Overshadowing and Blocking as Acquisition De cits: No Recovery After Extinction of Overshadowing or Blocking Cues

Transcription

1 q THE QUARTERLY JOURNAL OF EXPERIMENTAL PSYCHOLOGY, 1999, 52B (4), 307±333 Overshadowing and Blocking as Acquisition De cits: No Recovery After Extinction of Overshadowing or Blocking Cues Peter C. Holland Duke U niversity, Durham, N orth Carolina, U.S.A. The effects of extinction of the blocking or overshadowing stimulus on conditioned responding controlled by the blocked or overshadowed stimulus were examined in seven appetitive conditioning experiments with rats. The experiments differed in their designs, stimuli used, the amounts of conditioning and extinction training, and the levels of conditioned responding produced. In all cases, conditioned responding to the blocked or overshadowed cue was either unaffected or reduced by extinction of the blocking or overshadowing cue. These data are consistent with accounts of overshadowing and blocking that attribute those phenomena to acquisition de cits, rather than to retrieval failures. Animals are frequently faced with multiple stimuli, with different histories and relevance to current behavioural demands. A key problem is the selection of some stimuli to be used to guide learning and action, while deselecting or ``tuning out others. Issues of stimulus selection have been central to the study of simple associative learning processes for decades, and they have motivated the formulation of most current conditioning theories. Two widely studied conditioning phenomena, overshadowing and blocking, provide standard examples of stimulus selection. Overshadowing describes the common observation that when two stimuli are conditioned in compound, each (especially the less salient one) comes to control less conditioned responding than if it were conditioned separately (Pavlov, 1927). In blocking, if conditioning of a compound stimulus (AX) is preceded by conditioning of one of the elements (A), then the other element (X) is less able to control behaviour than in the absence of prior A training (Kamin, 1969). Thus, the ability to use a particular cue in the control of behaviour depends on the presence of other cues in training, as well as on the experimental histories of other cues in the array. Requests for reprints should be sent to Peter Holland, Department of Psychology: Experimental, Duke University Box 90086, Durham NC 27708±0086, U.S.A. pch@duke.edu This research was supported in part by Grants BNS±85±13603 and IBN±92±12267 from the National Science Foundation, and Grants MH±35554 and MH53667 from the National Institute of Mental Health. Thanks to John Chen, Marie Crock, Stephanie Nevels, and Flavia Saraceni for their technical assistance The Experimental Psycholog y Society

2 308 HOLLAND Early approaches to stimulus selection stressed the notion of limited capacity processing systems (e.g. Broadbent, 1958). Within this perspective, animals can not process all available stimuli simultaneously, and thus some preselection process is necessary to prevent overloading these systems. Early considerations of stimulus selection in animal conditioning also capitalized on the idea of limited capacity. Sutherland and Mackintosh (1971) suggested that the formation of associations between two events demanded the directing of attention to those events, and further more, that attentional processing was severely limited. Thus, overshadowing occurs with compound conditioned stimuli (CSs) because attention must be divided among the available cues, leaving less opportunity for each cue to be associated with the unconditioned stimulus (US). Blockin g was accommodated within this system by the further assumption that CS±US pairings not only strengthen the CS s association with the US, but also increase the tendency to direct attention toward that particular CS. Animals that had prior pairings of one CS (A) with the US come to compound conditioning directing most of their available attentional resources to A, leaving little capacity to attend to X. Because X commands relatively little attention, the formation of X±US associations is blocked. Later approaches to stimulus selection in conditioning questioned the applicability of such processing limitations to typical animal conditioning studies, in which the elements of a compound might be, for example, the illumination of a signal light and the presentation of a loud noise. Some of these theories (e.g. Mackintosh, 1975; Pearce & Hall, 1980) proposed that animals actively learn to exclude overshadowed and blocked cues from attentional processing. For example, Mackintosh (1975) claimed that on each conditioning trial, attention to a stimulus is increased if it is the best available predictor of the trial s outcome, and it is decreased if it is not. Thus, in a blocking experiment, prior training of A ensures that it is the best predictor of the US available when the AX compound is rst presented. Because the initially neutral X will then be a poorer predictor of the US, attention to that cue will be diminished, reducing the chance that it will be associated with the US. Similarly, in an overshadowing experiment, initial differences in salience or intensity cause one of the cues to be associated with the US more rapidly than others. Attention to that cue increases, while attention to other cues present decreases. Another approach to stimulus selection issues in conditioning rejected not only the notion of limited processing capacity, but also the idea that phenomena like overshadowing and blocking are mediated by changes in attention to the various CSs. Rescorla and Wagner (1972) proposed instead that outcomes like these were mediated by changes in the effectiveness of the US as a reinforcer. The effectiveness of a reinforcer on any given conditioning trial was de ned as the discrepancy between the conditioning strength supportable by a US and that already exhibited by all the CSs present on that trial. Thus, a US is effective as a reinforcer for further conditioning on a trial only if it is not already well predicted by the CSs present on that trial. In a blocking experiment, prior training of A endows it with conditioning strength nearly equal to that supportable by the US. Consequently, the AX compound is paired with a US that is ineffective as a reinforcer, and the initially neutral X has little opportunity for conditioning. Similarly, in an overshadowing experiment, the more salient element of a compound cue is rapidly associated with the US, rendering that US increasingly ineffective as a reinforcer. As a result the weaker element has less opportunity for conditioning than if it had been separately paired with the US.

3 OVERSHADOWING AND BLOCKING 309 Finally, some theorists interpreted many presumed stimulus selection phenomena as examples of stimulus compounding or con gural learning (e.g. Kehoe & Gormezano, 1980; Pearce, 1987, 1994). Typically, within these frameworks, the combination of individu al stimulus elements produces unique stimulus con gurations, which are distinct from those individu al elements. Effects of training of individual elements on responding to the compound, or vice versa, are derived from generalization functions, rather than attributed to variations in processing of the various CS or US events. Within the theory offered by Pearce (1987), for example, overshadowing is seen simply as the consequence of generalization decrement: Animals trained with one cue (AX) are then tested with another cue (X) that resembles it to some extent, but has no special status as an ``element of the original stimulus. Within Pearce s theory, the origin of blocking is somewhat more involved, demanding both generalization and variations in the effectiveness of USs, much as speci ed by the Rescorla±Wagner model. Initial training of A generalizes to the AX compound cue to some extent. Because that generalized strength allows AX to predict the US reasonably well, only small increments in the strength of AX are possible in the compound conditioning phase, making even less associative strength available for generalization to the test cue, X. All of these conditioning theories attribute stimulus selection phenomena like blocking and overshadowing to selective acquisition of conditioned responses (CRs). Within them, only the most salient or predictive stimuli available at the time of learning will acquire appreciable conditioned responding; little is learned about blocked and overshadowed cues. The apparent selectivity of animals responding at the time of performance is derived solely from these biases acquired at the time of learning; stimuli govern action only to the extent that they have acquired conditioned responses. Doubt is cast on these interpretations of overshadowing and blocking by studies that showed that various manipulations conducted after compound conditioning could reveal responding controlled by otherwise overshadowed or blocked cues. Kaufman and Bolles (1981), Matzel, Schachtman, and M iller (1985), and R.R. M iller, Barnet, and Grahame (1992), using aversive conditioned suppression procedures with rats, found that extinction of the overshadowing stimulus after compound conditioning substantially reduced overshadowingðthat is, it enhanced CRs to the otherwise overshadowed stimulus. Similarly, Kasprow, Cacheiro, Balaz, and M iller (1982) and Balaz, Gutsin, Cacheiro, and Miller (1982) found that overshadowed or blocked responding could be revived by presenting various ``reminder events, such as the US, prior to testing for evidence of these stimulus selection phenomena. Those investigators concluded that animals failure to respond to overshadowed and blocked cues could not re ect a failure to learn about those cues, but must instead indicate a performance or retrieval de cit, which prevented the translation of learned associations into action. Three general approaches to dealing with a reinterpretation of overshadowing and blocking as retrieval effects have been suggested. The rst approach (e.g. Balaz et al., 1982; Kasprow et al., 1982) simply shifts the role of attentional and informational variables to retrieval rather than acquisition. Variations in attention to particular CSs or variations in the effectiveness of USs, as speci ed by existing theories of stimulus selection, affect the future rerieval or expression of learning rather than the current acquisition of associations. For example, in the context of Mackintosh s (1975) theory, learning to

4 310 HOLLAND ignore a blocked stimulus impairs the ability to attend to that cue at the time of performance, rather than the ability to learn about it in acquisition. Or, within the Rescorla± Wagner model, a well-predicted US is dif cult to retrieve for the production of behaviour later, rather than being ineffective as a reinforcer for current learning. A second approach (M atzel et al., 1985) invokes comparator theory (e.g. R.R. Miller & Schachtman, 1985), originally proposed to deal with aspects of contingency learning in animals. Within comparator theory, CR performance in the presence of a CS depends not only on its own associative strength, but also on the current associative strength of ``comparator stimuli Ðcues that accompanied the CS in the past. Roughly, performance to a CS is a function of the difference between its own strength and that of its comparators. In an overshadowing experiment, the overshadowed CS acquires associations with the US, but fails to reveal them in performance because the associative strength of its comparator, the overshadowing stimulus, is at least as great. However, post-training extinction of the overshadowing stimulus lowers the strength of the overshadowed cue s comparator, which allows the former cue to generate greater performance. Finally, Kaufman and Bolles (1981) suggested that animals learn about both contiguity and causality when they are exposed to conditioning episodes, but that the performance of CRs re ects only the knowledge of causal relations. In an overshadowing experiment, animals learn the contiguity relation between both cues and the US, but assign causality primarily to the stronger cue. Thus, the overshadowed CS fails to control behaviour. However, if the causal relation between the overshadowing cue and the US is diminished by extinction, then, by contrast, the causal value of the overshadowed cue is raised, permitting it to control behaviour. The last two accounts relate Kaufman and Bolle s (1981), Matzel et al. s (1985), and R.R. Miller et al. s (1992) results to another puzzling nding from experiments of human contingency learning, backward blocking (e.g. Shanks, 1985; see J.S. Miller, McKinzie, Kraebel, & Spear 1996, for a discussion of conditions under which this phenomenon may be observed in rats). In backward blocking, subjects rst learn about causal relations between a compound cue and some consequence. Later, some subjects learn about the causal relation of one of its elements and that consequence. In a nal assessment, those subjects report weaker causal relations between the other element of the original compound than do subjects that did not receive exposure to the single element. Within Kaufman and Bolles s (1981) view, this outcome could be described as the result of subjects using the new causal information to re-evaluate the causal relations imputed in the compound stimulus phase. Within comparator theory (R.R. M iller & Schachtman, 1985), this outcome is easily described as resulting from an enhancement of the value of the (backward blocking) comparator cue, which reduces the ability of the other (backward blocked) cue to control behaviour. In any case, reinterpretation of many stimulus selection effects as re ecting processing carried out after the time of learning, rather than as variations in the acquisition of associations, represents a major change in the way learning theorists must view these phenomena and perhaps the very nature of conditioning itself. The seven experiments reported here grew from a wish to study these post-conditioning or retrieval processes and to reconcile the results of Kaufman and Bolles (1981), Matzel et al., (1985), and R.R. M iller et al. (1992) with those of other investigators who have

5 OVERSHADOWING AND BLOCKING 311 reported contradictory data. Rescorla and his colleagues (e.g. Rescorla & Cunningham, 1978; Speers, Gillan & Rescorla, 1980) found that extinction of avour aversion conditioning to one element of a previously conditioned compound reduced (rather than enhanced) CRs to the unextinguished element. Likewise, Holland (1984, Experiment 2B) reported no effect of extinction of a blocking or overshadowing cue on responding to the blocked or overshadowed cue in an appetitive conditioning preparation. The experiments described in this article examined the effects of non-reinforced or reinforced presentations of one element of a previously trained compound stimulus on responding to the other element of that compound in overshadowing and blocking procedures, using the appetitive conditioning preparation of Holland (1984). The experiments differed in their designs, stimuli used, the amounts of conditioning and extinction training, and the levels of conditioned responding produced, in an attempt to determine the boundary conditions for obtaining effects like those reported by Kaufman and Bolles (1981), Matzel et al. (1985), and R.R. Miller et al. (1992). None of the experiments showed evidence for enhancement of responding to overshadowed or blocked stimuli by post-training extinction of the overshadowing or blocking cues. EXPERIMENT 1 Experiment 1 examined the effects of post-conditioning extinction of an auditory overshadowing cue (A) on responding to a visual overshadowed cue (X), using a design modelled after that of Matzel et al. (1985, Experiment 3). Rats in two overshadowing groups rst received pairings of a compound cue (AX) with food, intermixed with pairings of another auditory cue (B) with food. Two groups of control rats received X±food and B±food pairings. After this training, one overshadowing group and one control group received repeated non-reinforced presentations of the overshadowing stimulus (A), and the other groups received comparable extinction of the other stimulus (B). Finally, responding to X alone was examined in all rats. (Table 1 shows outlines of the designs of Experiments 1±7.) If performance controlled by an overshadowed cue depends on a comparison of the CR strength of the overshadowed cue to that of the cue that accompanied it in compound training, then among the rats that received overshadowing (AX±food) training, the rats that received extinction of the overshadowing stimulus (A) should respond more to X in testing than the rats that received extinction of the separately trained cue (B). Method Subjects The subjects were 16 male and 16 female, 90±100-day-old, experimentally naive rats of Sprague- Dawley descent, bred in a Duke University psychology department facility. Equal numbers of male and female rats were assigned to each treatment condition. The rats were individu ally housed in stainless steel cages with ad lib access to water and were maintained at 80% of their ad libitum weights by limiting their access to food. The vivarium in which the rats were housed was maintained at 218 C, with the lights on from 0600 to 2000 hr daily. All experimental sessions were carried out during the light portion of the cycle, between 0800 and 1600 hr.

6 312 TABLE 1 Outlines of Procedures of Experiments 1± 7 Experiment Group Pretraining Compound Training Post-training Test 1 O±E AX+, B+ A2 X? O±N AX+, B+ B2 X? C±E X+, B+ A2 X? C±N X+, B+ B2 X? 2 O±E A2, B+ AX+ A2 X? O±N A2, B+ AX+ B2 X? B±E A+, B+ AX+ A2 X? B±N A+, B+ AX+ B2 X? 3 O±E A2, B+ AX+ A2, B+ X? O±R A2, B+ AX+ A+, B2 X? B±E A+, B+ AX+ A2, B+ X? B±R A+, B+ AX+ X+, B2 X? 4 O±E AX+ A2 X? O±N AX+ Ð X? C±E X+ A2 X? C±N X+ Ð X? 5 O±E + AX+ A2 X? O±N + AX+ Ð X? B±E A+ AX+ A2 X? B±N A+ AX+ Ð X? 6 C±E B+ X+ A2 X? C±N B+ X+ Ð X? O±E B+ AX+ A2 X? O±N B+ AX+ Ð X? B±E A+ AX+ A2 X? B±N A+ AX+ Ð X? 7 08±E 8 trials AX+ A2 X? 08±N 8 trials AX+ Ð X? 016±E 16 trials AX+ A2 X? 016±N 16 trials AX+ Ð X? 0128±E 128 trials AX+ A2 X? 0128±N 128 trials AX+ Ð X? C8±E 8 trials X+ A2 X? C8±N 8 trials X+ Ð X? C16±E 16 trials X+ A2 X? C16±N 16 trials X+ Ð X? C128±E 128 trials X+ A2 X? C128±N 128 trials X+ Ð X? Note: O = overshadowing, B = blocking, C = control, E = extinction, N = no extinction, + = food reinforcement, 2 = no reinforcement

7 OVERSHADOWING AND BLOCKING 313 Apparatus The apparatus consisted of eight individu al chambers ( cm) with aluminum front and back walls, clear acrylic sides and top, and a grid oor (0.48-cm stainless steel rods spaced 1.9 cm apart). A dimly illuminated food cup ( cm) was recessed in the centre of one end wall, behind a cm opening. Each experimental chamber was enclosed in a sound-resistant shell with an acrylic window for viewing the rats. A 6-W lamp was mounted on the inside wall of the shell, aligned with the end wall that contained the food cup, 10 cm above the top of the chamber. The illumination of this lamp served as the visual (light) CS. A speaker, used to present the auditory CSs, was mounted next to the CS lamp. Ventilation fans provided masking noise (70 db), and a 6-W, 110-V lamp (operated at 75 V) behind a red lens opposite the CS lamp provided continuous dim background illumination. Two low-light television cameras were mounted 2.1 m from the experimental chambers so that each included four chambers in its eld of view. Video cassette recorders were programmed to record behaviours that occurred during the 10-sec intervals before, during, and after CS presentations. Behavioural Observations All observations were made from videotapes and paced by auditory signals recorded on the tapes. For each rat, observations were made at 1.25-sec intervals during the 5-sec period im mediately prior to CS presentations and during CS presentations. At each observation, one and only one behaviour was recorded; the behavioural categories were de ned to be mutually exclusive. The primary measure of learning, food cup behaviour, is a composite of several kinds of behaviour directed toward the food cup: (1) head jerk: short, rapid, horizontal and/or vertical movements of the head, with the nose outside the recessed food cup, but usually directed toward the food cup; (2) magazine/ head jerk: head jerk behaviour, with the nose inside the recessed food cup, or active licking/chewing of the food cup; and (3) m agazine: nose within the recessed food cup, but no head jerk or licking/chewing behaviour. The rst two subcategories of behaviour are typically observed to occur to auditory CSs, whereas magazine behaviour occurs as a component of the CR to both auditory and visual CSs. Rearing behaviour (typically unique to visual CSs)Ðstanding on hind legs with front feet off the oor and not groomingðwas also recorded. Finally, all other behaviour was recorded as other. The index of behavioural frequency used was percentage total behaviour, obtained by dividin g the frequency of the target behaviour (food cup, rear, or other) in any observation interval by the total number of observations made in that interval. Because the rate of observations was constant within each observation interval throughout the experiments, this measure is an absolute frequency measure; it is not affected by the overall activity level of a subject. Two observers independently scored the behavioural data reported in each experiment. At least one observer was unaware of the rats conditioning histories when the data were scored. Using the behavioural categories reported here (food cup, rear, and other), each pair of observers agreed on 95% of over 12,000 joint observations. Judgements of the subcategories of food cup behaviour suffered from lower levels of interobserver agreement, which contributed to the decision to combine them. Visual and auditory CSs paired with food typically come to evoke distinct patterns of conditioned behaviour (e.g. Holland, 1977). With localizable visual CSs, rear behaviour occurs im mediately after CS onset, but is replaced by behaviour directed toward the food cup as the time of food delivery nears. Typically, only low levels of rear behaviour occur during the latter half of 10-sec visual cues, like those used in these experiments (e.g. Holland, 1977, 1980). In contrast, auditory CSs typically generate a rapid jump or startle response, followed by food cup behaviours that are more evenly

8 314 HOLLAND distributed across the CS±US interval. In an attempt to provide more comparable measures of conditioned responding during auditory and visual CSs, in these experiments the primary measure of conditioning used was the level of food cup behaviours during the last 5-sec period of all CSs. It is worth adding that in all cases, analysis of rear behaviour observed during the rst half of the visual CSs con rmed the results reported for second-half food cup behaviour; these data are not reported, however. Procedures All rats were rst trained to eat from the food cups in a single 20-min session in which deliveries of the food US were given on a variable-time 2-min schedule. The event used as the food US throughout these experiments was the delivery of two 45-mg food pellets, separated by 0.5 sec. Next, all rats received eight 90-min sessions in a compound training phase. In each of these sessions, the rats in group O±E (n = 8) and O±N (n = 8) received four 10-sec presentations of a compound of the light and one of the auditory cues, paired with food (AX+), and four 10-sec presentations of the other auditory stimulus paired with food (B+). The rats in groups C±E (n = 8) and C±N (n = 8) received four 10-sec presentations of each of two auditory cues, separately paired with food (A+ and B+), in each of these sessions. For half of the rats in each group, A was a 78-dB white noise stimulus, and B was a 72-dB, 8-Hz clicker; for the other half of the rats the identities of A and B were reversed. The trial sequences and intertrial intervals (mean = 675 sec, rectangularly distributed, 338±1,350 sec) were randomized daily. All rats then received eight 90-min sessions in which responding to one of the auditory CSs, A or B, was extinguished. Each session comprised eight 10-sec presentations of either the overshadowing stimulus (A; groups O±E and C±E) or the other, separately trained, CS (B; groups O±N and C±N). Finally, all rats received a single 90-min test session, which included eight 10-sec presentations of the light CS alone (X). Results and Discussion The rats rapidly acquired conditioned responding to the various CSs in the rst training phase. In the nal session, food cup behaviours comprised 72.7% and 76.6% of the total behaviour during AX in groups O±E and O±N, respectively, and 62.5% and 64.1% during X in groups C±E and C±N. Food cup responding during B ranged from 69.5% to 74.2% among the four groups. Responding to the noise and clicker CSs was equivalent. Pre-CS food cup behaviour ranged from 5 to 10% of total behaviour. Likewise, extinction proceeded rapidly; by the nal extinction session, responding during the A or B cues ranged from 6.6% to 11.7%, with pre-cs food cup behaviour less than 5%. Conditioned responding elicited by X extinguished rapidly in the test session in this and all subsequent experiments, so differences among the groups were more visible in the early stages of testing. Consequently, test data presented for this and subsequent experiments are from the rst four test trials only. Figure 1 shows performance averaged over the rst four test trials in Experiment 1. Overshadowing occurred: Responding to X was reliably lower in group O, which had received AX±food pairings, than in group C, which had received X±food pairings. However, unlike in Matzel et al. s (1985) studies, there was no reliable effect of post-training extinction of the overshadowing stimulus (A) on responding to X, relative to the effects of extinguishin g a separately trained excitor, B. Pre-CS behaviour was less than 5% in all groups.

9 OVERSHADOWING AND BLOCKING 315 A training (O or C) by extinction (A or B) analysis of variance (ANOVA) showed a reliable effect of group, F(1, 28) = 15.39, p <.001, but no overall effect of extinction, F(1, 28) < 1, or group by extinction interaction, F(1, 28) = 1.71, p =.202. Although Figure 1 suggests that extinction of the overshadowing stimulus (A) may have reduced responding to X, a planned comparison showed no reliable difference between the performance of groups O±E and O±N, F(1, 28) = 1.28, p =.268. It might be argued that the normally facilitatory effects of comparison processes after extinction of the overshadowing stimulus (A) were masked by larger, decremental effects of extinguishing associations between A and X (Rescorla & Cunningham, 1978). From this perspective, the substantial responding observed to X in the overshadowing groups may have been largely the consequence of the formation of X±A associations, rather than X±food associations, during compound training. That is, X s ability to elicit CRs may have been mediated by its activation of a representation of A, which in turn generated CRs. Extinction of A s ability to elicit CRs would then reduce X s ability to do so as well. Alternatively, as a consequence of A±X associations established during compound training, presentations of A in extinction might generate surrogate extinction of X (e.g. Holland & Forbes, 1982; Hall, 1996) by evoking a representation of X in the absence of reinforcement. In either case, it is possible that expression of X±US associations was indeed enhanced by extinction of A, but this enhancement was obscured by a larger, decremental effect of A extinction on responding mediated by X±A associations. Experiment 2 examined the effects of A extinction under conditions intended to reduce the importance of A±X associations. EXPERIMENT 2 Experiment 1 failed to reveal evidence for comparison processes in the expression of overshadowing. Experiment 2 examined the effects of extinction on both overshadowing and blocking of conditioning to a visual cue (X) after training designed to reduce the FIG. 1. Food cup behaviour during the light (X) in the test phase of Experiment 1. O = overshadowing treatment; C = control treatment; E = extinction of overshadowing cue (A) prior to the test; N = no extinction of overshadowing cue prior to the test.

10 316 HOLLAND contribution of within-compound associations. All of the rats received training in which a compound stimulus, AX, was paired with food, followed by extinction of either A or another cue, B. Prior to this compound training, rats in the blocking condition received both A±food and B±food pairings, and rats in the overshadowing condition received nonreinforced A presentations and B±food pairings. Thus, all rats received separate presentations of A prior to compound training, which Rescorla (1982) and Holland (1985) have shown to hinder the formation of within-compound associations involvin g that cue. With a reduced contribution of X±A associations to X s behavioural control, facilitatory effects of extinction of the overshadowing or blocking cues on responding to X might be more readily detected. Methods Subjects and Apparatus The subjects were 16 male and 16 female, 90±100-day-old, experimentally naive rats of Sprague- Dawley descent, bred in a Duke University psychology department facility, maintained in the same way as in Experiment 1. Equal numbers of male and female rats were assigned to each treatment condition. All experimental sessions were carried out between 0800 and 1400 hr. The apparatus was the same as that used in Experiment 1. Procedures After food cup training identical to that of Experiment 1, the rats received eight 90-min pretraining sessions, each of which included four 10-sec presentations of the clicker and four 10-sec presentations of the white noise, randomly intermixed. In the overshadowing groups (O±E and O±N), one of those cues (B) was paired with food, and one was presented without reinforcement (A). In the blocking groups (groups B±E and B±N), both A and B were (separately) paired with food. The identities of A and B (clicker or noise) were counterbalanced within each group. Then, all rats received four 90-min sessions of AX compound training. In each of these sessions, there were eight 10-sec presentations of a compound of the light and the auditory cue previously designated as A, each paired with food delivery. Next, all rats received eight 90-min extinction sessions in which there were eight non-reinforced 10-sec presentations of either A (groups O±E and B±E) or B (groups O±N and B±N). Finally, all rats received a single 90-min test session in which there were eight non-reinforced 10-sec presentations of the light. Results and Discussion As in Experiment 1, the rats rapidly acquired conditioned responding to the various CSs in the pretraining phase. In the nal session, responding to B+ averaged 75.9%, responding to A+ averaged 73.4%, and responding to A2 averaged 24.2%; there were no effects of counterbalancing. Pre-CS responding ranged between 5 and 10%. After some initial disruption, compound training was rapid; by the nal compound training day, responding to AX averaged 63.8% in groups O±E and O±N, and 74.6% in groups B±E and B±N, with pre-cs behaviour between 5% and 10% in both groups. Extinction proceeded rapidly; by the nal extinction session, responding during the A or B cues ranged from 5.1% to 9.4%, with pre-cs food cup behaviour less than 5%.

11 OVERSHADOWING AND BLOCKING 317 Figure 2 shows responding to X in the rst four trials of the test session of Experiment 2. Blocking occurred; that is, responding to X was lower when AX compound training was preceded by A±food pairings (groups B±E and B±N) than when it was not (groups O±E and O±N). However, extinction of the blocking stimulus (A, in group B±E) had no effect on responding to X, relative to extinguishing a separately trained excitor (B, in group B±N). Furthermore, as in Experiment 1, extinction of the overshadowing stimulus (A) after AX±food training had no reliable effect on responding to X, despite an attempt to reduce the potential contributions of X±A associations to that responding. Pre-CS behaviour was less than 5% in all groups. A group (B or O) by extinction (A or B) ANOVA showed a reliable effect of group, F(1, 28) = 13.95, p <.001, but no signi cant effect of extinction, F(1, 28) < 1, or group by extinction interaction, F(1, 28) < 1. Planned individ ual comparisons showed no reliable effects of the extinction treatment on responding to X in either the blocking or the overshadowing conditions, Fs < 1. An examination of the form of responding observed to the X and A cues also revealed no evidence for a substantial role of X±A associations in responding to X. Using visual and auditory cues similar to those used here, Holland (1985) and Morell (1997) found that the formation of light±noise associations produced responding to the light of a form normally produced by the noise (head jerk), rather than that normally generated by the light (rear and more passive food cup behaviours). No evidence for the occurrence of head jerk behaviour during the visual X cues was observed in this experiment (nor in Experiment 1 or subsequent experiments in this series). EXPERIMENT 3 It might be argued that the failure to observe facilitatory effects of extinction in Experiments 1 and 2 was the result of insuf cient extinction (64 extinction trials, after 32 compound conditioning trials). Matzel et al. (1985) gave 72 extinction trials, after only FIG. 2. Food cup behaviour during the light (X) in the test phase of Experiment 2. B = blockin g treatment; O = overshadowing treatment; E = extinction of blocking/overshadowing cue (A) prior to the test; N = no extinction of blocking /overshadowing cue prior to the test.

12 318 HOLLAND 8 compound conditioning trials. In Experiment 3, twice as many extinction trials (128) were given as in the previous experiments. In addition, in Experiment 3, the effects of post-training extinction trials with A were contrasted with the effects of post-training reinforced conditioning trials with A, as in backward blocking (Shanks, 1985). Thus, after pretraining and AX+ compound training like that given in Experiment 2, rats received either A2, B+ or A+, B2 post-training before testing of X. Within comparator theory, reinforced training of A might make X even less able to control responding, by increasing the value of X s comparison stimulus. Thus, a comparison of the effects of reinforced and non-reinforced post-training presentations of A might be an especially sensitive test of predictions from comparator theory. By contrast, to the extent that within-compound associations contribute to X s power to control responding (Rescorla & Cunningham, 1978; Speers et al., 1980), reinforced post-training presentations of A might have mixed eftfects. On the one hand, reinforcement of A might increase its value somewhat, and thus enhance responding to X, but on the other hand, repeated presentation of A in the absence of X might weaken the withincompound, X±A associations. Methods Subjects and Apparatus The subjects were 16 male, 240-day-old, and 16 female, 300-day-old, experimentally naive rats of Sprague-Dawley descent, bred in a Duke University psychology department facility, maintained in the same manner as in the previous experiments. Equal numbers of male and female rats were assigned to each treatment condition. All experimental sessions were carried out between 0800 and 1400 hr. The apparatus was the same as that used in Experiments 1 and 2. Procedures After food cup training identical to that of Experiments 1 and 2, the rats received a pretest of responding to the two auditory cues to be used as A and B. In the rst 30 min of a 60-min session, all rats received two non-reinforced 10-sec presentations of a 1,500-Hz tone (72 db) and two 10-sec presentations of the white noise. In the remaining half of that session, the rats received two 10-sec presentations of the tone and two 10-sec presentations of the white noise. In the overshadowing groups (O±E and O±R), one of those cues (B) was paired with food, and one was presented without reinforcement (A). In the blocking groups (groups B±E and B±R), both A and B were (separately) paired with food. The identity of A and B (tone or noise) was counterbalanced within each group. This pretraining continued for eight additional 60-min sessions, each of which included four A and four B trials, randomly intermixed, and the rst half of another 60-min session, identical to the last half of the initial session. Thus, the rats received a total of 36 A and 36 B presentations in the pretraining phase. The compound conditioning phase began immediately after the nal half-session of the pretraining. In the 30-min remainder of that session, all rats received four reinforced 10-sec presentations of a compound of the light (X) and the auditory A cue (tone or noise, counterbalanced). In each of four additional 60-min sessions, all rats received eight reinforced AX trials.

13 OVERSHADOWING AND BLOCKING 319 Next, all rats received min post-training sessions. In groups O±E and B±E, each of these sessions contained four non-reinforced A presentations and four reinforced B presentations; in groups O±R and B±R, each of these sessions included four reinforced A presentations and four non-reinforced B presentations. Finally, all rats received two 60-min sessions to test conditioned responding evoked by the light (X) cue. The rst half of each session comprised four non-reinforced presentations of the 10-sec light, as in the previous experiments, and the second half included four reinforced presentations of the light. Thus, the rst half of the rst test session provided a performance test identical to those used in Experiments 1 and 2, and the remaining trials provided a variation of a savings test, which permitted more extended assessment of conditioning to the light. Unlike in Experiment 2, the overshadowing groups (O±E and O±R) and the blocking groups (B±E and B±R) were not run concurrently. Thus, comparisons between the performances of blocking and overshadowing groups must be viewed cautiously. However, the primary comparisons in Experiment 3 were the contrasts of the effects of the two post-training manipulations within each of the two training procedures (blocking and overshadowing). Results and Discussion In the nal half-session of pretraining, responding to B+ ranged from 68.8% to 75.0%, responding to A+ averaged 75.0%, and responding to A2 averaged 32.5%; there were no effects of counterbalancing. Pre-CS responding was in the range 5±10%. After some initial disruption, compound training was rapid; by the nal compound training day, responding to AX averaged 66.3% in groups O±E and O±R, and 75.0% in groups B±E and B±R, with pre-cs behaviour between 5 and 10% in both groups. Extinction proceeded more slowly than in Experiments 1 and 2; recall that in this experiment, extinction of one auditory cue was interspersed with continued reinforcement of the other auditory cue. In the nal extinction session, responding ranged from 77.3% to 80.5% during the reinforced cue, and from 5.1% to 8.6% during the extinguished cue, with pre-cs food cup behaviour less than 10%. Figure 3 shows responding to X over the rst four trials of the test session in Experiment 3. In both the blocking groups (B±E and B±R) and the overshadowing groups (O±E and O±R), extinction of the blocking/overshadowing stimulus reduced responding to X, compared to separate reinforcement of that cue. Although comparisons of performance of the rats in the blocking groups with those in the overshadowing groups are somewhat tenuous because those groups were not run concurrently, Figure 3 suggests that blocking occurred: Responding to X was lower if AX compound training had been preceded by A±food pairings than if it had not. Separate one-way ANOVAs for the overshadowing and blocking groups showed a reliable effect of extinction of both the overshadowing stimulus, F(1, 14) = 5.36, p =.036, and the blocking stimulus, F(1, 14) = 6.36, p =.024. An ANOVA of the data from both sets of groups combined showed a reliable effect of both training condition, F(1, 28) = 33.36, p <.001, and extinction treatment, F(1, 28) = 10.66, p =.003, but no signi cant interaction, F(1, 28) < 1. The reintroduction of reinforcement in the second half of the test session restored responding brie y; the pattern of responding in the second test session (not presented here) was similar to that observed in the rst. There was no evidence for acquisition of higher levels of responding during the light over the course of the two test sessions.

14 320 HOLLAND FIG. 3. Food cup behaviour during the light (X) in the test phase of Experiment 3. B = blockin g treatment; O = overshadowing treatment; E = extinction of blocking/overshadowing cue (A) prior to the test; R = reinforcement of blocking/overshadowing cue prior to the test. As in Experiments 1 and 2, the effects of the post-training manipulations in Experiment 3 were not consistent with the predictions of comparator theory. Experiment 3 showed no evidence for backward blocking; post-training conditioning of one of the elements failed to reduce responding to the other element, relative to post-training extinction. Alternatively, extended extinction of the putative comparator stimulus failed to enhance responding to the overshadowed or blocked stimulus, even in comparison with additional conditioning of the comparator cue. Indeed, relative to the effect of reinforced presentations of the overshadowing or blocking cues, non-reinforced presentations of those cues generated signi cantly less responding to the overshadowed or blocked cue, as might be anticipated if responding was affected by within-compound associations (Rescorla & Cunningham, 1978; Speers et al., 1980). Note that, as in Experiment 2, compound training was preceded by separate presentations of A, intended to reduce the likeliho od of within-compound associations. Evidently, this treatment was not particularly effective, suggesting caution in interpreting the results of Experiment 2 as indicating the effects of extinction in the absence of within-compound associations. EXPERIMENT 4 Experiments 1±3 all used designs in which the effects of extinction of the overshadowing or blocking cue were compared to the effects of extinguishing another, separately trained excitor. A potential problem in interpreting the lack of facilitatory effects of extinction in Experiments 1 and 2 is the possibility that the rats failed to discriminate between the two excitors (a noise and a clicker) during the extinction and test phases. In that case, the failure to observe selective effects of extinguishing the overshadowing cue might re ect equivalent recovery in both extinction and no-extinction groups rather than the absence of recovery. Although these two cues were selected for use because they typically generalize to each other only minimally in my laboratory, in Experiment 1 and in the blocking

15 OVERSHADOWING AND BLOCKING 321 groups of Experiment 2, generalization of extinction may have been enhanced because both were previously paired with the same food reinforcer (Honey & Hall, 1989). By contrast, this criticism does not apply to Experiment 3, in which the post-training manipulation involve d training an explicit discrimination between the two excitors. Nevertheless, to reduce the chance of failing to observe recovery because of a lack of speci city, a simpler design was adopted (e.g. Matzel et al., 1995, Experiment 1) for the remaining experiments. In these studies, the performance of subjects that received extinction of the overshadowing cue was compared to that of subjects that were merely exposed to the experimental context. Experiment 4 was similar to Experiment 1, except that it compared the effects of post-training extinction of an overshadowing cue to the effects of simple post-training context exposure. In addition, Experiment 4 examined the effects of using other stimuli as the overshadowing and overshadowed cues. In Experiments 1±3, the overshadowing cues were always auditory, and the overshadowed cue was always a light. For half of the rats in Experiment 4, the overshadowing cue was the light and the overshadowed cue was a white noise; for the other half of the rats, the roles of these two cues were reversed. Methods Subjects and Apparatus The subjects were 48 male, 120-day-old, experimentally naive rats of Sprague-Dawley descent, bred in a Duke University psychology departmentfacility, maintained in the same manner as that in the previous experiments. All experimental sessions were carried out between 0800 and 1600 hr. The apparatus was the same as that used in Experiments 1±3. Procedures After food cup training identical to that of Experiments 1±3, the rats received four 64-min training sessions. In each of these sessions, the rats in groups O±E and O±N (ns = 12) received eight 10-sec reinforced presentations of a white noise + light compound, and the rats in groups C±E and C±N (ns = 12) received eight reinforced presentations of one of those cues (X) alone. For half of the rats in each group, the white noise was designated as X, and for the other half, X was the light; in each case, the other cue was designated as A. Next, the rats in groups O±E and C±E received eight 10-sec non-reinforced presentations of A in each of sixteen 64-min post-training sessions. The rats in groups O±N and C±N were placed in the experimental chambers for these sessions, but received no explicit cues. Finally, all rats received a single 64-min test session, which included eight nonreinforced presentations of X. Results and Discussion The rats rapidly acquired conditioned responding to X and AX in the rst training phase. In the nal session, food cup behaviours comprised 61.5% and 63.5% of the total behaviour during AX in groups O±E and O±N, respectively. In groups C±E and C±N, food cup responding during X averaged 63.0% among the rats for which X was the noise, and 39.1% among the rats for which X was the light. Pre-CS food cup behaviour ranged

16 322 HOLLAND from 5 to 10% of total behaviour. In the next phase, extinction proceeded rapidly; by the nal extinction session, responding averaged 4.7% and 6.8% during the noise and light, respectively, with pre-cs food cup behaviour less than 5%. Figure 4 shows performance on the rst four trials of the test session in Experiment 4; the top panel shows responding of the rats for which X was the light, and the bottom panel shows responding of the rats for which X was the noise. More conditioned responding was displayed during the noise than during the light, both when a single stimulus was paired with food (groups C±E and C±N) and when the AX compound was paired with food (groups O±E and O±N). Furthermore, there was a marginally reliable tendency for the noise to overshadow conditioning to the light more than the light overshadowed conditioning to the noise. Most im portant for the purposes of this study, extinction of the overshadowing stimulus had no reliable effect on responding to the overshadowed FIG. 4. Food cup behaviour during the light (top panel) or noise (bottom panel) X cue in the test phase of Experiment 4. O = overshadowing treatment; C = control treatment; E = extinction of overshadowing cue (A) prior to the test; N = no extinction of overshadowing cue prior to the test.

17 OVERSHADOWING AND BLOCKING 323 cue, regardless of the level of responding controlled by that stimulus, or the size of the overshadowing effect. A group (O or C) by extinction (E or N) by cue (light or noise) ANOVA showed a reliable effect of group, F(1, 40) = , p <.001, and cue, F(1, 40) = , p <.001, and a marginal group by cue interaction, F(1, 40) = 3.058, p =.088. No other effects or interactions were reliable, Fs < 1. Separate comparisons for each target stimulus showed reliable overshadowing (group) effects with both light, F(1, 40) = , p <.001, and noise, F(1, 40) = 4.545, p =.039, but no effects of extinction or group by extinction interactions with either cue, Fs < 1. Even with this simple design, then, there was no evidence for enhancement of responding to the overshadowed cue by extinction of the overshadowing cue, regardless of the choice of cues, the level of responding maintained by those cues, and the magnitude of overshadowing. EXPERIMENT 5 Experiment 5 used a simpli ed design like that of Experiment 4 to examine the effects of post-training extinction of the blocking stimulus on the expression of performance to the blocked cue. As in Experiment 4, the identities of the blocking and blocked cues were counterbalanced, making possible an examination of the effects of extinction when the blocking procedures generated very different levels of conditioned responding. Methods Subjects and Apparatus The subjects were 64 male, 120-day-old, experimentally naive rats of Sprague-Dawley descent, bred in a Duke University psychology department facility, maintained in the same manner as in the previous experiments. All experimental sessions were carried out between 0800 and 1600 hr. The apparatus was the same as that used in Experiments 1±4. Procedures After food cup training identical to that of Experiments 1±4, the rats received eight 64-min pretraining sessions. In each of those sessions, the rats in groups B±E and B±N received eight reinforced 10-sec presentations of A, whereas the rats in groups O±E and O±N received eight unsignalled presentations of the food US. For half of the rats (n = 8 in each group), Stimulus A was the white noise and Stimulus X was the light; for the other rats, the roles of the noise and light were reversed. Next, the rats received four 64-min compound training sessions. In each of those sessions, all received eight 10-sec reinforced presentations of a white noise + light (AX) compound. Then, the rats in groups O±E and B±E received eight 10-sec non-reinforced presentations of A in each of sixteen 64-min post-training sessions. The rats in groups O±N and B±N were placed in the experimental chambers for those sessions, but received no explicit cues. Finally, all rats received a single 64-min test session, which included eight non-reinforced presentations of X. Unlike in Experiment 4, the two stimulus conditions (with light or noise as X) were not run concurrently. Thus, direct comparisons between the performances of rats for which the light was the