Journal o] Experimental Psychology:Animal Behavior Processes1976, Vol. 2, No. 4, 313-322

Surprise and the Attenuation of BlockingAnthony Dickinson, Geoffrey Hall, and N. J. Mackintosh

University of Sussex, Brighton, England

Three experiments, employing conditioned suppression in rats, examined theextent to which pretraining on one element of a compound stimulus blockedconditioning to the other element when some feature of the reinforcer waschanged on compound trials. In Experiment 1, both the addition of anunexpected second shock 8 sec after the end of each trial and the post-ponement of an expected second shock from 4 to 8 sec attenuated blocking.Experiment 2 established that this latter unblocking effect was due to thesurprising postponement of the second shock on trials when the new ele-ment was added rather than to any generalization decrement from pre-training to compound training. Experiment 3 showed that blocking wasattenuated not only by the addition of an unexpected second shock but alsoby the omission of an expected shock.

In a series of experiments on conditionedsuppression in rats, Kamin (1968, 1969)showed that the amount of conditioning ac-cruing to one element of a compound condi-tioned stimulus (CS) over a series of rein-forced compound trials was affected by thesubject's prior experience with the other ele-ment. If one element had previously beenpaired with shock, it might severely attenuateor, in Kamin's words, block conditioning tothe other element.

Kamin interpreted his results to mean thatconditioning was dependent on the predict-ability of reinforcement: No conditioningaccrued to the added element on compoundtrials because reinforcement was already fullypredicted by the presence of the pretrainedelement. Only surprising reinforcers are re-inforcing. Rescorla and Wagner (1972)have proposed a model of Pavlovianconditioning which is intended to capture atleast part of this idea. They assume that agiven reinforcer can support only a certainamount of conditioning (represented by aparameter, X) ; and increments to the associa-

This research was supported by the United King-dom Science Research Council.

G. Hall is now at University of York, York,England.

Requests for reprints should be sent to N. J.Mackintosh, Laboratory of Experimental Psy-chology, University of Sussex, Brighton BN1 9QG,England.

tive strength of one element of a compoundCS are proportional to (A — V ) , where Vrepresents the current associative strengthof all stimuli present, including, of course,other elements of the compound CS. In theblocking experiment, sufficient initial train-ing with one CS will drive its associativestrength close to X, with the consequencethat on compound trials the quantity (A —V) will be very small and the added elementcan gain little or no associative strength.

Kamin (1969) presented further experi-mental evidence that is entirely consistentwith these accounts of blocking. He showed,for example, that significant conditioningwould occur to the added element (i.e.,blocking would be attenuated) if the com-pound CS signaled a stronger shock thanthat paired with the first element during pre-training. The stronger shock was unpre-dicted or surprising and hence acted as aneffective reinforcer. In Rescorla and Wag-ner's terms the amount of conditioning sup-portable by a strong shock will be greaterthan that supported by a weak shock, andwith an increase in X the quantity (X — V)will be sufficiently large to enable the addedelement to gain a significant amount of con-ditioning.

A second experiment reported by Kamin,however, although not necessarily creatingdifficulties for Rescorla and Wagner's model,may at least suggest that their analysis is

313

314 A. DICKINSON, G. HALL, AND N. MACKINTOSH

not identical to Kamin's. In this experiment,Kamin produced significant conditioning tothe added element, not by increasing the in-tensity of shock, but by programming asecond, surprising shock 5 sec after eachcompound trial. As this double shock didnot produce more rapid initial conditioningthan a single shock, Kamin argued that thissecond case of unblocking could not be at-tributed to an increase in the strength of re-inforcement. Whether or not one accepts thisargument (and Rescorla and Wagner, 1972,p. 79, do not), it serves to pinpoint the dif-ference between Kamin's and Rescorla andWagner's analyses. According to Kamin, un-blocking may require only that the addedelement signal some surprising event; ac-cording to Rescorla and Wagner, significantexcitatory conditioning to the added elementappears to require an increase in the magni-tude of reinforcement on compound trials.The present experiments sought to distin-guish between these two possibilities byscheduling changes in reinforcement on com-pound trials that could not plausibly be re-garded as an increase in magnitude.

EXPERIMENT 1That the addition of a second, delayed

shock on compound trials produces someconditioning to the added element is notsufficient to decide between Kamin's andRescorla and Wagner's analyses, since anadditional shock, even if delayed, presum-ably constitutes some increase in total aver-sive stimulation and may be regarded as in-creasing the overall magnitude of aversivereinforcement. In Experiment 1, therefore,we compared the effects of adding a second,surprising shock on compound trials to thoseproduced by 'simply delaying by a few sec-onds the delivery of an already expectedsecond shock. One group of rats was initiallytrained with a CS signaling a single shock,while a second group received comparabletraining except that a second shock was de-livered 4 sec after the end of each condition-ing trial. Both groups then received a seriesof compound trials on which a second shockwas delivered 8 sec after the end of eachtrial. The first group constitutes, of course,a partial replication of Kamin's experiment;

the second should show whether postponingthe delivery of a second shock, an operationunlikely to constitute an increase in the mag-nitude of effective reinforcement, is sufficientto produce unblocking. A third, control groupreceived a second shock, delayed by 8 secon all trials during both initial and com-pound training, while a fourth group wasgiven no initial training with one element inorder to provide some assessment of themagnitude of the blocking and unblockingeffects observed in the other three groups.

Method

Subjects and apparatus. The subjects were 32male hooded rats from the colony maintained atthe University of Sussex, weighing approximately300 g at the start of the experiment. They werehoused in individual cages with unlimited access towater, and gradually reduced to 80% of their free-feeding weights. They were maintained at thisweight for the remainder of the experiment, beingfed an appropriate number of pellets immediatelyafter each day's training session.

The apparatus consisted of four two-lever ratchambers manufactured by Campden Instruments.The right-hand lever was removed from the cham-bers, and a white Perspex ceiling substituted forthat usually fitted. A 60-W, 240-V strip light wasmounted above the Perspex ceiling. The chamberswere placed inside sound-attenuating shells, andthe experiment was controlled by automatic pro-gramming equipment in an adjoining room.

Throughout each session, the chambers werepermanently illuminated by a 24-V white jewellight mounted in the center of the front wall, andwhite masking noise was delivered from a speakermounted in the center of the rear wall. Two CSswere used, a light and a clicker. The light wasproduced by switching on the overhead strip light,the clicker by a Campden Instruments audio gen-erator delivering a 10-Hz train of clicks to thespeaker. All CSs were 100 sec long. Shocks weredelivered by Grason-Stadler shock generators andscramblers.

Procedure. The preliminary phase of the experi-ment involved magazine training and the establish-ment of lever pressing. On Day 1, subjects re-ceived a SO-min session, with the lever removed,and free food (45-mg pellets) delivered on a vari-able-time 1-min schedule. On Day 2, the lever wasinserted, and lever pressing was consistently rein-forced. This session lasted 30 min. Rats failing toeat pellets on Day 1 or to press the lever on Day2 were given additional training. Throughout theremainder of the experiment, all sessions were SOmin long. On Day 3, lever pressing was reinforcedon a variable-interval (VI) 30-sec schedule, andon Days 4 and 5 (and for the remainder of the ex-

SURPRISE AND THE ATTENUATION OF BLOCKING 315

9 O----O UlK A---A u/.t

S T A G E I

D A Y S

FIGURE 1, Experiment 1: Mean suppression ratiosto the light during Stage 1, to the clicker-lightcompound during Stage 2, and to the clicker duringthe test trials. (None — no conditioning trials dur-ing Stage 1; L+ = light paired with single shockduring Stage 1; L+4+ = light paired with doubleshock during Stage 1 with 4-sec intershock inter-val ; L+8+ = light paired with double shock duringStage 1 with 8-sec intershock interval.)

periment) on a VI 1-min schedule. On Days 6 and7 all subjects received nonreinforced pretest trialsto the light and to the clicker. On each day therewere two trials, beginning 20 and 37 min after thebeginning of the session, during which either thelight or the clicker was presented without pro-grammed consequence.

Conditioning sessions also contained two trials,beginning at the 20th and 37th min after the startof the session. On each trial, the 100-sec CSterminated with a .5-sec 1.0 tnA grid shock, and,where scheduled, a second shock of the same dura-tion and intensity was delivered 4 or 8 sec afterthe end of the first. There were 12 conditioningtrials in Stage 1, and 10 in Stage 2, followed bytwo nonreinforced test trials. In Stage 1, Group L+was conditioned to the light signaling a singleshock, while for Groups L+4+ and L+8+, asecond shock was delivered 4 or 8 sec after thefirst. Group N received no conditioning trials inStage 1, but continued to lever press on the VI 1-min schedule. In Stage 2, all groups received 10compound (clicker and light) trials, each followedby two shocks with an 8-sec interval between firstand second shocks. On the test day, all groups re-ceived two nonreinforced trials with the clicker.

Results

Response suppression during the CS wasexpressed as a suppression ratio to attenuatethe effects of individual differences in theoverall rate of responding. The ratio has theform A/(A + B) where A is the rate of re-sponding during the CS and B the rate dur-

ing a 100-sec period immediately prior to CSonset.

On pretest trials the group mean suppres-sion ratios to the light ranged from .40 to.47, F < 1, and to the clicker from .41 to.53, F(3, 28) = 1.55, p > .10.

Figure 1 portrays the mean suppressionratios of the different groups to the lightduring Stage 1, to the clicker-light com-pound during Stage 2, and to the clickerduring the test trials. Inspection of Figure 1suggests that the groups receiving a doubleshock (Groups L+8+ and L+4+) duringStage 1 suppressed more rapidly than thegroup receiving a single 'shock (Group L+).However, this conclusion was not supportedby statistical analysis. Neither the effect ofgroups (F < 1) nor the Groups X Days in-teraction (F < 1) was significant. An ex-amination of pre-CS response rates, how-ever, suggested that Group L+8+ re-sponded, at least on 'some days, at a slowerrate than the remaining groups. An analysisof variance performed on these pre-CS ratesrevealed a significant Groups X Days inter-action, F(10, 105) = 3.36, p < .01, and aseparate analysis of response rates on eachsession showed a significant effect of groupson Session 3, F(2, 21) = 7.01, p < .01, andSession 5, F(2, 21) = 11.21, p < .01. In-dividual comparisons using a Newman-Keulsprocedure indicated that Group L+8+ re-sponded significantly more slowly thanGroups L+ and L+4+ on both these ses-sions (p < .05).

In Stage 2 the three pretrained groupsmaintained comparable and constant levelsof suppression. Group N rapidly developedsuppression to the compound CS and afterthe first session suppressed to a greater ex-tent than the other three groups. An analysisof suppression ratios revealed a significantGroups X Trials interaction, F(12, 112) =11.95, p < .01. Separate analyses showed asignificant effect of groups on all sessions(p < .05 in all cases), with Group N beingsignificantly less suppressed than the otherthree groups on the first session and sig-nificantly more suppressed than Group L+on Sessions 3, 4, and 5, than Group L+4+on Sessions 3 and 5, and than Group L+8+on Sessions 2 and 3 (p < .05 in all cases).

316 A. DICKINSON, G. HALL, AND N. MACKINTOSH

There was no significant effect of groups onthe pre-CS response rates during Stage 2,F(3, 28) = 2.33, p > .05.

During the test trials the basic blockingeffect was demonstrated by the fact thatGroup N was more suppressed than GroupL+8+. The absence of suppression in GroupL+8+ suggests that the blocking of condi-tioning to the clicker was almost complete.The intermediate levels of suppression main-tained by Groups L+ and L+4+ show thatboth adding and delaying a second shock at-tenuated blocking to about the same extentwithout completely abolishing it. An analysisof suppression on the test trials revealed asignificant effect of groups, F(3, 28) = 17.35,p < .01. Individual comparisons showed thatGroup N was significantly more suppressedthan the three remaining groups (p < .01in all cases), and that Groups L+ andL+4+ were more suppressed than GroupL+8+ (p < .05 in both cases). The pre-CSresponse rates of the four groups did not dif-fer on test trials, F(3, 28) = 1.53, p > .10.Discussion

The results of Experiment 1 confirmKamin's (1969) finding that the addition ofa second shock during compound trainingattenuates blocking. More importantly, itappears that simply delaying the secondshock produces a comparable degree of un-blocking. Although it might be argued thatthe addition of a second shock could haveincreased the effective magnitude of theaversive reinforcer and so permit an explana-tion of unblocking in terms of Rescorla andWagner's (1972) theory, it is rather lesslikely that postponing the second shock couldhave had such an effect. There is good evi-dence that the longer the delay between CStermination and unconditioned stimulus on-set the more slowly conditioning occurs (e.g.,Kamin, 1965); it is not entirely plausible,therefore, to argue that a change from a 4-to 8-sec intershock interval could have repre-ented an increase in the total reinforcement•magnitude. Moreover, although the com-parison is made difficult by a difference inpre-CS response rates, there is no sugges-tion that Group L+8+ suppressed morerapidly in Stage 1 than Group L+4+.

Even if it is accepted that postponing thesecond shock from 4 to 8 sec could not haveincreased the overall magnitude of reinforce-ment, there is one possible interpretation ofthese results that makes no appeal to sur-prise as a cause of unblocking. In an earlierattempt to show that surprise was the im-portant factor, Feldman (1971) studiedblocking in appetitive instrumental discrimi-nation learning by rats, changing from aconsistent schedule of reinforcement in initialtraining with one element to a partial sched-ule of reinforcement during compound train-ing. In spite of the consequent reduction inthe overall probability of reinforcement cor-related with the addition of the new element,Feldman found that it gained significantcontrol over his rats' behavior, and attributedthis unblocking effect to the surprisingchange in reinforcement. Neely and Wagner(1974), however, suggested an alternativeinterpretation in terms of generalizationdecrement. The change in reinforcementschedule, they argued, might have disruptedthe conditioning produced by initial training,and thus, in Rescorla and Wagner's (1972)terms, have increased the magnitude of thequantity (X — V), not by increasing A but bydecreasing V. In the context of Feldman'sexperiment this is a plausible suggestion, forit amounts to saying that at the end of initialtraining, the subjects' behavior was con-trolled not only by the experimenter's nomi-nal discriminative stimulus but also bystimuli arising from the schedule of reinforce-ment, such as traces of the outcomes of pre-vious trials. A change in schedule of rein-forcement will alter this latter class of stimuliand therefore, by generalization decrement,reduce V.

It does not, perhaps, seem very probablethat in the present experiment a change froma 4-sec to an 8-sec interval between twoshocks on one trial should have significantlydisrupted conditioning on a subsequent trialoccurring not less than 17 min later. Never-theless, the argument is, in principle, im-portant enough to be taken seriously.

EXPERIMENT 2Neely and Wagner (1974) were able to

show that generalization decrement was the

SURPRISE AND THE ATTENUATION OF BLOCKING 317

cause of the unblocking observed in Feld-man's (1971) experiment by running addi-tional experiments in which trials to thecompound stimulus signaling a changedschedule of reinforcement were interspersedwith trials to the pretrained component alonestill correlated with its original schedule ofreinforcement. By thus maintaining the tracesof the initial schedule, this procedure shouldhave prevented significant generalizationdecrement. It also abolished the unblockingeffect.

Experiment 2 was designed first to con-firm the finding of Experiment 1 that theshift from a 4-sec to an 8-sec interval be-tween shocks was sufficient to produce un-blocking, and secondly, to see whether gen-eralization decrement was in any wayresponsible for this unblocking effect. Afterconditioning in Stage 1 to the light alonepaired with a double shock, all subjects re-ceived two trials per session in Stage 2, thefirst trial on each day being to the lightalone and the second to the clicker-lightcompound. A factorial design made it pos-sible to assess separately the effects of a shiftin intershock interval associated with com-pound trials and one associated with inter-spersed trials to the light alone. An analysisin terms of surprise predicts that unblockingwould depend on a shift in the intershockinterval on compound trials, while the prin-ciple of generalization decrement implies thata shift in the intershock interval on eachpreceding trial to the light alone would pro-duce unblocking.

MethodThe subjects were 32 male rats from the same

stock as those used in Experiment 1, and the ap-paratus and general procedure were both exactlythe same as before. The magazine and lever-presstraining was the same as in Experiment 1 exceptthat subjects received only two sessions of trainingon the VI 1-min schedule. Subsequently four pre-test sessions were administered, two with the lightand two with the clicker. Each session containedonly one stimulus presentation.

The design of the experiment is shown in Table1. As was the case for Groups L+4+ and L+8+ inExperiment 1, all subjects were conditioned to thelight signaling a double shock in Stage 1 (witheither a 4-sec or an 8-sec interval between shocks),and in Stage 2 they were all conditioned to theclicker-light compound signaling a double shock

TABLE 1DESIGN OF EXPERIMENT 2

Treatment

GroupStage I,

3 sessionsStage 2,

8 sessionsTest,

4 sessions

S 2L+4+ L+4 + , CL+8+ L+4+, CC 2L+8+ L+8+, CL+8 + L-f 8+, CS-GD 2L+4+ L + 8+, CL+8+ L+8 + , CC-GD 2L+8+ L+4+, CL+8+ L+4+, C

Note. Abbreviations: S =* surprise, C = control, GD =generalization decrement; L = light, C •» clicker, +4+ =double shock with 4-aec intershock interval. +8+ = doubleshock with 8-sec intershock interval.

with an 8-sec interval between shocks. In an at-tempt to increase the overall level of conditioningto the clicker, only 6, rather than 12 trials weregiven to the light in Stage 1. There were 8 com-pound trials in Stage 2, but these were spreadover eight days as the second trial of each day.The first trial of each day in Stage 2 was to thelight alone. For Groups S and C (surprise and con-trol), the shock signaled by the light alone wasexactly the same in Stage 2 as in Stage 1; forGroup C, of course, the compound also signaled thesame shock as the light, while for Group S, thecompound signaled a longer interval betweenshocks. For Groups S-GD and C-GD (surprise-generalization decrement, and control-generaliza-tion decrement) the light signaled a different in-terval between shocks in Stage 1 and Stage 2. Theintershock interval associated with compound trialswas the same as that experienced in Stage 1 forGroup C-GD, but longer for Group S-GD.

All subjects received four test trials to theclicker alone, one on each of four test days. Oneach day, the nonreinforced clicker trial was pre-ceded, as in Stage 2, by a trial to the light alonesignaling the same pair of shocks as in Stage 2.

ResultsOn pretest trials the group mean suppres-

sion ratios to the light ranged from .33 to.42, F(3, 28) = 2.05, p > .10, and to theclicker from .39 to .46, F < 1.

Figure 2 illustrates the acquisition of sup-pression to the light in Stage 1, suppressionto the clicker-light compound in Stage 2,and the suppression controlled by the clickeralone on test trials. During Stage 1, GroupS-GD appeared to condition more slowlyto the light than the remaining three groups.This difference cannot be due to the trainingconditions in effect during Stage 1, sinceGroup S-GD was treated in exactly the sameway as Group S (both receiving a 4-'secintershock interval). It presumably reflectsa sampling error. By the end of Stage 1, all

318 A. DICKINSON, G. HALL, AND N. MACKINTOSH

S T A G E 1 S T A G E a T E S T

T W O T R I A U B L O C K

FIGURE 2. Experiment 2: Mean suppressionratios to the light during Stage 1, to the clicker-light compound during Stage 2, and to the clickeralone during the test trials. (S = surprise; S-GD= surprise and generalization decrement; C =control; C-GD = control and generalization dec-rement.)

groups showed nearly complete suppressionto the light. An analysis over all three daysshowed a significant effect of groups, F(3,28) = 4.12, p < .025, but no significantGroups X Days interaction, F(6, 56)= 1.93,p > .05. Individual comparisons (Newman-Keuls test) showed that Group S-GD wasless suppressed than Groups S and C-GD,p < .05 in each case, while Groups C, C--GD, and S could not be differentiated. Pre-CS response rates did not differ betweengroups; for both the main effect of groupsand the Groups X Days interaction, F < 1.

All groups maintained virtually completesuppression both to the light alone and tothe clicker-light compound throughout Stage2. Suppression on trials to the light alonewas affected neither by whether the inter-shock interval was 4 or 8 sec, nor by whetherthe interval had been changed from Stage 1to Stage 2, nor by the interaction of thesefactors, F < 1 in all cases. Suppression tothe clicker-alight compound was similarlyunaffected by changing the intershock inter-val associated with the compound (surprisefactor), or that associated with the lightalone (generalization decrement), or by theirinteraction, F < 1 in all cases. Finally, therewere no differences between the four groupsin their pre-CS response rates, F(3, 28) =1.33, p > .10.

Inspection of Figure 2 shows that the twosurprised groups exhibited greater suppres-sion to the clicker than their respective con-trols throughout the test trials. By contrast,the effect of generalization decrement inter-acted with test trials. On Trials 1 and 2 thegroups experiencing generalization decre-ment showed less suppression than their re-spective controls. Although this differencewas preserved on Trials 3 and 4 for GroupsS-GD and S, it was reversed for the twononsurprised groups, with Group C exhibit-ing less suppression than Group C-GD.There was no difference between the fourgroups in the level of suppression maintainedon retraining trials to the light on these testdays.

These conclusions were confirmed by sta-tistical analysis. An overall analysis of sup-pression to the clicker revealed a significanteffect of surprise, F(l, 28) = 12.91, p < .01,and a significant Generalization DecrementX Trials interaction, F(3, 84) = 3.19, p <.05. Neither the main effect of generaliza-tion decrement, F(l, 28) = 3.58, p > .05,nor the Surprise X Generalization Decre-ment interaction, F(l, 28) = 2.44, p > .10,reached significance. An analysis of perform-ance on individual test trials revealed a sig-nificant effect of generalization decrement onTrial 2, F(l, 28) = 6.34, p < .025, and asignificant Surprise X Generalization Decre-ment interaction on Trial 3, /*(!, 28) =10.28, p < .01. An analysis of suppression ontrials to the light alone showed that perform-ance was affected neither by the intershockinterval associated with the light, nor bywhether this interval had been shifted fromStage 1 to Stage 2, nor by their interaction,F < 1 in all cases. The four groups showedsimilar pre-CS response rates throughouttesting, F < 1.Discussion

The results of Experiment 2 agree withthose of Experiment 1 in showing that thepostponement of a second shock on com-pound trials may attenuate blocking. Theyalso demonstrate that generalization decre-ment in no way contributes to this result.If the attenuation of blocking were entirelyattributable to generalization decrement, then

SURPRISE AND THE ATTENUATION OF BLOCKING 319

Group S would have shown no more sup-pression to the added element than Group C.This was clearly not so. If generalizationdecrement were even partially responsiblefor the attenuation of blocking, then GroupsS-GD and C-GD would have shown moresuppression to the clicker than Groups S andC, respectively. There was, however, no con-sistent difference between Groups C— GD andC, while Group S-GD was reliably less sup-pressed than Group S.

This latter difference is consistent withthe suggestion that the unblocking observedhere is entirely a consequence of correlatingclicker-light trials and light-only trials withdifferent intershock intervals. The explicitmethod of producing surprise in this experi-ment was to change the intershock intervalcorrelated with compound trials in Stage 2from that correlated with light trials in Stage1. Blocking should also be attenuated, how-ever, by a difference between the intershockintervals correlated with light and clicker-light trials within Stage 2 itself. Since GroupS was exposed to such a difference whileGroup S-GD was not, the greater sup-pression to the clicker shown by Group S isfurther evidence of the importance of sur-prise.

Groups C and C-GD both received thesame intershock interval on compound trialsas they had on light trials in Stage 1, butfor Group C-GD the intershock interval onlight trials was changed in Stage 2. Thereis no obvious reason why Group C shouldhave shown more suppression on early testtrials but less on later trials, and it does notseem worth speculating further on this pat-tern of results.

There were no reliable differences betweenthe various groups in suppression to thelight in Stage 1. and no suggestion of anydifference between groups receiving the 4-sec intershock interval and those receivingthe 8-sec interval. This again suggests thatthe attenuation of blocking we have observedcannot be a consequence of any increase inthe effective magnitude of reinforcement.Even if this is accepted, however, it remainstrue that the surprising postponement of asecond shock must involve the presentation ofan unexpected shock. It might be possible,

therefore, for Rescorla and Wagner's (1972)model to explain the unblocking observed inExperiments 1 and 2, not by arguing that alonger delay between two shocks was gen-erally more reinforcing than a shorter delaybut rather by supposing that the secondshock would be more reinforcing if it oc-curred at an unexpected time rather than atan expected time. It could be assumed thattemporal cues dating from the termination ofthe CS would come to signal the occurrenceof the second shock and serve to blockfurther conditioning to that shock, providedthat it continued to occur at the same intervalafter the CS. Any change in the time of thesecond shock, however, would mean that itsoccurrence was not fully predicted and wouldenable it to reinforce further conditioning.

EXPERIMENT 3Although these arguments are somewhat

speculative, the point they raise is sufficientlyimportant to justify further experimentalanalysis. The one condition that could beconstrued neither as increasing the overallmagnitude of reinforcement nor as involvingthe presentation of an unexpected secondshock would be the omission of an otherwiseexpected second shock. Experiment 3 em-ployed a simple factorial design in which theCS signaled either a single shock or a pairof shocks in Stage 1, and each of thesegroups was divided in Stage 2 for compoundtraining, with half receiving a single shockfollowing each compound trial and half re-ceiving a double shock. For one group,therefore, the new element added on com-pound trials signaled the addition of a secondshock, and for another it signaled the omis-sion of a second shock. The two controlgroups received the same number of shocks,either one or two, in both stages of the ex-periment. To add further generality to ourresults, the stimulus used in Stage 1 was theclicker, and the stimulus added in Stage 2was the light.

MethodThe subjects were 32 male rats from the same

stock as those used in Experiments 1 and 2. Theapparatus and general procedure were the same asbefore.

320 A. DICKINSON, G. HALL, AND N. MACKINTOSH

--a

FIGURE 3. Experiment 3: Mean suppression ratiosto the clicker (C) during Stage 1, to the clicker-light (CL) compound during Stage 2, and to thelight (L) during the test trials. (C+ CL++=single shock during Stage 1 and double shock dur-ing Stage 2; C++ CL++ = double shock duringing Stages 1 and 2; C++ CL+ = double shock dur-ing Stage 1 and single shock during Stage 2; C+CL+= single shock during Stages 1 and 2.)

In Stage 1, all subjects received two condition-ing trials to the clicker (C) on each day for 6days. For Group C+ each trial terminated witha single shock; for Group C++, a second shockwas delivered 8 sec after the first. In Stage 2there were two reinforced trials each day to theclicker-light compound (CL) for four days. GroupsC+ and C++ were each subdivided, and GroupsC+ CL+ and C++ CL+ received a single shockon each trial, while Groups C++ CL++ and C+CL++ received a second shock on each trial 8sec after the first. On the final day of the experi-ment there were two nonreinforced test trials tothe light.

ResultsOn pretest trials the group mean suppres-

sion ratios to the light ranged from .40 to .44,F < 1, and to the clicker from .44 to .48,F<1.

During Stage 1 the groups receiving twoshocks developed suppression more rapidly tothe clicker than the groups receiving a singleshock. This difference is illustrated in Figure3, which portrays the suppression to theclicker during Stage 1, to the clicker-lightcompound during Stage 2, and to the lighton test trials. Although there was no overallsignificant effect of the number of shocksduring Stage 1, F(l, 30) = 2.44, p > .10,the Number of Shocks X Days interactionwas significant, F(5, ISO) = 5.84, p < .01.

Separate analyses of suppression on eachday of Stage 1 showed that Groups C+ +CL+ and C++ and CL++ were signifi-cantly more suppressed than Groups C+CL+ and C+ CL++ on Day 2, F(l, 30) =20.20, p < .01. An analysis of pre-CS re-sponse rates showed that neither the maineffect of number of shocks, F < 1, nor theNumber of Shocks X Days interaction, F(5,150) = 1.26, p > .25, reached significance.

In Stage 2 Groups C+ CL+ and C+ +CL++, which were not surprised by eitherthe addition or omission of a shock, main-tained comparable levels of suppression tothe compound CS. During Days 2 and 3 thegroup surprised by the addition of a shock,Group C+ CL++, tended to be slightlymore suppressed than the nonsurprisedgroups, while Group C++ CL+, surprisedby the omission of a shock, was slightly lesssuppressed. However, by the end of Stage 2•these small differences had completely dis-appeared. Although neither the main effectof surprise, (F < 1), nor the effect of num-ber of shocks in Stage 2, F(l, 28) = 3.09,p > .05, significantly affected suppression,the Surprise X Number of Shocks interac-tion, F(l, 28)= 4.45, p < .05, was signifi-cant. There was no evidence that the pre-CSresponse rate varied as a function of surpriseor shock number. The F ratios for both theseeffects and for the Surprise X Number ofShocks interaction were all less than 1.

The two groups surprised during Stage 2,Groups C+ CL++ and C++ CL+, bothexhibited more suppression to the light ontest trials than the nonsurprised groups,Groups C+ CL+ and C++ CL+ + . How-ever, the level of suppression did not varyas a function of the number of shocks re-ceived during Stage 2. There was a sig-nificant effect of surprise, F(l, 28) = 16.16,p < .01, but the F ratios for the effect ofnumber of shocks and the Surprise X Num-ber of Shocks interaction were both less than1. An analysis of the pre-CS response ratesrevealed no significant effects with all Fratios less than 1.

DiscussionThe results of this experiment provide no

support for the suggestion that attenuation

SURPRISE AND THE ATTENUATION OF BLOCKING 321

of blocking requires an increase in the effec-tive magnitude of reinforcement from Stage1 to Stage 2. The differences observed in theacquisition of suppression in Stage 1 sug-gested that, if anything, a single shock actedas a less effective rein-forcer than a double•shock. Nevertheless, the change from adouble shock in Stage 1 to a single shock inStage 2 produced just as much unblockingas the addition of a .second shock.

GENERAL DISCUSSIONIn Kamin's (1968, 1969) blocking experi-

ments, little or no conditioning accrued toone element of a compound CS, if its addi-tion to the compound 'signaled no change inreinforcement from that already predictedby the first element. If the new elementsignaled an increase in shock intensity, orthe addition of a second shock shortly afterthe first, then it would acquire significantassociative strength. The present results sug-gest that it is not necessary that the newelement should signal any increase in shock,for either the temporary postponement of anexpected second shock or its complete omis-sion was as effective as the addition of anunexpected shock in producing conditioningto the new element. Rescorla and Wagner(1972) predict that any excitatory condi-tioning to the new element beyond the levelfound in control groups must depend eitheron a significant increase in the overall magni-tude of reinforcement on compound trials oron the occurrence of an otherwise unpre-dicted shock. The results of Experiment 3are particularly hard to reconcile with thisanalysis. The omission of an expected secondshock can hardly be regarded either as con-stituting an increase in the magnitude of re-inforcement or as equivalent to the presen-tation of an unpredicted shock. The criticalfinding of Experiment 3 appears sufficientlyreliable, since we have been able to replicateit in a somewhat different experimental situ-ation (Mackintosh, Bygrave, & Pickton,Note 1). Although Rescorla and Wagner'smodel may explain part of the unblockingobserved when shock intensity is increasedon compound trials, it does not appear thatit provides a complete account of unblocking.

This suggests that we should consider

more favorably Kamin's (1969) originalsuggestion that any change in reinforcementsufficient to surprise the subject on com-pound trials may produce significant condi-tioning to the added element. There is, how-ever, a certain ambiguity to this suggestion,which may be brought out by consideringKamin's original experiment in which asecond, surprising shock is added on com-pound trials. What is the precise role of thissecond shock? One possibility is that, be-cause it is expected, it can act as a rein-forcer in its own right: In spite of the traceinterval separating it from the terminationof the compound CS, it is able to support asignificant level of conditioning to the newlyadded element. An alternative possibility,however, is that the second shock does notitself support conditioning, but that, again byvirtue of its unexpectedness, its presencesomehow ensures that the first shock, whichwould not by itself have acted as an effectivereinforcer, is now able to play a normal re-inforcing role. Kamin's suggestion that thesurprising shock alerts the subject or causesit to engage in retrospective contemplation(Kamin, 1969) is consistent with this latterpossibility.

The present results, especially those ofExperiment 3, seem to require us to acceptthe second possibility. In that experiment,the surprising omission of a second shockon compound trials resulted in significantexcitatory conditioning to the added element.The omission of shock may support inhibi-tory conditioning, but it can hardly act as areinforcing event to produce excitatory con-ditioning. We must suppose that the role ofthe surprising event in this case was not it-self to reinforce conditioning but to enableconditioning to proceed normally to theotherwise ineffective first shock. This im-plication is consistent with the analysis ofblocking suggested by Mackintosh (1975).On this account, blocking is a consequenceof the fact that subjects learn to ignorestimuli that signal no change of consequence.In a typical blocking experiment, condition-ing may proceed normally to the new ele-ment on the first compound trial, but be-cause it signals no change in reinforcementfrom that already predicted by the pretrained

322 A. DICKINSON, G. HALL, AND N. MACKINTOSH

element, there will be a sharp decline in thevalue of a stimulus-specific learning-rateparameter, a, associated with the new ele-ment. For significant conditioning to occurto the new element, it is only necessary toprevent this decline in a. Provided that itsignals some event not already predicted bythe pretrained element, whether this be theaddition, postponement, or omission of asecond shock, the value of a associated withthe new element may be maintained at alevel sufficient to produce significant condi-tioning. The surprising event does not itselfreinforce that conditioning; it serves to main-tain attention to a stimulus that would other-wise be ignored and thus enables the rein-forcer actually paired with that stimulus (i.e.,the first shock) to have its normal effect.This viewpoint can then explain why Grayand Appignanesi (1973) were able to pro-duce unblocking by presenting a flash of thecompound trial stimuli following each com-pound trial. Such an unpredicted stimulusmaintained attention to the added elementand thus allowed it to be associated with thereinforcing shock.

The present results have a wider signifi-cance than their relevance to theories ofblocking. Whether or not one accepts theparticular account 'suggested here, thereseems no way of escaping the conclusionthat an event of motivational significancesuch as the addition or omission of the sec-ond shock is somehow able to affect the levelof conditioning to a particular CS withoutitself directly reinforcing that conditioning.If we assume that the effect of the surprisingevent is associative, rather than generallyarousing, we must suppose that it is associ-ated with the new element on compoundtrials, even though it is too far removed intime from the termination of that element to

support any change in performance. This im-plies a rather sharp distinction between theassociations that a reinforcer may enter intoand the change in the strength of the condi-tioned response that it may support.

REFERENCE NOTE1. Mackintosh, N. J., Bygrave, D. J., & Picton, B.

M. B. Locus of the effect of a surprising rein-forcer in the attenuation of blocking. Manuscriptsubmitted for publication, 1976.

REFERENCESFeldman, J. M. Added cue control as a function of

reinforcement predictability. Journal of Experi-mental Psychology, 1971, 91, 318-325.

Gray, T., & Appignanesi, A. A. Compound condi-tioning : Elimination of the blocking effect. Lcarn-ing &• Motivation, 1973, 4, 374-380.

Kamin, L. J. Temporal and intensity characteris-tics of the conditioned stimulus. In W. F. Pro-kasy (Ed.), Classical conditioning: A sym-posium. New York: Appleton-Century-Crofts,1965.

Kamin, L. J. "Attention-like" processes in clasiscalconditioning. In M. R. Jones (Ed.), Miamisymposium on the prediction of behavior: Avers-ive stimulation. Miami: University of MiamiPress, 1968.

Kamin, L. J. Predictability, surprise, attention,and conditioning. In B. A. Campbell & R. M.Church (Eds.), Punishment and aversive be-havior. New York: Appleton-Century-Crofts,1969.

Mackintosh, N. J. A theory of attention: Varia-tions in the associability of stimuli with rein-forcement. Psychological Review, 1975, 82, 276-298.

Neely, J. H., & Wagner, A. R. Attenuation ofblocking with shifts in reward; The involvementof schedule-generated contextual cues. Journalof Experimental Psychology, 1974 102, 751-763.

Rescorla, R. A., & Wagner, A. R. A theory of Pav-lovian conditioning: Variations in the effective-ness of reinforcement and nonreinforcement. InA. H. Black & W. F. Prokasy (Eds.), Classicalconditioning II: Current research and theory.New York: Appleton-Century-Crofts, 1972.

(Received January 14, 1976)