Animal Learning 1984,12 (1), Seriallearning, interitemassociations, … · 2017-08-26 · property...
Transcript of Animal Learning 1984,12 (1), Seriallearning, interitemassociations, … · 2017-08-26 · property...
Animal Learning & Behavior1984,12 (1), 7-20
Serial learning, interitem associations,phrasing cues, interference, overshadowing,
chunking, memory, and extinction
E. J. CAPALDI, DONNA R. VERRY, TIMOTHY M. NAWROCKI,and DANIEL J. MILLER
Purdue University, West Lafayette, Indiana
Two experiments indicated that two approaches to serial learning are too extreme-the classical view that it consists only of interitem associations and various recent views that it involves no interitem associations. The novel assumption introduced here was that phrasing cues,normally conceptualized as merely segregating long series into smaller units or chunks, mayalso enter into associations with items, thereby reducing interitem interference and facilitatingserial learning. It was found that one item could become a signal for another item, an interitemassociation, or be overshadowed by a phrasing cue, such as a brightness and temporal cue, alsosignaling that item. The items were ,045-g pellets. Rats traversed a runway for items arrangedin ordered series, 14-7-3-1-0 pellets (Experiment 1)or 10-2-0-10 (Experiment 2), Complete tracking of, for example, the 10-2-0-10 series would consist of fastest running to 10 pellets andslowest running to 0 pellets. In both investigations, the interitem association overshadowedwas that between 0 pellets and the subsequent rewarded item, 0-14 (Experiment 1) or 0-10(Experiment 2). Either repetitions of the 14·7·3-1-0 subpattem (Experiment 1) or merely the terminall0·pellet item (Experiment 2) were phrased, both methods producing identical results.Overshadowing the O-pellet item produced superior serial learning, more rapid extinction, and,in Experiment 1, considerable elevation of responding when the brightness phrasing cue wasintroduced in extinction, an effect said to be conceptually identical to spontaneous recovery andone demonstrating directly that phrasing cues are in reality overshadowing cues. It was suggested that many effects attributed to forgetting may be due to unrecognized overshadowing ofmemory cues by phrasing cues, giving rise to exaggerated estimates of forgetting.
Serial learning, employed by Ebbinghaus (1885/1964) to investigate learning and memory, produced a theory of learning, the chaining hypothesis,which went virtually unchallenged for over 50 years.Recently, however, its central assumption, that of interitem associations in which one item becomes thesignal for the next, has been rejected by a variety ofhuman serial learning models (e.g., Estes, 1972;Johnson, 1972; Restle & Brown, 1970). This retreatfrom Ebbinghaus has its counterpart in animal seriallearning, lately the object of considerable interest.For example, Hulse and his co-workers (e.g., Hulse,1978; Hulse & Dorsky, 1977, 1979) have explicitly rejected interitem associations in favor of a cognitiverule-encoding model based on human models of theRestle type.
Animal investigators normally speak in terms oftrials, for example, a nonreinforced trial rather than anonreinforced item. Because item terminology is moreuseful here, it is employed in this paper. According tothe chaining view, in the learning of a list of items, A-
This research was supported by NSF Grant BNS 80-001171 toE. J. Capaldi. The authors' mailing address is: Department of Psychological Sciences, Purdue University, West Lafayette, IN 47907.
7
B-C- ... , item A is the stimulus for the response B,then item B becomes the stimulus for the response C,and so on. In addition to adjacent associations, remote associations are postulated, A being associatedwith C but less strongly than with B, and so on. If thestimulus for an item is the preceding item, then pairing the two pairs in a subsequent paired-associate task(A-B; B-C) should produce positive transfer, a prediction not always confirmed (see Young, 1968).Various other failings of the chaining model (see,e.g., Restle & Brown, 1970) suggest that serial learning cannot be explained solely in terms of interitemassociations, even if they are conceptualized as S-Srather than S-R associations (Capaldi, Nawrocki, &Verry, 1983a). But we suggest that the complete rejection of iteritem associations that has occurred inrecent years may be unwarranted. We suggest this because, as has been emphasized elsewhere, serial learning appears to have much in common with varied reinforcement and discrimination learning, and certainfindings in those situations have been interpretedthus far only in terms of interitem associations(Capaldi & Molina, 1979; Capaldi, Nawrocki, &Verry, 1982; Capaldi & Verry, 1981; Capaldi, Verry,& Davidson, 1980).
Copyright 1984 Psychonomic Society, Inc.
8 CAPALDI, VERRY, NAWROCKI, AND MILLER
An example of an interitem association of directconcern in this report is that which may be established when a nonreinforced item is followed by areinforced item, an N-R transition. According to aninteritem view, in an N-R transition, cues associatedwith the nonreinforced item may become a signal forthe subsequent reinforced item (see, e.g., Capaldi,1964). If the "N" item becomes a strong signal forthe "R" item, then, as can be seen, whenever the "N"item is presented, as, for example, in extinction or asthe S- alternative of a discrimination task, responding will be elevated-resistance to extinction will beincreased, discriminative responding will be reduced,and so on (see, e.g., Capaldi, Nawrocki, & Verry, inpress; Haggbloom, 1980b). But in an N-R transition,cues produced by the nonreinforced item may not become a strong signal for reinforcement if some otherstimulus simultaneously signals reinforcement. Whentwo cues simultaneously signal reinforcement and thepresence of one reduces the behavioral control acquired by the other, we speak of overshadowing(e.g., Kamin, 1969). Overshadowing, we suggest, iscommonplace in serial learning. Our view, statedgenerally, is this: When the best predictor of an itemis the previous item, an interitem association will beformed. But if another cue is a better predictor of theitem, because it is either more intense or more valid,it will signal the item, and the previous item will bemore or less overshadowed.
It is generally agreed that people can learn a longseries of items more efficiently if they segment it intosmaller units or chunks on the basis of some commonproperty or relation (e.g., Bower & Winzenz, 1969;Restle & Brown, 1970). Thus, a 10-digit telephonenumber may be chunked into a 3-digit area code,a 3-digit prefix, and a 4-digit number. Two of thenumerous factors that have been shown to facilitatechunking (see, e.g., Bower, 1972) are similarityitems of the same color, shape, size, and so onand proximity-items occurring close together inspace or time. "Phrasing cues" is one term appliedto the numerous factors that have facilitated chunking. Recently, compelling evidence that animalschunk series, and on much the same basis as humansubjects, has been provided by Hulse and his coworkers (e.g., Fountain, Henne, & Hulse, in press;Hulse, 1978; Hulse & Dorsky, 1977).
Phrasing cues, we suggest, may not only segmentlong series into smaller chunks, but may also becomesignals for items. For example, a long temporal breakin a series can be used to signal the first item occurring after the break, overshadowing the previousitem as a signal. When an item is signaled weakly bythe prior item and strongly by a phrasing cue, thatitem and all subsequent items joined by interitem associations constitute, according to the present view, achunk. The hypothesis introduced and tested here isthat phrasing cues facilitate chunking, and thereforeserial learning, because they overshadow interitem
associations, thereby reducing interference amongitems. Consider an example of interference that results when an item signals two dissimilar items andhow such interference may be reduced by a phrasingcue. If item A signals the adjacent similar item A'(A- A') and A' signals the adjacent, but dissimilar,item B (A' - B), then generalization between thehighly similar items A and A' will give rise to considerable interference, item A signaling both A' andB. But if A signals A' , and B is signaled not by A'but by a distinctive phrasing cue having little in common with A, then interference will be reduced, Asignaling only A' ; in consequence, serial learning willbe facilitated.
Necessarily, some nonlist event, call it X, must signal the first item in the list. If each subsequent item inthe list A-B-C-D ... is signaled strongly by the preceding item, then we may have a list of only onechunk, X-A- B-C-D, the arrows indicating associations. But say the interval between items C andD is longer than those between items A-B-C. Thistemporal phrasing cue, call it Y, may completelyovershadow C as a signal for D. In that event, wewould have a list of two chunks, X-A-B-C;Y-D. This view was treated in the two investigations reported here by examining the effects of phrasing cues on extinction as well as serial learning. Theeffects of phrasing cues on extinction have not previously been investigated. Of major concern here wasthe condition in which, in an N-R transition, the Ritem was signaled not only by the N item, but by aphrasing cue, with brightness and temporal phrasingcues being employed here. It was expected that whenever a phrasing cue overshadowed N as a signal forR, serial learning would be facilitated because of reduced interference, and subsequent resistance to extinction would be reduced, the latter prediction perhaps being unique among serial learning hypotheses,none of which has as yet attempted to deal with extinction.
In each of the two investigations reported here, weemployed series containing a subpattern that terminated in a O-pellet item, the O-pellet item being followed by a rewarded item. The rewarded item eitherwas or was not signaled by some other cue, a changein brightness (Experiments 1 and 2) or a long temporal interval (Experiment 2). It was expected thatthe behavioral control exercised by the O-pellet itemwould be reduced when some other cue simultaneously signaled the subsequent rewarded item. Thisreduction in control would have two effects: it wouldresult in superior serial learning and it would reduceresistance to extinction.
EXPERIMENT 1
In order to facilitate a comparison of results, Experiment 1 employed procedures similar in many respects to those employed in investigations reported
by Hulse and his co-workers. In one preparationemployed in Hulse's laboratory (see, e.g., Hulse,1978), rats in a runway were given a long series ofitems (.04S-g food pellets) constructed by repeatingthe five-item subpattern of food pellets 14-7-3-1-0.On the first run down the runway, the rat would begiven 14 pellets, on the next run, 7 pellets, and so on.Runs within the subpattern were separated by a 202S-sec interval; repetitions of the subpattern wereseparated either by the same interval, the unphrasedseries, or by a IS-min interval, the temporallyphrased series. The assumption was that chunkingthe series into the 14-7-3-1-0 subpattern would bemore likely when the series was phrased rather thanunphrased. This should facilitate pattern learning, asindexed by running slowly to (tracking) the terminalO-pellet item of the subpattern. In agreement withthis reasoning, Hulse (1978) reported that trackingthe O-pellet item occurred much sooner in the temporally phrased series than in the unphrased series, 48 repetitions versus 80 repetitions of the subpattern, adramatic difference. Similar findings were obtainedin aT-maze, where, in addition, spatial phrasing wasalso shown to be effective (Fountain et aI., in press).Fountain et al. reported that tracking the O-pelletitem occurred much sooner in a spatially phrasedgroup that received its first repetition of thesubpattern in one arm of the T-maze, its next repetition in the opposite arm, and so on, than it did ineither an unphrased group, which received all repetitions on a given day in the same arm, or a mismatched group, for which the phase cue was locatedbetween the 3- and I-pellet items, segmenting theseries into a 1-0-14-7-3 subpattern, which is considered complex from a rule-learning viewpoint.
Experiment 1, like the investigations of Hulse andhis co-workers, employed a long series constructedfrom repeating the 14-7-3-1-0 subpattern. The interval between all items, within and between subpatterns, was 20-30 sec. The series was phrased usingbrightness cues. For Group GP (good phrasing), thefirst repetition of the 14-7-3-1-0 subpattern occurredin a runway of one brightness, black or white, thenext repetition in the runway of the other brightness,and so on. For Group BP (bad phrasing), the brightness cue was changed between the 3- and l-pelletitems, that is, 14-7-3/1-0-14-7-3/, and so on, theslashes indicating a change in alley brightness. ForGroup GP, except for the first daily presentation ofthe 14-pellet item, two cues signaled the 14-pelletitem, the terminal O-pellet item of the subpattern andthe change in alley brightness. For Group BP, againexcept for the initial 14-pellet item of the day, onlyone cue reliably signaled 14 pellets-the O-pelletitem.For Group GP, the change in alley brightness couldovershadow the O-pellet item as a signal for 14 pellets, but such overshadowing was not possible inGroup BP. According to the present hypothesis, as
SERIAL LEARNING 9
will be explained in detail later, with overshadowing,Group GP should track 0 pellets better than Group BPand should extinguish faster than Group BP, sinceboth effects are due to a common cause, 0 pelletsbeing a weaker signal for 14 pellets for Group GPthan for Group BP. In Experiment 1 there were 2days of extinction, and midway through the extinction trials on each day alley brightness was changed.If the change in alley brightness did indeed becomea strong signal for 14 pellets for Group GP, overshadowing the O-pellet item, then the change inbrightness should produce increased vigor of responding in Group GP but not in Group BP.
MethodSabJeeu. The subjects were eight naive male albino rats, 80 days
old on arrival at the laboratory, purchased from the HoltzmanCo., Madison, Wisconsin.
Apparatus. The apparatus consisted of two adjacent runways,identical except for brightness. One was black, and the other waswhite. The startboxes, also black or white, were 21.6 cm long; thegoalboxes were 33.95 em long. Each alley was 8.57 cm wide and atotal of 190.8 em long; each was enclosed by 11.43-cm sidescovered by a wire-mesh top on a hinged frame. Lowering the brassstartbox door started a completely silent .01-sec digital clock thatwas stopped when a photobeam, located 158.13 em beyond thestartbox door, was broken by the rat. Pellets (.045-g Noyes) couldbe placed in a goal cup (3.8 x 3.8 x .95 cm) at the end of the runway. The pellets were not visible before the photobeam wasbroken. A brass door confined the rat to the goalbox,
PretralniDI. On arrival at the laboratory, the rats were cagedindividually and allowed ad-lib food and water for 5 days. OnDays 1-9, each rat was handled for 1 min and then fed the dailyration in the home cage, 24-g Lab Blox, a laboratory rat chowmade by the Wayne Pet Food Company. On Day 10, each rat wasfed 10 .045-g pellets in the home cage. On Days 11-12, each rat wasexposed to each runway for 1 min and then returned to the homecage, where it was fed .045-g pellets, followed by the daily ration.Throughout all phases of the experiment, all pellets eaten by therat were subtracted from the 14-gdaily ration.
Experimental tralmnl. There were two groups of four rats each.Each rat received repetitions of the subpattern, 14-7-3-1-0, withtwo repetitions on Days 1 and 2 and four repetitions on Days 3-11.The interval between all items, within the subpattern and betweenrepetitions of the subpattern, was 20-30 sec. Four rats, two fromeach group, were taken into the experimental room from thecolony room. The rats were run in a different order each day, andeach rat received all items of the series before the next rat was run.After the last rat was run, the animals were returned to their homecages where, after 10 min, they were fed their daily ration.
About 3 sec after the rat was placed in the startbox, the startboxdoor was lowered. The rat was allowed 60 sec to reach the goalboxto receive whatever item was scheduled on that run. If the rat didnot make it in 60 sec, it was placed in the goalbox. Failures toreach the goalbox were rare. On o-pellet runs, the rat was confinedto the unbaited goalbox for 15 sec. On reinforced runs, immediately after eating the pellets, the rat was removed from thegoalbox and placed in a waiting cage, with water available. About20-30 sec later, the rat was given the next run of the series. ForGroup GP, the first repetition of the 14-7-3-1-0 subpattern occurred in an alley of one brightness, black or white, the next repetition in the runway of the other brightness, and so on. For Group BP,the brightness cue was changed between the 3- and l-pellet items,that is, 14-7-3-/1-0-14-7-3/, and so on, the slashes indicating achange in alley brightness. On Days 1 and 2, the black runway waspresented first. Thereafter, the white runway was presented first onodd days and the black runway was presented first on even days.
10 CAPALDI, VERRY, NAWROCKI, AND MILLER
On Days 12 and 13, each rat received 10 nonreinforced runs, allat 20-30-sec intervals. On Day 12, for each rat, the first five runsoccurred in the black runway, the next five in the white runway,the opposite order of runway presentation being used on Day 13.Except for the fact that all runs terminated in nonreinforcement,the conditions in extinction were as in the acquisition phase.
On rewarded trials, the pellets were placed noiselessly in the goalcup from a cloth-lined cup. To control for odors, an open cannisterof pellets was placed near the goalbox.
ResultsThere were two repetitions of the subpattern on
Days 1 and 2 and four per day thereafter. Speed ofrunning on each of the runs of the subpattern forDays 1 and 2 combined and for each of Days 8-11are shown in Figure 1. Figure 1 shows that, onDays 1-2, neither group tracked the O-pellet item, butthat from Day 8 on, Group GP, but not Group BP,did track the O-pellet item. An analysis of variancewasperformed over the entire serial learning phase, usingas factors groups, repetitions of the subpattern, runs,and days, Days 1 and 2 being combined into a single'day in the analysis. The analysis revealed that thegroups did not differ significantly (F < 1). However,they did perform differently over the runs of the subpattern, significant differences being obtained for theinteractions of groups X runs [F(4,24) =4.12, P <.01] and groups X day X runs [F(36,216)=1.54, p <.OS]. Subsequent Newman-KeuIs tests based on thesignificant triple interaction revealed that Group GPtracked the O-pellet item, running, with one exception, more slowly to it than to any other item onDays 8, 9, 10, and 11; on Day 9 the l-pellet and 0-
pellet speeds did not differ (other ps < .01). Statistically, there was no evidence at any point in trainingthat Group BP was able to anticipate the O-pelletitem. On Days 9-11, Group GP ran more slowly thanGroup BP to the O-pelletitem (ps< .01), but no otherdifference between the groups was significant.
Figure 2 shows running speed on each of the 10runs of each day of extinction for each group. Threeimportant relationships may be discerned in Figure 2.First, on each day of extinction, speed over runs declined more rapidly in Group GP than in Group BP.Second, on each day of extinction, when alley brightness was changed between Runs 5 and 6, Group GP,but not Group BP, showed a large and immediate increase in running speed. Third, on each day of extinction, Group BP, but not Group GP, showed an increase in running speed from Run 6 to Run 7. Therewere two reasons for taking the speed increases fromRuns 5 to 6 in Group GP and from Runs 6 to 7 inGroup BP seriously. These increases occurred oneach of the 2 extinction days and, for each group, ata point where in the prior acquisition phase 14 pelletshad been provided. In acquisition, Group GP had received 14 pellets coincident with a change in alleybrightness and Group BP had received them after aO-pellet run had occurred in a changed alley.
An analysis of variance over the speeds shown inFigure 2 revealed that the groups differed significantly [F(1,6) =52.34, p < .01], and that speed overruns declined more rapidly in Group GP than inGroup BP, the groups X runs interaction being significant [F(4,24) = 9.72, p < .01]. In order to evaluate
1110
DAY
,9
IN EACH
~._._.• B P
+------+ G P
0 120wtJ)
105<,::IE0 90
'-""
c 75ww 6011.tJ)
45....I
<C 30 ....-....-....I-~0
I- 15 ....
q1-2 8
TRIALS
Figure 1. Speed of running for Groups GP and BP on each of the runs of the subpattern for Days 1 and 2 combined and oneach of Days 8-11.
SERIAL LEARNING 11
+-------+ G P.-._.-.+ B P
.;-.-... _.+._.-+ .. -+-.-I \ I +._...
i "' Ii "' I
\. I\ i
..--..ow 120en<,
::E 100U--- 80cw 60wa.en 40
..J
« 20~
0 0....1 2 3 4 5 6 7 8 9 10
DAY 1
TRIALS IN
1 2 3 4 5 6 7 8 9 10
DAY 2
EAC H DAY
Figure 2. Speed of running on each of the 10 runs of each day of extinction for each group.
the effect of the brightness change from Runs S to 6,an analysis was performed using speeds from Runs Sand 6 for each group on each day of extinction. Importantly, the groups x runs interaction was significant in this analysis [F(l ,6) = 20.28, p < .01]. Subsequent Newman-Keulsposttests indicatedthat Group GP,but not Group BP, showed a significant increase inspeed from Run S to Run 6 (p < .01). A similaranalysis was performed for Runs 6 and 7. The groupsx run interaction was significant [F(1,6) = 11.80, p <.01], and subsequent Newman-Keuls tests indicatedthat Group BP, but not Group GP, showed a significant increase in speed from Run 6 to Run 7 (p <.01). The decline in speed from Run 6 to Run 7 inGroup GP was not significant.
Sinc~, in acquisition, Group GP ran more slowlythan Group BP, it was considered advisable to perform an additional analysis to better determine if, indeed, Group GP was less resistant to extinction thanGroup BP. The speeds of each group on the last dayof acquisition, Day 11, were compared with those inextinction, Days 1 and 2 of extinction being combined into a single day. Group BP, of course, ranmore rapidly than Group GP [F(I,6)=7.31, p< .03].Importantly, however, the groups x days interactionwas significant [F(l,6) =20.49, p < .01]. SubsequentNewman-Keuls tests based on the groups x daysinteraction revealed that-Group GP, but not Group BP,ran significantly more slowly in extinction than onthe last day of acquisition (p < .01), leaving littledoubt that Group GP was less resistant to extinctionthan Group BP.
DiscussionIn Experiment 1, 0 pellets was better tracked by
Group GP than by Group BP, a finding consistentwith that previously obtained by Fountain et al. (1983)using spatial and temporal phrasing cues rather thanthe brightness phrasing cues used here. However, twonovel and, we suggest, revealing findings were obtained in Experiment 1. First, Group GP was muchless resistant to extinction than Group BP. Second, inextinction a change in alley brightness was associatedwith significantly increased vigor of responding inGroup GP but not in Group BP. Consider these twoextinction findings in turn. First, according to thepresent hypothesis, the more strongly O-pellet items,which of course occur exclusively in extinction, signalreinforcement, the more vigorous will responding bein extinction. The capacity of 0 pellets to signalreinforcement was greater in Group BP than inGroup GP. This was so because, for Group BP, 0pellets in acquisition was a signal for 14 pellets.However, for Group GP, 14 pellets was signaled notonly by 0 pellets, but also by a change in alley brightness, the latter cue overshadowing the former, andapparently to a significant extent. Additional evidence that, for Group GP, the change in alley brightness overshadowed 0 pellets as a signal for 14 pelletswas provided in extinction. When, on each of 2 days,alley brightness was changed in extinction, Group GP,but not Group BP, showed a marked increase invigor of responding, demonstrating clearly that achange in alley brightness was a signal for 14 pelletsfor Group GP but not for Group BP. We may infer
12 CAPALDI, VERRY, NAWROCKI, AND MILLER
from this finding that the rats in Group GP remembered the alley brightness from the previous run,compared it with that on the current run, and, if thetwo differed, ran fast in anticipation of a 14-pellet reward. This would appear to involve the same generalsorts of processes of interest in the delayed matchingto-sample situation (e.g., Maki, Moe, & Bierley,1977). Whenever an external stimulus overshadows 0pellets as a signal for reward, extinction will be rapid.This explains why following well-established discrimination learning in which the positive stimulus is100010 rewarded and the negative stimulus is 0% rewarded, extinction is rapid-as rapid as it is in a consistently rewarded control group (e.g., Brown &Logan, 1965). In this instance, the positive brightnessstimulus rather than the O-pellet item is the signal forreward in the discrimination group. Such a discrimination group, then, is similar to Group GP here inthat some stimulus other than the O-pellet item is asignal for reward.
Above, we emphasized the strength of an interitemassociation, 0-14, in order to explain extinction.Emphasizing this same interitem association, but nowin describing how its strength contributes to interiteminterference, will serve to explain the better seriallearning achieved by Group GP than by Group BP.Let us see how the brightness phrasing cue, by becoming a signal for 14 pellets for Group GP, reducedinteritem interference and thus facilitated seriallearning. The capacity of an item to signal anothermay be acquired directly or indirectly through generalization from items similar to it (Capaldi & Molina,1979; Capaldi et aI., 1980). When one item precedesanother, the first may directly acquire a tendency tosignal the second. But the signal tendency acquireddirectly by an item may differ from the signal tendencies supplied indirectly to that item from otheritems, producing considerable interference. Thus, anitem that has acquired a strong direct tendency to signal 0 pellets may nevertheless produce poor trackingof 0 pellets if indirectly acquired tendencies suppliedby highly similar items strongly signal not 0 pellets,but reinforcement. This describes why tracking of 0pellets was poor in Group BP. In Groups BP andGP, the l-pellet item was the discriminative stimuluspreceding 0 pellets. We may assume, then, that inboth groups the I-pellet item was an equally strongdirect signal for 0 pellets, and to the extent that tracking of 0 pellets occurred in each group it was becausethe l-pellet item became a direct signal for 0 pellets.But this direct signal capacity, which produced tracking of 0 pellets, was opposed by indirect or generalized signal capacity of other items that signaled reward. Most notably, the l-pellet item would receiveconsiderable indirect signal capacity from the O-pelletitem. But, as we saw above in connection with extinction, the O-pellet signal for 14 pellets was muchstronger for Group BP than for Group GP. Thus, the
indirect tendency of the l-pellet item to signal 14pellets was much stronger for Group GP than forGroup BP, which explains the differences betweenthese groups in both tracking behavior and extinction.
Deserving comment is the finding, not directly relevant to the major concerns of this report, that oneach of 2 days of extinction Group BP showed a significant increase in running speed from Run 6 toRun 7. This finding was obtained presumably because, in acquisition, Group BP was rewarded with14 pellets after a O-pellet run had occurred in achanged alley. Thus, when these conditions prevailedin extinction, Group BP showed an elevation in speedfrom Run 6 to Run 7. For Group BP, then, althoughopellets was a strong signal for 14 pellets, a O-pelletrun after a change in alley brightness was an evenstronger signal. Of course, runs in another alleyalone did not predict 14 pellets for Group BP, precluding overshadowing of 0 pellets by brightness cuesas for Group GP.
EXPERIMENT :1
Experiment 1 provided strong evidence that phrasing cues are, in effect, overshadowing cues thatweaken interitem associations. But, although themethod of phrasing used in Experiment 1 (which, tofacilitate comparison of results, was similar to thatused in Hulse's laboratory) was adequate for testingthe present hypothesis, it was not completely optimalfor that purpose. Experiment 2 differed from Experiment 1 in two major respects. First, series werephrased in a manner more in conformity with thepresent hypothesis. Second, in order to provide certain additional tests of the present hypothesis, phrasing cues were removed for some groups prior to extinction.
Consider how the long series was phrased for GroupGP in Experiment 1. One complete repetition of the14-7-3-1-0 subpattern occurred in one brightness alternative, the next complete repetition occurred inthe other brightness alternative, and so on. Call thisprocedure, for convenience, phrasing completerepetitions of the subpattern. According to the present hypothesis, the findings obtained for Group GPwere not dependent upon phrasing complete repetitions of the subpattern, something which might,however, be emphasized in a rule-learning model(see, e.g., Fountain et al., 1983). Rather, accordingto the present hypothesis, Group GP behaved as itdid because the first item of the repetition of the subpattern, 14 pellets, occurred in a changed brightnessalternative. Such first-item phrasing, as we shall callit, is sufficient, according to the present hypothesis,to allow the brightness change to overshadow the 0pellet item as a signal for reinforcement. Experiment 2tested this view. It examined first-item phrasing by
employing what may be conceptualized as a shortfour-item series, 10-2-0-10. Experiment 2 employed abrightness phrasing condition, as in Experiment I,and a temporal phrasing condition. Phrasing occurred as follows. The series was segmented into twosubpatterns, the initial three items, 10-2-0, and theterminal item, 10. The initial three items occurred inone brightness alternative (or at a short interval between items), the terminal 10-pellet item in anotherbrightness alternative (or after a long interval). Alsoemployed in Experiment 2 was an unphrased groupwhich received all items at the same short intervaland, on a given day, in the same brightness alternative.
In the Fountain et al. (1983) investigation, which,as indicated, employed repetitions of the 14-7-3-1-0subpattern, for some groups repetitions were separated by a long interval (temporal phrasing) and forothers they occurred in different arms of aT-maze(spatial phrasing). Removal of the phrasing cues inthe Fountain et al. (1983) investigation disruptedtracking of the subpattern, with such disruptionbeing more severe in the temporal phrasing groupthan in the spatial one. Our interpretation of suchdisruption is that removal of the phrasing cue allowed the terminal O-pellet item ofthe 14-7-3-1-0subpattern to now become a strong signal for 14 pellets.Furthermore, where disruption was greater, as it wasfor the temporal group, the more strongly was 0pellets converted into a signal for 14 pellets. Fromthis it follows that the more disrupted groups shouldalso show greater subsequent resistance to extinction;as the capacity of 0 pellets to signal reinforcement increases, so too should resistance to extinction. Thishypothesis was tested in Experiment 2. In Experiment 2, half the rats that had received a brightness ortemporal phrasing cue in Phase 1 had this cue removed in Phase 2, thus being shifted to the unphrased series. All remaining rats, including those inthe unphrased group, were trained in Phase 2 as theyhad been in Phase 1. At the termination of Phase 2,all rats were extinguished.
MethodSubJeetI. The subjects were 30 rats of the same description as in
Experiment 1.Apparatal. The apparatus was the same as that used in Experi
ment 1.PretrwlDi.... Pretraining was identical to that used in Experi
ment 1. At the end of pretraining, the rats were divided into fivegroups of six rats each. There were three experimenters, each ofwhom ran two rats from each group. There were two replications,each experimenter running five rats in each replication, one fromeach group.
Phue 1 procedure. All rats received the series 10-2-0-10. Thisseries occurred twice each on Days 1 and 2 of experimental trainingand four times each day on Days 3-9. Thus, the interval separatingrepetitions of the series was 10-20 min. For four of the groups inPhase 1, the series was segmented into the initial three items, 102-0, and the terminal item, 10. For the fifth group, Group N-N,the series was unphrased, all items of the series occurring at a 20-
SERIAL LEARNING 13
sec interval in the black (B) runway on odd days and in the white(W) runway on even days. Groups T-T and T-N were treated exactly as was Group N-N in Phase 1, with one exception: A 10min interval (approximately) separated the third and fourth itemsof the series. Groups B-B and B-N were also treated as was GroupN-N, with one exception: The fourth item of the series occurred inan alley of a brightness that was different from that of the preceding three items, that is, BBBW on odd days and WWWB oneven days. All other aspects of the experimental procedure in Experiment 2 were as in Experiment 1, except for the following: Eachexperimenter brought five rats into the experimental room, onefrom each group. Running orders were randomized daily. Each ratwas run such that it received all four runs of a series before receiving the next repetition of the series 10-20min later.
Phue 2 procedure. Phase 2 training occurred on Days 10-17. Itwas identical to Phase 1 except for the following. On all days ofPhase 2, including the first 2 days, there were four repetitions ofthe series. Groups T-T, B-B, and N-N continued to be trained as inPhase 1, with Groups T-N and B-N being trained as was Group NN; that is, for Groups T-N and B-N, the temporal and brightness phrasing cues, respectively, were removed.
ExtInetIoD. Extinction occurred on Days 18-20. Each rat receiveda series of four nonreinforced runs in succession at a 2O-sec interval. There were four of these four-run series each day, the intervalseparating repetitions of the series being about 10 min. All runsoccurred in the white runway on Days 18 and 20 and in the blackrunway on Day 19. All other aspects of the procedure were as inExperiment 1.
ResultsFigure 3 shows speed of running on each of the
four runs of the 10-2-0-10 series on Days 1 and 2combined and on each day thereafter in Phase 1, forGroup N-N, for Groups T-T and T-N combined (T),and for Groups B-B and B-N (B). Perfect tracking ofthe 10-2-0-10 series would consist of slow running onO-pellet runs, somewhat faster running on 2-pelletruns, and fast and nondifferential responding on thetwo lO-pellet runs of the series. At one extreme wasGroup N-N, which deviated substantially from perfect tracking, and which, even when tracking inPhases 1 and 2, ran only slightly more slowly on 2and O-pellet runs than on 10-pellet runs. At the otherextreme, perfect tracking was achieved by Groups TT and T-N on the final days of Phase 1, these groupsrunning more slowly on 2- and O-pellet trials as earlyas Day 3 of Phase 1. In Phase 2, the nonshiftedgroup, Group T-T, maintained perfect tracking withlittle exception. In between these extremes was thenonshifted Group B-B, which finally achieved perfect tracking on some of the final days of Phase 2.In Phase 1, and on the earlier days of Phase 2, GroupB-B deviated from perfect tracking in runningequally slowly on 2- and O-pellet runs and in runningmore slowly to the terminal than to the initial 10pellet item.
An analysis of variance, using groups, runs, repetitions of the series, days, and experimenters as factors, was performed over Phase 1. In this analysis,Days 1 and 2 were treated as a single day and GroupsT-T and T-N were treated as a single group, as wereGroups B-B and B-N. This analysis revealed significant differences for groups [F(2,21) =S.88, p < .01],
14 CAPALDI, VERRY, NAWROCKI, AND MILLER
125-0
t!v\j\}~wtJ) 100<,
::::E0 75-Cw
~wa. 50
~- B
tJ) - N...J
<C 25 ~.... - T0.... a
1 - 2 3 4 5 6 7 8 9
TR IALS I N EACH DAY
Figure 3. Speed of running on each of the runs of the 10-2-0-10 series on Days 1 and 2 combined and on each of the daysof Phase 1 thereafter for each of the five groups.
for groups X runs [F(6,63) = 1.86, p < .005], and forgroups x runs X days [F(42,441) = 1.86, p < .005].Subsequent Newman-Keuls tests based on the significant groups X runs X days interaction indicated thatby the end of Phase 1 even Group N-N was givingsome evidence of tracking. Some important differences revealed by the Newman-Keuls tests were asfollows. The T groups ran more slowly on 2- and 0pellet runs than on 10-pellet runs as early as Day 3and, on each of the last 3 days of Phase 1, manifestedperfect tracking (ps < .05 or better). The B groups,by as early as Day 4, ran more slowly on 2- and O-pellet runs than on the initial lO-pellet run; however, onmany of the days of Phase 1 (3, 5, 7, 9), the B groupsran more slowly on the terminal than on the initial10-pellet runs (ps < .05 or better). Moreover, onsome days (3, 4, 5, 7), running was no faster on theterminal run than on 2- and O-pellet runs. By Day 8,Group N-N ran faster on its initiall0-pellet run thanon each of its subsequent runs (ps < .05). On Day 9,Group N-N was slower on its 2-pellet run than on itsterminallO-pellet run (p< .05).
Figure 4 shows speed on running on each of thefour runs of the 10-2-0-10 series on each of the daysof Phase 2 for Group N-N and the two unshiftedgroups, Groups T-T and B-B. Group T-T, withminor exceptions to be noted, showed perfect trackingon each of the days of Phase 2. In Group B-B, tracking improved over Phase 2 until, by the final days, itwas tracking all items very well indeed. Group N-N,although running much faster than Groups T-T andB-N on 2- and O-pellet trials, showed some evidenceof tracking, mainly in running more slowly on 2- andO-pellet runs than on lO-pellet runs.
Figure 5 shows speed of running on each of thefour runs of the 10-2-0-10 series on each of thedays of Phase 2 for Group N-N and the two shiftedgroups, Groups T-N and B-N. Tracking was disputedin both shifted groups, but earlier and more drastically in Group T-N than in Group B-N. For example,on Day 2 of postshift, Group T-N ran equally rapidly on all runs, whereas Group B-N continued totrack reasonably well except for running slowly onthe terminal 10-pellet run, a tendency that was alsostrong in Group B-N in Phase 1. Over days, GroupB-N sped up on the 2- and O-pellet runs and, by theend of Phase 2, was tracking about as well as GroupN-N. Group T-N, however, continued to track morepoorly than Groups B-N and N-N, mainly in failingto slow down on 2-pellet runs.
An analysis of variance performed over Phase 2revealed that the groups did not differ significantly[F(4,15) = 2.50, p > .05]. However, significant differences were obtained for groups x runs [F(12,4S) =6.74,p< .001]and for groups x runs x days [F(84,13s)=1.45, p < .05]. Some important findings revealedby subsequent Newman-Keuls tests, breaking downthe significant groups x runs X days interaction,were as follows. Group T-T showed perfect trackingon each of the days of Phase 2 except for Days 1 and4. On Day 1, the 2- and O-pellet runs failed to differ, and on Day 4, running on the terminal 100pelletrun was slower than it had been on the initial 10pellet run (p < .05). Group B-B showed perfect tracking on Days 7 and 8, missing it on Day 9 because ofslower running on the terminal than on the initial 10pellet run (ps < .05 or better). On most of the days ofPhase 2, Group T-N ran equally rapidly on all runs.
SERIAL LEARNING 15
8
, ,
--- B-B
- N-N
- T-T
, , , , , , , , , , , , , ,2 3 4 5 6 7
TRIALS IN EACH DAY
ow 125(J)<,
::IEo 100
Qw 75wQ.(J)
50...J
c:a:I- 250I- q: ! ,
1
Figure 4. Speed of running on each of the four runs of the 10-2-0-10 series on each of the days of Phase 2 for Group N-N andfor the two unshifted groups, T-T and B-B.
8
, , ,7
, ,
.'-·_·-4 T - N
6
DAY
, ,, ,5
EACH
4
IN
-N-N
2 3
TRIALS
"'1"1'"
~\ -.».-.r" _.A
\
\
\
\
..............
.-._._.• B - N
1
, ,
1 25.-oW(J)<, 100::IE0
.......-Qw 75wQ.(J)
...J 50c:a:I-0I-
25
~
Figure 5. Speed of running on each of the four runs of the 10-2-0-10 series on each of the days of Phase 2 for Group N-N and forthe two shifted groups, T-N and B-N.
By Day 8, however, Group T-N ran more rapidly onits initial to-pellet run than on any other run (ps <.OS). On each of the last 3 days of postshift, GroupsN-N and B-N ran more slowly on 2- and O-pellet runsthan on IO-pellet runs, indicating that by the end ofPhase 2 these groups were tracking about equallywelland better than Group T-N.
In extinction, all rats received four extinction seriesof four runs each on each of 3 days. Speeds on the
first run of each of the four series were summed eachday, as were speeds on Run 2, and so on. These meanspeeds are shown in Figure 6 for each group on eachday of extinction. Group T-N ran fastest in extinction, Groups N-N and B-N were intermediate, andGroups T-T and B-B were slowest, with Group T-Tbeing slower than Group B-B on the final runs ofDay I of extinction. An analysis of variance over the3 days of extinction revealed the following. There
16 CAPALDI, VERRY, NAWROCKI, AND MILLER
Figure 6. Speed of runmng for each of the five groups on eachof the four runs of the extinction series on each day of extinction.
tinction. Groups T-T and B-B of Experiment 2 did,indeed, show superior tracking and more rapid extinction than the remaining groups. In this, they werelike Group GP of Experiment 1, in which completerepetitions of the subpattern were phrased. Viewingthe findings of Experiments 1 and 2 together stronglysuggests that the behavior of Group GP, like that ofGroups B-B and T-T, was due to first-item phrasingonly. First-item phrasing clearly identifies the strengthof the interitem association between the O-pellet itemand the subsequent reinforced item as the major factor in the results reported here.
Although our phrasing procedure and series in Experiment 2 differed substantially from those employed by Fountain et a1. (1983), our results werehighly similar to theirs. In the Fountain et a1. (1983)investigation, tracking behavior was more disruptedby removal of a temporal phrasing cue than it was byremoval of a spatial phrasing cue. Similar findingswere obtained in our Experiment 2, except that theretemporal phrasing was compared with brightnessphrasing rather than with spatial phrasing. Our interpretation of such disruption is that removal of thephrasing cue, which overshadowed the O-pellet item,now allowed that item to become a signal for 10pellets in the 10-2-0-10 series. The signal strengthnow acquired by the O-pellet item then generalized tothe 2-pellet item, increasing speed and reducingtracking. The greater the disruption in tracking, according to the present hypothesis, the more stronglyis 0 pellets converted into a signal for 10 pellets andthe greater should resistance to extinction be. Consistent with this deduction, Group T-N, whichshowed greater disruption of tracking behavior thandid Group B-N, also showed greater resistance to extinction than did Group B-N. Indeed, the relationship between poor tracking performance and increased resistance to extinction included all fivegroups of Experiment 2. Group T-T showed the besttracking performance and, on Day 1, extinguishedmore rapidly than Group B-B, which showed the nextbest tracking performance and extinguished nextmost rapidly. Group N-N tracked about as well asGroup B-N and better than Group T-N, and extinguished about like Group B-N, both groups extinguishing more rapidly than Group T-N. This closerelationship between tracking performance, on theone hand, and extinction performance, on the other,supports the present view that the two behaviors areclosely related and due to the same mechanism, thestrength, or lack of strength, of the 0-10 interitemassociation.
It is suggested that the no-phrasing treatment ismore similar to the brightness phrasing treatmentthan to the temporal phrasing treatment and thatsimilarity relations among these treatments can explain the effects, on tracking behavior and extinction, of removing phrasing cues. Overshadowing of
- a-a.------. B-N
- N-N
.------.. T-N
- T-T
123
TRIALS IN EACH DAY
o
50
25
75
100
125
DiscussionIn Group GP of Experiment 1, complete repeti
tions of the subpattern were phrased. In contrast, inPhases 1 and 2 of Experiment 2, Groups B-B and T-Treceived first-item phrasing only. Such first-itemphrasing, it was suggested, should be sufficient toallow the change in brightness, Group B-B, or thelonger interitem interval, Group T-T, to overshadowthe O-pellet item as a signal for reinforcement. As indicated, such overshadowing should result in bettertracking of the subpattern and subsequent rapid ex-
were significant differences due to groups [F(4,15) =9.94, p < .01], runs [F(3,45) = 110.27, P < .001],groups X runs [F(12,45) =6.06, p< .001], and groupsx runs x days [F(24,96) = 1.99, p < .05]. SubsequentNewman-Keuls tests, breaking down the significantgroup effect, revealed that Group T-N ran fasterthan all other groups and that Groups T-T and B-B,which did not differ overall, ran faster than GroupsN-N and B-N, which did not differ overall (ps <.05 or better). Evidence that Group T-T extinguishedmore rapidly than any other group was provided bybreaking down the groups x run x days interaction.Newman-Keuls tests revealed that, on Day 1 of extinction, Group T-T ran more slowly on Runs 3 and 4than did any of the other groups (ps < .05 or better).An examination of Figure 6 indicates that on Runs 3and 4 of Day 1, Group T-Twas running about asslowly as it or any group ran at any point in extinction training, suggesting a floor effect.
4:I-
oI-
Clww"'V'l
.....
uw<I'l<,
::Eu
the O-pellet item was least in Group N-N, for whichno explicit overshadowing cue was provided, intermediate in Group B-B, and greatest in Group T-T,which showed the best tracking behavior and thefastest extinction. In terms of overshadowing, then,no phrasing is more similar to moderately effectivebrightness phrasing than to more completely effective temporal phrasing. On this basis, the shift no-cuetraining in Phase 2 was more similar to the Phase 1training received by Group B-N than to that receivedby Group T-N. Thus, more of what was learned inPhase 1 was relevant in Phase 2 for Group B-N thanfor Group T-N, and so Group B-N showed less disruption (greater transfer) in Phase 2 than did GroupT-N.
GENERAL DISCUSSION
The hypothesis introduced and tested in both investigations reported here was that phrasing cues canovershadow interitem associations, thereby decreasing resistance to extinction and facilitating seriallearning by reducing interitem interference. This hypothesis was confirmed in Experiments 1 and 2,which employed series containing subpatterns terminating in a O-pellet item, the O-pellet item beingfollowed by a rewarded item. In Experiment 1 complete repetitions of the sub pattern were phrased,whereas in Experiment 2 only the rewarded item wasphrased. In either event, when the rewarded item wasreliably signaled by some other cue, a change inbrightness, or a long temporal interval, tracking ofthe subpattern was facilitated and resistance to extinction was reduced. The close relationship betweenserial learning performance and extinction performance is compatible with the following view.If the O-pellet item is the best predictor of reward, itwill become a signal for reward and an interitem association will be formed. If, however, some other cuealso signals reward, the O-pellet item may be overshadowed by that cue, resulting in a weakened, perhaps even totally nonfunctional, interitem association. The weaker the interitem association betweenthe O-pellet item and the subsequent rewarded item,the more rapid the extinction and the better the seriallearning due to decreased interitem interference. Additional strong evidence that phrasing cues are actually overshadowing cues was provided in Experiment 1 when, in Group GP, a change in alley brightness midway through the extinction trials resulted ina dramatic increase in vigor of responding.
Granting that rapid extinction and facilitated seriallearning were due here to 0 pellets' being a relativelyweak signal for reward, there are two alternative accounts to overshadowing, both rejected here, as tohow such weakening might have occurred-generalization and retrieval, respectively. According to thegeneralization view, 0 pellets become a signal for re-
SERIAL LEARNING 17
ward in one context, for example, when the brightness alternative changed, but was not a signal for reward in anew, generalized context, for example,when the brightness alternative did not change. Thegeneralization view is highly incompatible with previous findings showing that context can be alteredmuch more extensively than was the case for GroupGP or Group B-B here, and 0 pellets will continue tobe a strong signal for reward (e.g., Capaldi, Capaldi, &Kassover, 1970; Ross, 1964). According to the retrieval view, the relatively weak capacity of 0 pelletsto signal reward was not due to overshadowing, butto a failure to retrieve O-pellet cues on rewarded runs.The retrieval approach is rejected here for two reasons. First, the dramatic increase in speed of responding shown by Group GP in extinction whenalley brightness changed indicates, as previouslymentioned, that events from the previous run wereretrieved and compared with events on the currentrun much as in delayed matching-to-sample experiments. Second, as will become clear below, there isconsiderable evidence from prior investigations thatindicates that a O-pellet event in one alley, say, a whiteone, is retrieved on the subsequent runs in anotheralley, say, a black one, and can acquire a strong tendency to signal reward.
Why did overshadowing occur here? One cuemay overshadow another because of its intensity orbecause it is the more valid predictor of reward (see,e.g., Hall, Mackintosh, Goodall, & Oal Martello,1977; Kamin, 1969). Considering the brightnessphrasing cue first, we recognize that, in Group GP ofExperiment 1 and Group B-B of Experiment 2, thechange in alley brightness could have overshadowedthe O-pellet cue as a signal for reward because ofgreater intensity. We suggest, however, that it ismore consistent with all available evidence to assumethat validity, rather than intensity, was responsiblefor overshadowing in Groups GP and B-B. In particular, evidence from brightness differential conditioning suggests that if brightness cues overshadow 0pellet cues, they do so for reasons of validity ratherthan intensity. This follows from a consideration ofthe effects of training level on N-R transitions thatoccur from a consistently nonrewarded alley of onebrightness, for example, black, to a consistently rewarded alley of another brightness, for example,white. Early in training, such transitions will result inO-pellet cues' acquiring strong control over responding; they will increase resistance to extinction in theS+ alternative or decrease vigor of responding in theS- alternative (see, e.g., Capaldi, Berg, & Morris,1975; Capaldi et al., 1984; Haggbloom, 1980b).Such evidence clearly suggests that brightness cuesare not intense enough to seriously overshadow 0pellet cues. However, it N-R transitions are introduced later in discrimination training, O-pellet cueswill fail to acquire control over behavior (Haggbloom,
18 CAPALDI, VERRY, NAWROCKI, AND MILLER
1980a). This finding suggests that the animal learnsthat the brightness cues are more valid predictors ofreward than the O-pellet cues, and as a result the 0pellet cues are overshadowed. How could validityproduce overshadowing here in Groups GP and B-B?Recognize that although the O-pellet item was itselfalways followed by reward for those groups, it washighly similar to other items signaling nonreward, 1pellet (Group GP), or 2 pellets (Group B-B) and wasin this sense invalid. To suggest that the validity ofthe O-pellet item was reduced because of its similarityto other items is not arbitrary; the alternative is toassume that validity is independent of similarity,which seems unreasonable.
It is generally recognized that organisms may employ various strategies that result in series' becoming easier to learn (e.g., Bower, 1972; Fountainet al., 1984; Restle & Brown, 1970). How organisms come to select one strategy over another isan interesting issue. The hypothesis introducedabove, that item validity decreases as interitem similarity increases appears to have some relevance forhow strategies come to be selected. If the reduction invalidity that accompanies increased item similaritywere to lead to a decreased tendency to form interitem associations and an increased tendency to associate items with phrasing cues, interitem interference would be reduced and the task of learning theseries greatly simplified. How could the organismcome to adopt such an effective strategy? Severaltheories suggest that the associability (a) of a stimulus decreases if it predicts less accurately than otherstimuli in the situation (e.g., Mackintosh, 1975;Pearce & Hall, 1980). Thus, given the relationshipbetween validity and similarity suggested above, theassociability approach is one mechanism capable ofexplaining why, as interitem similarity increases, thetendency to associate items with valid phrasing cues,rather than with relatively less valid items, shouldincrease.
Group T-T showed much better tracking and fasterextinction than Group B-B. On this basis, it may beconcluded that O-pellet cues were more effectivelyovershadowed in Group T-T than in Group B-B. Thismay be explained by assuming that although overshadowing in Group B-B (and Group OP) occurredfor reasons of validity rather than intensity, in GroupT-T overshadowing may have been due to both factors. For purposes of discussion, cues associated withthe first run after a long temporal interval are referred to as first-run cues. The greater effectivenessof first-run than of brightness cues as overshadowingcues may be related to their considerable intensity.This approach explains not only the effectiveness offirst-run cues here but also directly why long intervalsbetween series of massed extinction trials are associated with increased vigor of responding, so-calledspontaneous recovery. In acquisition, first-run cues,because they are intense, acquire a strong capacity to
signal reward. First-run cues are present only on thefirst run of a series of massed extinction trials andtherefore undergo little extinction. Thus, their reintroduction following a long temporal intervalleads to elevated responding, so-called spontaneousrecovery. But notice that the reasoning used in connection with first-run cues applies with little modification to the change in alley brightness experiencedby Group GP. Thus, the elevation of respondingshown by Group OP to a change in alley brightness inextinction is ascribed here to the same general mechanism that produces so-called spontaneous recovery,with the introduction of a cue in extinction-achange in alley brightness-itself had undergone littleextinction-being present on a few extinctiontrials-but which had acquired in acquisition astrong tendency to signal reward.
The present results suggest that the overshadowingof memory cues by phrasing cues may, in certain instances, be mistaken for forgetting. Generally speaking, the failure of some prior event to influence current performance is usually attributed to forgettingthat event. But it was found here that O-pellet cuesfailed to become a signal for reward not because offorgetting, but because of overshadowing. Overshadowing of memory cues has not been emphasizedin memory experiments. Thus, the possibility arisesthat in other experimental situations, effects attributed to forgetting may have been due to memorycues' having been overshadowed by phrasing cues.As a relevant specific example, N-R transitions arecapable of controlling responding when as much as24 h intervenes between the N item and the subsequent R item, but the extent of such control seems todecrease as the intervening interval increases (see,e.g., Capaldi, 1967; Jobe, Mellgren, Feinberg, Littlejohn, & Rigby, 1977). One interpretation of this is interms of increased forgetting of the O-pellet item withthe passage of time. Another possibility, the onebeing emphasized here, is that forgetting may beminimal, but that at relatively long intervals O-pelletcues are overshadowed by highly salient first-runcues. Clearly, overshadowing, like forgetting, suggests decreased behavioral control by O-pellet cues atlong intervals, but for entirely different reasons. Asmay be seen, however, overshadowing has the advantage of consistency in connection with the presentfindings. That is, we have ascribed the reduced behavioral control exercised by O-pellet cues at shortintervals (Group GP in Experiment 1 and Group B-Bin Experiment 2) and at long intervals (Group T-T inExperiment 2) to the same factor-overshadowing ofO-pellet cues by phrasing cues. The present analysissuggests that there may be a variety of cases in whichthe overshadowing of memory cues by phrasing cueshas been erroneously ascribed to forgetting.
Two general suggestions have been made as themost fruitful approach to animal learning. One isthat serial learning, including animal serial learning,
is unique and can best be understood within theframework of human cognitive rule-encoding models of serial learning (e.g., Fountain et al., 1984;Hulse, 1978; Hulse & Dorsky, 1977, 1979). The otheris that animal serial learning, although unique in certain respects, has much in cornmon with various moreorthodox instrumental learning situations, partialreinforcement, discrimination learning, and the like(e.g., Capaldi & Molina, 1979; Capaldi et al., 1980).The present results are relevant to this issue. Notethat commonly, in serial learning investigations, animals trained under one series are shifted to a secondseries in order to determine what was learned underthe first (e.g., Capaldi et al., 1980; Hulse & Dorsky,1979; Richardson & Kresch, 1983; Straub & Terrace,1981). This, of course, was the reason animals wereshifted to an extinction series here-to determine thestrength of an interitem association formed in a seriallearning task containing N-R transitions. The shift toan extinction series employed here, however, has amarked advantage: N-R transitions have been examined in a wide variety of more orthodox instrumental learning situations and with effects similar tothose obtained here. Thus, N-R transitions have beenshown to reduce discriminative responding (e.g.,Capaldi et al., 1975; Capaldi et al., 1984;Haggbloom,1980b, 1982), to retard reversal learning in discrimination tasks (e.g., Grosslight & Radlow, 1956;Haggbloom & Tillman, 1980), to reduce the simultaneous and successive negative contrast effects(Campbell & Meyer, 1971; Capaldi & Ziff, 1969),and to elevate resistance to extinction: in punishmentsituations (Capaldi & Levy, 1972), in escape situations (Seybert, lobe, & Eckert, 1974), in the S+alternative of discrimination tasks (e.g., Capaldiet aI., 1975; Haggbloom, 1980b), and in rewardschedule situations in animals (e.g., Capaldi, 1964,1967; Leonard, 1969) and people (e.g., Grosslight,Hall, & Murin, 1953). Similarities of the sort notedabove support the view that serial learning, whateverits unique characteristics, may be viewed as continuous with various more orthodox instrumentallearning situations.
As indicated earlier, the present report is the firstconcerned with extinction following serial learning.Models that do not postulate interitem associations,such as the rule-learning model, may, of course, provecapable of explaining extinction following seriallearning. This would tend to diminish the view beingexpressed here, that there is little basis at present forassuming a sharp distinction between serial learningand various more orthodox forms of instrumentallearning.
REFERENCES
BOWER, G. H. (1972). A selective review of organizational factors in memory. In E. Tulving & W. Donaldson (Eds.), Organization ofmemory. New York: Academic Press.
SERIAL LEARNING 19
BOWER, G. H., & WINZENZ, D. (1969). Group structure, coding,and memory for digit series. Journal ofExperimental PsychologyMonographs, 80(2, Pt. 2).
BROWN, R. T., & LoGAN, F. A. (1965). Generalized partial reinforcement effect. Journal of Comparative and PhysiologicalPsychology, 60, 64-69.
CAMPBELL, E. D., & MEYER, P. A. (1971). Effect of daily reward sequence on simultaneous and successive negative contrastin rats. Journal of Comparative and Physiological Psychology,74, 434-440.
CAPALDI, E. J. (1964). Effect of N-Iength, number of differentN-Iengths and number of reinforcements on resistance to extinction. Journal ofExperimental Psychology, 63, 230-239.
CAPALDI, E. J. (1967). A sequential hypothesis of instrumentallearning. In K. W. Spence & J. T. Spence (Eds.), The psychology of learning and motivation (Vol. 1). New York:Academic Press.
CAPALDI, E. J., BERG, R. F., & MORRIS, M. D. (1975). Stimulus control of responding in the early trials of differential conditioning. Learning and Motivation, 6, 217-229.
CAPALDI, E. J., CAPALDI, E. D., & KASSOVER, K. (1970). Aninstrumental partial reinforcement effect in the absence of anyovert instrumental acquisition training. Psychonomic Science,11,145-147.
CAPALDI, E. J., & LEVY, K. J. (1972). Stimulus control of punishedreactions. Learning and Motivation, 3, 1-19.
CAPALDI, E. J., & MOLINA, P. (1979). Element discriminability asa determinant of serial pattern learning. Animal Learning clBehavior, 7, 318-322.
CAPALDI, E. J., NAWROCKI, T. M., & VERRY, D. R. (1982). Difficult serial anticipation learning in rats: Rule encoding vs.memory. AnimalLearning cl Behavior, 10, 167-170.
CAPALDI, E. J., NAWROCKI, T. M., & VERRY, D. R. (1983). Thenature of anticipation: An inter- and intraevent process. AnimalLearning cl Behavior, 11, 192-198.
CAPALDI, E. J., NAWROCKI, T. M., & VERRY, D. R. (1984).Stimulus control in instrumental discrimination learning andreinforcement schedule situations. Journal ofExperimental Psychology: Animal Behavior Processes, 10,46-55.
CAPALDI, E. J., & VERRY, D. R. (1981). Serial order anticipationlearning in rats: Memory for multiple hedonic events and theirorder. Animal Learning cl Behavior, 9, 441-453.
CAPALDI, E. J., VERRY, D. R., & DAVIDSON, T. L. (1980).Memory, serial anticipation pattern learning, and transfer inrats. Animal Learning cl Behavior, I, 575-585.
CAPALDI, E. J., & ZIFF, D. R. (1969). SChedule of partial rewardand the negative contrast effect. Journal of Comparative andPhysiological Psychology, 61, 593-596.
EBBINGHAUS, H. E. (1964). Memory: A contribution to expertmentalpsychology. New York: Dover. (Original work published1885)
ESTES, W. K. (1972). An associative basis for coding and organization in memory. In A. W. Melton & E. Martin (Bds.), Codingprocesses in human memory. Washington, DC: Winston.
FOUNTAIN, S. B., HENNE, D. R., & HULSE, S. H. (1984).Phrasing cues and hierarchical organization in serial patternlearning by rats. Journal of Experimental Psychology: Animal Behavior Processes, 10, 30-45.
GROSSLIGHT, J. H., HALL, J. F., & MURIN, J. (1953). Patterning effect in partial reinforcement. Journal of ExperimentalPsychology, 46, 103-106.
GROSSLIGHT, J. H., & RADLOW, R. (1956). Patterning effect ofthe nonreinforcement-reinforcement sequence in a discrimination situation. Journal of Comparative and Physiological Psychology, 49, 542-546.
HAGGBLOOM, S. J. (1980a). Effects of training level and locus onN-R transitions on resistance to discrimination. PsychologicalRecord, 30, 419-422.
HAGGBLOOM, S. J. (1980b). Reward sequence and reinforcementlevel as determinants of S- behavior in differential conditioning. AnimalLearning cl Behavior, I, 424-428.
20 CAPALDI, VERRY, NAWROCKI, AND MILLER
HAGGBLOOM, S. J. (1982). Effect of N-R transitions during partialreinforcement pretraining on subsequent resistance to discrimination. AnimalLearning & Behavior, 10, 61-64.
HAGGBLOOM, S. J., & TILLMAN, D. J. (1980). Sequential effectson discrimination reversal. Learning and Motivation, 11,318-338.
HALL, G., MACKINTOSH, N. J., GOODALL, G., & DALMARTELLO,M. (1977). Loss of control by a less valid or less salient stimulus compounded with a better predictor of reinforcement.Learning and Motivation, 8, 14.5-1.58.
HULSE, S. H. (1978). Cognitive structure and serial pattern learning by rats. In S. H. Hulse, H. Fowler, & W. K. Honig (Eds.),Cognitive processes in animal behavior. Hillsdale, NJ: Erlbaum.
HULSE, S. H., & DORSKY, N. P. (1977). Structural complexityas a determinant of serial pattern learning. Learning andMotivation, 8, 488-.506.
HULSE, S. H., & DORSKY, N. P. (1979). Serial pattern learningby rats: Transfer of a formally defined stimulus relationship andthe significance of nonreinforcement. Animal Learning & Behavior, 7, 211-220.
JOBE, J. B., MELLGREN, R. L., FEINBERG, R. A., LITTLEJOHN,R. L., & RIGBY, R. L. (1977). Patterning, partial reinforcementand N-Iength effects at spaced trials as a function of reinstatement of retrieval cues. Learning and Motivation, 8, 77·97.
JOHNSON, N. F. (1972). Organization and the concept of memorycode. In A. W. Melton & E. Martin (Eds.), Coding processesin human memory. Washington, DC: Winston.
KAMIN, L. J. (1969). Predictability, surprise, attention and conditioning, I. In B. CampbelI & R. Church (Eds.), Punishment andaversive behavior. New York: Appleton-Century-Crofts.
LEONARD, D. W. (1969). Amount and sequence of reward inpartial and continuous reinforcement. Journal of Comparativeand Physiological Psychology,.63, 204-211.
MACKINTOSH, N. J. (197.5). A theory of attention: Variations inthe associability of stimuli with reinforcement. PsychologicalReview, 81, 276-298.
MAKI,W. S., MOE, J. C., & BIERLEY, C. M. (1977). Short-termmemory for stimuli, responses and reinforcers. Journal of Experimental Psychology: Animal Behavior Processes, 3, lS6-177.
PEARCE, J. M., & HALL, G. H. (1980). A model for Pavlovianlearning: Variations in the effectiveness of conditioned but notunconditioned stimuli. PsychologicalReview, 87, .532-.5.52.
RESTLE, F., & BROWN, E. (1970). Organization and serial patternlearning. In G. H. Bower (Ed.), The psychology of learning andmotivation. New York: Academic Press.
RICHARDSON, W. K., & KRESCH, J. A. (1983). Stimulus stringingby pigeons: Conditional strings. Animal Learning & Behavior,11,19-26.
Ross, R. R. (1964). Positive and negative partial-reinforcementextinction effects carried through continuous reinforcement,changed motivation, and changed response. Journal ofExperimental Psychology, 68, 492-.502.
SEYBERT, J. A., MELLGREN, R. L., JOBE, J. B., & ECKERT, E.(1974). Sequential effects in discrete trials instrumental escapeconditioning. Journal ofExperimental Psychology, 101,473-483.
STRAUB, R. 0., & TERRACE, H. S. (1981). Generalization of seriallearning in the pigeon. AnimalLearning & Behavior, 9, 4.54-468.
YOUNG, R. K. (1968). Serial Learning. In T. R. Dixon & D. L.Horton (Eds.), Verbal behavior and general behavior theory.Englewood Cliffs, NJ: Prentice-HalI.
(Manuscript received May 27, 1983;revision accepted for publication September 9, 1983.)