 displays mean choice accuracy across discrimination training for each of the eight pigeons up to and including the criterion day of training. These scores clearly show that the MNC task was highly effective in training the birds to attend to and discriminate stimuli along four different visual dimensions. 2.2. Dimensional stimulus control
Fig. 1. Half of the stimulus pairings when the S+ was created from circle, large horizontal, dark (top) or square, small, vertical, light (bottom) elements. For the other half of the pairings, the S+ was to the right of the S−. Surrounding the discriminative stimuli and occupying the full LCD display was a blue field. Fig. 1 shows only that portion of the blue field (the darker bordering region) that fell within the effective pecking area containing each of the two integral stimuli. That area was slightly larger (6.4 cm) than the large circle and square stimuli themselves (5.6 cm), but was the same for both the large (5.6 cm) and small stimuli (3.8 cm).
1.4. Procedure Daily discrimination training sessions comprised 120 trials involving the S+ and 1 of the 15 randomly chosen S−s simultaneously presented on a blue background. The 120 trials resulted from 4 presentations of each of the 15 S−s in 2 spatial locations (to the left and to the right of the S+). At the beginning of each trial, a start stimulus—a black plus sign in the center of a white background (6.7 cm × 6.7 cm)—was presented in the center of screen. Following one peck anywhere on the stimulus, the pigeons were presented with the S+ and one of the S−s on the screen. The pigeons’ task was to peck at the S+ stimulus and to withhold pecking at the S−. For half of the birds, the S+ was a large dark circle with a horizontal line superimposed; for the other half of the birds, the S+ was a small light square with a vertical line superimposed (Fig. 1). After a choice response was made, the selected stimulus
We next sought to determine whether and how strongly each of the four stimulus dimensions came to control the pigeons’ choice behavior; we did so by computing the probability of a correct choice response for each stimulus dimension regardless of the other three stimulus dimensions. Fig. 2 (top) depicts mean choice accuracy to each dimension averaged across all eight pigeons throughout all 40 days of training. Learning to all four dimensions was very fast, with control by orientation rising somewhat faster than control by the other three dimensions. A two-way ANOVA confirmed that the main effect of Days of training was statistically significant, F(39, 273) = 78.99, p < .001, as was the main effect of Dimension, F(3, 21) = 88.34, p < .001. The Dimension by Days interaction was not statistically significant. Table 2 shows mean choice accuracy to each dimension for each bird up to and including the criterion day of discrimination training. Generally, there were small disparities in discrimination learning among the four dimensions, with orientation again slightly surpassing the other three dimensions. Individual pigeons’ mean probability of correct responding to each of the four dimensions up to and including the criterion day ranged from .86 to .98; each pigeon’s lowest to highest dimensional discrepancy scores ranged from .02 to .10. A one-way ANOVA on the discrimination scores in Table 2 revealed a significant effect of dimension, F(3, 21) = 4.54, p < .05, confirming that the four dimensions of the visual stimuli differentially controlled the pigeons’ discrimination behavior. Tukey HSD analysis indicated that the accuracy of responding to orientation was significantly higher than accuracy to size and shape, but it was not significantly different from brightness; there were no statistically significant differences among accuracy levels to the
64
Y. Teng et al. / Behavioural Processes 112 (2015) 61–71
Table 1 Mean choice accuracy for each incorrect compound stimulus across discrimination training up to and including criterion day. Bird (days of training) Incorrect compound
CLHD+ condition
O− B− S− F− O− B− S− F+ O− B− S+ F− O− B− S+ F+ O− B+ S− F− O− B+ S− F+ O− B+ S+ F− O− B+ S+ F+ O+ B− S− F− O+ B− S− F+ O+ B− S+ F− O+ B− S+ F+ O+ B+ S− F− O+ B+ S− F+ O+ B+ S+ F−
SSVL+ condition
9W (6)
16W (9)
87Y (10)
34R (18)
81W (10)
89R (13)
40Y (21)
88B (21)
1.00 0.90 0.96 0.96 0.98 0.92 0.98 1.00 0.96 0.92 0.90 0.71 0.96 0.88 0.92
0.99 0.92 0.99 0.85 0.99 0.89 0.96 0.89 0.88 0.92 0.85 0.64 0.81 0.72 0.79
0.98 0.93 0.98 0.95 0.98 0.98 0.95 0.91 0.91 0.94 0.84 0.69 0.81 0.85 0.64
0.99 0.97 0.99 0.97 1.00 1.00 0.97 0.97 0.94 0.93 0.89 0.78 0.81 0.74 0.69
0.91 0.90 0.93 0.89 0.89 0.88 0.89 0.79 0.86 0.81 0.79 0.79 0.83 0.81 0.83
0.98 0.98 0.99 0.99 0.97 0.98 0.98 0.92 0.99 0.93 0.96 0.92 0.76 0.70 0.77
0.99 0.99 0.97 0.99 0.98 0.98 0.97 0.98 0.96 0.93 0.88 0.85 0.74 0.77 0.68
0.98 0.98 0.98 0.99 0.99 0.94 0.98 0.90 0.95 0.94 0.89 0.85 0.67 0.67 0.55
Note. Incorrect compound stimuli are denoted by whether the orientation, brightness, size, and form (shape) elements matched (+) or did not match (−) the correct compound stimulus in each correct compound condition (CLHD+ and SSVL+).
size, shape, and brightness dimensions. The data therefore suggest small, but reliable salience disparities among the four different dimensions. 2.3. Control by number of dimensional disparities We next sought to determine whether the pigeons’ choice accuracy depended on the number of dimensional disparities between S+ and the 15 S−s. We thus grouped the S−s in accord with the number of ways in which each S− differed from the S+: by four dimensional disparities (DD4), by three dimensional disparities (DD3), by two dimensional disparities (DD2), and by only one dimensional disparity (DD1). For example, with the CLHD S+, the four DD1 S−s were CLHL, CLVD, CSHD, and SLHD; the six DD2 S−s were CLVL, CSHL, CSVD, SLHL, SLVD, and SSHD; the four DD3 S−s were CSVL, SLVL, SSHL, and SSVD; and, the one DD4 S− was SSVL. Fig. 2 (bottom) depicts mean choice accuracy as a function of the number of dimensional disparities averaged across all eight pigeons throughout all 40 days of training. The speed of learning was directly related to the number of dimensional disparities. A two-way ANOVA disclosed that the main effect of Days of training was statistically significant, F(39, 273) = 18.28, p < .001, as was the main effect of Disparity, F(3, 21) = 382.93, p < .001. The Disparity by Days interaction, F(117, 819) = 1.24, p < .05, was also statistically significant. Dimensional disparity scores are shown in Table 3 for each individual pigeon up to and including the criterion day of discrimination training. Again, discriminative responding rose faster the more dimensional disparities existed between S+ and the 15 S−s, with discrimination learning being slowest with one dimensional
disparity and fastest with four dimensional disparities. A one-way ANOVA on the discrimination scores as a function of the number of dimensional disparities detailed in Table 3 revealed a statistically significant effect, F(3, 21) = 45.28, p < .001. Tukey HSD tests disclosed that discriminative performance on DD1 trials was significantly lower than on DD2, DD3, and DD4 trials; in addition, discriminative performance on DD2 trials was significantly lower than on DD3 and DD4 trials; performance on DD3 and DD4 trials did not significantly differ. 2.4. Attentional tradeoffs We next examined the individual pigeons’ dimensional accuracy scores for signs of attentional tradeoffs; this examination continued until the birds attained criterion discrimination performance. These scores are depicted in Figs. 3–6. Inspection of these figures yielded little evidence of attentional tradeoffs. By and large, changes in responding to the four different dimensions proceeded in close synchrony, with the possible exception of birds 40Y and 88B, the two slowest learning pigeons. For bird 40Y, Days 3, 4, and 5 suggest changes in control by size moving in opposite directions from control by brightness and shape; Days 6 through 18 suggest several instances where changes in control by brightness and shape moved in opposite directions. For bird 88B, Days 3 and 4 suggest the rise in control by brightness coming at the expense of control by shape and size; Days 19 through 21 show increases in control by shape and size being accompanied by a decrease in control by brightness. However, none of these changes were as dramatic as those shown in the successive discrimination performance of birds in the Vyazovska et al. (2014)
Table 2 Mean choice accuracy for each different dimension across discrimination training up to and including criterion day. Bird (days of training) Dimension
Orientation Brightness Size Shape
CLHD+ condition
SSVL+ condition
Mean
9W (6)
16W (9)
87Y (10)
34R (18)
81W (10)
89R (13)
40Y (21)
88B (21)
0.96 0.91 0.94 0.96
0.93 0.88 0.89 0.90
0.95 0.90 0.92 0.88
0.98 0.93 0.92 0.91
0.88 0.86 0.86 0.86
0.97 0.97 0.91 0.93
0.98 0.95 0.92 0.90
0.97 0.94 0.89 0.87
0.95 0.92 0.91 0.90
Y. Teng et al. / Behavioural Processes 112 (2015) 61–71
65
Fig. 2. (Top) Mean choice accuracy for each dimension averaged across all eight pigeons throughout the 40 days of discrimination training. (Bottom) Mean choice accuracy as a function of the number of disparities between the S+ compound and the S− compounds averaged across all eight pigeons throughout the 40 days of discrimination training.
Table 3 Mean choice accuracy for different numbers of dimensional disparities (DDs) across discrimination training up to and including criterion day. Bird (days of training) Number of DDs
One Two Three Four
CLHD+ condition
SSVL+ condition
Mean
9W (6)
16W (9)
87Y (10)
34R (18)
81W (10)
89R (13)
40Y (21)
88B (21)
0.88 0.94 0.95 1.00
0.76 0.88 0.94 0.99
0.77 0.91 0.95 0.98
0.79 0.93 0.98 0.99
0.80 0.85 0.89 0.91
0.83 0.93 0.98 0.98
0.82 0.92 0.97 0.99
0.74 0.90 0.97 0.98
0.80 0.91 0.95 0.98
66
Y. Teng et al. / Behavioural Processes 112 (2015) 61–71
Fig. 3. Probability of a correct response for each dimension for birds 9W (top) and 16W (bottom) in the CLHD+ condition.
study, as can be seen in Fig. 7, which depicts the behavior of the two slowest learning birds in that earlier investigation. Beyond visual inspection of the pigeons’ performance and as a way of quantifying possible attentional tradeoffs, we also calculated pairwise Pearson correlations among all four daily dimensional accuracy scores up to and including the criterion day of discrimination training. Low correlations between paired dimensional accuracy scores might suggest the presence of attentional tradeoffs, whereas high correlations between paired dimensional accuracy scores might suggest the absence of attentional tradeoffs. As detailed in Table 4, all of these pairwise dimensional correlations were extremely high, with the lowest being .83. By way of comparison, 80% of the dimensional correlation scores were lower in the successive discrimination study by Vyazovska et al. (2014) than the lowest dimensional correlation score in the present simultaneous discrimination study. The high pairwise dimensional correlations recorded here also provide little support for the occurrence of attentional tradeoffs among the four dimensions. Nonetheless, it should be noted that the pairwise dimensional correlations did trend downward the longer it took pigeons to learn the MNC discrimination in both the present simultaneous discrimination study (r = −.95, p < .001) and in the earlier successive
discrimination study of Vyazovska et al. (2014; r = −.88, p < .01). Comparing Figs. 3–6 with the correlations in Table 4 does suggest that slower learning was accompanied by both increasing dimensional discrimination reversals and lower interdimensional correlations. This overall pattern of responding is what would have been expected if our four-dimensional discrimination task were approaching some of the pigeons’ attentional limits. 3. Discussion Most theories of attention in discrimination learning (Kruschke and Johansen, 1999; Pashler, 1998; Riley and Roitblat, 1978; Sutherland and Mackintosh, 1971; Thomas, 1970; Trabasso and Bower, 1968) hypothesize that tradeoffs in discriminative responding are likely to arise from an organism’s limited capacity to simultaneously process all of the attributes of compound discriminative stimuli (Zentall, 2012). Such attentional tradeoffs follow from two key aspects of attention: limited capacity and selectivity. In our most recent successive discrimination study (Vyazovska et al., 2014), we found that all of the pigeons acquired the MNC discrimination, indicating that they had attended to all four dimensions of the stimuli. Learning rate was similar for all four
Y. Teng et al. / Behavioural Processes 112 (2015) 61–71
67
Fig. 4. Probability of a correct response for each dimension for birds 87Y (top) and 34R (bottom) in the CLHD+ condition.
dimensions, suggesting equivalent salience of the discriminative stimuli. The more dimensions along which the S−s differed from the S+, the faster was discrimination learning, implicating an added benefit from increasing perceptual disparities of the S−s from the S+. Finally, and critically clear evidence of attentional tradeoffs was seen during discrimination learning—correlated upswings and downswings in discriminative performance among one or more of the dimensions—especially for the slowest learning pigeons.
In the present simultaneous discrimination study, we obtained generally similar results. First, all of the present pigeons learned the MNC discrimination. Thus, four dimensions appear to be able to simultaneously control the pigeon’s discriminative responding regardless of the MNC task being simultaneously (choice task) or successively (go/no go task) programmed. More dimensions will have to be added in order to reach or to surpass the pigeon’s attentional limits.
Table 4 Pairwise correlations among all four dimensional accuracy scores across discrimination training up to and including criterion day. Bird (days of training) Dimensional pairing
Orientation and brightness Shape and brightness Shape and orientation Size and brightness Size and orientation Size and shape Mean
CLHD+ condition
SSVL+ condition
9W (6)
16W (9)
87Y (10)
34R (18)
81W (10)
89R (13)
40Y (21)
88B (21)
0.96 0.99 0.97 0.99 0.98 0.99 0.98
0.91 0.98 0.93 0.97 0.93 0.97 0.95
0.93 0.98 0.94 0.98 0.96 0.98 0.96
0.97 0.88 0.89 0.91 0.91 0.92 0.91
0.92 0.97 0.92 0.93 0.97 0.96 0.95
0.95 0.89 0.92 0.93 0.97 0.93 0.93
0.86 0.91 0.89 0.83 0.83 0.88 0.87
0.92 0.92 0.85 0.90 0.90 0.90 0.90
68
Y. Teng et al. / Behavioural Processes 112 (2015) 61–71
Fig. 5. Probability of a correct response for each dimension for birds 81W (top) and 89R (bottom) in the SSVL+ condition.
Second, in our selection of the specific stimuli for this project, we largely succeeded in equilibrating the salience of the four dimensions of the discriminative stimuli. No differences in dimensional salience were detected in the prior successive discrimination study and only a small difference in dimensional salience was detected in the present simultaneous discrimination study, with orientation here being a bit more salient than shape, size, and brightness. Future research can further adjust the exact values along multiple dimensions to create even more similar (or dissimilar) dimensional discriminability (cf., Gottselig et al., 2001). Third, just as in the prior successive MNC discrimination task, in the present simultaneous MNC discrimination task, the more dimensions along which the S−s differed from the S+, the faster the pigeons learned. Coupled with the birds’ final mastery of the MNC task, this finding again suggests that the pigeons were attending to all four dimensions of the stimuli. Finally, unlike our prior successive MNC discrimination task, the present simultaneous MNC discrimination task did not yield clear evidence of attentional tradeoffs among the four dimensions; even the slowest learning pigeons in the present project showed only weak signs of correlated upswings and downswings in dimensional discrimination performance (Fig. 6) compared to the very strong
signs of correlated upswings and downswings in dimensional discrimination performance of the slowest learning pigeons in our prior project (Fig. 7). If it were not for this evidence of attentional tradeoffs, then most non-attentional, associative learning theories would be quite able to account for the results of the many MNC discrimination studies that we have previously conducted, particularly those theories positing a unique configural cue emerging from the presentation of two or more elemental cues (e.g., Pearce, 2002; Rescorla and Wagner, 1972; Wasserman and Miller, 1997; also see Lamb and Riley, 1981). Such associative theories have largely held sway over attentional interpretations for many years. For just that reason, Vyazovska et al. (2014) explicitly asked whether a non-attentional, associative learning theory could explain the details of pigeons’ successive MNC discrimination behavior. We chose the Rescorla-Wagner (1972) model as the benchmark theory because of its widespread acceptance and influence. We examined a broad range of parameters across many simulations (using the R&W Simulator of Alonso et al., 2012) in an effort to assess the viability and generality of this model. And, we included a unique configural element to each of the 16 four-dimensional compound stimuli (Rescorla, 1973; Whitlow and
Y. Teng et al. / Behavioural Processes 112 (2015) 61–71
69
Fig. 6. Probability of a correct response for each dimension for birds 40Y (top) and 88B (bottom) in the SSVL+ condition.
Wagner, 1972); it is eminently plausible to expect some degree of configural stimulus control in the MNC task, as it resembles Pavlovian Positive Patterning, which also permits solution by both elemental and configural cues (Wasserman and Miller, 1997). Of course, the key question was whether the Rescorla-Wagner model can explain the dynamic session-to-session yin and yang of dimensional stimulus control that was exhibited by P8, P19, and other pigeons in Vyazovska et al. (2014) (Fig. 7). The answer here was a clear, “No,” because linear operator models like the RescorlaWagner model necessarily produce acquisition functions which are smooth and negatively accelerated (see Fig. 8 in Vyazovska et al., 2014). The failure for the Rescorla-Wagner model to provide evidence of attentional tradeoffs was not because it could not produce discrimination learning that was ordered in terms of dimensional salience (Fig. 2, top;); it was able to do so quite well (see Fig. 8 in Vyazovska et al., 2014). Nor was the model unable to capture the orderly unfolding of discrimination learning in terms of the number of dimensions along which the S−s differed from the S+ (Fig. 2, bottom); the same ordering was produced in the theoretical dimensional disparity scores (see Fig. 9 in Vyazovska et al., 2014). So, what may account for the empirical disparity between the simultaneous MNC discrimination task and the successive MNC
discrimination task? Of course, the simultaneous task allows the organism to directly compare the S+ with each of the S−s before deciding to which to respond; such comparison is impossible in the successive task where only one stimulus at a time is presented. That ability to compare the S+ and S− compounds in the simultaneous task might attenuate the attentional and/or memory demands imposed by the MNC discrimination and thereby lower the chances of detecting attentional tradeoffs. Other procedural disparities can be noted. The simultaneous MNC task involves correction trials following incorrect choices, whereas the successive MNC task does not. In addition, the probability of reinforcement on the simultaneous MNC task begins at a nominal value of .50 (and at a functional value below .50 because of repeated errors on correction trials) and rises toward 1.00 as learning progresses, whereas the probability of reinforcement on the successive MNC task remains constant at .0625. We hope in future research to further elucidate how these and other task disparities affect the dynamics of discrimination learning, particularly the magnitude and timing of attentional tradeoffs. In conclusion, we believe that the MNC discrimination task represents a useful method for clarifying the role of selective attention in discrimination learning; so too does the tracking of pecks at relevant and irrelevant stimulus attributes, as has recently been shown
70
Y. Teng et al. / Behavioural Processes 112 (2015) 61–71
Fig. 7. Discriminative key pecking to each dimension for birds P19 (CLHD+ condition; top) and P8 (SSVL+ condition; bottom) in the successive discrimination study of Vyazovska et al. (2014). To calculate the Discrimination Ratio (DR) scores depicted in this figure, two response rate scores were computed for each of the four dimensions: one for the S+ value and one for the S− value along that dimension. From those two response rates, a DR score was computed for each dimension: (8 S+ compounds)/[(8 S+ compounds) + (8 S− compounds)]. Scores near .50 represent near chance discrimination performance, whereas scores near 1.00 represent highly discriminative performance.
by Castro and Wasserman (2014). These and other new behavioral tools could help solve many longstanding mysteries of attention in organisms which cannot verbally share their phenomenological experiences with us. Acknowledgement We would like to thank Leyre Castro in the Department of Psychology, The University of Iowa for her help in the conduct and reporting of this research. References Alonso, E., Mondragón, E., Fernández, A., 2012. A Java simulator of Rescorla and Wagner’s prediction error model and configural cue extensions. Computer Methods and Programs in Biomedicine 108, 346–355.
Castro, L., Wasserman, E.A., 2014. Pigeons’ tracking of relevant attributes in categorization learning. Journal of Experimental Psychology: Animal Learning and Cognition 40, 195–211. Chatlosh, D.L., Wasserman, E.A., 1993. Multidimensional stimulus control in pigeons: selective attention and other issues. In: Zentall, T.R. (Ed.), Animal Cognition: A Tribute to Donald A. Riley. Erlbaum, Hillsdale, NJ, pp. 271–292. Gibson, B.M., Wasserman, E.A., Frei, L., Miller, K., 2004. Recent advances in operant conditioning technology: a versatile and affordable computerized touchscreen system. Behavior Research Methods, Instruments, & Computers 36, 355–362. Gottselig, J.M., Wasserman, E.A., Young, M.E., 2001. Attentional tradeoffs in pigeons learning to discriminate newly-relevant visual stimulus dimensions. Learning and Motivation 32, 240–253. James, W., 1890/1955. The Principles of Psychology, vol. 1. Dover, New York. Jennings, H.S., 1906/1976. Behavior of the Lower Organisms. Indiana University Press, Bloomington. Kirkpatrick-Steger, K., Wasserman, E.A., 1996. The what and the where of the pigeon’s processing of complex visual stimuli. Journal of Experimental Psychology: Animal Behavior Processes 22, 60–67. Kirkpatrick-Steger, K., Wasserman, E.A., Biederman, I., 2000. The pigeon’s discrimination of shape and location information. Visual Cognition 7, 417–436.
Y. Teng et al. / Behavioural Processes 112 (2015) 61–71 Kruschke, J.K., Johansen, M.K., 1999. A model of probabilistic category learning. Journal of Experimental Psychology: Learning, Memory, and Cognition 25, 1083–1119. Lamb, M.R., Riley, D.A., 1981. Effects of element arrangement on the processing of compound stimuli in pigeons (Columba livia). Journal of Experimental Psychology: Animal Behavior Processes 7, 45–58. Pashler, H.E., 1998. The Psychology of Attention. MIT Press, Cambridge, MA. Pearce, J.M., 2002. Evaluation and development of a connectionist theory of configural learning. Animal Learning & Behavior 30, 73–95. Rescorla, R.A., 1973. Evidence for the “unique stimulus” account of configural conditioning. Journal of Comparative and Physiological Psychology 85, 331–338. Rescorla, R.A., Wagner, A.R., 1972. A theory of Pavlovian conditioning: variations in the effectiveness of reinforcement and nonreinforcement. In: Classical Conditioning II: Current Theory and Research. Appleton-Century-Crofts, New York, pp. 64–99. Reynolds, G.S., 1961. Attention in the pigeon. Journal of the Experimental Analysis of Behavior 4, 203–208. Riley, D.A., Leith, C.R., 1976. Multidimensional psychophysics and selective attention in animals. Psychological Bulletin 83, 138–160. Riley, D.A., Roitblat, H.L., 1978. Selective attention and related cognitive processes in pigeons. In: Hulse, S.H., Fowler, H., Honig, W.K. (Eds.), Cognitive Processes in Animal Behavior. Erlbaum, Hillsdale, NJ, pp. 249–276. Soto, F.A., Wasserman, E.A., 2010. Integrality/separability of stimulus dimensions and multidimensional generalization in pigeons. Journal of Experimental Psychology: Animal Behavior Processes 36, 194–205.
71
Soto, F.A., Wasserman, E.A., 2011. Asymmetrical interactions in the perception of face identity and emotional expression are not unique to the primate visual system. Journal of Vision 11, 1–18. Sutherland, N.S., Mackintosh, N.J., 1971. Mechanisms of Animal Discrimination Learning. Academic Press, New York. Thomas, D.R., 1970. Stimulus selection, attention, and related matters. In: Reynierse, J.H. (Ed.), Current Issues in Animal Learning. University of Nebraska Press, Lincoln, NE, pp. 311–356. Trabasso, T., Bower, G.H., 1968. Attention in Learning. Wiley, New York. Vyazovska, O.V., Teng, Y., Wasserman, E.A., 2014. Attentional tradeoffs in the pigeon. Journal of the Experimental Analysis of Behavior 101, 337–354. Wasserman, E.A., Biederman, I., 2012. Recognition by components: a bird’s eye view. In: Lazareva, O.F., Shimizu, T., Wasserman, E.A. (Eds.), How Animals See the World. Oxford University Press, New York. Wasserman, E.A., Frank, A.J., Young, M.E., 2002. Stimulus control by same versus different relations among multiple visual stimuli. Journal of Experimental Psychology: Animal Behavior Processes 28, 347–357. Wasserman, E.A., Miller, R.R., 1997. What’s elementary about associative learning? Annual Review of Psychology 48, 573–607. Whitlow Jr., J.W., Wagner, A.R., 1972. Negative patterning in classical conditioning: summation of response tendencies to isolable and configural components. Psychonomic Science 27, 299–301. Zentall, T.R., 2012. Selective and divided attention in birds. In: Lazareva, O.F., Shimizu, T., Wasserman, E.A. (Eds.), How Animals See the World. Oxford University Press, New York, pp. 351–369.
Contents lists available at ScienceDirect
Behavioural Processes journal homepage: www.elsevier.com/locate/behavproc
Selective attention and pigeons’ multiple necessary cues discrimination learning夽 Y. Teng a , O.V. Vyazovska b , E.A. Wasserman a,∗ a b
The University of Iowa, USA Institute for Problems of Cryobiology and Cryomedicine of the National Academy of Sciences of Ukraine, Ukraine
a r t i c l e
i n f o
Article history: Available online 20 August 2014 Keywords: Discrimination learning Selective attention Attentional tradeoff Dimensional stimulus control Vision Touchscreen pecking Pigeon
a b s t r a c t We deployed the Multiple Necessary Cues (MNC) discrimination task to see if pigeons can simultaneously attend to four different dimensions of complex visual stimuli. Specifically, we trained eight pigeons on a simultaneous discrimination to peck only 1 of 16 compound stimuli created from all possible combinations of two stimulus values from four separable visual dimensions: shape (circle/square), size (large/small), line orientation (horizontal/vertical), and brightness (dark/light). Some pigeons had CLHD (circle, large, horizontal, dark) as the positive stimulus (S+), whereas others had SSVL (square, small, vertical, light) as the S+. All eight pigeons acquired the MNC discrimination, suggesting that they had attended to all four dimensions. Learning rate was similar to all four dimensions, with learning along the orientation dimension being a bit faster than along the other three dimensions. The more dimensions along which the S−s differed from the S+, the faster was learning, suggesting an added benefit from increasing perceptual disparities between the S−s and the S+. Of particular note, evidence of attentional tradeoffs among the four dimensions was much weaker with the simultaneous task than with the successive task. We consider several reasons for this empirical disparity. © 2014 Elsevier B.V. All rights reserved.
Everyone knows what attention is. It is the taking possession by the mind, in clear and vivid form, of one out of what seem several simultaneously possible objects or trains of thought. Focalization, concentration, of consciousness are of its essence. It implies withdrawal from some things in order to deal effectively with others. These famous lines of William James (1890/1950, pp. 403–404) emphasize the phenomenology of selectively attending to some stimuli and ignoring others. However obvious James’s mentalistic proposal may seem to us, it is utterly useless to researchers who are interested in the developmental and comparative psychology of selective attention—babies and baboons do not spontaneously relate their states of mind to us. Any hope of gaining objective insight into the nature of selective attention requires behavioral study—even in the case of verbal adults.
夽 This article is part of a Special Issue entitled: Tribute to Tom Zentall. ∗ Corresponding author at: Department of Psychology, The University of Iowa, Iowa City, IA 52242, USA. Tel.: +1 319 335 2445; fax: +1 319 335 0191. E-mail address: [email protected] (E.A. Wasserman). URL: http://www2.psychology.uiowa.edu/Faculty/Wasserman/ (E.A. Wasserman). http://dx.doi.org/10.1016/j.beproc.2014.08.004 0376-6357/© 2014 Elsevier B.V. All rights reserved.
In this connection, Herbert Spencer Jennings later pioneered a behavioral approach to the study of psychological phenomena, like attention. For Jennings, attention is not a conscious mental state; rather, “at the basis of attention lies objectively the phenomenon that the organism may react to only one stimulus even though other stimuli are present which would, if acting alone, likewise produce a response (1906/1976, p. 330).” The organism can then be said to attend to the particular stimulus to which it responds. This equating of attention with the stimulus control of overt behavior was most carefully and explicitly stated by George Reynolds (1961) in regard to his well-known study of pigeons’ visual discrimination learning, in which redundant relevant cues (color and shape) were associated with reinforced or nonreinforced key pecking. Reynolds proposed that, “an organism attends to an aspect of the environment if independent variation or independent elimination of that aspect brings about variation in the organism’s behavior (p. 203).” Reynolds’ pioneering experiments showed “that a pigeon may attend to only one of several aspects of a discriminative stimulus. Every part of the environment that is present when a reinforced response occurs may not subsequently be an occasion for the emission of that response (p. 208).” Therefore, according to Reynolds, “attention refers to the controlling relation between a stimulus and
62
Y. Teng et al. / Behavioural Processes 112 (2015) 61–71
responding. An organism attends to a stimulus when its responding is under the control of that stimulus (p. 208).” Many different experimental methods have been deployed to study selective attention in animals (Riley and Leith, 1976; Sutherland and Mackintosh, 1971; Thomas, 1970; Zentall, 2012). However, none of these methods affords researchers the opportunity to study the dynamics of selective attention while discrimination learning is actually unfolding. We therefore developed the Multiple Necessary Cue (MNC) task, which allows the concurrent monitoring of stimulus control by several different physical dimensions as learning progresses. Since 1993, we have deployed the MNC task to study a variety of issues in discrimination learning, with selective attention emerging as one of our prime concerns (Chatlosh and Wasserman, 1993; Gottselig et al., 2001; Kirkpatrick-Steger and Wasserman, 1996; Kirkpatrick-Steger et al., 2000; Soto and Wasserman, 2010, 2011; Vyazovska et al., 2014; Wasserman et al., 2002). For instance, in the course of our studies of stimulus control by geons and their spatial relations in object discrimination (reviewed by Wasserman and Biederman, 2012), we discovered that the MNC task could yield very useful information. In one project, KirkpatrickSteger and Wasserman (1996) arranged a successive go/no go version of the MNC discrimination procedure to teach eight pigeons to peck just 1 of 16 pictorial stimuli, each of which displayed two abutting shapes. As one example of a positive stimulus (S+), a wedge would be located to the right of a cube. The remaining three locations of the wedge relative to the cube (left of, above, below) were negative stimuli (S−s); so too were all four locations (right, left, above, below) of three different shapes (cylinder, cone, handle). The birds rapidly learned this go/no go discrimination task—pecking the 1 S+ at a much higher rate than any of the 15 S−s—thereby documenting stimulus control by both geon identity and spatial relation. Most interestingly, across all of the pigeons, there was an inverse relation between stimulus control by component shape and component location: that correlation was large and statistically significant, −.84. This strong negative correlation suggests that the more stimulus control was acquired by one aspect of the line drawings, the less control was acquired by the other—a classic attentional tradeoff. Our most recent MNC project (Vyazovska et al., 2014) studied the behavior of pigeons given a successive go/no go discrimination task involving four different dimensions of integral visual stimuli. Specifically, we trained nine birds to peck only 1 of 16 compound discriminative stimuli created from all possible combinations of two stimulus values from four separable visual dimensions: shape (circle/square), size (large/small), line orientation (horizontal/vertical), and brightness (dark/light). Some of the pigeons were assigned CLHD (circle, large, horizontal, dark) as the S+, whereas others were assigned SSVL (square, small, vertical, light) as the S+. All of the pigeons acquired the MNC discrimination, indicating that they had attended to each of the four dimensions of the stimuli. In addition, the more dimensions along which the S−s differed from the S+, the faster was discrimination learning, suggesting an added benefit from increasing the number of perceptual disparities between the S−s and the S+. Finally, clear signs of attentional tradeoffs among the four dimensions arose during the course of discrimination learning, with marked upswings in discriminating one dimension accompanied by marked downswings in one or more other dimensions, particularly for pigeons taking longer to master the MNC discrimination. Such attentional tradeoffs are believed to be the result of two basic and logically related aspects of attention (Pashler, 1998): limited capacity and selectivity. If an animal’s attentional capacities are overloaded, then selectivity is a necessary consequence of limited capacity. The notion that paying more attention to some
discriminative stimuli causes the loss of attention to others has been called the “inverse hypothesis” (Thomas, 1970). The present project asked how replicable the results of the Vyazovska et al. (2014) study would be if the MNC task were given as a simultaneous discrimination rather than as a successive discrimination; indeed, all of our prior MNC investigations had used successive discrimination procedures. One might expect some disparity in performance because the simultaneous task allows the organism to directly compare the S+ with each of the S−s before deciding to which to respond, a possibility that is precluded in the successive discrimination, where only one stimulus at a time is presented and the organism must decide whether or not to respond to it. The inability to compare the S+ and S− compounds in the successive task might amplify any attentional and/or memory demands imposed by the MNC discrimination. 1. Method 1.1. Subjects We studied 8 feral pigeons kept at 85% of their free-feeding weights by controlled daily feedings. The pigeons had served in unrelated studies prior to the present investigation and therefore needed no further pretraining before participating. 1.2. Apparatus The experiment used four 36-cm × 36-cm × 41-cm operant conditioning chambers detailed by Gibson et al. (2004). The chambers were located in a dark room with continuous white noise. Each chamber was equipped with a 15-in. LCD monitor located behind an AccuTouch® resistive touchscreen (Elo TouchSystems, Fremont, CA). The portion of the screen that was viewable by the pigeons was 28.5 cm × 17 cm. Pecks to the touchscreen were processed by a serial controller board outside the chamber. A rotary dispenser delivered 45-mg pigeon food pellets through a vinyl tube into a Plexiglas cup located in the center of the rear wall opposite the touchscreen. Illumination during the experimental sessions was provided by a houselight mounted on the upper rear wall of the chamber. The pellet dispenser and houselight were controlled by a digital I/O interface board. Each chamber was controlled by an Apple® iMac® computer. The program that ran the experiment was developed in MatLab® . 1.3. Stimuli and experimental design We prepared a total of 16 different integral shape/size/ orientation/brightness compound visual stimuli created from two possible values along four dimensions (circle/square, large/small, horizontal line/vertical line, dark/light)—(paired S+ and S− stimuli are depicted in Fig. 1). The width/diameter of the large stimuli was 5.6 cm, whereas the width/diameter of the small stimuli was 3.8 cm. The RGB value of the dark stimuli was (110, 110, 110), whereas the RGB value of the light stimuli was (160, 160, 160). We presented two compound stimuli on every training trial: an S+ and an S-, with the left-right positions of the two stimuli randomized across trials. The centers of the stimuli were placed 14.5 cm apart from each other, 6.5 cm from the edge, 5.0 cm from the top, and 9.0 cm from the bottom of the touchscreen area that was available to the pigeons. The discriminative stimuli were presented in the center of the touchscreen frame on a blue field (RGB values were 0, 0, 255) which filled the entire LCD display. The effective pecking area containing each discriminative stimulus (large or small) was 6.4 cm × 6.4 cm in order to equate the opportunity to record pecks from stimuli of both large and small sizes.
Y. Teng et al. / Behavioural Processes 112 (2015) 61–71
63
remained on the screen for 1 s, whereas the unselected stimulus disappeared. Differential food reinforcement was used to promote discrimination learning. If the birds made a correct response, then food was delivered (two or three 45-mg pellets) and an 8-s intertrial interval ensued, during which the screen blackened and the pigeons awaited the next trial. If the choice response was incorrect, then no food was delivered and the screen blackened; a 1-s timeout period ensued, followed by a correction trial. One or more correction trials were given until the correct response was made. Only the first choice response of a trial was recorded and used in data analysis. The training sessions lasted 40 days. 2. Results 2.1. Discrimination learning All eight pigeons learned the four-dimensional two-alternative forced-choice MNC discrimination task, although the speed of learning varied among the birds. Seven of the eight pigeons responded correctly to the S+ more than 85% of the time in the presence of each of the 15 different S−s in at least one (usually many more than one) session; bird 88B responded correctly to the S+ at least 75% of the time in the presence of each of the 15 different S−s in a single session. We deemed each bird to have first met these learning criteria when those performance levels were initially attained. The mean number of days to reach criterion was 13.5. Table 1 displays mean choice accuracy across discrimination training for each of the eight pigeons up to and including the criterion day of training. These scores clearly show that the MNC task was highly effective in training the birds to attend to and discriminate stimuli along four different visual dimensions. 2.2. Dimensional stimulus control
Fig. 1. Half of the stimulus pairings when the S+ was created from circle, large horizontal, dark (top) or square, small, vertical, light (bottom) elements. For the other half of the pairings, the S+ was to the right of the S−. Surrounding the discriminative stimuli and occupying the full LCD display was a blue field. Fig. 1 shows only that portion of the blue field (the darker bordering region) that fell within the effective pecking area containing each of the two integral stimuli. That area was slightly larger (6.4 cm) than the large circle and square stimuli themselves (5.6 cm), but was the same for both the large (5.6 cm) and small stimuli (3.8 cm).
1.4. Procedure Daily discrimination training sessions comprised 120 trials involving the S+ and 1 of the 15 randomly chosen S−s simultaneously presented on a blue background. The 120 trials resulted from 4 presentations of each of the 15 S−s in 2 spatial locations (to the left and to the right of the S+). At the beginning of each trial, a start stimulus—a black plus sign in the center of a white background (6.7 cm × 6.7 cm)—was presented in the center of screen. Following one peck anywhere on the stimulus, the pigeons were presented with the S+ and one of the S−s on the screen. The pigeons’ task was to peck at the S+ stimulus and to withhold pecking at the S−. For half of the birds, the S+ was a large dark circle with a horizontal line superimposed; for the other half of the birds, the S+ was a small light square with a vertical line superimposed (Fig. 1). After a choice response was made, the selected stimulus
We next sought to determine whether and how strongly each of the four stimulus dimensions came to control the pigeons’ choice behavior; we did so by computing the probability of a correct choice response for each stimulus dimension regardless of the other three stimulus dimensions. Fig. 2 (top) depicts mean choice accuracy to each dimension averaged across all eight pigeons throughout all 40 days of training. Learning to all four dimensions was very fast, with control by orientation rising somewhat faster than control by the other three dimensions. A two-way ANOVA confirmed that the main effect of Days of training was statistically significant, F(39, 273) = 78.99, p < .001, as was the main effect of Dimension, F(3, 21) = 88.34, p < .001. The Dimension by Days interaction was not statistically significant. Table 2 shows mean choice accuracy to each dimension for each bird up to and including the criterion day of discrimination training. Generally, there were small disparities in discrimination learning among the four dimensions, with orientation again slightly surpassing the other three dimensions. Individual pigeons’ mean probability of correct responding to each of the four dimensions up to and including the criterion day ranged from .86 to .98; each pigeon’s lowest to highest dimensional discrepancy scores ranged from .02 to .10. A one-way ANOVA on the discrimination scores in Table 2 revealed a significant effect of dimension, F(3, 21) = 4.54, p < .05, confirming that the four dimensions of the visual stimuli differentially controlled the pigeons’ discrimination behavior. Tukey HSD analysis indicated that the accuracy of responding to orientation was significantly higher than accuracy to size and shape, but it was not significantly different from brightness; there were no statistically significant differences among accuracy levels to the
64
Y. Teng et al. / Behavioural Processes 112 (2015) 61–71
Table 1 Mean choice accuracy for each incorrect compound stimulus across discrimination training up to and including criterion day. Bird (days of training) Incorrect compound
CLHD+ condition
O− B− S− F− O− B− S− F+ O− B− S+ F− O− B− S+ F+ O− B+ S− F− O− B+ S− F+ O− B+ S+ F− O− B+ S+ F+ O+ B− S− F− O+ B− S− F+ O+ B− S+ F− O+ B− S+ F+ O+ B+ S− F− O+ B+ S− F+ O+ B+ S+ F−
SSVL+ condition
9W (6)
16W (9)
87Y (10)
34R (18)
81W (10)
89R (13)
40Y (21)
88B (21)
1.00 0.90 0.96 0.96 0.98 0.92 0.98 1.00 0.96 0.92 0.90 0.71 0.96 0.88 0.92
0.99 0.92 0.99 0.85 0.99 0.89 0.96 0.89 0.88 0.92 0.85 0.64 0.81 0.72 0.79
0.98 0.93 0.98 0.95 0.98 0.98 0.95 0.91 0.91 0.94 0.84 0.69 0.81 0.85 0.64
0.99 0.97 0.99 0.97 1.00 1.00 0.97 0.97 0.94 0.93 0.89 0.78 0.81 0.74 0.69
0.91 0.90 0.93 0.89 0.89 0.88 0.89 0.79 0.86 0.81 0.79 0.79 0.83 0.81 0.83
0.98 0.98 0.99 0.99 0.97 0.98 0.98 0.92 0.99 0.93 0.96 0.92 0.76 0.70 0.77
0.99 0.99 0.97 0.99 0.98 0.98 0.97 0.98 0.96 0.93 0.88 0.85 0.74 0.77 0.68
0.98 0.98 0.98 0.99 0.99 0.94 0.98 0.90 0.95 0.94 0.89 0.85 0.67 0.67 0.55
Note. Incorrect compound stimuli are denoted by whether the orientation, brightness, size, and form (shape) elements matched (+) or did not match (−) the correct compound stimulus in each correct compound condition (CLHD+ and SSVL+).
size, shape, and brightness dimensions. The data therefore suggest small, but reliable salience disparities among the four different dimensions. 2.3. Control by number of dimensional disparities We next sought to determine whether the pigeons’ choice accuracy depended on the number of dimensional disparities between S+ and the 15 S−s. We thus grouped the S−s in accord with the number of ways in which each S− differed from the S+: by four dimensional disparities (DD4), by three dimensional disparities (DD3), by two dimensional disparities (DD2), and by only one dimensional disparity (DD1). For example, with the CLHD S+, the four DD1 S−s were CLHL, CLVD, CSHD, and SLHD; the six DD2 S−s were CLVL, CSHL, CSVD, SLHL, SLVD, and SSHD; the four DD3 S−s were CSVL, SLVL, SSHL, and SSVD; and, the one DD4 S− was SSVL. Fig. 2 (bottom) depicts mean choice accuracy as a function of the number of dimensional disparities averaged across all eight pigeons throughout all 40 days of training. The speed of learning was directly related to the number of dimensional disparities. A two-way ANOVA disclosed that the main effect of Days of training was statistically significant, F(39, 273) = 18.28, p < .001, as was the main effect of Disparity, F(3, 21) = 382.93, p < .001. The Disparity by Days interaction, F(117, 819) = 1.24, p < .05, was also statistically significant. Dimensional disparity scores are shown in Table 3 for each individual pigeon up to and including the criterion day of discrimination training. Again, discriminative responding rose faster the more dimensional disparities existed between S+ and the 15 S−s, with discrimination learning being slowest with one dimensional
disparity and fastest with four dimensional disparities. A one-way ANOVA on the discrimination scores as a function of the number of dimensional disparities detailed in Table 3 revealed a statistically significant effect, F(3, 21) = 45.28, p < .001. Tukey HSD tests disclosed that discriminative performance on DD1 trials was significantly lower than on DD2, DD3, and DD4 trials; in addition, discriminative performance on DD2 trials was significantly lower than on DD3 and DD4 trials; performance on DD3 and DD4 trials did not significantly differ. 2.4. Attentional tradeoffs We next examined the individual pigeons’ dimensional accuracy scores for signs of attentional tradeoffs; this examination continued until the birds attained criterion discrimination performance. These scores are depicted in Figs. 3–6. Inspection of these figures yielded little evidence of attentional tradeoffs. By and large, changes in responding to the four different dimensions proceeded in close synchrony, with the possible exception of birds 40Y and 88B, the two slowest learning pigeons. For bird 40Y, Days 3, 4, and 5 suggest changes in control by size moving in opposite directions from control by brightness and shape; Days 6 through 18 suggest several instances where changes in control by brightness and shape moved in opposite directions. For bird 88B, Days 3 and 4 suggest the rise in control by brightness coming at the expense of control by shape and size; Days 19 through 21 show increases in control by shape and size being accompanied by a decrease in control by brightness. However, none of these changes were as dramatic as those shown in the successive discrimination performance of birds in the Vyazovska et al. (2014)
Table 2 Mean choice accuracy for each different dimension across discrimination training up to and including criterion day. Bird (days of training) Dimension
Orientation Brightness Size Shape
CLHD+ condition
SSVL+ condition
Mean
9W (6)
16W (9)
87Y (10)
34R (18)
81W (10)
89R (13)
40Y (21)
88B (21)
0.96 0.91 0.94 0.96
0.93 0.88 0.89 0.90
0.95 0.90 0.92 0.88
0.98 0.93 0.92 0.91
0.88 0.86 0.86 0.86
0.97 0.97 0.91 0.93
0.98 0.95 0.92 0.90
0.97 0.94 0.89 0.87
0.95 0.92 0.91 0.90
Y. Teng et al. / Behavioural Processes 112 (2015) 61–71
65
Fig. 2. (Top) Mean choice accuracy for each dimension averaged across all eight pigeons throughout the 40 days of discrimination training. (Bottom) Mean choice accuracy as a function of the number of disparities between the S+ compound and the S− compounds averaged across all eight pigeons throughout the 40 days of discrimination training.
Table 3 Mean choice accuracy for different numbers of dimensional disparities (DDs) across discrimination training up to and including criterion day. Bird (days of training) Number of DDs
One Two Three Four
CLHD+ condition
SSVL+ condition
Mean
9W (6)
16W (9)
87Y (10)
34R (18)
81W (10)
89R (13)
40Y (21)
88B (21)
0.88 0.94 0.95 1.00
0.76 0.88 0.94 0.99
0.77 0.91 0.95 0.98
0.79 0.93 0.98 0.99
0.80 0.85 0.89 0.91
0.83 0.93 0.98 0.98
0.82 0.92 0.97 0.99
0.74 0.90 0.97 0.98
0.80 0.91 0.95 0.98
66
Y. Teng et al. / Behavioural Processes 112 (2015) 61–71
Fig. 3. Probability of a correct response for each dimension for birds 9W (top) and 16W (bottom) in the CLHD+ condition.
study, as can be seen in Fig. 7, which depicts the behavior of the two slowest learning birds in that earlier investigation. Beyond visual inspection of the pigeons’ performance and as a way of quantifying possible attentional tradeoffs, we also calculated pairwise Pearson correlations among all four daily dimensional accuracy scores up to and including the criterion day of discrimination training. Low correlations between paired dimensional accuracy scores might suggest the presence of attentional tradeoffs, whereas high correlations between paired dimensional accuracy scores might suggest the absence of attentional tradeoffs. As detailed in Table 4, all of these pairwise dimensional correlations were extremely high, with the lowest being .83. By way of comparison, 80% of the dimensional correlation scores were lower in the successive discrimination study by Vyazovska et al. (2014) than the lowest dimensional correlation score in the present simultaneous discrimination study. The high pairwise dimensional correlations recorded here also provide little support for the occurrence of attentional tradeoffs among the four dimensions. Nonetheless, it should be noted that the pairwise dimensional correlations did trend downward the longer it took pigeons to learn the MNC discrimination in both the present simultaneous discrimination study (r = −.95, p < .001) and in the earlier successive
discrimination study of Vyazovska et al. (2014; r = −.88, p < .01). Comparing Figs. 3–6 with the correlations in Table 4 does suggest that slower learning was accompanied by both increasing dimensional discrimination reversals and lower interdimensional correlations. This overall pattern of responding is what would have been expected if our four-dimensional discrimination task were approaching some of the pigeons’ attentional limits. 3. Discussion Most theories of attention in discrimination learning (Kruschke and Johansen, 1999; Pashler, 1998; Riley and Roitblat, 1978; Sutherland and Mackintosh, 1971; Thomas, 1970; Trabasso and Bower, 1968) hypothesize that tradeoffs in discriminative responding are likely to arise from an organism’s limited capacity to simultaneously process all of the attributes of compound discriminative stimuli (Zentall, 2012). Such attentional tradeoffs follow from two key aspects of attention: limited capacity and selectivity. In our most recent successive discrimination study (Vyazovska et al., 2014), we found that all of the pigeons acquired the MNC discrimination, indicating that they had attended to all four dimensions of the stimuli. Learning rate was similar for all four
Y. Teng et al. / Behavioural Processes 112 (2015) 61–71
67
Fig. 4. Probability of a correct response for each dimension for birds 87Y (top) and 34R (bottom) in the CLHD+ condition.
dimensions, suggesting equivalent salience of the discriminative stimuli. The more dimensions along which the S−s differed from the S+, the faster was discrimination learning, implicating an added benefit from increasing perceptual disparities of the S−s from the S+. Finally, and critically clear evidence of attentional tradeoffs was seen during discrimination learning—correlated upswings and downswings in discriminative performance among one or more of the dimensions—especially for the slowest learning pigeons.
In the present simultaneous discrimination study, we obtained generally similar results. First, all of the present pigeons learned the MNC discrimination. Thus, four dimensions appear to be able to simultaneously control the pigeon’s discriminative responding regardless of the MNC task being simultaneously (choice task) or successively (go/no go task) programmed. More dimensions will have to be added in order to reach or to surpass the pigeon’s attentional limits.
Table 4 Pairwise correlations among all four dimensional accuracy scores across discrimination training up to and including criterion day. Bird (days of training) Dimensional pairing
Orientation and brightness Shape and brightness Shape and orientation Size and brightness Size and orientation Size and shape Mean
CLHD+ condition
SSVL+ condition
9W (6)
16W (9)
87Y (10)
34R (18)
81W (10)
89R (13)
40Y (21)
88B (21)
0.96 0.99 0.97 0.99 0.98 0.99 0.98
0.91 0.98 0.93 0.97 0.93 0.97 0.95
0.93 0.98 0.94 0.98 0.96 0.98 0.96
0.97 0.88 0.89 0.91 0.91 0.92 0.91
0.92 0.97 0.92 0.93 0.97 0.96 0.95
0.95 0.89 0.92 0.93 0.97 0.93 0.93
0.86 0.91 0.89 0.83 0.83 0.88 0.87
0.92 0.92 0.85 0.90 0.90 0.90 0.90
68
Y. Teng et al. / Behavioural Processes 112 (2015) 61–71
Fig. 5. Probability of a correct response for each dimension for birds 81W (top) and 89R (bottom) in the SSVL+ condition.
Second, in our selection of the specific stimuli for this project, we largely succeeded in equilibrating the salience of the four dimensions of the discriminative stimuli. No differences in dimensional salience were detected in the prior successive discrimination study and only a small difference in dimensional salience was detected in the present simultaneous discrimination study, with orientation here being a bit more salient than shape, size, and brightness. Future research can further adjust the exact values along multiple dimensions to create even more similar (or dissimilar) dimensional discriminability (cf., Gottselig et al., 2001). Third, just as in the prior successive MNC discrimination task, in the present simultaneous MNC discrimination task, the more dimensions along which the S−s differed from the S+, the faster the pigeons learned. Coupled with the birds’ final mastery of the MNC task, this finding again suggests that the pigeons were attending to all four dimensions of the stimuli. Finally, unlike our prior successive MNC discrimination task, the present simultaneous MNC discrimination task did not yield clear evidence of attentional tradeoffs among the four dimensions; even the slowest learning pigeons in the present project showed only weak signs of correlated upswings and downswings in dimensional discrimination performance (Fig. 6) compared to the very strong
signs of correlated upswings and downswings in dimensional discrimination performance of the slowest learning pigeons in our prior project (Fig. 7). If it were not for this evidence of attentional tradeoffs, then most non-attentional, associative learning theories would be quite able to account for the results of the many MNC discrimination studies that we have previously conducted, particularly those theories positing a unique configural cue emerging from the presentation of two or more elemental cues (e.g., Pearce, 2002; Rescorla and Wagner, 1972; Wasserman and Miller, 1997; also see Lamb and Riley, 1981). Such associative theories have largely held sway over attentional interpretations for many years. For just that reason, Vyazovska et al. (2014) explicitly asked whether a non-attentional, associative learning theory could explain the details of pigeons’ successive MNC discrimination behavior. We chose the Rescorla-Wagner (1972) model as the benchmark theory because of its widespread acceptance and influence. We examined a broad range of parameters across many simulations (using the R&W Simulator of Alonso et al., 2012) in an effort to assess the viability and generality of this model. And, we included a unique configural element to each of the 16 four-dimensional compound stimuli (Rescorla, 1973; Whitlow and
Y. Teng et al. / Behavioural Processes 112 (2015) 61–71
69
Fig. 6. Probability of a correct response for each dimension for birds 40Y (top) and 88B (bottom) in the SSVL+ condition.
Wagner, 1972); it is eminently plausible to expect some degree of configural stimulus control in the MNC task, as it resembles Pavlovian Positive Patterning, which also permits solution by both elemental and configural cues (Wasserman and Miller, 1997). Of course, the key question was whether the Rescorla-Wagner model can explain the dynamic session-to-session yin and yang of dimensional stimulus control that was exhibited by P8, P19, and other pigeons in Vyazovska et al. (2014) (Fig. 7). The answer here was a clear, “No,” because linear operator models like the RescorlaWagner model necessarily produce acquisition functions which are smooth and negatively accelerated (see Fig. 8 in Vyazovska et al., 2014). The failure for the Rescorla-Wagner model to provide evidence of attentional tradeoffs was not because it could not produce discrimination learning that was ordered in terms of dimensional salience (Fig. 2, top;); it was able to do so quite well (see Fig. 8 in Vyazovska et al., 2014). Nor was the model unable to capture the orderly unfolding of discrimination learning in terms of the number of dimensions along which the S−s differed from the S+ (Fig. 2, bottom); the same ordering was produced in the theoretical dimensional disparity scores (see Fig. 9 in Vyazovska et al., 2014). So, what may account for the empirical disparity between the simultaneous MNC discrimination task and the successive MNC
discrimination task? Of course, the simultaneous task allows the organism to directly compare the S+ with each of the S−s before deciding to which to respond; such comparison is impossible in the successive task where only one stimulus at a time is presented. That ability to compare the S+ and S− compounds in the simultaneous task might attenuate the attentional and/or memory demands imposed by the MNC discrimination and thereby lower the chances of detecting attentional tradeoffs. Other procedural disparities can be noted. The simultaneous MNC task involves correction trials following incorrect choices, whereas the successive MNC task does not. In addition, the probability of reinforcement on the simultaneous MNC task begins at a nominal value of .50 (and at a functional value below .50 because of repeated errors on correction trials) and rises toward 1.00 as learning progresses, whereas the probability of reinforcement on the successive MNC task remains constant at .0625. We hope in future research to further elucidate how these and other task disparities affect the dynamics of discrimination learning, particularly the magnitude and timing of attentional tradeoffs. In conclusion, we believe that the MNC discrimination task represents a useful method for clarifying the role of selective attention in discrimination learning; so too does the tracking of pecks at relevant and irrelevant stimulus attributes, as has recently been shown
70
Y. Teng et al. / Behavioural Processes 112 (2015) 61–71
Fig. 7. Discriminative key pecking to each dimension for birds P19 (CLHD+ condition; top) and P8 (SSVL+ condition; bottom) in the successive discrimination study of Vyazovska et al. (2014). To calculate the Discrimination Ratio (DR) scores depicted in this figure, two response rate scores were computed for each of the four dimensions: one for the S+ value and one for the S− value along that dimension. From those two response rates, a DR score was computed for each dimension: (8 S+ compounds)/[(8 S+ compounds) + (8 S− compounds)]. Scores near .50 represent near chance discrimination performance, whereas scores near 1.00 represent highly discriminative performance.
by Castro and Wasserman (2014). These and other new behavioral tools could help solve many longstanding mysteries of attention in organisms which cannot verbally share their phenomenological experiences with us. Acknowledgement We would like to thank Leyre Castro in the Department of Psychology, The University of Iowa for her help in the conduct and reporting of this research. References Alonso, E., Mondragón, E., Fernández, A., 2012. A Java simulator of Rescorla and Wagner’s prediction error model and configural cue extensions. Computer Methods and Programs in Biomedicine 108, 346–355.
Castro, L., Wasserman, E.A., 2014. Pigeons’ tracking of relevant attributes in categorization learning. Journal of Experimental Psychology: Animal Learning and Cognition 40, 195–211. Chatlosh, D.L., Wasserman, E.A., 1993. Multidimensional stimulus control in pigeons: selective attention and other issues. In: Zentall, T.R. (Ed.), Animal Cognition: A Tribute to Donald A. Riley. Erlbaum, Hillsdale, NJ, pp. 271–292. Gibson, B.M., Wasserman, E.A., Frei, L., Miller, K., 2004. Recent advances in operant conditioning technology: a versatile and affordable computerized touchscreen system. Behavior Research Methods, Instruments, & Computers 36, 355–362. Gottselig, J.M., Wasserman, E.A., Young, M.E., 2001. Attentional tradeoffs in pigeons learning to discriminate newly-relevant visual stimulus dimensions. Learning and Motivation 32, 240–253. James, W., 1890/1955. The Principles of Psychology, vol. 1. Dover, New York. Jennings, H.S., 1906/1976. Behavior of the Lower Organisms. Indiana University Press, Bloomington. Kirkpatrick-Steger, K., Wasserman, E.A., 1996. The what and the where of the pigeon’s processing of complex visual stimuli. Journal of Experimental Psychology: Animal Behavior Processes 22, 60–67. Kirkpatrick-Steger, K., Wasserman, E.A., Biederman, I., 2000. The pigeon’s discrimination of shape and location information. Visual Cognition 7, 417–436.
Y. Teng et al. / Behavioural Processes 112 (2015) 61–71 Kruschke, J.K., Johansen, M.K., 1999. A model of probabilistic category learning. Journal of Experimental Psychology: Learning, Memory, and Cognition 25, 1083–1119. Lamb, M.R., Riley, D.A., 1981. Effects of element arrangement on the processing of compound stimuli in pigeons (Columba livia). Journal of Experimental Psychology: Animal Behavior Processes 7, 45–58. Pashler, H.E., 1998. The Psychology of Attention. MIT Press, Cambridge, MA. Pearce, J.M., 2002. Evaluation and development of a connectionist theory of configural learning. Animal Learning & Behavior 30, 73–95. Rescorla, R.A., 1973. Evidence for the “unique stimulus” account of configural conditioning. Journal of Comparative and Physiological Psychology 85, 331–338. Rescorla, R.A., Wagner, A.R., 1972. A theory of Pavlovian conditioning: variations in the effectiveness of reinforcement and nonreinforcement. In: Classical Conditioning II: Current Theory and Research. Appleton-Century-Crofts, New York, pp. 64–99. Reynolds, G.S., 1961. Attention in the pigeon. Journal of the Experimental Analysis of Behavior 4, 203–208. Riley, D.A., Leith, C.R., 1976. Multidimensional psychophysics and selective attention in animals. Psychological Bulletin 83, 138–160. Riley, D.A., Roitblat, H.L., 1978. Selective attention and related cognitive processes in pigeons. In: Hulse, S.H., Fowler, H., Honig, W.K. (Eds.), Cognitive Processes in Animal Behavior. Erlbaum, Hillsdale, NJ, pp. 249–276. Soto, F.A., Wasserman, E.A., 2010. Integrality/separability of stimulus dimensions and multidimensional generalization in pigeons. Journal of Experimental Psychology: Animal Behavior Processes 36, 194–205.
71
Soto, F.A., Wasserman, E.A., 2011. Asymmetrical interactions in the perception of face identity and emotional expression are not unique to the primate visual system. Journal of Vision 11, 1–18. Sutherland, N.S., Mackintosh, N.J., 1971. Mechanisms of Animal Discrimination Learning. Academic Press, New York. Thomas, D.R., 1970. Stimulus selection, attention, and related matters. In: Reynierse, J.H. (Ed.), Current Issues in Animal Learning. University of Nebraska Press, Lincoln, NE, pp. 311–356. Trabasso, T., Bower, G.H., 1968. Attention in Learning. Wiley, New York. Vyazovska, O.V., Teng, Y., Wasserman, E.A., 2014. Attentional tradeoffs in the pigeon. Journal of the Experimental Analysis of Behavior 101, 337–354. Wasserman, E.A., Biederman, I., 2012. Recognition by components: a bird’s eye view. In: Lazareva, O.F., Shimizu, T., Wasserman, E.A. (Eds.), How Animals See the World. Oxford University Press, New York. Wasserman, E.A., Frank, A.J., Young, M.E., 2002. Stimulus control by same versus different relations among multiple visual stimuli. Journal of Experimental Psychology: Animal Behavior Processes 28, 347–357. Wasserman, E.A., Miller, R.R., 1997. What’s elementary about associative learning? Annual Review of Psychology 48, 573–607. Whitlow Jr., J.W., Wagner, A.R., 1972. Negative patterning in classical conditioning: summation of response tendencies to isolable and configural components. Psychonomic Science 27, 299–301. Zentall, T.R., 2012. Selective and divided attention in birds. In: Lazareva, O.F., Shimizu, T., Wasserman, E.A. (Eds.), How Animals See the World. Oxford University Press, New York, pp. 351–369.
