The Auk 111(2):300-313, 1994

VOCAL LEARNING IN GREY (PSITTACUS ERITHACUS): EFFECTS OF SOCIAL INTERACTION, REFERENCE, AND CONTEXT

IRENE M. PEPPERBERG Departmentof Ecologyand EvolutionaryBiology, University of Arizona, Tucson,Arizona 85721, USA

ABSTR•cr.--Formany passerines,the extent,timing, and even presenceof allospecificvocal learning can be influencedby the form of input that is received.Little data exist,however, on vocal learning in parrots (). I have previously proposed that such vocallearning proceeds most readily when input is (1) referential,(2) contextuallyapplicable, and (3) interactive.The referentialaspect demonstrates the meaningof the codeto be taught, the contextualaspect demonstrates the use that can be made of the information contained in the code, and the interactive aspectprovides explicit training that is constantlyadjusted to the level of the learner. To obtain information on the relative importanceof thesethree aspectsof input on learning in a mimetic ,I used three different conditionsto train two juvenile Grey Parrots(Psittacus erithacus) to produceEnglish labelsto identify various commonobjects. Each experienced:(1) audiotapedtutoring, which was nonreferential, noninteractive,and did not demonstratecontextual applicability; (2) videotapes,which pro- vided reference and limited information about context, but which were noninteractive; and (3) live human tutors, who interactivelymodeled the meaning and use of the labelsto be learned.The birdslearned only from the live tutors.A third ,trained on a separateset of labelsby tutorswho provided only limited referenceand contextfor thosevocalizations, learnedto producethat setof labelswithout comprehension.The data suggestthat, even for known for their mimetic abilities, social interaction, reference, and full contextual experienceare important factorsin learning to produceand comprehendan allospecificcode. Received22 April 1993,accepted 10 October1993.

IN THE LASTDECADE, studies have shown how rot (Amazonaochrocephala), and a input affectsallospecific avian vocal learning (Psittacuserithacus) that experiencedlimited so- (review in Pepperberg 1991, 1993). For birds cial input (Mowrer 1952, 1954, 1958). In con- that favorconspecific learning (i.e. typicallydo trast, Grey Parrots that received modeled, in- not mimic otherspecies in the field), allospecific teractivehuman tutoring (Todt 1975,Pepperberg learning is often affectedby variationin social 1981, 1990a)acquired the targetedspeech pat- and environmental input. Such species(e.g. terns. Given the reputed easewith which these White-crowned Sparrows [Zonotrichialeuco- mimeticbirds are assumedto acquireany type phrys],Baptista and Morton 1981, Baptistaand of sound (see Amsler 1947), data on the effects Petrinovich 1984, 1986; Song Sparrows[Melo- of differentialinput were surprising.However, spizamelodia], Marler and Peters1977, 1987, Bap- becausedifferent laboratories tested separate sets tista 1988) may require visual and vocal inter- of conditions,the findings could have been a actionwith a live tutor for completeallospecific consequenceof interlaboratoryvariation, as well songlearning to occur.Less is known, however, as of the different learning conditions (Slater aboutthe effectof input on vocallearning in 1991). birds that are frequent mimics. The possibleconfounding factor of interlabo- Data on vocallearning in somemimetic birds ratory variation was counteredin a single study comefrom experimentson training thesebirds on mimetic EuropeanStarlings (Sturnus vulgaris) to reproducehuman speechin the laboratory that examined how competing forms of differ- (reviewin Pepperberg1988a). Little or nothing ential input affectedlearning (West et al. 1983). was learned by Indian Hill Mynahs (Gracula Sevenbirds, placedin three different groups, religiosa)exposed to tapes in social isolation did or did not experience, in various combi- (Grosslightet al. 1964,Grosslight and Zaynor nations:human care and vocal interaction; tapes 1967,Gossette 1969) or by an Indian Hill My- of humanspeech and whistles;and living quar- nah,Budgerigars (Melopsittacus undulatus), Black- ters sharedwith Brown-headed Cowbirds (Mol- billed Magpies(Pica pica), a Yellow-headedPar- othrusater) or other juvenile starlings.All birds

3OO April 1994] GreyParrot Vocal Learning 301 learnedvocalizations only from organismswith plicable, and noninteractive input; (2) refer- which they could interact (either human or ential, minimally contextuallyapplicable, and cowbird) or learned soundsclosely associated noninteractive input; and (3) referential, con- with the presenceof such organisms(e.g. the textually applicable,and interactive input. In opening of a door that preceded a morning the experimentwith the adult parrot, I studied greeting)and learned nothing from any non- the effectof input that was sociallyinteractive interactive source. Such data demonstrated the but that minimized reference and contextual effectof differentialinput on learning,but did applicability.At the time, I chosenot to study not isolatewhich aspectsof the input were crit- the effect of input that was referential, fully ical for learning. contextuallyapplicable, and noninteractivebe- To demonstratethe relative importance of causestudies on other speciessuggested that variousaspects of input for learning, one must suchconditions lead at bestto productionwith- first identify the relevant aspects.According to out comprehension(see Savage-Rumbaughet a psychologicalconstruct called "social-mod- al. 1980a, b). eling theory" (Bandura 1971, 1977), input can The resultsof theseexperiments not only pro- be characterizedby three main aspects(Pep- vide informationon Grey Parrots,but alsosug- perberg 1985, 1988b, 1991, 1992a, Pepperberg gestintriguing parallels between avian and hu- and Neapolitan 1988,Pepperberg and Schinke- man exceptional learning--learning that is Llano 1991):(1) degreeof referentiality,(2) scope unlikely in the normal courseof development of contextualapplicability, and (3) extent of so- but that can occur under certain conditions cial interaction.Reference and contextualap- (Pepperberg 1985, 1986, 1988a, 1993). Such be- plicability refer to the real-world useof the in- havior was first describedby human socialpsy- put, and socialinteraction is a potent meansof chologists (Bandura 1971, 1977). I discussthe highlightingvarious components of the input. resultsin the contextof my previousproposal Referenceis generally defined as the meaning that avian acquisitionof an allospecificcode is of an utterance(e.g. the relationshipbetween a particular form of exceptionallearning; I sug- a label and the objectto which it refers).Con- gestpossible parallels between such psittacine textual applicabilityinvolves the particularsit- learningand anotherform of exceptionallearn- uation in which an utterance is used and the ing, human second-languageacquisition (Pep- effectsof using the utterance. Socialinteraction perberg and Neapolitan 1988). actsto signal which componentsof the envi- ronmentshould be noted,emphasizes common attributes--and thus possible underlying METHODS FOR EXPERIMENT 1 rules--of diverse actions,and allows input to SUBJECTSAND HOUSING be continuouslyadjusted to the level of the learner.Interaction may alsoprovide a contex- Subjectswere juvenile Grey Parrots, Alo (female) tual explanation of the reasonsfor the actions and Kyaaro (male), that were 10 and 6.5 months, re- and demonstratethe consequencesof the ac- spectively,at the beginning of the experiment.They were hand raised and had been obtained from their tions (for detailed discussionof these points, breeder three months previously. They lived in sep- see Pepperberg1993). Researcherscan specifi- arate roomsand could not hear one another. Training cally design input that varies with respectto with live tutorsand testingoccurred while thesebirds theseaspects and then evaluatethe relativeef- were atop their cages,on "gyms" (branchesthat had fects of such variation. been nailed together),or on parrot stands.Birds were To carry out such an evaluation, I designed confined to Hoei cages(ca. 38 x 71 x 56 cm) when experimentsfor one adult and two juvenile Grey humanswere absent,and during sleepinghours. Wa- Parrots.I examinedhow variousforms of input ter and Harrison's Bird Diet were available continu- affectedthe amountand type of their acquisi- ously;fruit, vegetables,dried pastasand cerealswere tion of an allospecificcode, English speech. I provided when neither testing nor training were in studiedhow input might affectcompetence not progress. only with respectto physicalproduction, but TRAINING PROCEDURES alsowith respectto comprehensionand appro- priate use.In the experimentwith the juveniles, To provide input that varied with respectto social I studied the relative effectsof three types of interaction, reference, and contextual applicability, I input: (1) nonreferential, not contextuallyap- contrastedsessions of live, videotape, and audiotape 302 IREhIEM. PEPPERBERG [Auk, Vol. 111

T^BLœ1. Componentsof different types of tutoring Unlike othermodeling procedures (e.g. Todt 1975, used to train Alo and Kyaaro. Goldstein 1984), my protocol requiresrepeating an interactionwhile reversingroles of the humantrainer Contextual Social and model/rival, and includesthe parrot in the in- applica- inter- teractions.Thus, birds do not simply hear stepwise Reference bility action vocal duets, but observeand learn to engage in a Audiotapes No No No communicative process(i.e. a processthat involves Videotapes Yes Partial No reciprocityand can be usedby either party to request M/R protocol Yes Yes Yes information or effect environmentalchange). With- out role reversal,birds exhibit two typesof behavior that are inconsistent with interactive, referential com- tutoring.Table 1 depictsthe variationamong the types munication(Todt 1975):(1) they do not transferre- of tutoring. The birds received the three types of sponsesto anyone other than the human who poses training sessionson a rotating basis, two sessions/ the questions;and (2) they do not learn both partsof day (morningand afternoon)and five days/weekfor the interaction. approximately11 months beginning 28 June 1991, The M/R technique specificallydemonstrates ref- excluding student vacation periods. erenceand contextualapplicability by usingintrinsic Livetutoring and intrinsic rewards.--The live-tutoring reinforcers: Reward for each identification is the item system,called the model/rival (M/R) procedure,was to which the label refers.In contrast,some programs adapted from the work of Todt (1975). M/R training designedto teachcommunication skills, for both hu- involves three-way interactions between two com- mansand nonhumans,use extrinsic rewards (see Pep- petent human speakersand the avian student. M/R perberg 1990a). In these programs,all correct iden- training primarily introducesnew labels and con- tificationsof food or nonfood items or appropriate cepts,but also aids in shaping correct pronunciation. responsesto various specificcommands are rewarded Becausethis study is an in depth comparisonof train- with a single item (generally food) that neither di- ing protocols,I describethe M/R procedurein some rectlyrelates to the skill being taughtnor varieswith detail although the material is available elsewhere respectto the specifictask being targeted.Such ex- (Pepperberg1981, 1988b, 1990a,b, c). trinsic rewardsmay delay label acquisitionby con- During M/R training, humans demonstrateto a founding the label or conceptto be learnedwith some bird different types of targetedinteractions. Sessions aspectof the reward item (Greenfield 1978, Pepper- begin with a bird observingtwo humanshandling berg 1981, Miles 1983). My procedure,instead, pro- an object;the item is one of severalthat the bird has videsthe closestpossible association of the label that previouslychosen and used (e.g. as a preening im- is being taught and the item to which it refers (Pep- plement; Pepperberg1981) for three to five consec- perberg 1981). utive minutes in free-choice sessions over several Videotapepresentations.--To provide training that weeks. One human actsas a trainer, showing the item closelyfollowed the M/R procedurebut avoided so- to the second human, who is both a model for the cial interaction and minimized contextualapplica- bird's responsesand a rival for the trainer'sattention. bility, I videotaped the previously trained adult par- The trainer queries the model/rival about the item rot, Alex, during M/R sessionsand exposedthe ju- ("What's here?", "What matter?", "What toy?"), giv- venile parrotsto thosetapes. Although Alex already ing praiseand the objectto reward correctanswers. comprehendedand used the targeted labels refer- The techniquethus demonstrates referential and con- entially (e.g.Pepperberg 1990a, b), tapesdid not pres- textualuse of labelswith respectto observableobjects. ent the targetedmaterial as a review session,but fol- A trainer shows disapproval for incorrect responses lowed the lines of actual training. Thus, Alex (errorssimilar to thosemade by a bird, suchas partial occasionallyerred or interrupted with requestsfor identifications,unclear speech) by scoldingand tem- other objectsand changesof location (Pepperberg porarilyremoving the objectfrom sight.Thus, a bird 1983, 1987a; references to related behavior in other observesaversive consequences of errors.The model/ speciesare Davis 1984,Putney 1985,Moran et al. un- rival is askedto talk more clearly or try again when publ. data).As in the live M/R presentations,trainers a responseis incorrector garbled,thereby allowing would also occasionallyerr. Not only was the style the parrotto observe"corrective feedback" (see Gold- of vocalinteraction identical to regularM/R sessions, stein 1984, Vanayan et al. 1985). Becausea bird is but tapesalso retained patterns of breaksfor nonvocal rewarded for successiveapproximations to a correct exchanges(e.g. when trainerspreened Alex) and time- response,the protocoladjusts the level of training to outsby using,respectively, scenes of suchnonvocal the level of the bird. If a bird is inattentive or its interactions or a blank screen. Juveniles watched the accuracyregresses, trainers threaten to leave ("I'm videos in isolation, so that no direct social interaction gonna go away?); such behavior is aversive, as our with trainersoccurred. By watching a humanor Alex birdsbecome agitated and will beckon("Come here," producea particularsound and either receivean ob- "Want tickle") a departing trainer. ject or be scolded,the juveniles saw but did not ex- April 1994] GreyParrot Vocal Learning 303 perlericedirectly the effect of a vocalization.Videos, be equally easyfor eachbird to produce.Therefore, therefore, demonstrated reference but lacked clear I counterbalancedlabels, so that, with the exception contextualapplicability. of "paper" and "rock," labels used for one bird with Audiotapepresentations.--Audiotapes were designed one technique were used for the other bird with an- not only to parallel the M/R and video procedures, other technique.Both birds were exposedto "paper" but also to eliminate reference and context. Audio- via live tutors and to "rock" via audiotapein order tapesconsisted of the audio portion of the video of to compare their speedsof learning. I chose labels Alex's sessionson the targeted vocalizations;audio that the aforementionedAlex could clearly produce and video sessions, thus, did not differ in amount of (Pepperberg 1981, 1990a) to ensure that the vocali- auditory input. Juvenileslistened to the tapesin iso- zationswere within the capacityof the species. lation and, thus, no objects were associatedin any Previous work (Pepperberg et al. 1991) also had way with the soundspresented over the speaker. shown that motivation affectedAlex's label acquisi- tion. In situationsin which producingan objectlabel EQUIPMENT was, at leastinitially, equivalentto requestingthe object,Alex more frequently practicedlabels for fa- All tapeswere made in the laboratoryusing an AKG voreditems and acquiredthese labels more readily. CK8 microphone.Videos were madewith a Panasonic So that motivation would not affect the juveniles' SVHS AG-450 cameraand TDK HS tapesand were results,labels used in their training referredto items editedwith a PanasonicAG1960 proline deckand CT- with which they had chosento interact (see above) 2082y color monitor. A zoom lens enabled me to in- in free-playsessions. cludelife-size images of Alex andthe targetedobjects in addition to the somewhatsmaller images of the TESTING PROCEDURES entire training scenario (the object, Alex, and two humans).A previous study (unpubl. data) showed To evaluatewhat the parrotshad learned,I admin- that Alex couldaccurately label objects presented via isteredtests one to four times per week beginning a live video link; Grey Parrotsthus appearable to February1992. The protocol(quoted in part from Pep- recognize two-dimensional video representations. perberg 1990a) is summarized here. Specificdetails Audiotapeswere madeand presentedon eitherSony can be found in Pepperberg(1981). TCM 5000or MarantzPMD 221 recordersusing Max- Criterionprior to testing.--Thecriterion as to when ell XL-UDII tapes. to begin testing is based on the clarity of a bird's speech,and not on the accuracyof labelingin training LABELS TRAINED AND CONTROLS (Pepperberg1981). For testingto begin,a vocalization FOR S¾STF.•IAT•CBh•S must be recognizedby trainers in blind trials with better than 90% interobserveragreement. Thus, I Each bird was trained on six labels, two each in the separatethe effectof our procedureson a bird'sability three different conditions. Alo received "cork" and physicallyto emit a label from the effectof the pro- "paper" in M/R sessions,"nail" and "wood" on vid- cedureson the bird'sability to associatethe label and eo, and "key" and "rock" on audiotape.Kyaaro re- its referent. Only when the former skill is considered ceived"paper" and "nail" in M/R sessions,"key" and satisfactoryis the latterskill tested(Pepperberg 1981, "cork" on video, and "rock" and "wood" on audio- 1983). tape.Training sessionswere designedto preventbias Precautionsagainst trainer-induced cuing.--Test situ- that might arise from three sources:(1) differing ationsincluded specificprecautions to avoid trainer- amountsof exposureto training in each procedure, inducedcuing (Pepperberg 1981). One precautionwas (2) variableease of productionof labels,or (3) varying a design that prevented either the subjector exam- motivationto obtain the objectsthat the labelsrep- iners from predicting which questions(or answers) resented. would appearon a given day. Testswere constructed Becausediffering amountsof exposurecould influ- as follows: The principal trainer (I.M.P.) listed the ence the efficacyof the varioustraining techniques, objectlabels to be examined.The order of presenta- I closelymatched overall lengths of sessions.Videos tion wasassigned randomly by a studentnot engaged and audiotapeswere editedinto sectionsthat closely in testing.A small numberof questions(two to five) approximated natural breaks in live sessions,and was then presentedintermittently during training overall amountsof training were set to resemblean sessionson current (and thus unrelated) topics for averageof what was given per sessionto Alex (ca.30 several days until all questionswere asked. While min per 45-min session),who had received M/R tu- objectlabels were being tested,for example,students toringfor over 15years and hadacquired a repertoire and I were training sequentialnumber recognition. of morethan 80 referentialvocalizations (Pepperberg Training questions("How many?")were thus aslike- 1981, 1990a). ly during test sessionsas a test question ("What's Given individual differencesbetween parrots, I this?");also, a specifictest object might appear only could not be sure that each of the chosen labels would once or twice per sessionand its appearance could 304 IRENEM. PEPPERBERG [Auk, Vol. 111 not be predicted. A secondprecaution against cuing priate objectlabel), the bird is given praise and the was to ensurethat trials on a given label were con- objectand no additional presentationsoccur (i.e. only ducted by a student who never trained that label. a "first trial" responseexists). If a responseis incorrect While training is in progress,students test a number or indistinct, the examiner removesthe object,turns of labels that they do not train, so the presence of a his/her head (a brief "time-out"), and says"No!" The specificstudent could not cue a bird as to which label examinerthen implementsa correctionprocedure in would be tested. that the misnamedobject is immediately(re)presented Precautionsagainst "expectationcuing".--Intermin- until a correct identification is made; errors are re- gling different test questions (e.g. "How many?", corded. "What's this?", "What matter?") during training on Birdsthus find that an incorrectresponse (e.g. sub- other topicsalso ensures against "expectation cuing" stitution of the label of a more desired item for the that may occur if a subject "expects" queries on a one presented)is fruitless;instead, correct responses single topic. Contextual information in single-topic allow a bird to proceedto a preferred item. Because tests could be responsiblefor a better performance immediate representationof objectsduring a test oc- than would otherwisebe justifiedby a subject'sactual curs only when responseto the initial presentation knowledge of a topic, in that a homogeneousset of is incorrect,the protocolpenalizes a "win-stay" strat- questionsmight leada subjectto ignoreall but a small egy. Incorrect repetition of a previously correct re- subsetof responses.My birds, however, are never sponse(e.g. the nameof the previousexemplar) elicits queried on a single topic (e.g. objectlabels) in a ses- no reward. sion, nor, more importantly, tested successivelyin Scoringprocedure.--Test scores are reported two ways, one sessionon similar questions("What's here?")or for "first" and "all trials." First-trial data are the per- on more than three questionsthat have a particular centageof first trials that are correctand are usedfor correctresponse (e.g. "cork"). Moreover, only novel statisticalanalyses. For comparison,I report the all objectsare used for testing and identical exemplars trials score for each task. The all trials score is the are never used for similar questions(e.g. a bird is total number of correct identifications(i.e. the pre- askedabout shapes,sizes, and colorsof trucks,paper determined number of tested items) divided by the and piecesof corkthat differ from training exemplars total number of presentationsrequired. If a bird can- and differfrom questionto question).A question(with not producea correctresponse, I reportonly first-trial its exemplar)is repeatedin a sessiononly if the initial data. answeris incorrect(see below; Pepperberg 1981). Thus, although the range of correct responsesto, for ex- RESULTS OF EXPERIMENT ample, "What's here?", "What matter?", or "How many?" was limited initially to just a few object or number labels, in any sessiona bird had to choose The data demonstratedthat the form of input from among severalpossible responses to the object affected the parrots' acquisitionof labels. Nei- and numberquestions to be correct(Pepperberg 1981). ther juvenileacquired utterances after 11 months Maintainingthe subjects'attention.--Concurrent work of exposure to either the audio or the video on several tasks is also necessarybecause birds be- presentations,although two videotapesof them come restlessduring sessionsdevoted to a single task. during these sessionsdemonstrated that they They ceaseto work, preen, or interrupt with requests whistled, squawked,and appearedto attend to for preening (e.g. "Want tickle"). A detailed discus- the video. Both birds, however, learned vocal- sion of such behavior is in Pepperberg(1990a). izationsafter exposureto sevenmonths of M/R Correctionprocedure.--The number of times objects training, although the two labels that Kyaaro are presented to a parrot depends upon its accuracy, which is determined as follows: The examiner (a stu- acquiredwere still too garbledat the end of 11 dent trainer), presentsthe targetedobject to the bird. months to distinguish for testing purposes.Af- The principal trainer sits so that she cannot see the ter seven additional months of training, these bird, the examiner,or the objectbeing presented.The vocalizations could be tested. This additional examinerasks one of the testquestions, to which the exposuredid not, however, enable Kyaaro to bird responds.The principal trainer then repeatswhat acquirethe labelstrained in the audio or video she heard the parrot say.This repetition preventsthe conditions. examiner from accepting an indistinct, incorrect vo- calization that is similar to an expected,correct re- sponse(e.g. "or" for "cork"). Interpretationof a bird's ALO responseis unlikely to be influenced by hearing the type of question.Posttest transcriptions of contextless Alo never produced,in the presenceof train- tapesof Alex's responsesagree with the original eval- ers,labels she experienced via video (wood,nail) uations to within 98.2%(Pepperberg 1992b). If what or audio (key, rock).Tapes of her solitarysound the principal trainer heard is correct(e.g. the appro- productions also revealed that she did not April 1994] GreyParrot Vocal Learning 3O5

TABLE2. Resultsof object-identificationtests. Object Testscore Percent Erroneousidentifications (no.errors) A. Alo Paper 34/40 85 Cork (4), unintelligible (2) Cork 34/40 85 Paper (1), unintelligible (5) Wood 0/20 0 Paper(3), cork(8), four (4), unintelligible(5) Key 0/20 0 Paper(3), cork(4), four (5), unintelligible(8) Rock 0/20 0 Paper(1), cork(10), four (2), unintelligible(7) Nail 0/20 0 Paper(2), cork(4), four (8), unintelligible(6) B. Kyaaro Paper 34/40 85 Nail (5), unintelligible (1) Cork 0/20 0 Paper(8), nail (6), unintelligible(6) Wood 0/20 0 Paper(6), nail (7), unintelligible(7) Key 0/20 0 Paper(4), nail (13), unintelligible(3) Rock 0/20 0 Paper (6), nail (9), unintelligible (5) Nail 35/40 87.5 Paper (1), unintelligible (4)

"practice"(see Pepperberg et al. 1991)labels for beginningthe end of May 1993,Kyaaro scored the targetedobjects. On identificationtests of 0% on 20 trials for each label trained via audio theseobjects beginning in April 1992,Alo scored or video tape (binomial test, P = 0.0261, chance 0% on 20 trials for each label (binomial test, P 1/6). His total scorewas 0/80. As before,I did = 0.0261, chance of 1/6). Her total score was not re-presentthe objectson anytrial. On tests 0/80. I did not re-presentthe objectson any givenMay through July 1993 for labelstaught trialsbecause she never produced even an ap- via the M/R procedure,his firsttrial scoreswere proximation to the correct labels. 34/40 for paper(85%, binomial test, P < 0.0001, In contrast,on testsgiven February through chanceof 1/3; P = 0.0026,chance of 1/2) and April 1992on labelstaught via the M/R pro- 35/40 for nail (87.5%,binomial test, P < 0.0001, cedure,first trial scoreswere 34/40 for both cork for chanceof 1/3 and 1/2). and paper(85%, binomial test, P < 0.0001,chance Interestingly,in the first11 months,Kyaaro of 1/3 [possibleresponses were learnedlabels did acquirea few extremelyclear vocalizations or an unintelligibleutterance, possibly an at- from informal interactions with trainers: "Hi tempt at another label]; P = 0.0026, chance of Kyo," "Want tickle," "Kiss." These utterances 1/2 [possibleresponses of the learned labels are alwayscontextually appropriate; they are only]).Overall scores were 40/46 (87%)for cork used,respectively, when we enter his room but andfor paper.Table 2 givesa breakdownof her ignorehim (e.g. during cleaning or a "timeout"), errors. while he bows his head and stretchestoward our hands, and when he stretcheshis beak to- KYAARO wardour faces. As Kyaaro always accepts tickles or beak rubs, such utterancescannot be tested Kyaaroalso did not produce,either in the and no claims can be made for their referen- presenceof trainersor in privatepractice, labels tiality. that he experiencedvia audio(wood, rock) or video(cork, key). He attemptedto producela- DISCUSSIONOF EXPERIMENT1 belstaught via the M/R technique(paper, nail) but,at the end of 11months, ran them together Thedifference in performanceof the two ju- ("ail-er")in amanner too difficult to distinguish venile parrots must be examined before the re- by trainersfor testing.He did, however,pro- suitsof theexperiments can be interpreted. Al- duceclearly differentiatedversions of nail and thoughKyaaro's data eventually matched those paperduring privatepractice. After severalmore of Alo, he required several more months of monthsof training,his labelswere at criterion training.A studyof the literatureon develop- for testing.Because Kyaaro subsequently had mentaldelays in humanssuggested that Kyaaro severalsurgeries, we could not test until 1993. had manybehavior patterns in commonwith On identificationtests for eachof theseobjects thoseof "attention-deficitdisordered" (ADD) 306 IRENEM. PEPPERBERG [Auk, Vol. 111 children (Barkley 1990): short attention span, studiessince 1977. He had accessto all parts of the inability to wait one'sturn, difficultyin focus- laboratory(contingent upon his vocal requests;e.g. ing on an object,frequent task-irrelevantactiv- "Wanna go gym") when trainers were present (8 ity (e.g. self-stimulation,which for Kyaaro is h/day); trials thus occurredat various locations.He wasconfined to a desktop and a wire cage(62 x 62 obsessivefoot chewing), erraticperformance on x 73 cm) at other times (e.g. sleeping hours).He had repetitive tasks,and physicalclumsiness. After no regular accessto other parrots,as this study was this study was completed, my students and I completedbefore I acquired the juveniles. Water, a found that Kyaaroresponded to one of the tech- psittacineseed mix, and a limited selectionof chew- niques used for working with ADD children able objects(e.g. wooden plant stakes)were available (Sherman pets. comm.). We found that Kyaaro continuously;fruits, vegetables, whole nutsand other would attend more closelyto all training if he objects(keys, variously shaped pieces of wood,paper, receivedintense physical stimulation (stroking, wool, etc.) were provided at his vocal request(e.g. "I tickling) for several minutes before a session. want cork."). Such intervention would not, however, have When this study began,Alex had been trained ex- clusivelywith the M/R procedureand referential re- affected the overall results of this study. In a wards. He thereby had learned referential use of la- subsequent series of experiments, Kyaaro bels for 50 different objects,6 shapes,7 colors,and learned via the M/R technique and not from quantifiers up to 6 to identify, request, refuse, and other forms of video training even when phys- categorizeobjects (Pepperberg 1981, 1983, 1987a). He ical stimulation preceded all types of sessions had been tested on conceptssuch as the presenceor (unpubl. data). absenceof samenessand difference,and on the ability Thus, the data suggestthat referential, con- to categorizeobjects with respectto color, shape, or textually applicable, and socially interactive in- matter (Pepperberg1987b, 1988c).Other tests(Pep- put best facilitateslearning. Even Kyaaro'sre- perberg1990b) showed that he couldcomprehend as well as produce all of his color, shape, material, and suits suggest the effectivenessof social input that contains reference and contextual infor- categorylabels. He alsohad functionaluse of several phrases (e.g. "Come here," "You tickle," "What's mation. The datado not, however, tell anything that?,""I'm sorry,""You tell me," "Wannago X," and about the effect of input that is socially inter- "Want Y," where X and Y are location and object active but that lacks reference and contextual labels). applicability. I attempted to examine the effect of such in- put aspart of a study on numerical competence. TRAINING PROCEDURES While the juvenile parrots were receiving au- As part of two other studies,on numericalcom- dio, video, and M/R training on object labels, petenceand serial learning, Alex was taught a se- they were alsoexposed to different numbersof quenceof eight number labelsthat had no reference sequentialnotes on a synthesizerwhile humans either to specificobjects in the laboratoryor to pre- modeled correct numerical responses.Neither viously acquiredlabels. The set of labels,il eebam ba bird would attend to training (instead, they ooyuk chilgal, was derived from Koreancount labels would preen, chew their feet, request tickles, both to facilitate comparisonswith children (Fuson etc.) until I used chewable Arabic numerals as 1988) and to be maximally different from English numberlabels already in his repertoire."Barn" (pro- rewards. Such behavior suggested,but did not nounced \baem\) and "ba" were substituted for the prove, the importanceof referenceand context Korean "sam" and "sa" because Alex sometimes had for learning. Data from the juveniles' learning, difficultyproducing an initial "s." Training occurred therefore,must be comparedwith resultsfrom four to five timesper week, beginning October1988 a separateexperiment in which the adult parrot, and continued until Alex produceda modified form Alex, was taught, via the M/R protocol,a non- of the sequencein June1989; we eventuallyaccepted referential set of vocalizations that lacked ex- his insistenceon nukin placeof yuk. plicit contextualapplicability. The usual M/R procedurewas amended to elimi- nate as much referenceand contextas possible.Two humans still engagedin training, but did not em- METHODS I•OR EXPERIMENT 2 phasizethe connectionbetween labels and specific (PAI•T 1) objectsor collections.Initially, one human would state SUBIECT•ND HOUSING "Saynumber" and the otherwould producethe string in the absenceof any objects.Correct responseswere The experimentalsubject, a GreyParrot named Alex, rewarded with vocal praise and the opportunity to has been the focusof cognitiveand communicative requestany desiredobject (Pepperberg 1987a); errors April 1994] GreyParrot Vocal Learning 307 were punished by scolding and time-outs. Trainer TABLE3. Condensedjournal entries describingorder and model/rival reversed roles periodically. of Alex's acquisition of adapted Korean count la- Alex did not attend to training (e.g. would preen bels. or requesttoys or many changesof location)until we included a minimal point of reference: a "number 19 Oct. 1988--Began training on rote sequenceof number labels. board" consistingof a piece of paper with Arabic 8 Nov.--Alex first producedchil in his trainers' symbols1 to 8 tracedalong the diagonal.The trainer presence;he occasionallypraised model with then held the paperin front of the model/rival while "Yeah!That's right" for a correctresponse during asking him/her to "Say number," but did not point training. to the numbers,refer to them in any manner,or trans- 14 Nov.--Clearly vocalizedchil and ba independent fer the paper. Training, therefore, lacked the usual of all other labels when given the count direc- contextand lacked all but minimal referentialit'y. tive, "Say number." Even with the number board, Alex still often ig- 14-29 Nov.--Gave continuous and more frequent nored (e.g. turned his back to) the trainers. He would vocalizations of chil ba. 30 Nov.--First vocalization of chil ee bam ba. say "No" during his turn or would request various 30 Nov.-2 Dec.--Tapes of solitary practiceshowed items before replying. Trainersresponded to such frequent production of chil,chil eebam ba, and oc- behaviorwith a "time-out":They would say"I'm go- casionallyil eebam ba. ing to go away" and leave the room. They would 2-7 Dec.--Tapesof solitarypractice show frequent return if Alex said"Come here," "I'm sorry,"or if he productionof il eebam ba. attemptedthe targetedvocalizations. If, afterlearning 9 Dec.--First vocalized il ee bam ba in a trainer's part of the sequence,Alex regressedto an earlierver- presence. sion (e.g. omitteda label from the string), he would 19 Dec.--Began to place chil later in list; he fre- not be scoldedbut wasasked to 'Saybetter." quently producedil ee bamba (pause)chil. 23 Dec.--Said il ee bam ba chil bail; first evidence of label subsequentto chil. TESTING 9-25 Jan. 1989--Tapes during solitary practice showed frequent production of il eebam bak Alex's testswere identical to the juveniles'with (pause)chil. The "k" sound at the end of bak is respectto precautionsagainst cuing, maintaining his pronounced. attention,correction procedures, and scoring.The tests 16 Jan.--First attempt by Alex at a longer, ordered differedin only two ways.First, Alex wasbeing tested sequence:il eebam bak ss-uck chil. No evidence for on moretasks than the juveniles,so his testsincluded gal. more topics. Second,because the labels did not refer 6-24 Feb.--Often placeda "k" soundon ba;also of- to particularitems, correct responses were rewarded ten inserted sn-uckafter ba in presenceof train- ers. Alex also interrupted with the word chil in with praiseand a chanceto requesta desiredobject its appropriateplace when trainer recited com- rather than the item to which a label referred. plete sequence. Initial testingoccurred mid-June to mid-December 2 Mar.--Continued attempt to recite sevenlabels: il 1989, excludingstudent vacation and exam periods. ee bam basn uck chil. Testsoccurred two to four timesper week but, given 9 Mar.--Alex added an eighth label, although it the numberof topicscovered, trials on Koreanlabels was not correct: il ee bam basn uck chil ee-bail. couldoccur on averageless than oncea week.During 20-24 Mar.--First attempt by Alex at vocalizing tests Alex was shown the number board and asked to eight labelsof similarphonology and order to "Say number." training sequence:il ee bamba oo nuk chil wool. Still no evidence for gal. Alex wasalso tested on his ability to make 1:1cor- 24 Apr.--Responded to "Say number" with il ee respondencesbetween the number labelsand various bam ba oo nuk chil. quantitiesof objects(i.e. on hiscomprehension of the 9 May--Regressed to il ee bamba lookchil. labels).The intentionwas to seeif he would say,for 10 May--Produced il ee bamba oo nuk chil, some- example,"il eebam" to threeitems. During suchtests, times without oo. he wasshown one to sevenobjects on a trayand asked 13 June--Producedentire sequenceil ee bamba oo to "Say number."Such tests occurred early January nuk chilgal. to late March 1990.

in a few days or overnight (Pepperberg 1983, RESULTS OF EXPERIMENT 2 1990c, Pepperberg et al. 1991). In this study, (PART 1) however,he acquiredthese presumably "easy" labelssix weeksand five months,respectively, Training on Part 1 took 9 months, which was after he producedthose that lessclosely resem- unusuallylong (Pepperberg1981, 1983).Nor- bled familiar sounds(e.g. "chil"; Table 3). mally,labels containing sounds that are already Alex eventuallyproduced, in order,the string in Alex'srepertoire (e.g. "ee," "oo") are learned of labelsthat were modeled,substituting "nuk" 308 IR•E M. PEPPERBERG [Auk, Vol. 111 for "yuk." In 20 tests, he made 4 errors (80% erence and contextual applicability. His correct);he omitted "oo" twice, "gal" once,and behaviorwas not necessarilya consequenceof both "oo" and "gal" once. The probability of his training history, becausethe juveniles, de- obtaining this scoreby chancewas less than spitetheir lackof a similar training history,also 0.0001 (binomial test, chance 0.0002). His over- failed to attend to sociallyinteractive input that all score was 20/24 (83%). lacked reference and contextualapplicability Alex could not, however, produce shorter (e.g. the first attempt at training sequential stringswhen presentedwith setsof fewer than numbers).Lack of thesefeatures may, however, eight items.His responseswere similar to those have affectedAlex's motivation to learn. In pre- when asked to "Say number" to sets of eight vious studies,his desire for the objectto which items. In 10 of 14 trials, he produced the entire a label referred affectedthe speedof his acqui- string; he omitted "oo" on two trials (during sitionof that label(Pepperberg et al. 1991).Thus, presentationsof 1 and 4 items), and omitted the lack of referentialreward and contextuality "gal" on two trials (during presentationsof 3 likely contributed to his lack of interest in ses- and 5 items). sionsand his delay in acquisition. Alex's failure to comprehend the labels in Mm-r•OOS FOR EXPERIMENT 2 Part 1 was not surprising. His training inten- (P•R? 2) tionally failed to provide either any under- standingof the individual meaningsof the la- Alex was again the experimental subjectand was bels he heard or any clear connectionbetween housedas in Part I. After testing on Part 1, students use of the labels and his environment (i.e. did and I began to train Alex, with limited referenceand contextuality,on use of the string of labels to refer not provide referenceor specificcontextual ap- to quantity. We modeled 1:1 correspondencesbe- plicability).He hadbeen shown and had learned tween the entire string of labels and several different merely an associationbetween events (i.e. to setsof eight objects(e.g. toy cars,pompons). A trainer produce a rote seriesin responseto a patterned pointed to each objectas the model/rival responded sheet of paper and the command "Say num- with the relevant number label. We intentionally ber"). never modeled quantities less than eight. Training Even the demonstration in Part 2 of 1:1 cor- began in late March 1990 and continued for approx- respondencebetween the full set of labels and imatelythree months,except for studentvacation and the full set of objectsprovided conditionsthat exam periods. were inadequate for a Grey Parrot to learn to Alex was tested as in Part 1. Testing began May 1990 and ended August 1990. comprehendas well as produce a set of labels. Suchtraining provided only a limited senseof context and no direct information about the RESULTS OF EXPERIMENT 2 meaningof individual labels.Previous data (e.g. (PART 2) Pepperbergand Brezinsky 1991) suggestthat, had his training in Part 2 been explicitly ref- Even after our modeling of 1:1 correspon- erential,he might have succeeded.For example, dences for sets of eight items, Alex failed to had Alex been referentially taught the relation- respondappropriately to smallersets. Whatever ship between a few subsetsof the labels and the quantity on the tray, he most often pro- the appropriatesubsets of objects(e.g. the con- duced the entire set of labels. Thus, his behavior nection between "il ee" and two keys, and "il resembled that of Part 1. In 11 of 14 trials, he ee bam" and three cups), he could have trans- producedthe entire string;he omitted "oo" on ferred such learning to the remaining subsets. a trial for five items, omitted "ba" on a trial for The point of the study, however, was to ex- seven items, and omitted "gal" on a trial for amine what he would acquirewith only limited four items. referenceand contextuality. Two inferences can be made from these find-

DISCUSSION OF EXPERIMENT 2 ings. First, social interaction can, to a degree, (PARTS1 AND 2) compensatefor input with only minimal ref- erenceand contextuality,but learningfrom such Although the amountof time that Alex need- input will occur more slowly than from input ed to acquire these labels was somewhatun- that is also referential and contextuallyappli- usual, his training had never before lacked ref- cable.Second, input that is sociallyinteractive, April1994] GreyParrot Vocal Learning 309 but that provides only limited reference and mineexactly what aspectof Alex'sbehavior was contextual applicability, enables a learner to actuallycausing the transferof the desiredob- producebut not comprehendthe code that is jects;or (3) they simply stoppedresponding to trained. what they saw on tape becausethey received no encouragementfor what could have been GENERAL DISCUSSION their first approximationsto the targeted vo- calization(i.e. training wasnot adjustedto their Social interaction, reference, and full contex- level of competence).Clearly, merely watching tual experienceare all important factors in another individual receive objectsfor produc- learning to produceand comprehendan atto- ing particularsounds provided insufficientin- specificcode even for a mimic suchas the Grey put for learning.These data are consistentwith Parrot. Absence of some of these factors affects findings that demonstratethat verbal interac- whether and how learning occurs.The effects tionsare necessaryfor a child to learn how and of each condition can be described in some de- why to use a secondlanguage (a form of ex- tail. ceptionallearning; Pepperberg and Neapolitan When all three factors are missing, as in au- 1988). Children who are exposedto a second diotapepresentations, allospecific vocal learn- languagesolely through television fail to ac- ing doesnot occur.Although the juvenilesre- quire morethan a few words(Snow et at. 1976), acted to the soundsof the tape recorder with even if the showshave a tutorial style (Larsen- parrot noises(e.g. whistlesand squawks),they Freeman 1979). madeno attemptto reproducethe humansounds In the experimentwith Alex, the presenceof containedon the tape. In a sense,the sounds social interaction and severely limited context may havebeen no more than interestingback- and reference provided, at best, conditionsfor ground noises. The birds were given no op- productionbut not comprehensionof an allo- portunity to deduceexplicit meaningsfor the specificcode. In this case,the parrot received sounds and were not shown the purpose for positive feedbackmerely for making a partic- which the soundscould be used.Their response ular soundin responseto a specificcue; the bird to the soundshad no effect on what they sub- was given no reason to work towards under- sequentlyheard or received,either vocally or standing what it was saying or the appropri- physically.Therefore, they had no reasonto atenessof the vocatization. If acquired, such a acquire the sounds.Even so, they might have vocalizationis unlikely to be generalizedto re- learned the sounds from the tapes and then lated situations. Such training representsmost either produceda sound at random or in con- learningsituations of mimeticbirds that arepets, nection with some irrelevant cue, or made some and explainswhy parrots were once thought association between the novel sound and the incapableof doing more than randomly mim- novel objectthat, subsequently,was presented icking human speech sounds (e.g. Lenneberg to them. They did not, however,behave in any 1973).Note that Alex's data are consistentwith of theseways. Accordingto social-modeling those of studies on the use of nonreferential, theory(e.g. Bandura 1971, 1977), acquisition of sociallyinteractive input for otherforms of ex- any form of exceptionallearning (especially ceptionallearning (e.g. with childrenwho are second-languageacquisition; see Snow et al. [also] learning a second language). Krashen 1976) is unlikely for humans under such con- (1982) has shown that children may fail to ac- ditions,and the sameappears true for Grey Par- quire (or acquire very slowly) foreign lan- rots. guagesspoken in their homeswhen the input The presenceof referenceand limited context provides few referencesto the objectsor situ- in the absence of interaction is also not suffi- ationsto which the child is attending and more cient for allospecificvocal learning. Although referencesto other aspectsof the environment. the birds attended to the videos, they did not Similarly, adult humans whose foreign-lan- acquire the soundsthat they saw modeled.In guagecompanions do not provide referenceor the absenceof socialinteraction, the juveniles demonstratecontextual applicability will likely could have failed to learn for at least three rea- fail in their attempt to acquire the secondlan- sons:(1) they failed to realize that the interac- guage(see Burling 1981,Winitz 1981,Klein 1986, tion that they observedcould be transferredto alsoPepperberg and Neapolitan 1988). their own situation; (2) they could not deter- Most likely, social interaction and context 310 IRENEM. PEPPERBERG [Auk, Vol. III without reference will also engender produc- No information existsconcerning Grey Par- tion without comprehension.Such is probably rots' use of referential vocalizations in the wild the casefor pet birds that learn to produce,for (Forshaw 1989). Limited data for other psitta- example,"Hello" or "Bye,bye" routines("Good cine species,however, suggestthat referential- night dear," "Good-bye, and thank you"; Am- ity is a characteristicfor which one might fruit- sler 1947)appropriately but do not comprehend fully search.Some parrots have intrapair duets the use of the individual words in these rou- that are distinct from interpair or other inter- tines.These birds may have a moregeneral sense parrot vocalizations;these duets may mediate of the situations in which their vocalizations interactions among flock members (Gwinner can be used than do birds taught without con- and Kneutgen 1962, Mebes 1978, Wickler 1980). text, but cannot be said to have acquired full Such duets are more complex and take longer functionaluse of the part of the allospecificcode to learn than those of passerines(e.g. wrens; they have acquired. Farabaugh 1982). If complex communication Basedon the predictionsof social-modeling developed in responseto, and in order to me- theory (Bandura1977, Pepperberg1991), I sug- diate, complexsocial interactions(Humphrey gest that Grey Parrotsare also unlikely to ac- 1978, Crook 1983, Burling 1985), such vocali- quire comprehensionof elementsof an allospe- zations could be referential. So far, Yamashita cific code from input that is referential, fully (1987)reported cooperative vocal "sentinel" be- contextually applicable, but noninteractive. havior in flocksof Indigo Macaws (Anodorhyn- Thus, the presenceor absenceof an item that chusleari) as part of their extensive social or- could be considereda reward is likely to be less ganization,Gnam (1988)suggested that callsare important than the presenceor absenceof social used for individual recognition of mated pairs interaction. In the M/R training, for example, within groups of Bahama Amazons (Amazona the presenceof a reward is unlikely to be the leucocephalabahamensis), and Pepper'sdata (pers. most critical factor in learning becausea bird comm.)suggest individual vocalrecognition and is rewarded only after it has made an attempt contextual calls in Glossy Black Cockatoos at the targetedlabel; that is, reward occursonly (Clyptorhynchuslathami; note Saunders 1983). after somelearning has taken place. Moreover, Moreover, a Grey Parrot would not likely ac- the reward primarily reinforces referentiality. quire referential communication in the labo- Data on nonvocalallospecific learning in a non- ratory unless such behavior (e.g. Pepperberg human primate provide somecorroboration. For 1990a, 1992b, in press)were basedon a preex- example, a chimpanzee (Pan troglodytes),who istent cognitive architecture (Rice 1980, Pre- learned non-interactively to produce symbols mack 1983). basedon human language to answer questions Comparinglaboratory and fieldwork on Grey or make requests,was unable to generalize to Parrots is, however, difficult because no studies related situations(see Savage-Rumbaughet al. exist that describe the conditions under which 1980a, b). Like this chimpanzee, parrots might Grey Parrotsacquire or use allospecificvocali- learn to produce appropriate elements of the zationsin nature. A recent paper (Cruickshank code, but comprehension would similarly be et al. 1993)provides the only evidencethat such lacking. Becauseit is conceivable(although un- allospecificlearning occurs.Conceivably, re- likely) that parrots could learn under condi- search such as mine will provide an impetus tions unfavorable for chimpanzees, such con- for the appropriate field studies. ditions remain to be tested. In sum, I have sought to determine the con- All of these experiments, however, involve ditions necessaryfor the acquisitionof a ref- teachinga referential allospecificcode to sub- erential, allospecificcommunication code by jects in a laboratory situation. Thus, two ques- Grey Parrots. Even though mimetic birds are tions arise as to the generalvalidity of the ex- characterizedby their extensive capacitiesto periments:(1) To what extent is natural parrot acquireallospecific vocalizations, Grey Parrots communication referential (i.e. why examine (at least)seem to learn sucha codemost readily referentialityif parrotsdo not normallyengage under certain environmental conditions. Al- in referential communication)? (2) Under what though some combinations of conditions re- conditionsmight allospecificlearning occurin main to be tested for these birds (e.g. reference nature (i.e. how does this study relate to con- and full contextualapplicability in the absence ditions in the real world)? of social interaction; reference and limited con- April 1994] GreyParrot Vocal Learning 311 text with either full or limited interaction; the guage learning. Pages 279-290 in Native lan- effectiveness of two- versus three-dimensional guageand foreign languageacquisition (H. Win- referents), input that is fully referential, con- itz, Ed.). New York Academy of Sciences,New York. textuallyapplicable, and sociallyinteractive en- BURLING,R. 1985. The selectiveadvantage of com- suresthat theseparrots cannot only producebut plex language.Ethol. Sociobiol.7:1-16. alsoeventually comprehend allospecific vocal- CROOK,J. H. 1983. On attributing consciousnessto izations (Pepperberg 1987a, b, 1990b, 1992b). . Nature 303:11-14. Lack of some or all of these aspectswill affect CRUICKSHANK,A. J., J.-P. GAUTIER,AND C. CHAPPUIS. the courseof acquisitionand will likely prevent 1993. Vocal mimicry in wild African Grey Par- full allospecificlearning from occurring. rots Psittacus erithacus. Ibis 135:293-299. DAVIS, H. 1984. Discrimination of the number three

ACKNOWLEDGMENTS by a raccoon(Procyon lotor). Anim. Learn. Behav. 4:121-124. This work wassupported by grantsto I.M.P. from FARABAUGH,S.M. 1982. The ecologicaland social the National Science Foundation (BNS 88-20098 and significanceof duetting.Pages 85-124 in Acoustic 91-96066)and the NSF ResearchExperience for Un- communicationin birds. Vol. 2, Song learning dergraduatesprogram, and to studentsin the Uni- and its consequences(D. E. Kroodsmaand E. H. versity of Arizona UndergraduateBiology Research Miller, Eds.). Academic Press, New York. Program.I thank Madonnaand RussLaPell of VIP FORSHAW,J.M. 1989. The parrots of the world. Lans- Aviaries for donation of Alo and Kyaaro, Gloria Dolan down Editions,Sydney. of Emerald Bird Caddy for donation of cages,gyms, FUSON,K. 1988. Childlren's counting and concepts and stands, and Pam Banta, Mike Brezinsky, Robyn of number.Springer-Verlag, New York. Bright, SeanGarcia, Stacey Humble, EricJackson, Jane GINSBURG,N. 1963. Conditioned talking in the my- Lewis, Sharon Marconi, Diana May, Dianne Patter- nah bird. J.Comp. Physiol. Psychol. 56:1061-1063. son,Jayme Silverstone, Norma Quevedo,Susan Tur- GNAM,R. 1988. Preliminary resultson the breeding hey, Audrey Warren, Denice Warren, and Rebecca biology of BahamaAmazon. Parrot Letter 1:23- Zweifel for help as secondarytrainers. I thank Pam 26. Banta,Luis Baptista,Donald Kroodsma,Lewis Petri- GOLDSTEIN,H. 1984. The effects of modeling and novich, Meredith West and two unknown referees corrected practice on generative language and for commentson an early version of the manuscript. learning of preschoolchildren. J. SpeechHear. Disord. 49:389-398. GOSSETTE,R.L. 1969. Personal communication to O. LITERATURE CITED H. Mowrer, 1980. Pages105-106 in Psychology of language and learning (O. H. Mowrer, Ed.). AMSLER,M. 1947. An almost human Grey Parrot. Plenum Press, New York. Avicult. Mag. 53:68-69. GREENFIELD,P.M. 1978. Developmental processesin BANDURA,A. 1971. Analysisof modeling processes. the languagelearning of child and chimp. Behav. Pages 1-62 in Psychologicalmodeling (A. Ban- Brain Sci. 4:573-574. dura, Ed.). Aldine-Atherton,Chicago. GROSSLIGHT,J. H. AND W. C. ZAYNOR. 1967. Verbal BANDURA,A. 1977. Socialmodeling theory. Aldine- behavior in the mynah bird. Pages5-19 in Re- Atherton, Chicago. BAPTISTA, L.F. 1988. Imitations of White-crowned searchin verbal behavior and some neurophys- iological implications (K. Salzinger and S. Sal- Sparrow songsby a Song Sparrow. Condor 90: 486-489. zinger, Eds.).Academic Press, New York. GROSSLIGHT,J. H., W. C. ZAYNOR, AND B. L. LIVELY. BAPTISTA,L. F., AND M. L. MORTON. 1981. Interspe- 1964. Speechas a stimulus for differential vocal cific song acquisitionby a White-crowned Spar- row. Auk 98:383-385. behavior in the mynah bird (Graculareligiosa). BAPTISTA,L. F., AND L. PETRINOVICH. 1984. Social Psychon.Sci. 1:7-8. interaction, sensitivephases, and the song tem- GWINNER,E., ANDJ. KNEUTGEN.1962. Uber die biolo- plate hypothesisin the White-crownedSparrow. gischeBedeutung der 'zweckdienlichen'Anwen- Anim. Behav. 32:172-181. dung erlernterLaute bei Vogeln.Z. Tierpsychol. 19:692-696. BAPTISTA,L. F., ANDL. PETRINOVICH.1986. Song de- velopment in the White-crowned Sparrow: So- HUMPHREY,N. 1978. Nature's psychologists.New cial factors and sex differences. Anim. Behav. 34: Sci. 78:900-903. 1359-1371. KLEIN,W. 1986. Secondlanguage acquisition. Cam- BARKLEY,R. A. 1990. Attention deficit disorder: A bridgeUniv. Press,Cambridge. handbookfor diagnosisand treatment.Guildford KRASHEN,S. 1982. Principlesand practicein second Press, New York. languageacquisition. Pergamon Press, Oxford. BURLING, R. 1981. Social constraints on adult lan- LARSEN-FREEMAN,D. 1979. The importanceof input 312 IRENEM. PEPPERBERG [Auk,Vol. 111

in secondlanguage acquisition. Pages 87-93 in ical perspectives(T. R. Zentall and B. G. Galef, Pidginization and creolizationas languageac- Eds.).Erlbaum Associates, Hillsdale, New Jersey. quisition (R. W. Andersen,Ed.). Newbury House, PEPPERBERG,I. M. 1988b. An interactivemodeling Rowley, Massachusetts. techniquefor acquisitionof communicationskills: LENNEBERG,E. H. 1973. Biologicalaspects of lan- Separationof "labeling" and "requesting"in a guage.Pages 49-60 in Communication,language, psittacinesubject. Appl. Psycholinguist.9:59-76. and meaning (G. A. Miller, Ed.). BasicBooks, New PEPPERBERG,I. M. 1988c. Comprehension of "ab- York. sence"by an AfricanGrey Parrot:Learning with IV[ARLER,P., AND S. PETERS. 1977. Selective vocal respectto questionsof same/different. J. Exp. learning in a sparrow.Science 198:519-521. Anal. Behav. 50:553-564. MARLER,P., ANDS. PETERS.1987. A sensitiveperiod PEPPERBERG,I. M. 1990a. Somecognitive capacities for song acquisitionin the Song Sparrow,Mel- of anAfrican Grey Parrot (Psittacus erithacus). Pages ospizamelodia: A case of age-limited learning. 357-409 in Advancesin the study of behavior, Ethology 76:89-I00. vol. 19 (P. J. B. Slater, J. S. Rosenblatt, and C. MEBEs,H.D. 1978. Pair-specificduetting in the Peach- Beer, Eds.). Academic Press, New York. faced Lovebird, Agapornisroseicollis. Naturewis- PEPPERBERG,I. g. 1990b. Cognition in an African senschaften 65:66-67. Grey Parrot(Psittacus erithacus): Further evidence MILES,H.L. 1983. Apes and language.Pages 43-61 for comprehensionof categoriesand labels.J. in Language in primates (J. de Luce and H. T. Comp. Psychol.104:41-52. Wilder, Eds.).Springer-Verlag, New York. PEPPERBERG,I. M. 1990c. Referentialmapping: A MOWPER,O. H. 1952. The autismtheory of speech techniquefor attachingfunctional significance development and some clinical applications. J. to the innovativeutterances of an African Grey SpeechHearing Disor& 17:263-268. Parrot(Psittacus erithacus). Appl. Psycholinguist. MOWPER,O. H. 1954. A psychologistlooks at lan- 11:23-44. guage.Am. Psychol.9:660-694. PEPPERBERG,I. M. 1991. Learning to communicate: MOWPER,O. H. 1958. Hearing and speaking:An Theeffects of socialinteraction. Pages 119-164 in analysisof languagelearning. J. SpeechHearing Perspectivesin ethology,vol. 9 (P. P. G. Bateson Disord. 23:143-152. and P. H. Klopfer, Eds.).Plenum Press,New York. PEPPERBERG,I. M. 1981. Functionalvocalizations by PEPPERBERG,I.M. 1992a. What studieson songlearn- an African Grey Parrot (Psittacuserithacus). Z. ing can teach us about playback experiments. Tierpsychol.55:139-160. Pages47-57 in Playback and communi- PEPPERBERG,I. M. 1983. Cognition in the African cation:Problems and prospects(P. K. McGregor, Grey Parrot:Preliminary evidence for auditory/ Ed.). Plenum Press, New York. vocal comprehensionof the classconcept. Anim. PEPPERBERG,I. M. 1992b. Proficient performanceof Learn. Behav. 11:179-185. a conjunctive,recursive task by an African Grey Parrot(Psittacus erithacus). J. Comp. Psychol.106: PEPPERBERG,I. M. 1985. Social modeling theory: A 295-305. possibleframework for understandingavian vo- PEPPERBERG,I. M. 1993. A review of the effects of cal learning. Auk 102:854-864. social interaction on vocal learning in African PEPPERBERG,I. M. 1986. Acquisition of anomalous Grey Parrots (Psittacuserithacus). Neth. J. Zool. communicatorysystems: Implications for studies 43:104-124. on interspeciescommunication. Pages 289-302 in PEPPERBERG,I. g. 1994. Numerical competence in an Dolphin behavior and cognition: Comparative African Grey Parrot.J. Comp. Psychol.108:1-9. and ecologicalaspects (R. Schusterman,J. Thom- PEPPERBERG,I. M., K. J. BRESE,AND B. J. HARRIS. 1991. as, and F. Wood, Eds.). Erlbaum Associates,Hills- Solitarysound play during acquisition of English dale, New Jersey. vocalizationsby an AfricanGrey Parrot (Psittacus PEPPERBERG,I. M. 1987a. Evidence for conceptual erithacus): Possible parallels with children's quantitative abilities in the African Grey Parrot: monologuespeech. Appl. Psycholinguist.12:151- Labeling of cardinalsets. Ethology 75:37-61. 178. PEPPERBERG,I. M. 1987b. Acquisition of the same/ PEPPERBERG,I. M., AND M. V. BREZINSKY. 1991. Ac- different conceptby an African Grey Parrot (Psit- quisitionof a relativeclass concept by an African tacuserithacus): Learning with respect to color, Grey Parrot (Psittacuserithacus): Discriminations shape,and material.Anim. Learn. Behav.15:423- basedon relativesize. J. Comp. Psychol. 105:286- 432. 294. PEPPERBERG,I. M. 1988a. The importance of social PEPPERBERG,I. M., AND D. M. NEAPOLITAN. 1988. Sec- interactionand observationin the acquisitionof ond language acquisition: A framework for communicative competence: Possible parallels studyingthe importanceof input and interaction between avian and human learning. Pages279- in exceptionalsong acquisition. Ethology 77:150- 299 in Sociallearning: Psychologicaland biolog- 168. April1994] GreyParrot Vocal Learning 313

PEPPERBERG,I. M.,AND L. SCHINKE-LLANO.1991. Lan- in three socialclasses. J. Psycholinguist.Res. 5:1- guageacquisition and form in a bilingual envi- 20. ronment:A frameworkfor studyingbirdsong in TODT,D. 1975. Sociallearning of vocalpatterns and zones of sympatry.Z. Tierpsychol.89:1-28. modes of their applicationsin Grey Parrots. Z. PREMACK,D. 1983. The codes of man and beast. Be- Tierpsychol. 39:178-188. hay. Brain Sci. 6:125-167. VANAYAN, M., H. A. ROBERTSON,AND G. B. BIEDERMAN. PUTNEY,R.T. 1985. Do willful apesknow what they 1985. Observationallearning in pigeons:The ef- are aiming at? Psychol.Record 35:49-62. fects of model proficiency on observer perfor- PacE,M. 1980. Cognition to language:Categories, mance.J. Gen. Psychol.112:349-357. word meanings,and training. University Park WEST, M. J., A. N. STROUD,AND A. P. KING. 1983. Press, Baltimore. Mimicry of human voice by EuropeanStarlings: SAUNDERS,D.A. 1983. Vocal repertoireand individ- The role of social interaction. Wilson Bull. 95: ual vocal recognitionin the short-billedWhite- 635-660. tailed Black Cockatoo,Calyptorhynchus funereus WICKLER,W. 1980. Vocalduetting and the pairbond. latirostrisCarnaby. Aust. Wildl. Res. 10:527-536. I. Coynessand the partner commitment.Z. Tier- SAVAGE-RUMBAUGH,E. S., D. M. RUMBAUGH, AND S. psychol.52:201-209. BOYSEN.1980a. Do apesuse language?Am. Sci. WINITZ,H. 1981. Input considerationin the com- 68:49-61. prehensionof first and secondlanguages. Pages SAVAGE-RUMBAUGH,E. S., D. M. RUMBAUGH,S. T. SMrrH, 296-308in Native languageand foreignlanguage ANDJ. LAWSON. 1980b. Reference:Thelinguistic acquisition(H. Winitz, Ed.). New York Academy essential. Science 210:922-925. of Sciences, New York. SLATER,P. J. B. 1991. Learned song variations in Yaxt•sHrrA, C. 1987. Field observations and com- British storm-petrels?Wilson Bull. 103:515-517. mentson the IndigoMacaw (Anodorhynchus leari), SNOW,C. E., A. ARLMAN-RuPP,Y. HASSING,J. JOBSE,J. a highly endangeredspecies from northeastern JOOSTEN,AND J. VOP, STER. 1976. Mother's speech Brazil. Wilson Bull. 99:280-282.