Monkey Octave Generalization vs. Music Perception: Childhood Songs vs. Synthetic Melodies , Study notes of Psychology

Download Monkey Octave Generalization vs. Music Perception: Childhood Songs vs. Synthetic Melodies and more Study notes Psychology in PDF only on Docsity! lounul of Experimental Psychology: General 2000. Vol. 129. No. 3. 291-307 Copyright 2000 by the American Psychologic.) Association, Inc. C096-3445/OOrt5.00 DOI: 10.1037S0096-344S.129.3.29I Music Perception and Octave Generalization in Rhesus Monkeys Anthony A. Wright and Jacquelyne J. Rivera University of Texas Medical School at Houston Stewart H. Hulse Johns Hopkins University Melissa Shyan Butler University Julie J. Neiworth Carleton College Two rhesus monkeys were tested for octave generalization in 8 experiments by transposing 6- and 7-note musical passages by an octave and requiring same or different judgments. The monkeys showed no octave generalization to random-synthetic melodies, atonal melodies, or individual notes. They did show complete octave generalization to childhood songs (e.g., "Happy Birthday") and tonal melodies (from a tonality algorithm). Octave generalization was equally strong for 2-octave transpositions but not for 0.5- or l.S-octave transpositions of childhood songs. These results combine to show that tonal melodies form musical gestalts for monkeys, as they do for humans, and retain their identity when transposed with whole octaves so that chroma (key) is preserved. This conclusion implicates similar transduction, storage, processing, and relational memory of musical passages in monkeys and humans and has implications for nature-nurture origins of music perception. Music is considered among cultures' highest achievements. Nevertheless, music from different cultures shares many charac- teristics. Among these common characteristics is that all music uses a limited number of possible notes. A limited number of possible notes helps to make songs memorable and reproducible. Other factors contribute to their memorability, reproducibility, and general appeal. Take, for example, the familiar tune "Happy Birth- day." There is no doubt about its memorability. The first four notes readily identify it. Furthermore, different sets of four notes sepa- rated by whole octaves suffice equally well to identify "Happy Birthday." Preverbal infants as well as adults can identify a transposed melody as the same melody while at the same time recognizing that the notes are different, that is, different pitch heights (e.g., Demany & Armand, 1984; Pick & Palmer, 1993, p. 199; Trehub, Morrongiello, & Thorpe, 1985). Thus, the melody becomes a whole or gestalt unto itself, somewhat independent of the notes used to produce it. Anthony A. Wright and Jacquelyne J. Rivera, Department of Neurobi- ology and Anatomy, University of Texas Medical School at Houston; Stewart H. Hulse, Department of Psychology, Johns Hopkins University; Melissa Shyan, Department of Psychology, Butler University; Julie J. Neiworth, Department of Psychology, Carleton College. This research was partially supported by National Institutes of Health Grants MH-35202 and DA-10715 and National Science Foundation Grant IBN 9317868. We gratefully acknowledge the assistance of Annie Takeuchi in analysis of the childhood song and synthetic melodies and supplying the tonal and atonal melodies for Experiments 7 and 8. We also thank Jim Bartlett and Jeffrey Katz for their careful and thoughtful suggestions on earlier drafts of this article. Correspondence concerning this article should be addressed to Anthony A. Wright, University of Texas Medical School at Houston, Department of Neurobiology and Anatomy, PO Box 20708, Houston, Texas 77225. Elec- tronic mail may be sent to anthony.a.wright@uth.tmc.edu. Among critical variables in the recognition of melodies are contour (the pattern of up and down changes in pitch), interval (sequence of pitch distances in semitones), and chroma (key; e.g., Dowling & Hollombe, 1977). All three of these variables contrib- ute to melody perception and can change it in complex ways. When all three variables are controlled, there is almost no decre- ment in melody identification, even novel melodies (e.g., Massaro, Kallam, & Kelly, 1980). For example, a 2-octave change in the melody where a starting note of C at 262 Hz is changed to a C at 1,048 Hz preserves the original melody almost perfectly. The importance of this "octave" effect can be demonstrated by com- paring changes that are other than whole octaves. For example, changing the starting note (and melody) from a C at 262 Hz to an F* at 370 Hz is a much smaller frequency change. However, this much smaller frequency change has a much greater effect on changing the melody from its original. The octave scale is based on frequency doubling and is fundamental to music perception and the auditory system generally (e.g., Dowling & Hanvood, 1986). This special status of the similarity of octave-transposed notes can be modeled as a three-dimensional helix of ascending pitch where notes of the same key (e.g., C) are vertically aligned and are thus in close proximity (Shepard, 1982). In the present article, tests with frequencies doubled, halved, or higher multiples of two are re- ferred to as octave-generalization tests. Octave generalization is a special case of contour transposition, where, in addition to contour, intervals (sequence of pitch dis- tances) and chroma are also preserved. An octave-generalized melody can sound so similar to the original, particularly after a time delay, that the change may not even be noticed. This is not to say that notes of the same key from different octaves are indistin- guishable. Not only is the pitch change noticeable when melodies from different octaves are played in rapid succession, but (random) note substitutions from different octaves destroy the melody (Deutsch, 1972; Dowling & Hollombe, 1977). Thus, a melody can be shifted all over the music scale and still retain its gestalt 291 292 WRIOHT, RIVERA, HULSE, SHY AN, AND NE1WORTH (melody) qualities. But substitutions from different octaves, even notes with the same letter name (same chroma), tend to destroy the gestalt. The purpose of the studies presented in this article was to explore the possibility that some animal species might perceive music in the same gestalt-preserving way that humans do. When these studies began, there was little encouragement to be found for this possibility. Most human researchers have tended to regard music perception as a human experience. The small amount of animal research on this topic has done little to dispel this notion. Monkeys failed octave-generalization tests (e.g., D'Amato, 1988; D'Amato & Salmon, 1984), and so did songbirds (e.g., Cynx, 1993; Dooling, Brown, Park, Okanoya, & Soli, 1987; Hulse & Cynx, 1985, 1986). In a review, D'Amato (1988) concluded that "monkeys can't hum a tune . . . Because they don't hear them" (p. 478). And this deficiency "may be widespread among animals" (p. 453). Furthermore, two widely referenced claims of octave generali- zation in animals do not stand scientific scrutiny. A claim of octave generalization in the white rat was based on one point for one test frequency displaced from one generalization gradient for one train- ing frequency (Blackwell & Schlosberg, 1943). Many more tests with other training and test frequencies would be necessary for this to be a definitive test of octave generalization today. A claim was made for octave generalization in the bottlenosed dolphin (Ralston, Herman, Williams, Gory, & Jerger, 1988). It is difficult to tell the basis for this claim because it appeared only as a meeting abstract. This laboratory had previously shown that dolphins would spon- taneously mimic computer generated sounds and would occasion- ally transpose their mimics by an octave (Richards, Wolz, & Herman, 1984). It is unclear how these two references are related without more information. Experimental control of such sponta- neous octave shifts undoubtedly would be difficult, and maybe this is the reason why these intriguing preliminary results were not followed up. In any case, when the research of this article began, there were several failures, few positive indications, and no defin- itive evidence in support of octave generalization in nonhuman animals. Dauntless, we reasoned that contour and octave generalization should depend on relating two musical passages. Relational learn- ing, in this case, means that the subject relates the two musical passages and has learned the same-different concept. Concept learning is assessed by testing with novel stimuli. If the concept has been learned, then accurate same-different judgments should be made to any pair of sounds or music, whether they be familiar or novel. Furthermore, training and testing should be conducted in a musical context. A musical context means that judgments should be based on a sequence of notes, that is, a melody, not individual notes or sine-wave tones. Simply training with melodies instead of notes does not always ensure that judgments will be based on the whole melody. Individual notes of a melody can come to control the behavior. Such a result becomes most likely when only a few melodies are used in training and when absolute identification of a particular melody is required (e.g., go/no-go tasks) rather than a relationship between melodies. Go/no-go tasks typically have a single S+ stimulus in which responses are rewarded. In a study by D'Amato and Salmon (1984), for example, only certain notes, not the whole tune, were shown to control the go/no-go discrimination of monkeys. With these requirements of relational concept learning and a musical context, two relations emerge. The first relation is among the notes—the melody. The other relation is between melodies— the same-different judgment. Therefore, the task might be thought of as a relation between relations or analogical-reasoning task (e.g., Gillan, Premack, & Woodruff, 1981). Unfortunately for the prognosis of these experiments, only language-trained chimpan- zees and humans are thought capable of learning analogical- reasoning tasks (Premack, 1983a, 1983b). Such a task might be- come simplified if the notes configure together as gestalts (melodies). Simplifying the analogical-reasoning task by resolving the relationship among notes into a gestalt (melody) might allow species less sophisticated than language-trained chimpanzees to learn such a task. For a variety of reasons, rhesus monkeys were used in these studies, not the least of which was that rhesus monkeys learn relationships among visual stimuli more readily than do some other species such as pigeons (Wright, Cook, & Kendrick, 1989; Wright, Santiago, Sands, & Urcuioli, 1984). Despite their relative predisposition for learning visual relationships, researchers had not had much success in training rhesus monkeys in auditory relational tasks (Mishkin, personal communication; Thompson, 1980). Nev- ertheless, after several task modifications, including the require- ment that monkeys touch the speakers (sound sources), two rhesus monkeys eventually did learn an auditory same-different task. In this auditory same-different task, the sample (first) sound was presented from a center speaker. A second sound was then presented simultaneously from two side speakers. If this second sound matched the first sound, then a touch to the right-hand speaker was required to obtain reward. If the second sound was different from the first sound, then a touch to the left-hand speaker was required to obtain reward. A fading procedure was used in the early training stages of this task. Initially, the second sound came only from the correct side speaker. The intensity of second sound from the incorrect side speaker was gradually increased in inten- sity until both speakers were equal. Natural (e.g., coyote howl) and environmental (e.g., car crash) sounds were used in the original training and testing and in maintaining accurate performance on this task in the experiments of this article. The monkeys' relational learning in this same-different task was shown by their abstract concept learning; they performed as well with novel stimuli as with training stimuli (Wright, Shyan, & Jitsumori, 1990). The experiments presented in this article capi- talized on the monkeys' auditory same-different concept learning as the basis for testing music perception and octave generalization. The order in which the experiments are presented is the order in which they were conducted. Experiment 1 The purpose of Experiment 1 was to test contour transpositions of "synthetic" melodies in a same-different task with monkeys. These melodies were essentially a random collection of notes within a restricted frequency range. Music training began by gradually introducing music trials while maintaining accurate per- formance with natural/environmental sound stimuli. Stable perfor- mance with natural sounds assured a stable baseline and main- tained the same-different concept. As was done in training with natural sounds, melody training was conducted with a large num- MONKEY OCTAVE GENERALIZATION AND MUSIC PERCEPTION 295 (DIFF) response is plotted because the two melodies were different even though their notes were the same. These melodies were judged to be different 74% of the time (73.3% for BW and 75.9% for FD) even though they shared the same six notes. This result shows that the monkeys were not under stimulus control of indi- vidual notes, which adds to the evidence that their responses were based on the relationship among notes, the melody. For transposition tests (SAME-TUNE, DIFF-FRBQ), percent same response was plotted because the melodies were the same even though the frequencies (FREQ) of the notes were different. These are the critical transposition tests, and therefore this histo- gram is filled in the figure. The monkeys did not generalize to these transposed melodies (46.7% for BW, 60% for FD). They judged them to be the same melody only 53.3% of the time. This performance was not (significantly) different from (50%) chance—<(1) = 0.35, p > .5, single means test: Hays, 1963, p. 311—and was (significantly) different from training-trial (SAME and DIFF) performance, F(l, 4) = 31.7, p < .01. In conclusion, there is no evidence that these monkeys generalized to these transposed synthetic melodies. Experiment 2 The lack of generalization to transposed melodies in Experi- ment 1 was puzzling. One possibility was that monkeys do not have this ability, as D'Amato (1988) had concluded. Another possibility was that the synthetic melodies were inadequate in some way for testing octave generalization. Their random compo- sition did not result in what most listeners would refer to as a melodic tune. Indeed, 44% of these synthetic melodies had notes outside the appropriate diatonic scale, similar to what Trehub (1993) and Trehub, Thorpe, and Trainer (1990) referred to as "bad" melodies. Tunes that are not musical tend to be less mem- orable (e.g., Frances, 1958, Experiment 9; Krumhansl, 1979, Ex- periment 3; Trehub, Cohen, Thorpe, & MorrongieDo, 1986). Any detriment to the memorability of a melody would, in turn, be expected to adversely affect transposition. How could a transposed melody be matched to one previously heard, if the previous one cannot be remembered? In Experiment 2, we explored transpositions of melodies that would be expected to be more musical, more scalar, more tonal, and more memorable than the synthetic melodies of Experiment 1. The rationale was that by maximizing these qualities we might maximize the chances of finding same-melody recognition in these monkeys. Among melodies considered to be most musical are childhood songs, such as "Old MacDonald," "Yankee Doodle," and "Happy Birthday" (e.g., Dowling, 1988; Kallman & Massaro, 1979; Unyk, Trehub, Trainor, & Schellenberg, 1992). Most people find such childhood songs memorable, occasionally to the point of annoyance. In addition, (whole) octave transpositions were tested rather than partial octave transpositions. The rationale was that these two factors, childhood songs and octave transpositions, might work together to increase the chances of finding octave generalization and melody recognition in monkeys. Method Subjects and Apparatus The subjects and apparatus were identical to Experiment 1. Procedure and Stimuli The procedure was similar to Experiment 1, except that childhood songs were tested. These childhood songs were the first six notes of "Row Row Row Your Boat," "Oh Susanna," "London Bridge," "Old MacDonald," "Yankee Doodle," "Ring Around the Rosey," "Skip To My Lou," 'Two Bits," "Camptown Races," "Happy Birthday," and "How Much Is That Doggie In The Window." For one melody, "Skip To My Lou n," the last six notes were used. They were played in dieir original rhythms by a musician on a Yamaha DX7 electronic keyboard with the same instruments as in Experiment 1: Flute, Accordion, Spanish Guitar, Oboe, Electric Piano, Bag Pipe, Clarinet, Grand Piano, Honky-Tonk Piano, and Glock- enspiel. Melodies were recorded in three different octaves spanning a frequency range of 131 Hz to 988 Hz. There were 12 melodies, 10 instruments, and three octaves for a total of 360 recordings. Test melodies were novel renditions when instrument and octave were taken into account. Selection was quasirandom and as many aspects as possible were coun- terbalanced widiin the limits of the experiment. Training trials were similar to Experiment 1. The natural-sound stimuli were selected quasirandomly from the same 520 stimulus pool used in Experiment 1. The synthetic melodies used in training were the same as Experiment 1, except that the order of first versus second melody on different trials varied quasirandomly. Each training melody was unique to a particular trial during the entire experiment. The sequence of trials was different from Experiment 1. On two test trials, octave transpositions of two melodies (childhood song) were tested. On each of these test trials, there was an octave difference in frequency between presentations of a melody. Thus, these trials are referred to as Same-Melody, Different-Frequency (SAME-TUNE DIFF-OCTAVE). These were the octave-generalization test trials and were the trials of primary interest in this experiment. Within each test trial, both presentations of the childhood song were played by the same instrument Selection of song, instrument, octave, and transposition direction (higher vs. lower frequency) was quasirandom. On the remaining two test trials, two different melodies (childhood songs) were presented from the same frequency range or octave. Thus, these trials are referred to as Different-Melody, Same-Frequency trials (DIFF-TUNE SAME-FREQ) and are essentially tests for control by fre- quency (as opposed to melody). Like the octave-generalization test trials, both melodies were played by the same instrument and selection of song, instrument, and octave to be tested was quasirandom. Reward on test trials was given for either response. Reward probability was set equal to training-trial accuracy from the previous session for each subject. This change in reinforcement probability was made so that test trials would be unlikely to be distinctive from training trials on the basis of outcome. Results and Discussion The results are shown in Figure 3. Training-trial performance is shown on the left. Training-trial accuracy was reasonably good with natural sounds (BW = 78.7%, 78.2%; FD = 75.3%, 79.6% for DIFF & SAME, respectively) and with synthetic melodies (BW = 80.0%, 86.6%; FD = 86.7%, 86.7% for DIFF & SAME, respectively). The most striking result was the substantial generalization to octave-transposed childhood songs, compared with the lack of any generalization to transposed synthetic melodies of Experiment 1. Octave generalization was 76.7% (for both monkeys) and is sig- nificantly different from chance performance, 1(1) > 100, p < .0001. An analysis of variance (ANOVA) showed significant effects across training and testing conditions, F(5, 6) ** 6.90, p < .02. 296 WRIGHT, RIVERA, HULSE, SHYAN, AND NEIWORTH TRAINING TESTING 100 _ 90 o o: 80 or 8 70 g 60 o Q; 1 1 RnUJ DU D_ 40 Octave Generalization DIFF SAME NATURAL SOUNDS DIFF-TUNE SAME-TUNE DIFF-FREQ SAME-FREO DIFF-TUNE SAME-TUNE SAME-FREQ DIFF-OCTAVE Figure 3. Mean training and testing performance of two monkeys with natural/environmental sounds and childhood songs such as "Happy Birthday." For DIFF-TUNE SAME-FREQ (DUFF = different; FREQ = frequency) testing trials, the percent different responses is plotted because the two songs were different. For SAME-TUNE DIFF-OCTAVE testing trials, the percent same response is plotted because the two melodies of each trial were the same song but the frequencies of their notes differed by an octave (factor of two). The dashed line is chance (50% correct) performance. 100 90 80 70 60 50 40 Planned comparison tests (Sokal & Rohlf, 1969, p. 234) using the MSB (mean square error/residual) term from the ANOVA showed that octave generalization was not significantly different from training performance with natural sounds, fs(6) = 0.05, 0.38, p > .5, for DIFF & SAME, respectively. Thus, it can be said that generalization was complete because it was equivalent to training- trial performance. This is the first definitive evidence (in our assessment) for octave generalization from a (nonhuman) animal species and the first evidence from a task where animals were required to compare pairs of musical passages. The generalization with childhood songs is interesting from a human developmental perspective. It might be reasonable to assume that human recognition of such octave trans- posed childhood songs would depend on a lifetime of exposure to these melodies (cf. Dowling, 1978). However, the monkeys showed good generalization to (octave) transposed childhood songs with little or no experience with them, let alone a lifetime of experience. The much greater generalization to childhood songs than synthetic melodies means that the critical difference is to be found in the melodies themselves. Before considering manipulations that might affect octave gen- eralization, it should be pointed out that there was some improve- ment on training trials with synthetic melodies relative to Exper- iment 1. This slight improvement could have been due to the greater experience with the musical passages. The next experiment was conducted to evaluate any effect this greater experience might have on generalization to the transposed synthetic melodies of Experiment 1. Experiment 3 We conducted Experiment 3 to test the possibility that greater experience with music stimuli might have influenced the differ- ence in generalization between Experiments 1 and 2. Thus, Ex- periment 3 was essentially the third "leg" of an A-B-A experi- mental design, where the first test (A) is repeated to control for the possibility that experience has influenced the results. Synthetic melodies of Experiment 1 were retested with some minor modifi- cations to make Experiment 3 more similar to Experiment 2 than was Experiment 1. Method Subjects and Apparatus The subjects and apparatus were the same as in Experiments 1 and 2. Procedure Trials and stimuli in this experiment were similar to Experiment 1, except for a reduction in the total number of melodies to be tested on transposition tests. Training trials and training stimuli were similar in terms of stimulus sets and composition. Pairings (natural sounds), melody order on melody training, and trial sequences varied daily and were different than those of previous experiments. On transposition tests, 12 of the 30 synthetic melodies tested in Exper- iment 1 were retested in this experiment. The 12 melodies were matched in terms of note repetitions to the 12 childhood songs of Experiment 2. Thus, six melodies contained three different notes (plus three repetitions), five melodies contained four different notes, and one melody contained five different notes. These 12 melodies were recorded by the same 10 instru- ments and were tested at the same transpositions as in Experiment 1. The selection of instruments used to play the test melodies varied quasiran- domly. When instrument and octave are considered, these renditions of the melodies had not been previously heard by the monkeys. The number of semitones transposed, direction of transposition, and the order of testing for the first block of the 12 melodies was 11, 40, -32, 9, -19, -9, -19, -13, 32, 28, -23, 10. Other randomized blocks of these same melodies and same transpositions (but different synthetic instruments) were also tested. As in Experiment 2, the reinforcement probability for either re- sponse on test trials was equal to the reinforcement rate from the previous session. The same number of test sessions (15), training trials, and testing trials (2 per session) were conducted. MONKEY OCTAVE GENERALIZATION AND MUSIC PERCEPTION 297 Results The results are shown in Figure 4. As in Experiment 1, training trial accuracy with synthetic melodies was somewhat less accurate (BW = 66.7%, 73.3%; FD = 60.0%, 73.3% for DIFF & SAME, respectively) than with natural sounds (BW = 81.2%, 82.8%; FD = 82.0%, 84.6% for DEFF & SAME, respectively). Once again, the most important result was the lack of general- ization (51.7%) to transposed melodies (60.0% for BW, 43.3% for FD), which was not significantly different from chance perfor- mance, f(l) = 0.19, p > .5, single means test. An ANOVA showed significant effect across training and test- ing conditions, f(5,6) = 8.75,p = .011. Planned comparison tests (using the MSE from the ANOVA) showed that generalization to transposed melodies differed significantly from training perfor- mance with natural sounds, (,(6) = 5.02, 5.37, p > .01, for DIFF & SAME, respectively. Experiment 3 essentially replicated Experiment 1 but with a smaller set of 12 synthetic melodies matched in terms of note repetitions to the childhood songs of Experiment 2. The lack of generalization to the transposed melodies was essentially similar (51.7% vs. 53.4% from Experiment 1) and contrasts with the considerable generalization (76.7%) to transposed childhood songs of Experiment 2. Thus, this experiment establishes that generali- zation to the childhood songs was not due to greater experience on the part of the monkeys. Experiment 4 Experiment 3 retested the synthetic melodies tested in Experi- ment 1 but with the number of melodies, note repetitions, and reinforcement contingencies of Experiment 2. There were, how- ever, three remaining differences other than the melodies them- selves (childhood songs vs. synthetic melodies). A minor differ- ence was the frequency range. The synthetic melodies covered a four-octave range, whereas the childhood songs covered only three octaves, the lower three octaves of the synthetic melodies. Another difference was the larger transposition distance for some synthetic melodies. Three synthetic melodies were transposed more than two octaves and one more than three octaves. A third difference was the partial octave transpositions of the synthetic melodies. Partial octave transpositions change the key (chroma) of the melody. These differences, particularly the latter two, could have influ- enced the lack of generalization to transposed synthetic melodies and thus were tested in Experiment 4. The purpose of Experiment 4 was to test the same synthetic melodies of Experiment 3 but with one octave (12 semitone) transpositions and the same three octave range as used with the childhood songs. Method Subjects and Apparatus The subjects and apparatus were the same as in previous experiments. Procedure Training trials and training stimuli in this experiment were similar to those of previous experiments. But the sequences, environmental-stimulus pairings, and order of presentation varied quasirandomly and were different from previous experiments. As in previous experiments, sessions con- tained 26 training trials and four testing trials, and the experiment was conducted for 15 sessions. Transposition tests were octave-generalization tests (SAME-TUNE DIFF-CCTAVE). The same 12 synthetic melodies tested in Experiment 3 were tested with one-octave transpositions. The 12 synthetic melodies were recorded with the same 10 instruments, over the same three octave range as the childhood songs in Experiment 2. They were tested in randomized blocks of the 12 melodies with direction of transposition, particular octave, and instrument counterbalanced within the limits of the experiment. With instrument and octave taken into account, the renditions tested in this experiment were novel. Melodies on control tests (DIFF-TUNE SAME-FREQ) were also novel renditions. These tests were similar to those of Experiment 2. The two different melodies presented on a trial shared the same frequency range and were recorded with the same instrument. Reward was given for either response on test trials with reward probability equivalent to the accuracy on training trials from the previous session for each subject. TRAINING o u 100 90 80 70 60 40 TESTING . Transposition Generalization • 1 1 (XXJOO 88 8» 1OO 90 80 70 60 40 DIFF SAME NATURAL SOUNDS DIFF-TUNE SAME-TUNE DIFF-FREO SAME-FREQ DIFF-TUNE SAME-FREO SAME-TUNE DIFF-FREO Figure 4. Results of a replication of Experiment 1 (Figure 2). Mean some/diffemit (DBFF) training and testing performance of two monkeys with trials composed of natural/environmental sounds (left) and random-synthetic melodies (right). FREQ = frequency. z o o. 300 WRIGHT, RIVERA, HULSE, SHYAN, AND NEIWORTH themselves. Notes were transposed and tested in Experiment 6 just like the melodies in the previous experiment. If the monkeys can discriminate these pitch changes (and report them accordingly), then such a result would show that pitch changes of melodies were discriminable despite the melodies being judged to be the same. Additionally, such a result would show that individual notes are inadequate for tests of octave generalization. It would show that melodies, and in particular musically strong melodies, are neces- sary for tests of octave generalization. The melody provides the musical context for octave generalization, and the melody is ef- fectively removed when individual notes are tested. The transpo- sition test of melodies (Experiment 5) was replicated in this experiment in a mixed design so that note (pitch) and melody generalizations could be compared within the same session. Method Subjects and Apparatus The subjects and apparatus were the same as in previous experiments. Procedure The procedure was similar to that of Experiment 5, except that transpo- sitions of individual notes were tested in addition to the whole melodies. Notes were tested in the same manner as the melodies. Because of the additional pitch-testing trials, there were no melody-training trials. There were 26 natural-sound training trials, two pitch-testing trials, and two melody-testing trials each session. For the two pitch-testing trials, notes were taken from the childhood songs and tested individually. The particular notes tested were not necessarily notes from the particular melodies tested in the same session. Selection of notes was quasirandom, counterbalanced for the melody from where they originated, position in the six-note melody, instrument, and octave within the limits of the experiment. During sample (center speaker) presentations, notes were repeated to fill the 3-s sample listening time, like the melodies. Each note was presented for 200 ms with 200 ms between presentations. Notes were presented eight times during sample presentations and 5-15 times during test presentations (depending on when the choice response was made). Notes were tested at 0.5, 1.0, 1.5, or 2.0 octave transpositions, as were melodies. The reinforcement probability for either choice response on all test trials was equivalent to the reinforcement rate on the previous session's training trials for each subject. The two melody-testing (transposition) tests were similar to those of Experiment 5. The particular renditions of the childhood songs rested were novel (with respect to instrument and octave). Selection and counterbal- ancing of melodies was similar to the previous experiment. Results and Discussion The results from the experiment are shown in Figure 7. Accu- racy on the natural-sound training trials remained high (BW = 86.0%, 76.3%; FD = 75.8%, 83.7% for DIFF & SAME, respectively). Test results for octave transposed melodies are shown in the right-hand portion of the figure. As in the previous experiment, there was good generalization (77.5%) to 1-octave transpositions (85% for BW, 75% for FD) and even better generalization (85%) to 2-octave transpositions (80% for BW, 90% for FD). Also like Experiment 5, there was no significant generalization (50%) for 0.5-octave transpositions (50% for both monkeys) and no generalization (50%) for 1.5-octave transpositions (45% for BW, 55% for FD). An ANOVA showed a significant effect across training and testing conditions, F(9,10) = 17.6, p < .001. Planned comparison tests (using the ANOVA MSB term) showed no significant differ- ence relative to natural-sound performance for 2-octave generali- zation—f5(10) = 0.9, 0.7, p > .4—or 1-octave generalization, rs(10) = 0.16, 0.0, p > .5, for DIFF & SAME, respectively. By contrast, comparisons showed significant differences relative to natural-sound performance for 1.5-octave generalization— rs(10) = 5.6, 5.4, p < .001—and for 0.5-octave generalization, r,(10) = 5.6, 5.4, p < .001, for DIFF & SAME, respectively. The test results from the pitch test with transposed notes are shown in the middle portion of Figure 7. Results from the pitch test show that with increasing distance (pitch) between the sample and test notes, the percentage of same responses decreased (70%, 55%, 45%, 45% for BW; 80%, 60%, 50%, 45% for FD). This is a monotonically decreasing similarity (increasing discrimination) effect. Comparisons to natural-sound training performances showed no significant difference at 0.5-octave pitch transpositions, rs(10) = 1.1, 0.9, p > .3, for DIFF & SAME, respectively. Other pitch transpositions were significantly (p < .01) different from PITCH TESTING MELODY TESTING DIFF SAME NATURAL SOUNDS 0.5 1.0 1.5 NUMBER OF OCTAVES 1.0 1.5 2.0 NUMBER OF OCTAVES Figure 7. Training performance with natural/environmental sounds (left), testing performance with childhood songs (right), and testing performance for pitch of individual notes taken from the childhood songs (middle). The percent "SAME" response is the degree of generalization to childhood songs or notes transposed 0.5, 1.0, 1.5, and 2.0 octaves. DIFF = different. MONKEY OCTAVE GENERALIZATION AND MUSIC PERCEPTION 301 natural-sound training performances, f.(10) = 4.2, 4.1 for 1.0- octave; <s(10) = 6.1, 5.9 for 1.5-octaves; ts(lO) = 6.5, 6.3 for 2.0- octaves, for DIFF & SAME, respectively. The generalization trends were very different for childhood songs than for the individual notes from these songs. At transpo- sitions of six semitones (i.e., to the tritone), melodies were not judged (50%) to be the same. Paradoxically (relative to individual notes), increasing the distance, for example by doubling frequency, makes childhood songs more similar. Indeed, it makes them so similar that they are judged to be as similar as identical natural sounds. The trend goes through another inflection as the distance is further increased to 1.5 octaves. At 1.5-octave separations (transpositions), the melody is judged to be the same only 50% of the time. There is another recapitulation with a further separation to two octaves. Childhood songs with note frequency differences as great as 1,110 Hz (370 to 1,480 Hz for a two-octave transpo- sition) are judged to be the same melody 85% of the time, even a slightly higher percentage than for one-octave transpositions or identical natural sounds. These results highlight some differences between psychophys- ical discriminability and gestalt perception, and procedures that can influence octave generalization. One conclusion from this experiment is that notes, the subunits of a melody, do not result in octave generalization. This is a conclusion similar to the one based on octave transposition of notes with humans (e.g., Kallman, 1982) and raises additional concerns over studies claiming to show octave generalization to pure tones (e.g., Blackwell & Schlosberg, 1943). From a gestalt standpoint, these same notes when presented as melodies configure to overshadow substantial pitch differences in the transpositions. Most interesting, on the basis of the evidence we have presented, is that not all melodies result in octave gener- alization. Only the childhood songs produced octave generaliza- tion, not the random-note synthetic melodies. It now remains to identify the properties of the melody that makes it a generalizable melody for rhesus monkeys. This is the topic of the last two experiments. Experiment 7 The strong octave generalization for childhood songs of Exper- iments 2, 5, and 6 contrasts with the lack of octave generalization for random synthetic melodies of Experiments 1, 3, and 4. The childhood songs were originally selected because they were thought to be musical and memorable by contrast to the synthetic melodies (cf. Trehub, 1993). Certainly one would not expect generalization of melodies that could not be remembered. Quan- tifying of musical tonality, however, is complicated. Two ap- proaches may be relevant Key Distance One index of tonality is key distance on a scale of the circle of fifths (e.g., Bartlett, 1993). Beginning with the note C, successive fifths (7 semitone steps) produces the series C, G, D, A, E, B, F#, C#, G#, D#, A#, F, which is joined in a circle. Adjacent notes are structurally, musically, and tonally related for infants (Schellen- berg & Trehub, 1996) as well as adults (Bartlett, 1993; Takeuchi & Hulse, 1992). Take, for example, a melody in the key of C major. The tonic, C, forms a point of departure and arrival. The note, G, is a "perfect" fifth (3/2 frequency ratio) above the tonic and is the secondary point of arrival. The perfect fifth above the note of G is D, and so on. Adjacent notes (e.g., C and G) are most strongly tonally related. Notes on opposite sides of the circle (e.g., B and F*) are least related. The childhood songs and synthetic melodies were analyzed for key distance with fewer steps (on the circle of fifths) indicating stronger tonality. The mean interval between notes (on the circle of fifths) was 2.71 steps for childhood songs and 3.06 steps for synthetic melodies. This difference, while not large, is in the same direction as the octave-generalization results. The smallest range (on the circle of fifths) is also related to tonality (Bartlett, 1993; Bartlett & Dowling, 1988). The total range was 5.16 steps for childhood songs and 6.80 steps for synthetic melodies, which is also in the same direction as the octave- generalization results. A more thorough evaluation of these circle- of-fifths measures as they relate to octave generalization is pre- sented in the General Discussion section. Maximum Key Profile Correlation (MKC) The ultimate goal of indexing tonality would seem to be an algorithm whereby musical passages could be varied along a continuum of tonality. At one extreme a melody would be atonal (like the random synthetic melodies) and at the other extreme a melody would be tonal (like the childhood songs). Takeuchi (1994) has supplied just such an algorithm. This algorithm uses human judgments of the stability of individual pitch classes in a tonal context of a particular key (Kramhansl & Kessler, 1982). In the key of C major, for example, the perfect fifth, G, as well as the tonic, C, are judged highly stable and occur frequently. On the other hand, the tritone, F*, is judged unstable in C major and is therefore unlikely to occur. Kramhansl (1990) presents the list of stability ratings for scale values in the key of C major. The Takeuchi MKC index is based on the evidence that the most stable pitch classes (i.e., the tonic, the perfect fifth, etc.) occur most often in the musical passage. The more frequently they occur, the more the key is established, and the more tonal is the musical passage. The MKC algorithm correlates the frequency of the 12 pitch classes (scale values) with the stability of the 24 possible major and minor key profiles. The largest correlation identifies the most likely key and indexes tonality. The mean MKC tonality values for the childhood songs and the random-synthetic melodies of this article was 0.69 and 0.57, re- spectively. This difference in tonality values is in the same direc- tion as the difference in octave generalization for these melodies. The MKC tonality algorithm can also be used to select the melodies to be tested. The purpose of Experiment 7 was to test melodies that were generated using the MKC tonality algorithm so that tonal and atonal melodies would be related on the same continuum. A secondary purpose of Experiment 7 was to evaluate the role of rhythm in octave generalization. Although rhythmic structure has little or no effect on human octave generalization (Massaro et al., 1980), rhythm and rhythm differences were present in the childhood songs. Therefore, the tonal and atonal melodies in Experiment 7 (and Experiment 8) were isochronous with notes and internote intervals of the same duration. 302 WRIGHT, RIVERA, HULSE, SHY AN, AND NETWORTH Method Subjects and Apparatus The subjects and apparatus were the same as in previous experiments. Stimuli There were 24 tonal and 24 atonal melodies. Twelve of each type were used in training and the other 12 in testing. Each melody contained seven notes of 250 ms each. All melodies were recorded with the same keyboard instrument. The atonal melodies were selected so that they contained two or three pitch-contour changes and had MKC tonality values of less than 0.57, resulting in a range of 0.57 to 0.32. Although one cannot say that an atonal melody is truly in any key, the Krumhansl and Kessler (1982) key-finding algorithm does indicate the most likely key. The most likely key of the atonal melodies was identified to make tonal melodies from the atonal melodies. Certain notes were then modified, but the original con- tours of the melodies were preserved. In cases where the first or last notes were whole tones or semitones removed from the tonic, these notes were changed to the tonic. Tritones were changed to the perfect fifth. The resulting tonal melodies had MKC values of greater than 0.75 with a range of 0.75 to 0.93. The melodies were from the three-octave range used previously. Procedure The training procedure was similar to that of previous experiments. There were 26 natural-stimulus training trials and one tonal and one atonal training trial per session. There were 12 different tonal and 12 different atonal melodies used in training. Melodies were presented once as sample from the center speaker and 1-3 times from side speakers, depending on when the choice response occurred. Same and different responses were rewarded according to whether the two stimuli (natural sounds or melo- dies) were identical or different, respectively. There were two test trials each session. There were 12 different atonal and 12 different tonal melodies used in testing. Test melodies were differ- ent in contour from training melodies. Transpositions were one octave. The direction of transposition (to higher or lower frequencies) was different on the two test trials each session. Stimulus pairings and trial sequences varied daily and were counterbalanced within the limits of the experiment. The melodies were tested in two randomized blocks for a total of 24 sessions in the experiment On test trials, either response (same/different) was rewarded with the reward probability equal to the subject's training-trial accuracy from the previous session. Results and Discussion The results are shown in Figure 8. Training-trial performance was slightly better for natural-sound stimuli (BW = 87.8%, 85.5%; FD = 89.1%, 85.9% for DIFF & SAME, respectively) and tonal melodies (BW = 91.7%, 91.7%; FD = 100%, 75% for DIFF & SAME, respectively) than for atonal melodies (BW = 75.0%, 83.0%; FD = 66.7%, 75.0% for DIFF & SAME, respectively). The slightly less accurate performance with atonal melodies may have been due to the atonal melodies being less memorable than natural sounds or tonal melodies. Test results for one-octave transpositions are shown in the right-hand portion of Figure 8. There was good generalization (81.3%) to one-octave transposed tonal melodies (83.3% for BW, 79.2% for FD). There was no significant generalization (47.9%) to one-octave transposed atonal melodies (50% for BW, 45.8% for FD). An ANOVA showed a significant effect across training and testing conditions, F(7, 8) = 12.9, p = .001. Planned comparison tests (using the ANOVA MSB term) showed no significant differ- ence of tonal melodies for 1-octave transpositions relative to natural-sound performance, t,(S) - 1.21,0.76, p > .2, for DIFF & SAME, respectively. Comparison tests showed significant differ- ences for the atonal melodies relative to natural-sound perfor- mance, »s(8) = 7.1,6.6,p < .001, forDIFF& SAME, respectively. Because octave generalization for tonal melodies was equivalent to training-trial performance, it can be concluded that octave shifts of tonal melodies result in melodies that are unchanged in this essential aspect from the original for rhesus monkeys. This good octave generalization rules out rhythm as an important variable in octave generalization with rhesus monkeys, a result similar to a 100 90 80 70 40 TRAINING TESTING DIFF SAME NATURAL SOUNDS DIFF SAME DIFF SAME ATONAL TONAL MELODIES 1.0 1.0 ATONAL TONAL NUMBER OF OCTAVES 100 90 80 70 60 50 40 Figure S. Training performance (left, unfilled histograms) with natural/environmental sounds, atonal melodies, and tonal melodies. Testing performance (right, filled histograms) with atonal and tonal melodies transposed an octave. DIFF = different. MONKEY OCTAVE GENERALIZATION AND MUSIC PERCEPTION 305 Implications for Species Similarity in the Perception of Tonality Such a functional similarity to humans warrants further com- ment. At a first glance, it might seem implausible that rhesus monkeys would respond to tonality in the same way that humans do. The tonal melodies were tonal according to one particular musical scale—the Western diatonic scale. Most of us have had a lifetime of experience with music that differs in tonality on the Western diatonic scale. The monkeys' experience with such music was limited by comparison. Why should monkeys be sensitive to tonality on the basis of a Western diatonic scale when there are many other musical scales on which tonality could vary, for example, the North Indian that or the Balinese slendro or pelog scales? An answer may be provided by research with human listeners who are unfamiliar with these other scales. Krumhansl (1990) summarized a substantial body of literature and came to the conclusion that people are sensitive to tonal hierarchies of unfa- miliar as well as familiar musical systems. For example, Western listeners were able to rate accurately the stability of pitches of North Indian music on the basis of that scale just as well as listeners with a lifetime of familiarity with this music (Castellano, Bharucha, & Krumhansl, 1984). Similarly, Western as well as Balinese listeners were able to rate Balinese gamelan musical passages on the basis of the slendro or pelog tuning systems in a manner that suggested strong tonal hierarchies (Kessler, Hansen, & Shepard, 1984). Thus, although musical scales can vary sub- stantially across cultures there is a commonality in the perception of tonality and tonal hierarchies. This same common perception of tonality appears to be also shared by rhesus monkeys. Thus, the profound conclusion emerges that sensitivity to tonal hierarchies and the perception of tonality is not unique to humans. This trait is shown to be shared by rhesus monkeys and may also be shared by some other species as well (e.g., monkeys, apes, avians, etc.). Furthermore, because sensitivity to tonal hierarchies is a relatively sophisticated musical ability, other principles of musical organization may also be shared. Critical Conditions in Testing Octave Generalization These conclusions regarding an animal's music perception awaited the development of adequate animal testing procedures. Humans can be asked, "Is this the same tune that you heard before?". Somehow we all know what the term "the same tune" means. It is another matter for animals to learn what "the same tune" means. There have been different approaches in training animals to identify "the same tune." One way to classify these approaches is in terms of whether they are relational tasks or absolute-identification tasks. The Problem: Absolute-Identification Tasks Absolute-identification tasks have the advantage in that they are comparatively easy tasks for animals to learn. Typically, one or more "tunes" are designated as S+ (in which responses are re- warded) and others S— (in which responses are not rewarded). These are typically labeled go/no-go tasks where S+ means "go" or respond. Many animal species learn absolute-identification (go/ no-go) tasks in just a few weeks. By contrast, many months or even years are required for animals to learn a relational task such as a same-different task. Thus, for practical reasons absolute- identification tasks are most often used. Unfortunately, absolute-identification tasks have not been suc- cessful in testing melody perception and octave generalization. For example, in one absolute-identification task, monkeys were trained with one positive melody (S+) and one negative melody (S—). What was earlier thought to be octave generalization (D'Amato & Salmon. 1982), was later found to depend on the presence or absence of certain individual notes (D'Amato, 1988; D'Amato & Salmon, 1984). Even an increase from 1 to 3 in the number of S+ and S— tunes did not alter the outcome. To learn the (absolute) identification of S+, the tune must be heard hundreds, if not thousands, of times. Thus, control by the parts (e.g., notes) might override the whole (melody). It is apparent from the results pre- sented in the present article that D'Amato's (1988) conclusion that monkeys cannot perceive tunes (music) was incorrect. The Solution: Relational Tasks and Many Stimuli The experiments of this article are relational tasks. In this case, the same-different task requires that the subject relate two musical passages. On same trials, the two musical passages were composed of the same notes. On octave-generalization trials, where the two musical passages also were judged to be the same, there were no common notes. Thus, the aspect that made these two musical passages sound the same was the melody. Because there were no common notes, it can be concluded that the melody transcended the particular notes used to play it. A critical aspect of the experiments of this article was that the melody to be learned changed from trial to trial. This feature of changing melodies is made possible by relational learning. By continuously changing the melodies, minutia. and features (e.g., notes)—those absolute aspects of the stimuli—are less likely to gain control of behavior than if only a few melodies were used. With a large number of melodies, control is more likely to be established by the relationship between two melodies. Learning this relationship is what is meant by higher-order concept learning. It is higher order because the relationship or concept, same- different concept in this case, transcends the individual stimuli (natural sounds, musical passages) used in training and testing. A Musical Context Another critical aspect of this research was testing octave gen- eralization in a musical context (cf. Kallman. 1982). As previously discussed, notes tested individually do not show octave generali- zation because the melody, or musical context, has been elimi- nated. The melody is the musical context and provides the basis for octave generalization. Some Implications for Nature-Nurture Issues in Music Perception It is not uncommon for music and language development to be considered closely related cognitive skills (e.g., Trehub & Trainor, 1994). Proposals that music and language are closely related tend to reinforce the common belief that music and language set hu- mans apart from other animal species. Debate often moves on to 306 WRIGHT, RIVERA, HULSE, SHYAN, AND NEIWORTH consider whether music abilities are inherited or learned with the implicit assumption that the music abilities being considered are unique to humans (e.g., Attneave & Olson, 1971; Burns & Ward, 1982; Dowling, 1978; Ward, 1970). The results presented in this article question the premise of this debate, and impact both nurture and nature arguments of music perception. Although the monkeys were exposed to music, their exposure was much less than that of most humans. Unlike humans, the monkeys probably never heard any of the childhood songs before being tested with them in these experiments. As to inherited aspects of music perception (the nature part), it is clear that we are not unique in our ability to perceive tunes and identify them irrespective of different notes. Rhesus monkeys apparently transduce, store, process, and relate musical passages much in the same way that we do. The matching of octave-transposed tunes depends on a memory system that stores higher-order representations of tunes. They are higher order because they transcend the original notes. It is the relationship among notes, the gestalt, that forms the essence of the melody or tune. References Attneave, F., & Olson, R. K. (1971). Pitch as a medium: A new approach to psychological scaling. American Journal of Psychology, 84, 147-166. Bartlett, J. C. (1993). Tonal structure of melodies. In T. 1. Tighe & W. J. Dowling (Eds.), Psychology and music (pp. 39-61). Hillsdale, NJ: Erlbaum. Bartlett, J. C, & Dowling, W. J. (1980). The recognition of transposed melodies: A key-distance effect in developmental perspective. Journal of Experimental Psychology: Human Perception & Performance, 6, 501-515. Bartlett, J. C, & Dowling, W. J. (1988). Scale structure and similarity of melodies. Music Perception, 5, 285-314. Blackwell, H. R., & Schlosberg, H. (1943). Octave generalization, pitch discrimination, and loudness thresholds in the white rat Journal of Experimental Psychology, 33, 407-419. Burns, E. M., & Ward, W. D. (1982). Intervals, scales, and tuning. In D. Deutsch (Ed.), The psychology of music (pp. 241-270). New York: Academic Press. Castellano, M. A., Bharucha, J. J., & Krnmhansl, C. L. (1984). Tonal hierarchies in the music of North India. Journal of Experimental Psy- chology: General, 113, 394-412. Cohen, A. J. (1982). Exploring the sensitivity to structure in music. Canadian University Music Review, 3, 15-30. Cuddy, L. L. (1993). Melody comprehension and tonal structure. In T. J. Tighe & W. J. Dowling (Eds.), Psychology and music (pp. 19-38). Hillsdale, NJ: Erlbaum. Cynx, J. (1993). Auditory frequency generalization and a failure to find octave generalization in a songbird, the European starling (Stumus vulgaris). Journal of Comparative Psychology, 107, 140-146. D'Amato, M. R. (1988). A search for tonal pattern perception in Cebus monkeys: Why monkeys can't hum a tune. Music Perception, S, 453- 480. D'Amato, M. R., & Salmon, D. P. (1982). Tune discrimination in monkeys (Cebus apella) and in rats. Animal Learning & Behavior, 10, 126-134. D'Amato, M. R., & Salmon, D. P. (1984). Processing of complex auditory stimuli (tunes) by rats and monkeys (Cebus apella). Animal Learning & Behavior, 12. 184-194. Demany, L., & Armand, F. (1984). The perceptional reality of tone chroma in early infancy. Journal of the Acoustical Society of America, 76, 57-66. Deutsch, D. (1972). Octave generalization and tune recognition. Percep- tion & Psychophysics, 11, 411-412. Dooling, R. i.. Brown, S. D., Park, T. J., Okanoya, D., &. Soli, S. D. (1987). Perceptual organization of acoustic stimuli by budgerigars (Melopsitta- cus undulatus). Journal of Comparative Psychology, 101, 139-149. Dowling, W. i, (1978). Scale and contour Two components of a theory of memory for melodies. Psychological Review, 85, 341-354. Dowling, W. J. (1988). Tonal structure and children's early learning of music. In J. Sloboda (Ed.), Generative processes in music (pp. 113-128). Oxford, England: Clarendon. Dowling, W. I. (1991). Tonal strength and melody recognition after short and long delays. Perception & Psychophysics, 50, 305-313. Dowling, W. J., & Harwood, D. L. (1986). Music cognition. New York: Academic Press. Dowling, W. I., & Hollombe, A. W. (1977). The perception of melodies distorted by splitting into several octaves: Effects of increasing proxim- ity and melodic contour. Perception & Psychophysics, 21, 60-64. Frances, R. (1958). Laperception de la musigue [The perception of music]. Paris: J. Vrin. Gillan, D. J., Premack, D., & Woodruff, G. (1981). Reasoning in the chimpanzee: I. Analogical reasoning. Journal of Experimental Psychol- ogy: Animal Behavior Processes, 7, 1-17. Hays, W. L. (1963). Statistics for psychologists. New York: Holt, Rinehart, & Winston. Homa, D., Cross, J., Cornell, D., Goldman, D., & Shwartz, S. (1973). Prototype abstraction and classification of new instances as a function of number of instances defining the prototype. Journal of Experimental Psychology, 101, 116-122. Homa, D., & Vosburgh, R. (1976). Category breadth and the abstraction of prototypical information. Journal of Experimental Psychology: Human Learning & Memory, 2, 322-330. Hulse, S. H., & Cynx, J. (1985). Relative pitch perception is constrained by absolute pitch in songbirds (Mimus, Molothus, and Stumus). Journal of Comparative Psychology, 99, 176-196. Hulse, S. H., & Cynx, J. (1986). Interval and contour in serial pitch perception by a passerine bird, the European starling (Stumus vulgaris). Journal of Comparative Psychology, 100, 215-228. Idson, W. L., & Massaro, D. W. (1978). A bidimensional model of pitch in the recognition of melodies. Perception & Psychophysics, 24, 551-565. Kallman, H. J. (1982). Octave equivalence as measured by similarity ratings. Perception & Psychophysics, 32, 37-49. Kallman, H. J., & Massaro, D. W. (1979). Tone chroma is functional in melody recognition. Perception & Psychophysics, 26, 32-36. Kessler, E. J., Hansen, C., & Shepard, R. N. (1984). Tonal schemata in the perception of music in Bali and the West. Music Perception, 2,131-165. Krumhansl, C. L. (1979). The psychological representation of musical pitch in a tonal context. Cognitive Psychology, 11, 346-374. Krumhansl, C. L. (1990). Cognitive foundations of musical pitch. New York: Oxford University Press. Krumhansl, C. L., Bharucha, J., & Castellano, M. A. (1982). Key distance effects on perceived harmonic structure in music. Perception & Psycho- physics, 32, 96-108. Krumhansl, C. L., & Kessler, E. J. (1982). Tracing the dynamic changes in perceived tonal organization in spatial representation of musical keys. Psychological Review, 89, 334-368. Massaro, D. W., Kallmaji, H. J., & Kelly, J. L. (1980). The role of tone height, melodic contour, and tone chroma in melody recognition. Jour- nal of Experimental Psychology: Human Learning and Memory, 6, 77-90. Pick, A. D., & Palmer, C. F. (1993). Development of the perception of musical events. In T. J. Tighe & W. J. Dowling (Eds.), Psychology and music (pp. 197-213). Hillsdale, NJ: Erlbaum. Premack, D. (1983a). Animal cognition. Annual Review of Psychology, 34, 351-362. Premack, D. (1983b). The codes of man and beasts. The Behavioral and Brain Sciences, 6, 125-167. MONKEY OCTAVE GENERALIZATION AND MUSIC PERCEPTION 307 Ralston, J. V., Herman, L. M, Williams, N. S., Gory, J. D., & Jerger, K. K. (1988). Melody recognition by an Atlantic bottlenosed dolphin. Journal of the Acoustical Society of America (Suppl.), 77. Richards, D. G., Wolz, J. P., & Herman, L. M. (1984). Vocal mimicry of computer-generated sounds and vocal labeling of objects by a bottle- nosed dolphin, Tursiops truncatus. Journal of Comparative Psychol- ogy, 98, 10-28. Schellenberg, E. G., & Trehub, S. E. (1996). Natural musical intervals: Evidence from infant listeners. Psychological Science, 7, 272-277. Shepard, R. N. (1982). Structural representations of musical pitch. In D. Deutsch (Ed.), The psychology of music (pp. 344-390). New York: Academic Press. Sokal, R. R., & Rohlf, p. J. (1969). Biometry. San Francisco: Freeman. Stebbins, W. C. (1973). Hearing of old world monkeys. American Journal of Physical Anthropology, 38, 357-364. Stevens, S. S., & Volkmann, J. (1940). The relation of pitch to frequency: A revised scale. American Journal of Psychology, 53, 329-353. Takeuchi, A. H. (1994). Maximum key-profile correlation (MKC) as a measure of tonal structure in music. Perception & Psychophysics, 56, 335-346. Takeuchi, A. H., & Hulse, S. H. (1992). Key-distance effects in melody recognition reexamined. Music Perception, 10, 1-24. Thompson, R. K. R. (1980). Auditory cued reversal and matching-to- sample learning by rhesus monkeys. Antropologia Contemporanea, 3, 284-292. Thompson, W. P., & Cuddy, L. L. (1992). Perceived key movement in four-voice harmony and single voices. Music Perception, 9, 427-438. Trehub, S. E. (1993). The music listening skills of infants and young children. In T. J. Tighe & W. J. Dowling (Eds.), Psychology and music (pp. 161-176). Hillsdale, NJ: Erlbaum. Trehub, S. E., Cohen, A. J., Thorpe, L. A., & Morrongiello, B. A. (1986). Development of the perception of musical relations: Semitone and diatonic structure. Journal of Experimental Psychology: Human Percep- tion and Performance, 12, 295-301. Trehub, S. E., Morrongiello, B. A., & Thorpe, L. A. (1985). Children's perception of familiar melodies: The role of intervals, contour, and key. Psychomusicology, 5, 39-48. Trehub, S. E., Thorpe, L. A., & Trainor, L. J. (1990). Infants' perception of good and bad melodies. Psychomusicology, 9, 5-15. Trehub, S. E., & Trainor, L. J. (1994). Listening strategies in infancy: The roots of language and musical development. In S. McAdams & E. Bigand (Eds.), Cognitive aspects of human audition (pp. 278-327). London: Oxford University Press, Unyk, A. M., Trehub, S. E., Trainor, L. J., & Schellenberg, E. G. (1992). Lullabies and simplicity: A cross-cultural perspective. Psychology of Musk, 20, 15-28. Ward, W. D. (1970). Musical perception. In J. V. Tobias (Ed.), Founda- tions of modem auditory theory (Vol. 1, pp. 405-447). New York: Academic Press. Wright, A. A., Cook, R. G., & Kendrick. D. F. (1989). Relational and absolute stimulus teaming by monkeys in a memory task. Journal of the Experimental Analysis of Behavior, 52, 237-248. Wright, A. A., Cook, R. G., Rivera, J. J., Sands, S. F., & Delius, 1. D. (1988). Concept learning by pigeons: Matching to sample with trial- unique video picture stimuli. Animal Learning and Behavior, 16, 436- 444. Wright, A. A., Santiago, H. C, Sands, S. F., & Urcuion, P. J. (1984). Pigeon and monkey serial probe recognition: Acquisition strategies and serial position effects. In H. L. Roitblat, T. Sever, & H. S. Terrace (Eds.), Animal cognition (pp. 353-374). Hillsdale, NJ: Erlbaum. Wright, A. A., Shyan, M. R.. & Jitsumori, M. (1990). Auditory samel different concept learning by monkeys. Animal Learning & Behav- ior, 18, 287-294. Received July 30, 1998 Revision received July 6, 1999 Accepted July 7, 1999 Correction to Friedman and Brown (2000) The article "Reasoning About Geography," by Alinda Friedman and Norman R. Brown (Journal of Experimental Psychology: General, 2000, Vol. 129, No. 2, pp. 193-219), contained an error. On page 211, the Table 6 note incorrectly reads "V(270) for all first estimates; /(90) for second estimates with Luna as the seed fact; 1(81) for second estimates with Rio as the seed fact." The correct number of degrees of freedom for all tests is 29.