GENERALIZATION PEAK SHIFT FOR AUTOSHAPED AND OPERANT KEY PECKS

JOURNAL OF THE EXPERIMENTAL ANALYSIS OF BEHAVIOR

GENERALIZATION PEAK SHIFT FOR AUTOSHAPED ANDOPERANT KEY PECKS

STANLEY J. WEISS AND RICHARD D. WEISSMAN

THE AMERICAN UNIVERSITY

Pigeons acquired discriminated key pecking between 528- and 540-nm stimuli by either a response-reinforcer (operant group) or a stimulus-reinforcer (autoshaped group) contingency, with other train-ing-schedule parameters comparable over groups. For the birds in the operant group, key pecksintermittently produced grain in the presence of one hue on the key (positive stimulus) but not in theother (negative stimulus). For the birds in the autoshaped group, pecking emerged when grain wasintermittently presented independently of key pecking during one key color but was not presentedduring the other key color. Two independent contingency assays, peck-location comparisons andelimination of differences in reinforcement rate, confirmed the effectiveness of the two training pro-cedures in establishing operant or respondent control of key pecking. After reaching a 10:1, or better,discrimination ratio between key pecks during the two key colors, the birds received a wavelengthgeneralization test. Criterion baseline key-peck rates were comparable for operant and autoshapedgroups prior to testing. On the generalization test, performed in extinction, all birds pecked most ata stimulus removed from the positive training stimulus in the direction away from the negative stimulus.In testing, autoshaped "peak" rates (24.5 to 64.9 pecks per minute) were from 33% to 80% higherthan rates in the presence of the training stimuli. Respondent peak shift rarely has been reportedheretofore, and never this consistently and robustly. These results further confirm the similarity ofperceptual processing in classical and operant learning. They are discussed in terms of Spence'sgradient-interaction theory and Weiss' (1978) two-process model of stimulus control.Key words: stimulus generalization, peak shift, autoshaping, operant-respondent comparisons, con-

tingency assays, intradimensional color discrimination training, key peck, off-key peck, pigeons

The slope of a generalization gradient, aswell as where it peaks, is frequently used toreveal the nature of the stimulus control gen-erated by the discrimination training preced-ing the test. When a response is under inter-dimensional stimulus control, responseprobability progressively decreases as the teststimulus is removed from the training stimulusalong the continuum being studied. This ex-citatory stimulus generalization has been re-ported in classical (e.g., Hovland, 1937; Pav-lov, 1927) and instrumental (e.g., Guttman &Kalish, 1956; Hanson, 1959) conditioning. InHanson's study, pigeons' key pecks were re-inforced when the key was illuminated at ahue of 550 nm (S+) in a procedure that in-

This research was supported by NIMH Grant R03-MH42721 awarded to the first author. It was presentedat the Psychonomic Society Convention in New Orleans,1990, and the Pavlovian Society of North America inBaltimore, 1991. We thank Matthew Yarczower for theprojection apparatus and filters, Clarence Seiter of BRS/LVE for his help in designing and building the pigeonkey, and Leigh Panlilio for assistance in making the fig-ures. For reprints, write to Stanley J. Weiss, Departmentof Psychology, The American University, Washington,D.C. 20016.

cluded brief periods of darkness during whichresponding was ineffective (S-). In a subse-quent generalization test, in which wave-lengths within the range of 480 nm and 620nm were presented, a roughly symmetricalgradient of responding was produced with amaximum at 550 nm (S+).Hanson (1959) was among the first to ex-

plore the effect of intradimensional discrimi-nation training on stimulus control. His in-terdimensional group was described above. Inone of his intradimensional groups, pigeons'key pecks were reinforced at 550 nm (S+) andnot reinforced at 555 nm (S-). This discrim-ination training displaced the gradient's peakfrom S+ in a direction away from S -, thestimulus associated with extinction (EXT), ina subsequent generalization test. This peakshift was profound, with approximately fourtimes as many pecks emitted at 540 nm thanat S+ (550 nm).

Peak shift has been demonstrated many timesin operant conditioning with positive rein-forcement (see Purtle, 1973, and Weiss, 1978,Table 3, for reviews) and has also been re-ported in situations in which responding was

127

1992, 57, 127-143 NUMBER 2 (MARCH)

STANLEYJ. WEISS and RICHARD D. WEISSMAN

maintained by shock avoidance by treadlepressing in pigeons (Bushnell & Weiss, 1980)and by bar pressing in rats (Weiss & Schindler,1981). A notable contrast exists between theoften-demonstrated robust and reliable peakshift following intradimensional operant train-ing and the rarity, weakness, and inconsistencyof peak shift following respondent intradi-mensional training. Attempts to show peak shiftin respondent procedures are described in de-tail below.

In one classical conditioning study, Liu(1971) gave 8 rabbits intradimensional differ-ential conditioning in which a 1200-Hz tone(CS+) preceded a brief shock to the infraor-bital region of the eye, activating the nictitatingmembrane. A 1600-Hz tone (CS-) was neverfollowed by shock. A generalization test pro-duced symmetrical gradients, with a peak atCS+, similar to those produced by a group ofrabbits that received interdimensional (CS+only) training.

Hupka, Liu, and Moore (1969) employedintradimensional classical discriminationtraining in a design essentially similar to thatof Liu (1971), with the addition that CS+/CS- separations were manipulated over sixgroups of rabbits. These counterbalancedstimulus separations were 400 versus 1600 Hz,1600 versus 2800 Hz, and 2800 versus 4000Hz. Of the 12 mean stimulus generalizationgradients reported (two for each group over 2days), only one showed a peak shift away fromCS-. This occurred for the group for whichCS+ was 1600 Hz and CS- was 400 Hz,and occurred only in the first test. Unfortu-nately, with neither individual gradients norstatistical comparisons reported, it is difficultto evaluate the significance of this one instanceof peak shift.Moore (1972, pp. 214-217) sought addi-

tional information on the effects of differentialclassical conditioning on generalization byrunning small squads of rabbits under a widerange of intradimensional training and testingprocedures. However, only 5 of 30 subjectsproduced peak shift in their first generalizationtest, with this number increasing to 11 whenall tests (ranging from one to three per subject)were considered. Moreover, even when Mooreand his associates did obtain peak shift withthe rabbit's nictitating membrane reflex, itusually occurred when S+ was a higher fre-quency tone than S-, and not vice versa.Cowan (1968) also reported this asymmetry

when using counterbalanced CSs of 2000 and3000 Hz in differential eyelid conditioning withrabbits. Thus, in what would be considered atraditional classical conditioning preparation,the peak shift phenomenon is suggested, butthe findings are inconsistent, weak, and uni-directional.

Keller (1974) applied an analytical tech-nique introduced by Catania (1973) to identifycontrol by stimulus-reinforcer and response-reinforcer contingencies in the analysis of be-havioral contrast. In this "topographical tag-ging" paradigm, the operant manipulandum,whose operation produces reinforcers, is sep-arate and distinct from the stimulus presen-tation key that comes to control stimulus-di-rected pecks. These latter pecks are monitoredbut have no scheduled consequences. The gen-eralization of intradimensionally controlledstimulus-key directed pecks was investigatedby White and his associates (Skelton & White,1982; White & Braunstein, 1979; White &Thomas, 1979) and Weiss and his associates(Bushnell & Weiss, 1980; Weiss & Dacanay,1982). Line-tilt and wavelength dimensionswere represented that can be distinguished ac-cording to whether the stimulus-directed be-havior and operant behavior were "homoge-neous" (both key pecks) or "heterogeneous"(stimulus-directed key pecks and operanttreadle presses). The topographical taggingstudies using homogeneous behavior were per-formed by White and his associates, whereasWeiss and his associates performed their stud-ies with heterogeneous operant and stimulus-directed behavior. In general, the control ofstimulus-key responding roughly paralleledthat of operant-key responses, but was weakerand less consistent. Being concerned here withrespondent stimulus control, our discussion ofthese studies will emphasize the stimulus-keyresponses.White and Braunstein (1979) trained 4 pi-

geons on an intradimensional discriminationin which a 67.50 line (S+) was projected onthe stimulus key when operant-key responsesproduced food on a variable-interval (VI) 30-sschedule. A 22.50 line (S-) was projected onthe stimulus key during EXT. In a line-tiltgeneralization test, 2 birds produced peak shifton the stimulus key and 1 produced clear areashift. (Area shift refers to the generalizationtest outcome in which S+ controls the highestrate, but the gradient is asymmetrical becausefewer responses are produced by the stimuli

PEAK SHIFT OF AUTOSHAPED AND OPERANT KEY PECKS

on the S- side of S+ than by the stimulusvalues on the other side.) However, stimulus-key rates were exceedingly low in testing.Among the birds showing peak shift, the fastestrate was under three pecks per minute, lessthan 1/min higher than the rate to S+.

White and Thomas (1979) conducted a sys-tematic replication of White and Braunstein's(1979) study that was designed to produce morestimulus-key directed behavior in testing. Theirbirds received wavelength discriminationtraining with reinforcement maintained in S+during generalization tests. Peak shift was pro-duced in 60% of the three tests presented toeach of 2 pigeons, with peak rates in the rangeof 10 to 15 pecks per minute. The remaindershowed clear area shift.

Skelton and White (1982) replicated theWhite and Braunstein (1979) study with 3birds. The stimulus-key wavelengths used intraining were the same as those used by Whiteand Thomas (1979). None produced peak shifton the operant or stimulus keys, although allshowed area shift on the operant key and 2showed this shift on the stimulus key. Thedesign of this experiment also revealed that". . . stimulus-key responding was suppressedwhen brief timeout periods were contingent onresponses directed at the stimulus signallingreinforcement for the operant key" (p. 275),demonstrating that stimulus-key pecks weresensitive to control by an operant contingency.Skelton and White also suggested that stim-ulus-key responding in the topographic tag-ging paradigm may, to some extent, reflectgeneralization of operant-key responding.

In the heterogeneous topographical taggingstudies by Weiss and his associates, the operantwas treadle pressing with the S+ and S-wavelengths presented on a key directly abovethe treadle at about the height of a pigeon'sbeak. Bushnell and Weiss (1980) used mul-tiple VI EXT training similar to that of Whiteand his associates, and reported peak shift forthe operant response (treadle pressing) for all6 pigeons. However, only 1 of their birds pro-duced peak shift on the stimulus key, with 3others showing clear area shift.

Weiss and Dacanay (1982) trained 3 birdson a chained schedule in which treadle press-ing in the initial-link wavelength (S2) pro-duced the terminal-link wavelength (S1). Re-inforcers were delivered in Si according toa differential-reinforcement-of-other-behavior(DRO) schedule applied to the treadle press.

During S1, stimulus-directed key pecking wasrobust, with rates ranging from 17.2 to 136.3pecks per minute and S2:S1 peck discrimi-nation ratios ranging from 16:1 to 146:1. Inspite of these powerful intradimensional dis-criminations with autoshaped pecks, only 1bird produced area shift. The other 2 birdsproduced very steep, symmetrical, "knife-edge"-like gradients around the food-associ-ated stimulus.What might be concluded from these to-

pographic tagging studies about the stimuluscontrol of stimulus-directed responding? It ap-pears that peak, or at least area, shift wasproduced by these responses in the homoge-neous and heterogeneous training paradigms,but these shifts were rather weak and fragilephenomena. Taking the first generalization testof the total of 18 birds used in these studies,only 4 produced peak shift, with 8 others show-ing area shift. However, area shift and peakshift do not necessarily reveal comparable pro-cesses. Area shift can result from the fact thatS- differentially reduces the number of re-sponses on the S- side of S+ while havinglittle effect on the generalized behavior to stim-ulus values on the other side of S +. Therefore,unless specifically evaluated in relation to in-terdimensional control, area shift does not offerconvincing evidence of increased excitatorycontrol to stimuli beyond S+ in a directionaway from S -, as peak shift does.One should also consider, when analyzing

results from the topographic tagging design,that it is rather difficult to characterize con-fidently the contingencies controlling the stim-ulus-directed pecks as purely respondent, be-cause (a) the stimulus-directed pecks aresensitive to operant punishment (timeout)contingencies (Skelton & White, 1982); (b) atall stimulus values, but particularly at S+ andbeyond, there can inevitably be competitionbetween responses on the operant manipulan-dum and the stimulus key; and (c) to the extentthat operant responses can generalize to thestimulus key, the characterization of these lat-ter types of behavior as respondent may bechallenged. Although the topographical tag-ging studies have produced occasional in-stances of peak shift for the autoshaped peck,a more traditional intradimensional autoshap-ing study would be less susceptible to theseproblems, if performed adequately.

Nallan, McCoy, Pace, and Welch (1979)reported differential intradimensional auto-

129

STANLEY J. WEISS and RICHARD D. WEISSMAN

shaping along the line-tilt dimension. Theytrained 3 pigeons on a multiple variable-time(VT) extinction (EXT) schedule in which re-sponse-independent reinforcers were deliveredin the presence of one line orientation (450)but not during an orientation 30° removed (150).This was meant to establish the discriminationsolely on the basis of a respondent contingency.After this training, peak shift was not obtainedin a line-tilt generalization test. In fact, all 3birds responded at substantially lower rates tothe stimulus one step removed from their VTstimulus, in the direction away from the EXTangle, than to the VT stimulus itself. Thismakes it unlikely that the generally appliedgradient interaction dynamics of peak shift(Rilling, 1977; Spence, 1937; Weiss, 1978)were responsible for the area shift reported byNallan et al. Their findings are consistent withprevious difficulties in producing reliable peakshift with autoshaped pecks or with traditionalclassical conditioning paradigms.The generalization test data reported by

Nallan et al. (1979) would have been moreinformative if they had included a conditionin which the line-tilt discrimination was es-tablished through an operant contingency. Thisis an important control in light of previousdifficulties in obtaining peak shift on a line-tilt dimension (Hearst, 1968, 1971) and thedegree to which peak shift is stimulus andparameter bound (Purtle, 1973). Moreover,even if this operant conditioning group existed,independent assays of the control actually gen-erated in the two groups would be essential ifthe generalization results were to be linkedunambiguously with operant or respondentdiscriminative control.Most of the stimulus control phenomena

produced after operant training have also beendemonstrated with autoshaping. For example,excitatory and inhibitory gradients for auto-shaped pecks following interdimensional dis-crimination training have been reported(Tomie, Davitt, & Engberg, 1976; Wessells,1973). Further, the effects of extradimensionaltraining on the shape of generalization gra-dients were shown to be similar in the auto-shaping paradigm when respondent contin-gencies were manipulated (Tomie et al., 1976)to those obtained when operant contingencieswere also changed (Thomas, Freeman, Svi-nicki, Burr, & Lyons, 1970). This compara-bility was demonstrated under differential,

nondifferential, and single-stimulus extradi-mensional training. On the basis of these par-allels in stimulus control for respondent andoperant contingencies, as well as for most otherconditioning phenomena (Kimble, 1961;Mackintosh, 1974), generalization peak shiftshould be the likely outcome after intradimen-sional respondent discrimination training whenstimulus parameters are appropriately chosenand care is taken to insure control only by therelevant stimuli.The present experiment compared the effect

of discrimination training under operant andrespondent contingencies on postdiscrimina-tion generalization gradients. Except for thecontingency in the component associated withgrain delivery, stimulus parameters and ratesof reinforcement were kept comparable be-tween operant and respondent procedures. The12-nm conditioned stimulus (CS+:CS-) dif-ference employed is close to that proven effec-tive in producing operant peak shift in thislaboratory (Bushnell & Weiss, 1980; Weiss &Dacanay, 1982, Experiment 2), but to ourknowledge pure autoshaping has not yet beenreported with such small wavelength differ-ences.The instrumental procedure was a multiple

VI EXT schedule under which pecking inter-mittently produced grain when the key wasilluminated with one hue but not when it wasilluminated with the other. The autoshapingprocedure was a multiple VT EXT scheduleunder which grain was presented indepen-dently of the pigeons' behavior in one hue andno grain was presented in the other. The goalof this training was to establish comparableintradimensional key-peck discriminations fortwo groups of pigeons. However, for the in-strumental procedure an operant contingencywas programmed in S+, whereas for the au-toshaped procedure a respondent contingencyoperated in CS+. Prior to testing, componentdurations, reinforcement frequency, peck rates,and discriminative control were comparablebetween the two conditions.

Special precautions were taken to insurecontrol by key color and not by intensity of thestimulus, temporal parameters of the stimuluscomponents, apparatus sounds related to stim-ulus presentations, or extraneous stimuli. Inaddition, the effectiveness of these trainingprocedures in establishing operant and auto-shaped key pecks in the two conditions, re-


spectively, was evaluated with a variety of as-says traditionally used to distinguish betweenoperants and respondents. These included key-peck accuracy (Barrera, 1974; Jenkins, 1981)and manipulation of the stimulus-reinforcercorrelation (Gamzu & Williams, 1971, 1973;Rescorla, 1968).

METHODSubjectsThe subjects were 11 experimentally naive

Silver King or White Carneau pigeons 1 to 3years old at the start of training. The pigeonswere trained and tested at 75% to 85% of theirfree-feeding weights. They were maintainedon a diet of mixed grain that was also used forreinforcement. They were fed whatever extragrain was needed to maintain them at theirpredetermined weights directly after a trainingsession. Water and grit were continuouslyavailable in the home cage.

ApparatusThe training chamber measured 30.5 cm

high, 34.3 cm long, and 24.1 cm wide. Its sideand back walls were made of white 0.6-cmtranslucent plastic. The hinged top was 0.6-cm clear plastic perforated with ventilationholes, and the floor was composed of 1-cmstainless-steel hardware cloth. A 2-cm-diam-eter clear plastic stimulus key, made translu-cent by light sanding, was centered on the whitealuminum front wall 19.5 cm above the floor.The hue stimuli were projected on the back ofthis key. Surrounding this key was a 0.5-cmwhite plastic doughnut-shaped disc that wasalso reactive to pecks. Pecks that contacted thisarea were counted as "off-key" pecks. Thismade the on-key area 3.1 cm2 and the sur-rounding off-key area 3.9 cm2. "Border" pecksthat activated both keys were not differentiallycounted. A force of approximately 0.15 N op-erated the normally closed switches behind eachkey.A food magazine aperature, 5 cm wide and

6 cm high, was located below the key, withthe lower edge of this opening 3.5 cm from thefloor. When the hopper was elevated, the ap-erture was illuminated by a 28-V GE 509Kbulb. The chamber was illuminated by a 6-in.120-V showcase bulb centered vertically be-hind the rear white plastic wall. The intensityof this bulb was adjusted to make the rear wall

just bright enough to eliminate keylight pro-jections. To accomplish this, the bulb was op-erated through a variac at 75 V. The illumi-nation of the front and rear walls, measuredby a Simpson Model 408-2 Illumination LevelMeter, whose sensor was located midway be-tween them, was 0.233 and 1.153 lux, re-spectively. The chamber was enclosed in asound-attenuating chest based on the designreported by Weiss (1970). An exhaust fan wasmounted on an outside wall of this chest overthe ventilation sound baffle. A PDP-8A® com-puter, located in an adjacent room, was pro-grammed with SUPERSKED® to presentevents according to planned contingencies. Datawere recorded by the computer and a cumu-lative recorder.The visual stimuli projected on the stimulus

key were generated by Bausch and Lomb in-terference filters (44-78 series) whose peakwavelength transmissions were nominally in-dicated as 500, 510, 520, 530, 540, 550, 560,and 570 nm. However, independent measure-ment of the peak transmission values of thesefilters with a Varian Series 634 Spectropho-tometer showed them to be 498, 510, 522, 528,540, 548, 558, and 566 nm, respectively. Ko-dak 0.1 and 0.2 neutral density filters (No. 9series) were also used. The illumination sourcewas a GE 18A/TlO/2P-6-v microscope illu-minator bulb with ribbon filament and anoutput of 1,800 lumens. The collimated mon-ochromatic light transilluminated the trans-lucent stimulus key. The lamp source was 50cm from the key. Each of the 10 filters (8wavelength and 2 neutral density) was placedin a seat that could be raised by activating asolenoid. When a seat was raised, the filter itheld was placed in the path of the light beam.This is the same projection system used byBushnell and Weiss (1980) and Weiss andDacanay (1982), with Bausch and Lomb fil-ters substituted for Ditric filters.

ProcedureThe pigeons were first trained to eat grain

from the food hopper. The hopper was raisedand the aperture illuminated until the pigeonlocated the grain and ate it for approximately5 s, after which the hopper was lowered andthe aperture light shut off. This hopper train-ing continued until the bird was eating fromthe hopper promptly after aperture illumi-


nation. Of the 11 birds, 5 were assigned to theoperant group and 6 to the autoshaped group.Training progressed for both groups so as toreach a terminal baseline schedule with similarcharacteristics except for the reinforcementcontingency operating in the presence of theCS+ or S+ stimulus. However, during earlytraining, component durations were used thatencouraged acquisition of the target contin-gency (Terrace, Gibbon, Farrell, & Baldock,1975). Thus, components with reinforcementwere initially shorter for the autoshaped thanfor the operant group, and those without re-inforcement were longer.

Special precautions were taken to insurecontrol by key color and not by intensity of thestimulus, solenoid activations, or other regu-larities of the training procedure as follows:

1. When going from one component to thenext, the key was darkened for 1 s by raisinga black slide that blocked the light beam whilerelevant solenoids operated to get the appro-priate filter in the light path. Thus, a brieflydarkened key, or the solenoid activations nec-essary for filter change, could signal a com-ponent change. To control for this possibility,during both S+ and S- components the keywas occasionally darkened for 1 s, with allassociated solenoid activations. These "pseu-dotransitions" were presented, randomly, onaverage of once every 40 s. They were intro-duced into the S- components early in train-ing and into the S+ components after the dis-crimination began to stabilize.

2. Although the two colors projected on thekey looked comparably bright to the experi-menters, one could not conclude that this wasalso true for the pigeons. To control for po-tential discriminations on the basis of bright-ness rather than hue, and the real possibilitythat test hues would differ in brightness fromtraining stimuli, brightness was made an un-correlated dimension. From early in training,the 0.1 and 0.2 neutral density filters wereraised and lowered independently of each otheron average of every 15 s. This served to changethe brightness of the keys many times duringeach session without regard to the wavelengthstimulus present, while further making sole-noid operation an irrelevant factor in stimuluscontrol. When the illumination sensor wasplaced directly in front of, and almost touching,the key for both training hues (528 and 540nm), the reading was 0.186 lux when neither

neutral density filter was raised and 0.139 luxwhen both were raised.

3. To minimize the likelihood of subjects'behavior coming under temporal control of thestimulus components, the durations of thesestimuli were varied around mean values. Someof the final baseline components were com-parable to the 45-s durations presented in test-ing.

4. When a pigeon's discrimination began tostabilize, the percentage of positive compo-nents in which reinforcement was received wasgradually reduced to 75% on average. This wasdone to minimize the behaviorally disruptiveeffects the transition from training to the gen-eralization test performed in extinction mightproduce, to give us an indication in trainingof the influence on stimulus control of rein-forcement presentations, and to produce moretest behavior by increasing resistance to ex-tinction.

Operant group. After hopper training, thesepigeons were placed in the chamber with theS+ (528 or 540 nm) keylight on continuously.Key pecking was generated through differ-ential reinforcement of successive approxi-mations, usually within one session. After asession of 35 to 45 key-peck-contingent rein-forcements on a continuous schedule, a mul-tiple variable-interval 1 0-s extinction schedule(mult VI 10 s EXT) was introduced. S+ andS- were counterbalanced across subjects overkeylight wavelengths of 528 and 540 nm. Ini-tial S+ durations averaged 30 to 40 s, withthese components ranging from 50% to 150%of the mean within a session. Initial S- com-ponent durations were chosen that did notoverly disrupt a bird's pecking in S+. Thesewere 80, 120, and 40 s for Birds 2, 5, and 9,respectively. S- component durations weretypically within 50% to 200% of the mean,with only occasional shorter durations.As the cumulative records and peck rates

revealed that a discrimination was developing,the VI schedule was gradually increasedthrough VI 15 and VI 22 until VI 30 wasreached. On all VI schedules, interval dura-tions were arranged to avoid systematic se-quencing among durations. The limits of allVI schedules were 1 s to three times the mean.S+ and S- durations were also graduallymodified as discriminative control developeduntil they were approximately in a 1:1.5-2ratio. To reduce Bird 2's pecking in S-, late

132


in training a contingency was introduced thatrequired no response during the final 10 s ofeach S- component before the component wasterminated.

Final baseline. On the final operant-trainingbaseline schedule, the mean S+ duration was60 s for all birds, and the mean S- durationwas 120, 90, and 90 s for Birds 2, 5, and 9,respectively. A VI 30-s food schedule operatedin S+. Training continued until the S+ key-peck rate was at least 10 times the S- rate forfour consecutive sessions, with no trend in rates.Then a stimulus generalization test was ad-ministered. Birds 2, 5, and 9 received 18, 30,and 32 mult VI 30-s EXT sessions, respec-tively. One bird in the operant group was elim-inated prior to testing because, after 18 sessionsof intradimensional training, its S+/S- dis-crimination ratio had reached only 1.6:1.

Stimulus generalization test. This test con-sisted of 12 randomized blocks of eight stimuli.Each block contained 45-s presentations ofstimuli of 498, 510, 522, 528, 540, 548, 558,and 566 nm. The order was made unsystematicwithin each block of eight stimuli, with caretaken to avoid long series of positive or negativestimuli that might produce induction or con-trast effects. A 1-s dark key separated succes-sive stimulus presentations, making each testblock 368 s long. Testing commenced after asubject was run on its baseline schedule, withreinforcement, for 15 to 20 min. Reinforce-ment, as well as pseudotransitions and the op-eration of neutral density filters, was discon-tinued during testing. The complete 73.6-mintest was administered in one session. After test-ing, a bird was returned to its final operantbaseline schedule for several days.

Autoshaped group. Following hopper train-ing, the pigeons were placed on a schedule inwhich, on the average of once every 165 s, theCS+ replaced the CS- for 15 to 21 s. Tenseconds into the CS, the hopper was raised forapproximately 3 to 5 s, with the CS remainingon for 2 to 6 s after the hopper was loweredand the aperture light extinguished. The pi-geons received about 36 CS+-grain pairingsduring each session, and were exposed to thiscontingency until pecking emerged. For theinitial birds in this group, including Bird 3who met criterion, this was entirely automated.However, our observations revealed that withthe automated procedure, CS+-grain pairingswere often presented when the bird wasn't

facing the front wall (where the key was lo-cated). This seriously interfered with training,perhaps because of the small wavelength dif-ferences employed in this experiment. There-fore, the contingency described above wasslightly modified such that CS+ (key-colorchange)-grain pairings were initially pre-sented when the pigeon was facing the frontwall. Prior to instituting this modified initialtraining procedure, 2 birds were eliminatedfrom this group because pecks were not elicitedby the CS+ within eight sessions. They en-gaged in "bobbing" behavior that appeared tobe under the control of some combination ofthe auditory and visual aspects of the 1-s dark-key stimulus transitions and the hopper op-eration.

After key pecking emerged on the contin-gency described above, which usually took onlytwo to four sessions on the modified procedure,the CS was lengthened to average 20 s, notcounting hopper durations, and the extinctioncomponent continued to average 165 s. Duringthe CS a VT 10-s grain presentation scheduleoperated. As differential control emerged, thelength of the CS+ and the value of the VTschedule were gradually increased while theinter-CS period (CS-) was gradually de-creased. The terminal training schedule wasa multipleVT 30-s EXT schedule where CS+saveraged 60 s and CS-s averaged 105 s, withthe within-session range of these componentdurations like those described earlier for theoperant group. When key pecks occurred at10 times the rate in CS+ as in CS- for fourconsecutive sessions, the same stimulus gen-eralization test administered to the operantgroup was given to the autoshaped group. Af-ter testing, a bird was returned to its finalautoshaping baseline schedule for several days.

Birds 3, 18, and 19 received 52, 7, and 9sessions on the mult VT 30-s EXT schedule,respectively, prior to testing. At least some ofBird 3's additional training sessions can beattributed to the fact that many of the controlprocedures, such as pseudotransitions andneutral density filter operation, were intro-duced and perfected during its sessions.Two birds did not form a stable discrimi-

nation between CS+ and CS- and thereforewere not tested. One of these reached only a3.8:1 discrimination between CS+ and CS-after 34 sessions on mult VT 30 EXT underthe automated acquisition procedure because

133

STANLEY I. WEISS and RICHARD D. WEISSMAN

Table 1

Response rates (pecks per minute) for the last 4 days ofdiscrimination training for birds of the operant and au-toshaped groups.

8+ s- Off keyas % of

Subject On key Off key On key Off key on key

Operant group2 98.2 33.5 7.4 5.8 34.15 88.8 37.0 8.8 6.9 41.79 74.6 1.6 5.3 1.0 2.1M 87.2 28.8 7.2 4.6 26.0

Autoshaped group3 64.4 40.7 2.9 2.9 63.2

18 102.9 68.0 8.2 4.3 66.119 60.0 66.9 5.6 2.9 115.5M 75.8 58.5 5.6 3.4 80.3

pecks appeared to be prompted by pseudo-transitions in CS-. Bird 17, who acquiredpecking on the modified acquisition procedure,reached a CS+:CS- discrimination ratio of15:1 after only 3 days on mult VT 30 EXTwhen the hopper bulb burned out during a

training session. The equipment malfunctionseverely disrupted this bird's discrimination.With nine additional training sessions, its dis-crimination ratio returned to no better than 5:1, and its behavior in CS- was clearly lessorderly than before the mishap.

Behavioral assays for contingency control. Toassist in interpreting the results of these gen-eralization tests, independent assays were in-troduced to distinguish whether our trainingprocedures were indeed successful in gener-ating different types of control in the auto-shaped and operant groups. First, the distri-bution of on- and off-key pecks by membersof the two groups were compared. Also, fol-lowing the generalization test and return tobaseline sessions, a VT 30-s schedule of grainpresentation was introduced in what were pre-viously the extinction components. Thus, theprocedure for the autoshaped pigeons was nowa mult VT 30-s VT 30-s schedule and theprocedure for the operant pigeons was a multVI 30-s VT 30-s schedule. Birds were run on

these schedules until they pecked at less thantwo pecks per minute for 2 days or their peck-ing showed no downward trend. Next theywere returned to their pregeneralization testbaselines for several days, either mult VT EXT(autoshaped group) or mult VI EXT (operantgroup).

RESULTS AND DISCUSSION

Baseline Training and Generalization Tests

The terminal, pregeneralization test, base-line key-peck performance is presented in PanelI of Figure 1 for the birds of the operant groupand in Panel I of Figure 2 for the birds of theautoshaped group. The discriminations be-tween positive and negative stimuli are com-parable for the two groups, with their absolutekey-peck rates overlapping considerably. Ta-ble 1 presents these response rates averagedover the last 4 days of discrimination trainingfor each subject. Rates for the operant groupranged from 74.6 to 98.2 pecks per minute (M= 87.2), and the rates for the autoshaped groupranged from 60.0 to 102.9 pecks per minute(M = 75.8). However, these groups differedin the number of off-key pecks relative to on-key pecks, but these data will be discussed laterwhen contingency assays are considered.

Figure 3 presents the absolute generaliza-tion gradients for the operant and autoshapedgroups. Peak shift was unequivocally producedby all subjects in both groups for both on- andoff-key pecks. This is the most robust and con-sistent report of peak shift for an autoshapedresponse to date, and one of the few instancesof operant peak shift with training stimuli inthis range of the spectrum. The highest peckrate to a stimulus removed from S+, in a di-rection away from S -, was significantly greaterthan the rate to S+ for the autoshaped, t (2)= 4.42, p = .048, and the operant, t(2) = 6.63,p = .022, groups. The relative gradients (Fig-ure 4) show that for each bird, irrespective ofgroup, the shapes of the on- and off-key peckgradients are overlapping and indistinguish-able from each other.

Contingency AssaysKey-peck accuracy. A number of investiga-

tors have observed that autoshaping is oftencharacterized by off-key pecking (Barrera,1974; Jenkins, 1981; LoLordo, McMillan, &Riley, 1974). In fact, a majority of Jenkins'pigeons produced over four times as many off-key pecks, relative to on-key pecks, under au-tomaintenance than under intermittent oper-ant reinforcement. Table 1 presents off-keyand on-key peck rates for operant and auto-shaped birds during criterion baseline sessions.The difference between groups is dramatic. Inabsolute terms, as well as relative to on-key

134


160-

140r | l #2 (Operant)120- 1

2 I] II ] 11~~~1ON +

100- # On80 "0 OFF+

60-

40- UJp20-

-4 -3 -2-1 1 2 1 23456 12 3

140-

120 ~~~~~#5(Operant)100 I III80-

160-

40- 0- 0

20

-4-3 -2-1 1 23 1 23 4 56 7 891011 12 3

100- #9 (Operant)

80-

60-

40-

20-

0~~~~~~~~~~~~~~~~~~r-

1 2 3 4 5 6 7 8

VI VT

Session

4-3-2-1 1 2

VI EXT VI EXT1 2

VI EXT

Fig. 1. Responses per minute for the birds trained in the operant procedure, for the last four baseline sessions (I),posttest sessions (II), contingency assay sessions with the mult VI 30-s VT 30-s schedule (III), and sessions with themult VI 30-s EXT baseline (IV). In each panel, on-key and off-key peck rates are shown.

rates, all autoshaped birds had more off-keypecks than any operant bird. On average, forthe autoshaped group off-key peck rates inCS+ were 80.6% of on-key rates. In compar-ison, during S+ this figure was only 28.6% for

the operant group, a 52% difference betweengroups.A measure of the behavioral stability of each

subject's key-pecking localization could in-crease our confidence in this measure of con-

C',

0

C')

CD,U,N

0U,


-4 -3 -2 -1 1 2 1 2 3 4 5 6 7 8 9 10111213 1 2 3

160-#18 (Autoshaped)

140-120- I II III IV100-

80#60-

40-

20 -I

0 TrrTI ritI-4-.3-.2-1 1 2 12 34 56 7 89 12 3

100#19 (Autoshaped)

80-

60-

40-

20-

0-4 -3 -2 -1

VT EXT1 2 1 2 3

VT EXT VTVT1 2 3

VT EXT

SessionsFig. 2. Responses per minute for birds trained in the autoshaping procedure, for the last four baseline sessions

(I), posttest sessions (II), contingency assay sessions with the mult VT 30-s VT 30-s schedule (III), and sessions withthe mult VT 30-s EXT schedule (IV). In each panel, on-key and off-key peck rates are shown.

tingency control. Table 2 presents each bird'soff-key pecks, as a percentage of its on-keypecks, to S+ during criterion baseline sessionsand during the generalization test at its S+value and the stimulus value controlling its

peak rate (S+ +). The stability of each sub-ject's key-peck accuracy over these three crit-ical instances is striking, as is the consistentoverall difference between operant and auto-shaped groups. Over these three conditions, the


#2 (Operant)90-

75 -

60-

45 -

30 -

#3 (Autoshaped)

S+ S-I II II I

498 510 518 528 540 548 556 566

S+ S-IIIJt I

* On-Key

o Off-Key)

90-

75.

60

45

30

15

566

90 -

75 -

60 -

45 -

30 -

15

498 510 518 528 540 548556 566

#18 (Autoshaped)

S- S+I II

I /

498 510 518 528 540 548 556 566

#19 (Autoshaped)

S+ S-IIII

v I I I I

498 510 518 528 540 548556 566

Stimulus Wavelength (nm)Fig. 3. Stimulus generalization gradients for birds in the operant group (left) and birds in the autoshaped group

(right). Reinforced and nonreinforced training stimuli are indicated for each bird with their on-key (filled circles) andoff-key (open circles) rate at each test value.

autoshaped group produced a significantlygreater proportion of off-key pecks than theoperant group, F(1, 8) = 7.87, p < .05. Thedifferential effects on peck location of operantand respondent contingencies were dramati-cally revealed. Taken in conjunction with thedifferential-nondifferential contingency assaypresented below, the findings of the current

experiment further validate key-peck accuracyas a measure of contingency control.

Differential-nondifferential association assay.The finding that a differential correlation be-tween CS+ and the unconditioned stimulus isnecessary for conditioned suppression to bemaintained (Rescorla, 1968; Wagner, 1969)has been extended to the analysis of auto-

#5 (Operant)90

75

60'45

30

15

0

U)

0

o)#9 (Operant)

S- S+I I AIIII


#2 (Operant) #3 (Autoshaped)

#18 (Autoshaped)

S- S+I I

III /I

498 510 518 528 540 548 556 566

#9 (Operant)

498 510 518 528 540 548 556 566

40

30

498 510 518 528 540 548556 566

#19 (Autoshaped)

S+ S-I Ik I

i\

498 510 518 528 540 548 556 566

Stimulus Wavelength (nm)Fig. 4. Relative generalization gradients of on-key pecks and off-key pecks for birds in the operant group (left)

and the autoshaped group (right). Each point on a gradient represents the percentage of total pecks, on-key (filledcircles) or off-key (open circles), produced at that stimulus value by the subject. Two points are presented at eachstimulus value. Overlapping points are represented as open circles.

shaped key pecks (Gamzu & Williams, 1971,1973). According to this analysis, elicited keypecks should disappear when the stimulus-reinforcer contingency was eliminated for theautoshaped birds in the present experiment byprogramming VT 30-s grain presentation in

the presence of both 528 and 540 nm (Schwartz& Gamzu, 1977). That is precisely what hap-pened to the pecking of all birds in the auto-shaped group (Panel III of Figure 2). Theessentially complete elimination of key peckingin the autoshaped birds by the mult VT VT

g

0o

1 40

qQ 30

20

C 10

0 0E*

40

40

30

20

10

I

s-IIII


schedule confirms that their key pecks were

elicited as a function of the differential stim-ulus-reinforcer contingency, and that adven-titious operant contingencies did not contributeto these pigeons' key pecks. Further, the re-

emergence of key pecking when the stimulus-reinforcer contingency was again differentiallyprogrammed for these birds (Panel IV of Fig-ure 2) supports their elicited respondent na-

ture. The relatively long, variable durations ofthe VT components in the present experiment,coupled with programmed reinforcement inonly 75% of them, did not lend our autoshapingbaseline to the traditional omission test of re-

spondent control (cf. Sheffield, 1965; Williams& Williams, 1969) in which short, fixed-du-ration CSs are usually employed.

Panel III of Figure 1 shows what happenedwhen a VT 30-s schedule was programmed inthe former S- component of the operant group.Key-peck rates in S+, where the VI 30-sschedule continued to operate, were essentiallyunaffected by this manipulation for Birds 2and 9 (except initially for the latter), whereasthose of Bird 5 restabilized at approximately50 per minute. However, the apparent positivecontrast effect in S+ when the birds were re-

introduced to the mult VI 30-s EXT schedule(see Panel IV of Figure 1) suggests that themult VI 30-s VT 30-s schedule influencedstimulus control, as an additivity theory of con-trast would predict (Schwartz & Gamzu,1977). Overall, this assay is quite convincingin affirming that the behavior of birds of bothgroups was, in fact, under the control of theirprogrammed contingencies.

Bird 3 received completely automated initialautoshaping. However, due to several condi-tioning failures described earlier, until peckingemerged, initial autoshaping sessions weremodified such that the CS was presented whenthe pigeon was oriented towards the front wall.The 12-nm CS+:CS- separation (the small-est for which autoshaping has been producedto our knowledge) probably made this modi-fication more important than if stimulus dif-ferences had been more pronounced. In anyevent, for conditioning to occur the CS shouldbe in the subject's receptive field. There are

two sources of evidence clearly indicating thattiming initial CS presentations to occur withinthe pigeon's visual receptive field did not com-promise the integrity of the classical condi-tioning procedure. First, the generalization test

Table 2

Off-key S+ response rates as a percentage of on-key S+response rates for autoshaped and operant groups (pecksper minute).

S+ S+ S++Group Subject baseline test testa

Autoshaped 3 63.2 77.3 89.118 66.1 55.6 59.919 115.5 95.8 96.3M 81.6 76.2 81.6

Operant 2 34.1 28.5 27.65 41.7 61.0 50.49 2.1 2.2 7.7M 26.0 31.0 28.6

a S+ + refers to the stimulus controlling the highest ratein testing.

results of Bird 3 were comparable to those ofBirds 18 and 19. Second, both contingencyassays revealed stimulus-reinforcer control forall autoshaped birds.

Extremely effective conditioning and dis-criminative performance were produced by themodified autoshaping procedure. Birds 18 and19 were ready for testing after less than 10mult VT 30-s EXT sessions. Nevertheless,their generalization test results were compa-rable to those of Bird 3, who received morethan 50 sessions on that schedule. Thus, num-ber of training sessions within this rather widerange (7 to 52) was not critical in shifting thepeak of autoshaped key pecks.

Analyses of Peak ShiftSpence's (1937) gradient interaction theory

predicts the occurrence of peak shift wheneverthe intradimensional values of the S+ and S-stimuli are similar and their collateral excit-atory and inhibitory gradients are appropri-ately contoured. According to this formulation,the summative interaction of the S+ centeredexcitatory gradient and S- centered inhibitorygradient results in peak shift (see Rilling, 1977,Figure 8, p. 448). Spence's theory made nodistinction between gradients resulting fromoperant or respondent contingencies. There-fore, because excitatory and inhibitory gradi-ents for autoshaped pecks following interdi-mensional discrimination training have beenreported (Tomie et al., 1976; Wessells, 1973),peak shift would be predicted following theappropriate intradimensional autoshaping.The robust and consistent peak shift producedin the present experiment, with autoshaped

STANLEY I. WEISS and RICHARD D. WEISSMAN

60

Operant

Autoshaped

S+ S++Generalition Test Stimulus

Fig. 5. Absolute difference between on-key minus off-key peck rates for operant group (open circles) and au-

toshaped group (filled circles) at S+ and S + + general-ization test stimulus values. S+ is the stimulus reinforcedin training (528 or 540 nm). S+ + is the stimulus valuethat controlled a subject's peak rate in testing. S+ + was

always removed at least one stimulus value from S+ inthe direction away from S-.

key pecks clearly under stimulus-reinforcercontrol, confirms respondent predictions fromSpence's theory.

Although Spence (1937) made no distinctionbetween operant and respondent control here,Weiss' (1971, 1977, 1978) two-factor modelof stimulus control deals with the contributionof these different contingencies to the likeli-hood, and robustness, of peak shift. He ob-served that in the large majority of operantdiscrimination studies, the presence of S+ iscorrelated with an increase in both responseand reinforcement frequency and argued thatthis covariation obscures the contributions ofeach contingency to the resulting stimulus con-

trol. Weiss' experiments separately measuredand compared the contributions to stimuluscontrol of the response-reinforcer and stimu-lus-reinforcer contingencies found in operantdiscrimination situations. Extrapolations fromthis research indicated that either source ofcontrol alone should be sufficient to producediscrimination-dependent phenomena such as

peak shift. However, according to Weiss' for-mulation, peak shift would be enhanced whenboth contingencies produced excitation at S+,in comparison to the situation in which onlyone contingency was operating (see also Weiss& Schindler, 1987).

For the operant group in the present ex-

periment, the increased key pecking in S+

produced by the response-reinforcer contin-gency created a correlated stimulus-reinforcercontingency in that stimulus. In comparison,the key pecking of the autoshaped group wassolely the product of S+ being associated withan increased frequency of reinforcement. Thispermits a direct test of Weiss' two-factor al-gebraic combination model. When key-peckrates in S+ + (peak-rate stimulus) were com-pared over groups, the additive prediction fromthe model was supported. The operant group's73.8 pecks per minute to S+ + was signifi-cantly higher than the 48.5 pecks per minuteto S+ + by the autoshaped group, t(4) = 2.29,p < .05, one-tailed. This significant superi-ority of the operant group at S+ + was threetimes their 8.4 pecks per minute superiority(not statistically significant) at S+.

This superiority in S+ + rates by the op-erant group, coupled with the between-groupdifference in off-key peck performance, pro-duced the interaction profile shown in Figure5. This profile is based on the absolute differ-ences between on-key and off-key peck ratesfor each group at S+ and S+ +. The groupeffect was significant on this difference mea-sure, F(1, 4) = 9.42, p = .037. Moreover, thesignificant interaction reveals that the on-keyminus off-key difference did not change fromS+ to S+ + for the autoshaped pigeons,whereas this difference almost doubled for theoperant pigeons, F(1, 4) = 40.44, p < .003.This is another indication of greater respond-ing to the stimulus controlling maximum shiftby the operant group than by the autoshapedgroup because off-key rates were a comparablepercentage of on-key rates at S+ and S+ + forboth groups (Table 2).As an algebraic combination model, Weiss'

formulation also predicts a decrease in the like-lihood of peak shift when, as a by-product ofdiscrimination training, (a) the response-re-inforcer contingency produces an increase inthe operant at the stimulus value at which theimplicit stimulus-reinforcer contingency pro-duces respondent inhibition or (b) the re-sponse-reinforcer contingency produces a de-crease in the operant at the stimulus value atwhich the implicit stimulus-reinforcer contin-gency produces respondent excitation.

Weiss and Dacanay (1982) developed abaseline schedule that permitted them to testboth of these predictions simultaneously in thesame subjects. Their experiment was briefly

140

PEAK SHIFT OF AUTOSHAPED AND OPERANT KEY PECKS 141

described in the introduction. To recapitulate,they trained pigeons on a chained schedule inwhich treadle pressing in the initial-link wave-length (S2), presented on a key, produced theterminal-link wavelength (Si), where rein-forcers were delivered according to a DROschedule applied to the treadle press. Here, theoperant and respondent by-products of dis-crimination training were conflicting at bothS1 and S2 (one excitatory and the other in-hibitory).On this chain VI DRO schedule, over 95%

of the treadle presses were emitted during S2,and most (94% to 99.3%) nonfunctional keypecks were elicited in S1 where all the foodwas delivered. However, in spite of these ex-cellent intradimensional discriminations, gen-eralization tests revealed no peak shift for thetreadle-press operant, with even area shift at-tenuated. Moreover, their subjects producedsharp, symmetrical autoshaped key-peck gra-dients that peaked at S1, with area shift en-tirely absent for 2 of the 3 birds.

Consistent with predictions from Weiss'model, in the Weiss and Dacanay (1982) study,peak shift did not occur for the operant (treadlepress) when primary reinforcement was absentduring its discriminative stimulus (S2), andpeak shift did not occur for the respondent (keypeck) when during its CS (SI) operant be-havior, maintained by the same reinforcer as-sociated with the CS, was suppressed by theDRO contingency. That is a perfectly symmet-rical, bidirectional, two-factor application. Itcontrasts with traditional unidirectional two-factor theory that considers only the effects ofclassical mediation on instrumental behavior(Rescorla & Solomon, 1967). The implicationsof a bidirectional two-factor theory to an anal-ysis of stimulus control deserve further explo-ration.

CONCLUSIONThis experiment has been successful in

clearly and unambiguously producing signif-icant generalization peak shift with auto-shaped key pecks. Peak shift that can be con-sidered respondent in nature has rarely beenreported heretofore, and never this consistentlyand robustly. In addition, the contingency as-says demonstrated, independently of the train-ing procedures used, that the autoshaped birdsbehaved as classically conditioned subjects.

They produced significantly more off-key pecksthan did the operant birds, and their peckingceased when CS+ was no longer differentiallycorrelated with reinforcement. This suggeststhat with the appropriate stimulus and train-ing parameters, robust generalization peak shiftshould be forthcoming after intradimensionalclassical training with responses such as therabbit's nictitating membrane or elicited jawmovement. However, if this classical trainingis to be functionally comparable to that of theautoshaped group in the present experiment,a two-component paradigm is necessary, inwhich the CS+ is an intradimensional changedirectly from the CS- stimulus condition. Intraditional classical conditioning studies, theCS+ and CS- are separated by an interstim-ulus interval. Generalization peak shift of anautoshaped key peck remains to be demon-strated in a paradigm in which CS+ and CS-wavelengths projected on the key are separatedby, for example, a dark key.The generalization test results of the auto-

shaped group confirmed predictions of re-spondent peak shift that can be derived fromSpence's (1937) gradient interaction theory. Inaddition, comparisons of the magnitude of theshift for operant and autoshaped groups in-dicated that in the former group the discrim-inative by-products of both the stimulus-re-inforcer and response-reinforcer contingenciescontribute to the strength of peak shift. Thissupported Weiss' (1978) two-factor model ofstimulus control. The qualitative comparabil-ity of the gradients produced after stimulus-reinforcer and response-reinforcer discrimi-nation training further confirms the similarityof perceptual processing in classical and op-erant learning.

REFERENCES

Barrera, F. J. (1974). Centrifugal selection of signal-directed pecking. Journal of the Experimental Analysisof Behavior, 22, 341-356.

Bushnell, M. C., & Weiss, S. J. (1980). An investigationof peak shift and behavioral contrast for autoshapedand operant behavior. Journal ofthe Experimental Anal-ysis of Behavior, 33, 101-118.

Catania, A. C. (1973). Self-inhibiting effects of rein-forcement. Journal of the Experimental Analysis of Be-havior, 19, 517-526.

Cowan, R. A. (1968). Peak shift following classical con-ditioning. Unpublished master's thesis, University ofMissouri, Columbia.

142 STANLEY J. WEISS and RICHARD D. WEISSMAN

Gamzu, E. R., & Williams, D. R. (1971). Classicalconditioning of a complex skeletal response. Science,171, 923-925.

Gamzu, E., & Williams, D. R. (1973). Associative fac-tors underlying the pigeon's key pecking in auto-shap-ing procedures. Journal of the Experimental Analysis ofBehavior, 19, 225-232.

Guttman, N., & Kalish, H. I. (1956). Discriminabilityand stimulus generalization. Journal of ExperimentalPsychology, 51, 79-88.

Hanson, H. M. (1959). Effects of discrimination train-ing on stimulus generalization. Journal ofExperimentalPsychology, 58, 321-334.

Hearst, E. (1968). Discrimination learning as the sum-mation of excitation and inhibition. Science, 162, 1303-1306.

Hearst, E. (1971). Differential transfer of excitatoryversus inhibitory pretraining to intradimensional dis-crimination learning in pigeons. Journal of Comparativeand Physiological Psychology, 75, 206-215.

Hovland, C. I. (1937). The generalization of conditionedresponses: I. The sensory generalization of conditionedresponses with varying frequencies of tone. Journal ofGeneral Psychology, 17, 125-148.

Hupka, R. B., Liu, S. S., & Moore, J. W. (1969). Au-ditory differential conditioning of the rabbit nictitatingmembrane response: V. Stimulus generalization as afunction of the position of CS+ and CS- on the fre-quency dimension. Psychonomic Science, 15, 129-131.

Jenkins, P. E. (1981). The determiners of keypeck du-ration. Animal Learning & Behavior, 9, 501-507.

Keller, K. (1974). The role of elicited responding inbehavioral contrast. Journal ofthe Experimental Analysisof Behavior, 21, 249-257.

Kimble, G. A. (1961). Hilgard and Marquis'conditioningand learning (2nd ed.). New York: Appleton-Century-Crofts.

Liu, S. S. (1971). Differential conditioning and stimulusgeneralization of the rabbit's nictitating membrane re-sponse. Journal of Comparative and Physiological Psy-chology, 77, 136-142.

LoLordo, V. M., McMillan, J. C., & Riley, A. L. (1974).The effects on food-reinforced pecking and treadle-pressing of auditory and visual signals for response-independent food. Learning and Motivation, 5, 24-31.

Mackintosh, N.J. (1974). Thepsychology ofanimal learn-ing. New York: Academic Press.

Moore, J. W. (1972). Stimulus control: Studies of au-ditory generalization in rabbits. In A. H. Black & W.F. Prokasy (Eds.), Classical conditioning Il: Current re-search and theory. New York: Appleton-Century-Crofts.

Nallan, G. B., McCoy, D. F., Pace, G. M., & Welch, R.(1979). Generalization gradients following differen-tial intradimensional autoshaping. Perceptual and Mo-tor Skills, 48, 671-677.

Pavlov, I. P. (1927). Conditioned reflexes. New York:International Publishers.

Purtle, R. B. (1973). Peak shift: A review. PsychologicalBulletin, 80, 408-421.

Rescorla, R. A. (1968). Probability of shock in the pres-ence and absence of CS in fear conditioning. Journal ofComparative and Physiological Psychology, 66, 1-5.

Rescorla, R. A., & Solomon, R. L. (1967). Two-processlearning theory: Relationships between Pavlovian con-ditioning and instrumental learning. Psychological Re-view, 74, 151-182.

Rilling, M. (1977). Stimulus control and inhibitory pro-cess. In W. K. Honig & J. E. R. Staddon (Eds.),Handbook of operant behavior (pp. 432-480). Engle-wood Cliffs, NJ: Prentice-Hall.

Schwartz, B., & Gamzu, E. (1977). Pavlovian controlof operant behavior: An analysis of autoshaping andits implications for operant conditioning. In W. K.Honig & J. E. R. Staddon (Eds.), Handbook ofoperantbehavior (pp. 53-97). Englewood Cliffs, NJ: Prentice-Hall.

Sheffield, F. D. (1965). Relation between classical con-ditioning and instrumental learning. In F. W. Prokasy(Ed.), Classical conditioning: A symposium (pp. 302-322). New York: Appleton-Century-Crofts.

Skelton, A., & White, K. G. (1982). Stimulus-gener-alization of signal-directed responding under a negativecontingency. Behavior Analysis Letters, 2, 275-282.

Spence, K. W. (1937). The differential response in an-imals to stimuli varying within a single dimension.Psychological Review, 44, 430-444.

Terrace, H. S., Gibbon, J., Farrell, L., & Baldock, M.D. (1975). Temporal factors influencing the acqui-sition and maintenance of an autoshaped keypeck. An-imal Learning & Behavior, 3, 53-62.

Thomas, D. R., Freeman, F., Svinicki, J. G., Burr, D. E.S., & Lyons, J. (1970). The effects of extradimen-sional training on stimulus generalization. Journal ofExperimental Psychology, 83 (1, Part 2).

Tomie, A., Davitt, G. A., & Engberg, L. A. (1976).Stimulus generalization of auto-shaped keypecking fol-lowing interdimensional and extradimensional train-ing. Learning and Motivation, 7, 240-253.

Wagner, A. R. (1969). Stimulus validity and stimulusselection in associative learning. In N. J. Mackintosh& W. K. Honig (Eds.), Fundamental issues in associativelearning (pp. 90-122). Halifax: Dalhousie UniversityPress.

Weiss, S. J. (1970). An effective and economical sound-attenuation chamber. Journal of the Experimental Anal-ysis of Behavior, 13, 37-39.

Weiss, S. J. (1971). Discrimination training and stim-ulus compounding: Consideration of non-reinforce-ment and response differentiation consequences of S".Journal of the Experimental Analysis of Behavior, 15,387-402.

Weiss, S. J. (1977). The isolation of stimulus-reinforcerassociations established with multiple schedules. Ani-mal Learning & Behavior, 5, 421-429.

Weiss, S. J. (1978). Discriminated response and incen-tive processes in operant conditioning. A two-factormodel of stimulus control. Journal of the ExperimentalAnalysis of Behavior, 30, 361-381.

Weiss, S. J., & Dacanay, R. J. (1982). Incentive pro-cesses and the peak shift. Journal of the ExperimentalAnalysis of Behavior, 37, 441-453.

Weiss, S. J., & Schindler, C. W. (1981). Generalizationpeak shift in rats under conditions of positive rein-forcement and avoidance. Journal of the ExperimentalAnalysis of Behavior, 35, 175-185.

Weiss, S. J., & Schindler, C. W. (1987). The composite-stimulus analysis and the quantal nature of stimuluscontrol: Response and incentive factors. The Psycho-logical Record, 37, 177-191.

Wessells, M. G. (1973). Errorless discrimination, au-toshaping, and conditioned inhibition. Science, 182, 941-943.

PEAK SHIFT OF AUTOSHAPED AND OPERANT KEY PECKS 143

White, K. G., & Braunstein, S. B. (1979). Stimuluscontrol of topographically tagged responding. AnimalLearning & Behavior, 7, 333-338.

White, K. G., & Thomas, D. R. (1979). Topographi-cally tagged stimulus control: Maintained generaliza-tion and stimulus-specific gradients. Bulletin of the Psy-chonomic Society, 13, 275-278.

Williams, D. R., & Williams, H. (1969). Auto-main-tenance in the pigeon: Sustained pecking despite con-tingent non-reinforcement. Journal of the ExperimentalAnalysis of Behavior, 12, 511-520.

Received January 8, 1991Final acceptance November 14, 1991

GENERALIZATION PEAK SHIFT FOR AUTOSHAPED AND OPERANT KEY PECKS

Documents