Anda di halaman 1dari 17

Journal of Experimental Psychology: Learning. Memory, and Cognition IWS.Vol. 21. Nn.'3, 7.

17-751

Copyrisht 1995 by the American Psychological Association. Inc. 0278-7393/95/S3.00

Perceiving Implied Harmony: The Influence of Melodic and Harmonic Context


Susan Holleran, Mari Riess Jones, and David Butler
Ohio State University Two experiments investigated the influence of melodic and harmonic context on the obviousness of pitch changes in 3 Western tonal melodies. In Experiment 1, listeners with different levels of musical skill heard standard melodies followed by comparison melodies in which single pitch changes could occur. In-key pitch changes that were not implied by the local harmony were judged more obvious than in-key pitch changes that were so implied. This effect was influenced by the surrounding context of melodic line and chords and was absent when the context contained chords that were not consistent with the harmony implied by the melodic line. In Experiment 2, with similar listeners, implied harmony effects were eliminated by removing parts of the melodic line, indicating the relative importance of melodic context as a determinant of implied harmony. Questionnaire data suggested that knowledge of tonal relationships was tacit, even in many trained listeners.

The influence of context on linguistic inference and its role in resolving phonological, syntactic, and lexical ambiguity have been extensively explored in research on language comprehension. Although the mechanics and time course of context effects remain controversial, their existence and strength are well established (see, for example, Gorfein, 1989; Graesser & Bower, 1990). Less is known about the influence of context on musical inference. This is surprising because, at any level of analysis (e.g., note, chord, and key), the interpretation of a musical event seems eminently context dependent. For example, in music the individual sound unit (which might parallel a speech phoneme), namely a tone, takes on different meanings in different contexts. An elementary example of a musical context, one often termed a tonal context, is found in the ordered subset of pitches known as a musical scale.1 Typically, the first note of a scale is its keynote, meaning that it functions for a listener as a cognitive anchor with respect to other pitches. Among other things, identification of a keynote defines for the listener a particular tonal context (or key). For example, the familiar C major scale has ordered pitches C, D, E, F, G, A, B, C , with a keynote pitch of C. The context dependency of music quickly becomes apparent, however, with another common scale, the F major scale (F, G, A, Bb, C, D, E,

F'). Notice that in this context, the cognitive anchor is now the pitch F. Further, in this tonal context, C becomes the fifth scale member, meaning that its function has been transformed from a keynote to a dominant pitch (i.e., the fifth scale note of F major). In these two tonal contexts, the pitches F and C change their meanings. This can also be seen in special three-note groups drawn from the two scales, for example, C-E-G and F-A-C. Musically, such groups correspond to a level of analysis analogous to morphemic or word units in language. When such pitches are sounded together, as a simultaneous group, they form harmonic relationships known as chords (see Footnote 1). Again the issue of context arises in interpreting
1 A fundamental musical scale is the equal-tempered chromatic set, which consists of twelve pitches spaced in equal logarithmic steps (semitones, for example, adjacent notes on the piano keyboard) bounded by a 2:1 (or octave) frequency ratio. Diatonic scales can be conceived as seven-pitch subsets of the chromatic scale in which distances between successive pitches are prespecified by patterns of semitones (s; half steps) and whole steps (w = 2 s). Major diatonic scales are named according to their first scale degree, and subsequent scale degrees are defined by the pattern w w s w w w s. Scale degrees are given special names reflective of their importance in the scale: for example, thefirstscale degree is the tonic, and the fifth, the dominant. In the text, we refer to the tonic scale degree as the keynote. The terms tonic and dominant are also assigned to chords built on thefirstand fifth scale members (degrees), respectively. As defined in this study, explicit chords are groups of tones that are separated in pitch by prescribed scale intervals and played simultaneously, implicit chords are sequences of tones that are heard as perceptually grouped. Even though a scale may contain notes that are not found in a given musical passage and, conversely, a tonal passage may contain notes not found in a given musical scale, musical passages in which pitches largely conform to a particular scale and follow certain conventions (to be discussed later) are commonly said to be in the key with the same name as that scale. Thus, a tonal passage is one in which a single pitch functions as a psychological reference frame for other pitches within both that scale set and related scales. The system of tonal harmony aims to describe the interactions of simultaneous and successive pitches (chords and melody) within this framework.

Susan Holleran and Mari Riess Jones, Department of Psychology, Ohio State University; David Butler, School of Music, Ohio State University. This results of Experiment 1 were reported at the November 1991 meeting of the Psychonomic Society in San Francisco. This research was supported in part by a grant from the Center for Cognitive Science of Ohio State University to David Butler and Mari Riess Jones. We are indebted to Angie Fankhauser, Lisa Schmakel, Rocki Strader, Elizabeth Zinober, Mark Peter, and Leo Schmidt for data collection and also to Caroline Palmer for comments on a version of this article. Correspondence concerning this article should be addressed to Mari Riess Jones, Department of Psychology, Ohio State University, 1827 Neil Avenue, Columbus, Ohio 43210. Electronic mail may be sent via Internet to mrjones@magnus.acs.ohio-state.edu.
737

738

S. HOLLERAN. M. JONES. AND D. Bl'Tl.ER

these harmonic units. Although both of the chords C-E-G and F-A-C appear in C and F major scales, they mean different things in the two tonal contexts. When the C of the C-E-G chord is identified as a keynote in the C major scale, then C-E-G is a I chord.2 This chord thus implies that C is its cognitive anchor. However, when the pitch C is the fifth note of the F major scale, then C-E-G becomes a V chord. In this context, F functions as a cognitive anchor. In short, context dependency is evident at several levels of musical analysis. Meaning conventionally arises from inferences an individual makes about something to effectively place it within a larger domain. Musical meaning, as we use it here, refers to inferences listeners make about events, such as single tones or chords, that place them within a larger musical context. Such inferences are possible because these events consist of units at one level of analysis that carry implications about specific referents at another level: for example, what key (if any) is implied by a particular chord (e.g., C-E-G), or what chord (if any) is implied by a given tone (e.g., C). As previously discussed, a tone or chord by itself is ambiguous and can participate in several different chords or keys; its meaning must derive from aspects of its context. However, in actual music, this context seldom consists solely of sequences of chords, or of tones arranged in musical scales. Rather, musical context is dynamic and relational, comprising patterned sequences of melodic pitches and chords that permit listeners to anticipate certain forthcoming chords and pitches (Butler & Brown, 1984; Jones, 1981, 1990; Lerdahl, 1988; Narmour, 1990, 1992). In Western tonal music, various chord patterns realize changing harmonies, with each chord constituting a local harmonic unit. Some chord sequences are common, such as I-V-I, whereas others are quite uncommon, such as V-FV-IH (Piston, 1987). Listeners generate harmonic expectancies conforming to the more common chord patterns (Schmuckler, 1989). Together with the musical conventions outlined above which relate pitches of individual tones to chords, and chords to keys, the principles governing the patterning of such harmonic relationships can be considered a tonal "grammar." Applying such a grammar to a simple C major melody reveals how its dynamic context permits meaningful anticipations of forthcoming elements. Imagine a tune that opens with a three-note melodic sequence, say C-E-G; recall that these pitches form the I chord in the key of C major. If this melody continues, such that the next three pitches are drawn from the set B, D, G (from the V chord), then knowledgeable listeners might expect the following ones to be derived from the I chord (C, E, G). This is because the example melody is based on a succession of pitches that is consistent with the common chord sequence I-V-I. Thus, if listeners with knowledge of the tonal grammar can infer from the melody one (or more) of these initial chords, then they may also be able to infer that subsequent tones are consistent with this chord sequence, in spite of the fact that such pitches, taken out of context, logically imply several alternative chords or keys. There is some evidence that listeners' expectancies about underlying chords do influence their judgments about the strength of a sense of key in melodies (e.g., Boltz, 1989a, 1989b, 1991; Brown, 1988; Brown & Butler, 1981; Butler, 1989; Butler & Brown, 1984). However, this example assumes that listeners

knowledgeable about musical convention1; should actually he able to track the pattern (if harmonic relationships implied by a melody, inferring various chords at different points within it. On this topic there is no current evidence. Our research considered whether people could infer the appropriate chord from pitch elements of an unfolding melody. We had three goals. The first was to assess whether listeners could make reliable harmonic inferences on the basis of individual tones within a changing tonal melody. Our second goal was to examine contextual properties, both harmonic and melodic, that might affect the strength of these inferences, if they exist. The final goal was to determine how listeners with different levels of musical knowledge differed in their inferencemaking abilities. To achieve the first goal we used typical Western tonal melodies to assess listeners' sensitivity to melodic changes that conform to or conflict with commonplace local harmonic implications. Local harmonic implications are those generated by a small group of neighboring tones within the overall melody; here, we call such groups measures, with each measure comprising three or four tones and suggesting a chord consistent with its tones. We can distinguish two forms of local harmony: explicit and implicit. Explicit local harmony refers to the case in which a chord is actually sounded as an accompaniment to the melody in that measure. Implicit local harmony refers to the case in which a chord is not sounded, but the melodic pitches within that measure imply it; that is. if those pitches had occurred together, they would form that chord. Some theorists contend that listeners routinely make local harmonic inferences, even when the melodic line does not contain all members of an implied chord (e.g., Dahlhaus, 1965-1966/1990; Piston, 1987: Schenker, 1979). Others have questioned this time-honored opinion (e.g., Povel & van Egmond, 1993). We used a standard-comparison task to determine listeners' sensitivity to changes in local implied harmony. Listeners heard two patterns, a standard and a comparison, that were similar in that they contained the same basic melody and explicit chords. In fact, the comparison pattern differed from its standard only with respect to the pitch of one tone. This pitch change was always an in-key pitch change. The catch is that the new pitch either conformed to or conflicted with the melody's local (implied) harmony in that measure. When it conformed to the harmonic implications of other pitches in that measure, the pitch change was an implied (I) one; when it did not fit with these harmonic implications, the pitch change was nonimplied (NI). If listeners monitor patterns of changing harmonies (implicit or explicit) that occur in the surrounding context, then they should judge comparison patterns with NI pitch changes to differ more from their standards than comparisons with I pitch changes. The reason for this is that if listeners infer the chord implied at this point in the melody, then the NI pitch change will "stand out" as a violation of their expectancies. To achieve our second goal, we considered aspects of the larger context that influence listeners' abilities to monitor

Symbols for musical chords appear in bold to distinguish them from abbreviations in the text.

IMPLIED HARMONY changing harmonic information. We assumed that tracking a melody is affected by its overall tonal clarity, namely the degree to which it follows conventions (described above) that specify chord sequences within an established key. When a tonal melody is accompanied by explicit chords that reinforce its key, this should provide a clear tonal context; by contrast, tonal clarity is reduced if the melody is accompanied by chords that do not reinforce the key. We also considered the possibility that a melodic line, taken by itself, can convey important harmonic information, and hence tonal clarity should be greater when a melody is intact than when it is not. A number of current approaches address the influence of tonal context on musical inferences at various levels (e.g., tone, chord, and key levels; Krumhansl, 1990; Krumhansl & Kessler, 1982; Shepard, 1982). Krumhansl has maintained that a "tonal hierarchy" scheme summarizes acquired musical knowledge. With this knowledge, listeners can make inferences about a given pitch and its relative stability within a contextually established key level (Krumhansl, 1990; Krumhansl & Kessler, 1982). Others have relied on associative networks and patterns of network activations to model different levels of pitch relationships that figure in harmonic implications (e.g., Bharucha, 1987, 1991; Gjerdingen, 1990, 1992). Thus, Bharucha proposed a tonal network scheme in which tones activate the lowest level units, which are tuned to pitch equivalence classes. Harmonic inferences about specific chords come about when lower level units (e.g., single pitches) activate associative links with higher level chord units. These units are, in turn, linked hierarchically to still higher key levels in a way that also permits downward activations from a key-based tonal scheme. All of these views suggest that a tonal scheme, which links units across hierarchical levels, underlies tonal clarity; thus, tonal clarity should be greater when melody and chords are linked by the same key. Further, a tonal scheme should more readily assimilate to itself those elements that are strongly linked across levels (i.e., implied) than ones that are weakly linked (nonimplied). Supporting evidence for these views comes largely from research that addresses inferences at the level of key (e.g., Krumhansl, 1990; Krumhansl & Kessler, 1982) and chord (e.g., Bharucha & Stoeckig, 1986; Tekman & Bharucha, 1992). In addition, findings that out-of-scale pitch changes are more noticeable than in-scale pitch changes in musical contexts (e.g., Cuddy, Cohen, & Miller, 1979; Dowling, 1978; Frances, 1958/1988) are explained in terms of scheme assimilation at the level of key. Nevertheless, Povel and van Egmond (1993) disputed the tonal scheme hypothesis, and the idea of harmonic inference. They presented to their listeners standard-comparison pairs of nine-tone melodies; in some trials, only the melodic tones were played, whereas in others, each of the nine melodic tones was played with an accompanying chord. Each of the chords contained the melodic tone that it accompanied, but some of the sequences of chords used as accompaniment were arbitrarily chosen and did not consistently follow the conventions of the tonal grammar. Listeners were asked to detect pitch changes in the melodies of the comparison patterns, and their performance was not affected by the type of chordal accompaniment used. Accordingly, Povel and van Egmond proposed an independence hypothesis, which asserts that melodies are coded

739

strictly in terms of steps along a musical scale in a given key. and independently of explicit or implicit chords. Thus, the independence hypothesis casts doubt on the existence of local harmonic inference in any tonal context. Some findings of Platt and Racine (1994) are relevant to the independence hypothesis. They used nine-tone melodies in which the first eight tones all derived from the same three-note chord; thus, local harmonic implications of successive pitch groups remained constant across these eight pitches. Listeners were to decide if a ninth (target) pitch resulted in a change in harmony. When the target pitch was a member of a different chord, most listeners were able to determine that a change in harmony had occurred. Thus, these data are contrary to expectations that are based on an independence hypothesis because they suggest that in certain melodic contexts listeners are sensitive to harmonic implications. However, because a given chord can occur in multiple keys, these melodies may not have been heard as occurring in a specific key, a prerequisite for the independence hypothesis. Our design used explicitly tonal melodies (i.e., in a specific key). In addition, unlike the design of Povel and van Egmond (1993). in our design these melodies were built on conventional, not arbitrary, implied chord progressions. We therefore not only provide a test of the independence hypothesis but also assess whether listeners can draw timely harmonic inferences from a tonal melody with lawfully changing harmonic implications. Our third goal involved comparisons of the performance of listeners with different levels of musical training. Most theories imply that sensitivity to implicit chords should increase with training; therefore, it seems reasonable that musicians will show greater sensitivity to pitch changes (including harmonically implied ones) than nonmusicians. In fact, the literature is not unequivocal on this point. In some cases, less musically skilled participants produced patterns of discrimination similar to those of musicians, although performance was poorer and more variable (e.g., Cohen, Trehub, & Thorpe, 1989; DeWitt & Samuel, 1990; Monahan, Kendall, & Carterette, 1987; Smith & Cuddy, 1989). In other tasks, level of musical training did not significantly affect performance (Cuddy et al., 1979). Where differences did emerge, the favored interpretation assumes that musicians possess better developed tonal schemes (Monahan et al., 1987) or more efficient higher level encodings (Cohen et al., 1989). In these cases, such interpretations raise questions about the role of acquired tonal schemes in harmonic inference, as the recent work of Trainor and Trehub (1992) illustrated. They compared detection of diatonic (in-scale) versus nondiatonic (out-of-scale) pitch changes in brief melodies by using both infant listeners and adults varying in musical skill. Diatonic changes in a comparison melody formed a greater pitch interval from the standard than did the nondiatonic changes, yet adult listeners had greater difficulty detecting the in-scale changes; 8-month-old infants tested on the same patterns detected both change types equally well. (In fact, in conditions in which both groups were given training with nondiatonic changes, adults were actually worse than infants at detecting diatonic pitch changes.) Overall, there was no relationship between adults' performance and their level of musical skill. Trainor and Trehub suggested that all adults had more difficulty discriminating diatonic changes

740

S. HOLLERAN. M. JONES. AND D. BUTLER The type of pitch change was manipulated relative to the melodic structure of standard and comparison patterns. All melodies followed conventional tonal grammar rules in that common chord arrangements dictated the harmony implied by pitches within successive measures. For example, a standard melody in A major is shown in Figure 1 (top); it opens with three pitches in the first measure (C#5, E5, and A4, where the numeral indicates the octave; C4 is middle C) that imply a I chord. Over the eight measures of the melody, successive measures contain similar groups of pitches, each implying a plausible (missing) chord. The key and the chord sequence specify "what" chords should occur (i.e., those in A major) and "when" they should occur (given a common chordal sequence). In the comparison versions of this melody, a single pitch change can occur in one of several test measures. In this example, the third test measure illustrates the target change manipulation. It comprises three pitches (B4, E5, and G#4), with G#4 as the to-be-altered target. Together these pitches imply the V chord in A major. Given this, one of three types of target change can occur (Figure 1, top). A target can be the same (S) pitch as in the standard, or a new pitch that reinforces the harmony of the V chord, an I change, or a new pitch that conflicts with that harmony, an NI change. With the I change, G#4 becomes E4, another pitch in the V chord, and with the NI change, it becomes F#4, a pitch that is not in the V chord. Note that both F#4 and E4 are in the key of A major. However, the I pitch change results in a comparison pattern that preserves the local harmony of its standard, whereas the NI change does not. If listeners perceive a melody in terms of its harmonic implications, then NI changes should seem more obvious than I changes; furthermore, the extent to which the obviousness of the NI changes exceeds that of the I changes should reflect the strength of harmonic implications. Conversely, if listeners perceive the melodic line independently of any chordal implications, then they may respond only to differences in the pitch intervals created by the target changes. In this case, because NI changes always result in a smaller change in pitch interval from the standard than do I changes, NI changes may seem less obvious than NI changes. Two aspects of context are considered to contribute to tonal clarity: Melodic context, which consists of all melodic pitches forming a continuous melodic line (excluding the test measure), and harmonic context, which refers to explicit chords that surround the test measure. Figure 1 shows the melodic line (top) and the same melody in three harmonic contexts (bottom); in this figure, melodic context does not vary. In the first context (control), the melody has no chordal accompaniment; in the other two contexts (which we refer to as consistent and nonconsistent), explicit chords are present. In the consistent chord condition, the chords that accompany the melody are in A major, the key of the melody, and they support its changing local harmonies: Each chord is implied by the pitches appearing within that measure of the melody, and in most cases actually consists of those same melodic pitches. Notice that chords are never present in test measures; harmonies in these measures are always implicit. In the nonconsistent chord condition, the accompanying chords realize the harmony of a different key, G major; in this condition, many pitches in these chords do not correspond to those in the melody. However, in

because these "fit" readily into tonal schemata acquired through exposure to Western musical structure. Because infants lack such schemes, they showed no difference in sensitivity to diatonic and nondiatonic pitch changes. These studies raise the following question: What role does musical training actually play in the ability of skilled listeners to detect certain tonally consistent implications? If the increased exposure to Western tonal structures afforded by musical training results in better developed tonal schemes, then in the present research high-skill participants may actually find the harmonically implied changes less obvious than low-skill listeners. Alternatively, if both groups of listeners have acquired these tonal schemes (low-skill listeners presumably by listening to Western music), then they should perform similarly in judging I versus NI changes. Another facet of acquired knowledge concerns whether it is tacit knowledge. If listeners' performance is shaped, in part, by principles of tonal grammar, are tonal inferences verbalizable? Most of us who enjoy music cannot articulate its principles. However, research on tacit knowing has not addressed the musical domain; instead it typically relies on categorization or well-formedness (grammaticality) judgments of artificially generated serial patterns (e.g., Reber, 1989; Reber, Kassin, Lewis, & Cantor, 1980; Reber, Walkenfeld, & Hernstadt, 1991). The grammars of choice have been Markovian finite state grammars, in part because of their potential for generating complex sequences (for a discussion see Jones, 1974). The usual finding is that people's categorizations reveal abstract knowledge of these probabilistic grammars, with postsession questionnaires indicating that they cannot articulate this knowledge. Reber maintained that implicit knowledge of complex rule systems is acquired "in the absence of conscious, reflective strategies to learn" (Reber, 1989, p. 219). The rules of Western tonal harmony are not Markovian, but they must be at least as complex because they incorporate elements of uncertainty while remaining basically relational in nature. As such they should offer a basis for implicit knowledge. The typical music listener presumably has not engaged in a conscious and reflective strategy to learn these tonal grammars. Therefore, if average listeners have acquired a tacit knowledge of the tonal grammars through general enculturation, then they should be able to reliably detect differences between changes that are harmonically implied and those that are not implied; at the same time, they should be unable to describe the rules they use. On the other hand, it is possible that listeners with more training will have greater ability to articulate grammatical rules, but they will not necessarily have enhanced ability to detect harmonically consistent pitch changes. Accordingly, we questioned listeners in both skill groups about the properties of the pitch changes they heard. Experimental Rationale In these experiments we used a standard-comparison task in which an in-key pitch change could occur somewhere in a comparison pattern, with this change representing either a shift in local harmony or no such shift. Two primary manipulations corresponded respectively to the type of pitch change and to the context surrounding a pitch change.

IMPLIED HARMONY Test Measure 1 Test Measure 2 Test Measure 3

741

Implied Chords (A Major):

VI

V7

TYPES OF TARGET CHANGE: Same: No Target Change (S) I f' (j(J)li G Implied Target Change (I) | P' [? A |

CONTEXT CONDITIONS:

Nonlmplied Target Change (NI)

Ft

gnr Li
Consistent Chords

(A Major)( I NonConsistent Chords i*

I i
i i melody and its chords, should provide greatest tonal clarity. For example, in this condition both melodic and chord context unambiguously specify the key of A, meaning that a major chord, such as E - G # - B , will function in a test measure as a V chord and not, say, a I chord. This reduction in tonal ambiguity suggests that in such contexts pitch changes that do not fit (NI targets) within this local harmony should be much more obvious than ones that do fit (I targets), relative to contexts with greater tonal ambiguity. The nonconsistent chord conditions supply greater ambiguity about key and hence about the tonal function of a given chord; accordingly, if tonal clarity influences harmonic inferences, then differences between NI and I targets should be smallest in this condition. Finally, the control condition, which presents an unaccompanied but continuous melodic line, is interesting because it implies the

(0 Mjor)(

Figure 1. A standard melody along with three target changes (S, I, and NI) is shown. Also shown are control, consistent chord, and nonconsistent chord contexts. Asterisk denotes an inverted chord.

both cases, within their respective keys the consistent and nonconsistent chords follow identical chord sequences typical of the Western tonal grammar (Piston, 1987). Furthermore, target pitch changes (I and NI), when they occur, are always pitches that are in the key of the melodic line (A major) and also in the key of the nonconsistent chords (G major). This design permits assessment of listeners' use of context to disambiguate a melody's local harmonic implications. Recall that the independence hypothesis predicts that listeners use neither explicit nor implicit chords to do this. By contrast, we assume that tonal clarity provided by these contexts should systematically affect the strength of harmonic implications. Tonal clarity is greatest in conditions predicted to be least tonally ambiguous. Thus, the consistent chord conditions, in which key-consistent information is provided jointly by a

742

S. HOLLERAN. M. JONES. AND D. BUTLER

same information about key and chord progression that is explicit in the consistent chord condition. This condition permits an evaluation of the claim that a melodic line conforming to the rules of the tonal grammar is alone sufficient to imply its "missing" chords. In two experiments we considered the relative influence of harmonic and melodic context on listeners' tendencies to make harmonic inferences. In both studies we sampled from populations of listeners with different degrees of musical training. In the first experiment, tonally coherent melodies contained in-key pitch changes that were congruent with harmonic implications of the melodic line (S or I) or that conflicted with these implications (NI). Different groups of listeners were asked to judge the obviousness of such changes in three respective chord contexts. In the consistent chord context, melodies were accompanied by chords that reinforced the melodic line; in the nonconsistent chord context, accompanying chords did not reinforce the melodic line; and in a control condition, no accompanying chords were presented. In this experiment, we anticipated that listeners would find NI changes most obvious, especially in the consistent chord context. In the second experiment, the coherence of the melodic line was disrupted by removing some parts of it that did not contain the relevant pitch changes. In all other respects the experimental design paralleled that of Experiment 1. In Experiment 2, we anticipated that if listeners rely more heavily on surrounding chords to make harmonic inferences than on a continuous melodic line they would perform as in Experiment 1. In both experiments, if musical training desensitizes listeners to changes that fit a tonal scheme, then listeners with more musical training should find the I changes less obvious than would listeners with less training. Experiment 1 In the first experiment we presented musical patterns that differed in harmonic context to different groups of listeners. Some listeners (controls) heard standard-comparison pairs that were strictly based on monophonic (unaccompanied) melodies; others heard the same melodic line accompanied either by consistent chords or by nonconsistent chords (see Figure 1). Explicit chords never appeared in potential test measures, which were always Measures 2, 4, and 6. Both the test measures of a given melody and its entire melodic line were identical in all three context conditions. We addressed several questions in this experiment. First, we asked: Are listeners sensitive to local harmonic implications in these contexts? If so, then we would expect to find that NI targets are judged more obvious than I targets, overall. Second, does the tonal clarity of a musical context affect the strength of harmonic implications? If so, then we would expect to find an interaction of type of target change with harmonic context: The greatest difference between obviousness ratings of I and NI target changes should occur in the consistent chord conditions, and the smallest difference between judgments of NI and I targets should occur in the nonconsistent conditions. Third, do these effects, if they exist, hold for all listeners? Specifically, are listeners with more musical training more responsive to implied harmonies than those with less training,

perhaps because they are less sensitive to pitch changes that preserve the local harmony? If so, then we would expect to find interactions of skill level with these variables. Alternatively, if higher skilled listeners perform equivalently to less skilled listeners with respect to target changes, we should find no such interactions. Finally, assuming that obviousness ratings reflect sensitivity to the implications of tonal grammar, we assessed the extent to which these ratings, together with after-session questionnaire responses, reveal tacit knowledge. Method
Participants. A total of 66 participants served in Experiment 1; 36 qualified as low-skill listeners, and 30 as high-skill listeners. Low-skill listeners from introductory psychology classes at Ohio State University volunteered to participate in return for course credit. All had at least 5 years of musical experience (defined as playing in a band or orchestra, singing in a choir or chorus, or taking private lessons on an instrument), but none had extensive formal musical training, and none were music majors. The mean number of years of private lessons on the participants' primary instrument was 5.7. Fifteen of the participants had additional experience on a second instrument, with an average of 3.9 years of formal training. High-skill listeners participated in the study in return for monetary reimbursement (each received $6.00). All were professional musicians from the Columbus, Ohio, area, graduate students in music at Ohio State University, or undergraduates who had completed or were currently enrolled in the second year in a two-year sequence of aural training courses required for the music major at Ohio State University. Participants had a mean of 8.74 years of formal training on their primary instrument; 23 had formal training on a second instrument, with a mean of 6.41 years. Design. The design was a 2 x 3 x 2 x 3 x 3 mixed factorial. The two within-subject variables were target change (same, implied, and nonimplied) and melodic instance (1, 2, and 3). The three betweensubjects variables were musical skill level (low and high), context (control, consistent chords, and nonconsistent chords), and counterbalance order (1 and 2), yielding a total of 12 (2 x 3 x 2) different betweensubjects conditions. Participants from each skill level were randomly assigned to each of the remaining 6 between-subjects conditions. Apparatus. Stimuli were generated by the MIDILAB software system (Todd, Boltz, & Jones, 1989), through a Yamaha TX81Z FM tone generator controlled by an AT-2000 computer. Stimuli were recorded on a Nakamichi LX-3 cassette deck onto professional-quality cassette tapes, and played to participants on a Nakamichi 550 cassette system, amplified by a Kenwood KA-5700 stereo integrated amplifier, through AKG K240 headphones at a comfortable listening level. These levels could vary according to listener, but none were less than 65 dBA or greater than 75 dBA (GenRad sound-level meter model 1982). Materials and conditions. Three different core melodies (melodic instances 1, 2, and 3) were developed and then modified to include chordal accompaniment (context manipulations) and target note changes (target change). All three melodies were generated by following elementary rules of the tonal grammar (Berkowitz, Frontrier, & Kraft, 1986; Piston, 1987). One of these melodies appears in Figure 1; all three are shown in the Appendix. They were generated to represent a selection of common chordal arrangements, with each using a distinctive serial pattern of chords (implicit or explicit). They observed the following constraints: (a) All were eight measures long; (b) all contained the I chord, thus implying the keynote, in thefirstand last measures; (c) all were in the same key (A major); (d) the melodic sequence of each measure implied some chord (not necessarily the same chord) in the key of A major; (e) all chord progressions (explicit

IMPLIED HARMONY
and implicit) followed conventional rules over the eight measures; (f) melodic lines included no pitch skips greater than an octave and minimized the number of unisons (immediately repeating notes); and (g) all pitches were realized as sine tones. With respect to timing constraints, all melodies manifested (a) a V, time signature; (b) nonisochronous rhythms with the possibility of five different note durations: eighth note, quarter note, dotted quarter note, half note, and dotted half note; and (c) a final lengthened note (dotted half). These core melodic instances served as the basis for developing the following three context conditions (see Figure 1): 1. Control context involved strictly monophonic melodies; that is, no explicit chords were present. 2. Consistent chord context involved homophonic (accompanied) melodies in which three- or four-pitch chords (sometimes in inverted forms) from the key of A major were sustained for the duration of a measure in Measures 1, 3, 5, 7, and 8 of all three instances. For each measure, the chords presented were consistent with the key of the melodic line and in fact reinforced the implied harmony of each measure. 3. Nonconsistent chord context involved homophonic tunes in which three- or four-pitch chords from the key of G major occurred for the duration of a measure in Measures 1, 3, 5, 7, and 8. In this condition, explicit chords were not consistent with the key of the melodic line. All chords were exact transpositions of the A major chords of the consistent chord condition to G major for that melodic instance. Several precautions were taken in introducing the chords in the consistent and nonconsistent chord conditions: (a) All tones within chords were sine tones, (b) onsets and offsets of tones within each chord were slightly asynchronous (i.e., beginning 10 ms later than melodic pitches at the beginning of the measure) to create a more natural sound, and (c) chord intensities were adjusted to be softer (by between 4 to 6 dB) than the melodic line, but always distinctly present. The core melodic instances served as standard melodies in a standard-comparison task for a given context condition. To increase task difficulty and to encourage listeners to attend to transformational aspects of the tunes, we played all standard sequences faster (.8 beat period) and higher (one octave) than their respective comparison sequences. Specifically, quarter notes in standard and comparison sequences corresponded, respectively, to stimulus onset asynchronies (SOAs) of 320 (312 ms on and 8 ms off time) and 400 ms (390 ms on and 10 ms off time). Standard sequences began in the fifth octave and comparisons in the fourth (where C4, or middle C, is located). Three different kinds of comparison tunes, corresponding to levels of the target change variable, were created for each context condition: 5, I target change, and NI target change. The S comparisons incorporated no target changes. The other two conditions incorporated a single in-key changed pitch (target) that could occur in one of the three measures in tunes that contained no explicit chords: Measures 2, 4, or 6. The I comparisons included in-key pitch changes that were in the implied chord of the test measure. The NI comparisons used in-key pitch changes that were not in the implied chord of the test measure. Local melodic context was identical across all manipulations of target change and global context; together with the changed pitch, other test measure tones always formed a prominent in-key chord in the I conditions but not in the NI conditions. In the NI conditions, test measure and target tones did not form a chord in all melodic instances and test measures except one: In melodic instance 2, test measure 2, the NI change formed a part of a II chord in A major, a highly unlikely chord given this key and chord context (Piston, 1987). (In fact, the latter might be expected to reduce the predicted I versus NI difference, thus working against predictions about implied harmony.) Finally, I targets always formed a larger pitch interval difference from the standard than did NI targets. Additional constraints determined targets: (a) A target could not break the melodic contour of the sequence; (b) a target could not form an immediately repeating tone within the sequence; (c) targets were

743

always quarter notes (400-ms SOA); (d) targets always replaced the final note in a test measure; (e) targets and test measures were identical across the three context conditions; and (f) for each of the three melodic instances, only two of the three potential test measures were functional test measures. Thus, for a given melody, a pitch change could occur in either of two of the three measures that were designated test measures. (The unused test measure in each melody is labeled as such in the Appendix.) This constraint insured greater uncertainty with respect to target location. Eighteen standard-comparison pairs were constructed: 6 S pairs, fi I pairs, and fi NI pairs. Each melodic instance had two different target locations for each of the two changed-note conditions (3 x 2 = 6pairs); for S pairs, each melodic instance occurred unchanged twice. Two different counterbalance orders (1 and 2) arranged sequence pairs with the following constraints: (a) No melodic instance followed itself; (b) pairs that occurred in the first half of one counterbalance order were assigned to the second half of the other, minimizing the occurrence of the same sequence in both counterbalance orders; and (c) no more than two S or changed target sequences could occur in a row. Procedure. Participants were run in groups of 1 to 4. Recorded instructions explained that they should rate each of the standard and comparison pairs of sequences. Listeners heard 21 trials in all; the first 3 trials consisted of all three melodic instances played with same comparisons. These trials were not scored. Participants were told that the second sequence could sometimes contain a changed note in its melodic line; in consistent and nonconsistent context conditions it was stressed that if a pitch change occurred it would never occur in the chordal accompaniment but always in the higher pitched melodic line. Participants used a 7-point scale (6 = very obvious change, 1 = less obvious change, and 0 = same or no change) to indicate detection of a melodic change and to rate its obviousness. They were specifically told to use a 0 rating whenever they thought the comparison melody was the same as the standard. Following this, the participants received two practice trials using a different melody in a similar style (comparisons were respectively same and different) and received feedback with regard to their responses; at this time, experimenters clarified that uniform shifts of a comparison's tempo and overall pitch level were not relevant to the task. Each trial began with a high-pitched (D7; 2349 Hz) 1-s warning signal followed by a 2-s silence and then a standard pattern. The comparison sequence occurred 2 s after the offset of its standard. Following this, listeners were given 5 s to respond, in writing, before the warning tone sounded for the next trial. After the session listeners filled out two questionnaires. The first comprised six questions probing their understanding of the musical structure manipulations. They were asked (a) if they noticed more than one kind of (target) change and if so to explain the differences, (b) how the change affected the sound of the comparison, (c) to state specific strategies they used to perform the task, and (d) for general comments. The second questionnaire solicited information regarding the participants' musical background.

Results and Discussion Overall, the results of Experiment 1 suggest that all listeners were sensitive to harmonic implications and were heavily reliant on context to make these inferences. Details are discussed in two sections, which consider, respectively, obviousness ratings and questionnaire responses. Obviousness judgments. Ratings from the two test measures per melodic instance were averaged, yielding one obviousness score for each melodic instance. Figures 2a and b present mean obviousness ratings for listeners of low and high musical skill levels, respectively, as a function of context (control,

744

S. HOI.LERAN. M. JONES, AND D. BUTLER

a)
Low Skill

b)
High Skill
TARGET CHANOE

H
gH I

Sun* (S)
Implied (I) I Nonlmplitd (Nl)

Control

NonContlaMit

Control

Conltont

NonConlt*nt

CHORD CONTEXT CHORD CONTEXT Figure 2. Mean obviousness ratings for (a) low- and (b) high-skill listeners as a function of target change (S, I, and NI) and chord context (control, consistent, and nonconsistent) in Experiment 1 are shown. Figure 2, speaks to the question of whether context affects the strength of harmonic implication. The answer here appears to be "Yes" as well. Planned nonorthogonal contrasts (each evaluated at a of .025) indicated that differences between I and NI targets did not differ significantly between the control and consistent chord conditions, but together the mean differences between I and NI, averaged over control and consistent chord conditions, were significantly greater than corresponding differences (I vs. NI) in the nonconsistent chord conditions, F(l, 108) = 8.38. Thus, the strength of harmonic implication, as indexed by differences in obviousness between I and NI targets, was reliably greater in conditions predicted to have greater tonal clarity, namely the control and consistent chord conditions. A final issue concerning obviousness ratings pertains to differences arising from musical training. Overall there was no reliable difference in rating scores as a function of skill, F(l, 54) < 1.00. More important, as already indicated, the pattern of mean ratings across target change and context was fairly similar for listeners of both skill levels. High-skill listeners yielded somewhat higher obviousness ratings (2.63 vs. 2.49) and appeared to make clearer differentiations among S versus I and NI target changes (Figure 2), producing a significant interaction of skill with target change, F(2, 108) = 7.26, p < .002, MSE = 2.75. With an a per contrast of .025, the three pairwise comparisons, considered together, verify that the main source of this interaction stems from fact that high-skill listeners found NI targets much more obvious relative to S targets than did the low-skill listeners, F(l, 108) = 14.57. The two skill groups did not significantly differ with respect to either I versus NI targets or I versus S targets. Again, if I versus NI differences index strength of harmonic inference, then the latterfindingsconfirm that the two skill groups do not differ in terms of the strength of their harmonic inferences. This observation is reinforced by thefindingthat the two-way interaction of target change with context was not qualified by skill; that three-way interaction was not significant, F(4, 108) < 1.00. Thus, in terms of broad outlines of the data, there is no

consistent chords, and nonconsistent chords) and target change (same, implied, and nonimplied). Our first question concerns whether listeners were sensitive to harmonic implications. The main effect of target change suggests that they were, F(2,108) = 114.84,p < .0001, MSE = 2.75. Overall, NI changes were most obvious (M = 3.80) with I (2.53) and S (1.33) following in the anticipated order. Pairwise comparisons (a = .025 per contrast) confirmed the significance of all three differences: I versus NI, F( 1,108) = 6.86; I versus S, F(l,108) = 6.50; and NI versus S, F(l,108) = 13.39. This general ordering was true of listeners in both skill groups. Overall, for all listeners, comparison pitch changes that did not fit with the local harmony established in the standard pattern seemed to pop out. It is possible that the main effect of target change arises simply from local melodic context effects, an issue we pursued in Experiment 2. However, note that if greater obviousness of NI targets were due to this factor in Experiment 1, then only a main effect for target change should occur. However, the influence of chordal context was also evident: chord context provided not only a significant main effect, F(2, 54) = 9.69, p < .0005, MSE = 4.03, but also a significant interaction with target change, F(4, 108) = 2.66, p < .05, MSE = 2.75. Overall, obviousness ratings were highest with the control condition (M = 3.08) and lowest with the consistent chord condition (M = 2.25). Planned pairwise contrasts on the context main effect indicated that ratings in the control condition were significantly higher than in either of the chord conditions, F(l, 54) = 15.45, for consistent chord versus control, and F(l, 54) = 12.81, for the parallel nonconsistent comparison, but ratings in consistent chord conditions did not differ significantly from those in nonconsistent. However, when target type is considered, the interaction reveals that differences in ratings of I versus NI targets were greatest in the consistent chord conditions and least in the nonconsistent chord condition. Because differences between I and NI are diagnostic of harmonic inference, we consider this interaction in more detail next. The interaction of context with target change, evident in

IMPLIED HARMONY evidence that listeners with more musical training were significantly less sensitive to target changes that preserved the local harmony than were listeners with modest amounts of training.' Questionnaire responses. How aware were listeners of these influences on their judgments? We assessed questionnaires probing listeners' descriptions of differences between standard and comparison sequences. Each listener was assigned to one of four mutually exclusive categories on the basis of his or her responses, as determined by two scorers: (a) Harmony participants mentioned changes related to harmony, tonality, key, or chords; (b) Interval listeners heard changes in interval size or in melodic motion (movement by step or by leap); (c) Character participants reported general character changes regarding changed notes or the comparison melody (e.g., the note "stuck out" or was "sour," or the melody seemed different or "off"); and (d) Other listeners made only miscellaneous comments (e.g., about the sine tones, register, or tempo changes between standard and comparison). One of the authors and an undergraduate student with moderate musical training who was naive as to the purpose of the experiment served as scorers. Scorers agreed on category assignment for 82% of responses. Discrepancies were decided by a third scorer, another author. Table 1 (top) presents the distribution of low- and high-skill listeners as a function of the categories listed above. The two skill groups assessed the stimuli differently, x2(3, N = 36) = 10.97, p < .025. Most low-skill listeners fell into the third (character) and fourth (other) categories, reflecting a lack of awareness of the tonal and harmonic principles at play. By contrast, more high-skill listeners appealed to harmonic principles and were assigned to the harmony category. No differences as a function of context appeared in either skill group. As for explicit articulation of I versus NI distinctions in target changes, only 1 of the 36 low-skill listeners (under 3%) was able to clearly articulate this manipulation, whereas 7 of the 30 high-skill listeners (under 24%) could do this. Thus, although skilled listeners did better in this respect than less skilled listeners, in neither group could most of the listeners volunteer precisely correct information about the tonal grammaticality of target changes. In summary, given melodies with dynamically changing harmonies, all listeners were sensitive to changes in local implied harmony, as indexed by differences between their NI and I ratings. These differences were large even when the melodies were unaccompanied by chords in the surrounding context (control). By contrast, differences between I and NI targets were attenuated in conditions assumed to convey lower tonal clarity, namely where accompanying chords were not consistent with the melodic line (nonconsistent). These effects held for both skill groups. Although high-skill listeners were generally more articulate about underlying harmonic principles governing the melodies, relatively few listeners in either skill group could accurately explain differences between grammatical (I targets) and ungrammatical (NI) pitch transformations. Experiment 2 Experiment 1 demonstrated that all listeners made local harmonic inferences and that the strength of these inferences was influenced by context. However, the way in which context Table 1 Percentage of Listeners in Questionnaire Categories
Skill level Experiment 1 Lowa Highb Experiment 2 Lowa Highb "n = 36.
b

745

Harmony 5.6 33.3 5.6 26.7

Interval 19.4 20.0 11.1 40.0

Character 41.7 36.7 36.1 16.7

Other 33.3 10.0 47.2 16.7

/i = 30.

affected harmonic inference remains unclear. Although harmonic (chord) context heightened implied harmony effects, it is also evident that listeners were sensitive to harmonic implications, even in melodies lacking explicit chordal accompaniment (controls). The latter finding suggests that a melodic line, by itself, may be an important determinant of implied harmony. In Experiment 2 we pursued the hypothesis that the melodic line has a significant influence on a listener's sensitivity to local harmonic implications. Specifically, we used patterns from Experiment 1 in which the melodic line was rendered less coherent by removal of melodic information that connected the test measures. Thus, in both standard and comparison patterns only a broken melodic line remained. Figure 3 depicts a sample stimulus pattern. Three different context conditions (control, consistent, and nonconsistent chords) corresponded to those of Experiment 1. This means that listeners in control conditions received only three melodic fragments (potential test measures) and no accompanying chords (top line only of Figure 3), whereas those assigned to consistent and nonconsistent chord conditions received the same intervening chords used in these conditions in Experiment 1, but with no concurrent melodic line connecting test measures. In this experiment harmonic context consisted of the same explicit chord progressions used in Experiment 1. Chord information is redundant with that supplied by the melodic line (in Experiment 1) insofar as key and changing harmonies are concerned; therefore removal of the melodic line (in Experiment 2) should not necessarily change the pattern of results. If harmonic context alone determines harmonic inference, then in Experiment 2 tonal clarity should continue to be greatest in the consistent chord condition and least in the nonconsistent chord condition. In short, in Experiment 2 we should replicate the pattern of findings of Experiment 1 if surrounding chord progressions alone effectively determine for listeners the "what" and "when" of the implied chords in test measures. The role of local melodic context, namely the pitches within a given test measure, can be more directly assessed in this Significant differences that were due to melodic instance and interactions of melodic instance with target change, context, and counterbalance order were also found. In none of these cases did the findings reverse conclusions reported with respect to the order of obviousness ratings: NI > I > S. Complex interactions suggested that for skilled listeners the magnitude of the implied harmony effect was greater in some melodies and experimental conditions than in others.
3

746

S. HOLLERAN, M. JONES, AND D. BUTLER Consistent Chords No Target Change (S)

16

i
with silence (top musical line of Figure 3). In the two chord conditions, the same three melodic fragments (test measures) occurred and intervening measures were filled either with consistent chords or nonconsistent chords. Figure 3 depicts a new standard melody corresponding to the same, consistent chord condition of Experiment 1 (cf. Figure 1). These manipulations insured that test measures remained identical across all contexts in both experiments. Procedure. The procedure was identical to Experiment 1, with the following exceptions. Participants in the two chord conditions were instructed that the stimuli consisted of a "series of chords interrupted periodically by higher pitched tones without chordal accompaniment" and that pitch changes (targets) could only occur in these upper, higher pitched melodic notes. Stimuli in the control condition were described as "three brief pitch sequences which are separated by silences; each sequence in turn consists of three pitches." To keep retention and response periods equivalent across all three conditions, we did not incorporate in the control condition the time periods corresponding to the final measures (7 and 8) in the chord conditions. Thus, as in Experiment 1, in the two chord conditions, the comparison pattern began 2 s after the offset of Measure 8 of the standard, and the 5-s participant response period began after the offset of Measure 8 of the comparison pattern; in the control condition, however, the comparison pattern began 2 s after the offset of the third pitch sequence (test measure) of the standard, and the 5-s participant response period began immediately after the offset of the third pitch sequence (test measure) of the comparison pattern. Results and Discussion Overall, the findings of Experiment 2 suggest the importance of melodic context as a source of tonal clarity. Relevant data from ratings and questionnaires are discussed separately. Obviousness judgments. Obviousness ratings were averaged over the two test measures for each melodic instance. Figures 4a and b present the mean ratings for low- and high-skill listeners, respectively, as a function of target change and context averaged over counterbalance order and melodic instance. The pattern of results from an analysis of variance (ANOVA) on ratings in Experiment 2 differs markedly from that of Experiment L The most important difference is the disappearance of evidence for harmonic implication. Although target change exerted a significant main effect on ratings, F(2,108) = 117.57, p < .0001, MSE = 3.49, this does not indicate sensitivity to harmonic implication, as reflected in NI versus I differences. Instead, the effect rests entirely on differences between S target changes, which were judged not especially obvious, and the other two kinds of target changes (I and NI), both of which were judged significantly more obvious than S

Figure 3. Example stimulus for Experiment 2, in which melodic line is absent in Measures 1, 3, 5, 7. and 8, is shown. Asterisk denotes an inverted chord.

design. The control condition of Experiment 2 presented test measures in isolation (i.e., lacking more extended melodic and harmonic context). If listeners are primarily biased by local context, then the differences in I versus NI ratings observed in Experiment 1 should appear in the control condition of Experiment 2. In summary, in Experiment 2 we broke the melodic line. If melodic context contributes to listeners' dynamic expectancies about forthcoming chords, then the diagnostic evidence for local harmonic implication, namely differences between I and NI targets, should disappear in all contexts. However, if conventional tonal chord progressions alone can provoke expectancies about forthcoming chords, then we should replicate the findings of Experiment 1 for the two chord context conditions. Method Participants. A total of 66 listeners participated in Experiment 2. Thirty-six qualified as low-skill listeners, and 30 as high-skill listeners. The low-skill listeners were volunteers from introductory psychology classes at Ohio State University who participated in return for course credit. All met the criteria of Experiment 1, and none had participated in the previous experiment. They had an average of 4.13 years of private instruction on a primary instrument. Eighteen had experience on a second instrument, with an average of 1.89 years of formal training. The high-skill listeners met the same criteria as those in Experiment 1, and none had participated in that experiment; each participated in return for monetary reimbursement ($6.00). They had an average of 7.95 years of formal instruction on their primary instrument. Twentythree had experience on a second instrument, with a mean of 4.5 years of formal training. Design. The design was a 2 x 3 x 2 x 3 x 3 mixed factorial. The three between-subjects variables were skill (high and low), context (control, consistent chord, and nonconsistent chord), and counterbalance order (1 and 2). The two within-subject variables were target change (same, implied, and nonimplied) and melodic instance (1,2, and 3). Participants of each skill group were randomly assigned to each of the six context by counterbalance order conditions. Apparatus. This was identical to Experiment 1. As in that study, listening levels of recorded events were adjusted for each participant to a comfortable level; these ranged between 64 and 70 dBA. Materials and conditions. In all respects, save one, the context conditions were identical to those of Experiment 1. The one change involved removal of all sine tones corresponding to the upper (melodic) line in Measures 1, 3, 5, 7, and 8 for all melodic instances. In control conditions, this resulted in patterns containing three melodic fragments corresponding exactly in timing and pitch to potential test measures (i.e., Measures 2, 4, and 6); intervening measures were filled

IMPLIED HARMONY

747

a)
wry abvloua Chang*

b)
Low Skill High Skill
TARGET CHANGE M H Sama (3) US Impliad (I)

C D Nonlmpllad (NO

obvloua ehang* Control Cwialalaiil NonCmalalanl Control Contlatant NonConl(*nl

CHORD CONTEXT

CHORD CONTEXT

Figure 4. Mean obviousness ratings for (a) low- and (b) high-skill listeners as a function of target change (S, I, and NI) and chord context (control, consistent, and nonconsistent) in Experiment 2 are shown.

changes. Three pairwise comparisons (a .025 per contrast) verified that the I versus S difference, F(l, 108) = 12.09, and the NI versus S difference, F(l, 108) = 13.45, were both significant, but the I versus NI difference was not. Although target change produced a significant interaction with context in this experiment, F(4, 108) = 3.73, p < .005, MSE = 3.49, this interaction does not reflect context-determined harmonic inference. Instead, it arises entirely from differences between NI and S, rather than NI and I, in the control and nonconsistent chord conditions, F(l, 108) = 7.98,p < .025. This higher order contrast indicates that the difference between NI and S target changes was larger in the nonconsistent chord condition than in the control condition. However, most important from the perspective of harmonic implication effects is the finding that pairwise contrasts of I versus NI were not statistically significant in any of the three contexts. The absence of I versus NI differences in all context conditions, including the control condition, indicates not only that harmonic context was a weak determinant of harmonic inference in this task but also that local melodic context cannot explain the harmonic inference effects observed in the control condition of Experiment 1. Skill differences were again evident in Experiment 2. A main effect of skill, F(l, 54) = 12.447, p < .001, MSE = 4.54, indicates that overall high-skill listeners produced higher obviousness ratings than did low-skill listeners. However, these overall means really reflect the fact that high-skill listeners tended to judge any pitch change (I or NI) as highly obvious in all contexts. This produced a Target x Skill interaction, F(2, 108) = 20.36, p < .0001, MSE = 3.49. Ratings of I versus NI did not differ significantly as a function of skill level, but high-skill listeners were better at distinguishing patterns with no changes from ones with true targets; hence contrasts involving I versus S, F(l, 108) = 22.23, and NI versus S, F(l, 108) = 25.33, did change significantly with skill (a per contrast = .025). These findings indicate that although low-skill listeners had more difficulty than high-skill ones in distinguishing comparison sequences that contained true pitch changes from ones

that did not, neither skill group relied heavily on implied harmony to do this. Questionnaire responses. Questionnaire results were tabulated according to the same four categories as in Experiment 1. The two scorers agreed on category assignment for 67% of the questionnaires, with discrepancies again decided by a third scorer. Table 1 (bottom) presents the number of participants from each skill group assigned to each category. As in Experiment 1, the two skill groups differed significantly in their assignment to the four categories, x2(3, N = 36) = 17.30,/? < .001. Relative to low-skill listeners, the high-skill listeners were more likely to be assigned to the harmony category. As in Experiment 1, chord context (control, consistent, and nonconsistent) did not significantly affect category assignment in either skill group. Finally, with respect to specific articulation of I versus NI differences, none of the low-skill listeners volunteered such an explanation, whereas six of the high-skill listeners (20%) did. Nevertheless, although listeners with more training are better than those with less training in pinpointing grammatical violations of the tonal structure, the majority of high-skill listeners did not do this. In summary, the results of Experiment 2 indicate that, regardless of musical skill level, listeners were less likely to respond differentially to information about implied harmonic relations in situations that lacked a coherent melodic line. Skilled listeners were more adept than others at detecting true pitch changes when they occurred, regardless of their harmonic implications. Finally, no low-skill listener reported noticing harmonic differences among different target changes, and only afifthof the high-skill listeners did.

General Discussion
The answer to our first question, "Can listeners infer local harmonic meaning within a changing tonal melody?" is, quite simply, "Yes." Even in unaccompanied melodies, all listeners

748

S. HOLLERAN. M. JONES, AND D. BUTLER insure that listeners heard the melodies as tonal but also to minimize variability with respect to the particular key involved. As a result, it is most unlikely that listeners heard these melodies in unintended keys. At least in control and consistent chord conditions, this means that I targets and NI targets truly reflected changes consistent and inconsistent, respectively, with the local harmony of A major. Other aspects of the design confirm that listeners responded to manipulations of target change in terms of harmonic implications. For example, it is possible that listeners would respond to these changes strictly in terms of log frequency melodic differences between standard and comparison patterns and not to tonal harmonic implications. In these designs, comparison patterns containing NI targets turned out to be more similar to their standards with respect to pitch interval structure than did comparisons containing I targets, in that they realized the smaller pitch interval change. If listeners were responding only to pitch interval differences, then the I targets should have seemed more obvious than NI targets. In fact, there is some evidence that larger melodic intervals may be taken (incorrectly) as evidence of harmonic change (Platt & Racine, 1994), a circumstance that might even have contributed to higher obviousness ratings of I targets in the present studies and thus worked against the NI versus I differences we predicted (and found). Furthermore, because low-skill listeners appear to be more responsive to interval size manipulations than are highskill listeners (e.g., Platt & Racine, 1994), it could be argued that they might have found the I targets more obvious than the NI targets in the present studies. This was not the case. In fact, in Experiment 1 low-skill listeners did not rate the I targets as more obvious than the NI targets in any context condition. Another interpretation of the target change effect could be based on the spectral structure of the target tones. Because sine tones were used, we can eliminate an explanation in which the virtual pitch associated with targets within test measures might fit more readily with I target changes than with NI changes. With pure tones, the mode of analytic listening proposed by Terhardt (1984) applies; in this case, there is no theoretical basis for predicting I versus NI differences. Because moderate listening levels (around 70 dB SPL) were used, we cannot entirely rule out aural nonlinearities (e.g., Green, 1976; Wegel & Lane, 1924), but it is highly unlikely that such phenomena played any significant role in the pattern of observed results. The most telling evidence against any strictly psychoacoustic-peripheral interpretation is the finding that the implied harmonic effect associated with the same set of sine tones (i.e., melodic test fragment) did not occur when only the local melodic context was present (control conditions in Experiment 2), but it did appear when additional melodic context was present (Experiment 1). Other possible explanations of the target change effect appeal to principles that are not directly related to tonal grammars. One maintains that the I pitch changes are less obvious merely because they might belong to the same pitch class as another pitch in the test measure. However, if this were the case, then we would expect NI versus I differences to be evident in all context conditions of Experiment 2 (and in Experiment 1), but this was not found. A more sophisticated

found in-key target pitches that conflicted with the local harmony implied by a melody much more obvious than in-key pitches that did not. This finding extends previous research that suggests that even listeners with fairly modest degrees of formal musical training respond to general aspects of diatonicity in Western tonal music. It also confirms a longstanding premise of music theory that individual pitches of a melody can indeed imply certain underlying chords. At the same time, it undermines the hypothesis that people encode melodies independently of their harmonic implications. A second goal of this research concerned the impact of context on harmonic inferences. In particular, we considered whether the tonal clarity of a surrounding context determined the strength of local harmonic inferences drawn from a melody. Together the findings of Experiments 1 and 2 permit us to conclude that context is important and that a clear and unambiguous tonal context is probably necessary for observation of a strong harmonic inference. In particular, when supplied with only a local melodic context (control conditions of Experiment 2), listeners showed no clear evidence of having made harmonic inferences. The fact that local melodic context did not have strong effects on judgments of I versus NI targets is not entirely surprising because, taken in isolation, the tones within a single test measure did not logically specify the key; therefore they could not, by themselves, tonally disambiguate an implied chord. The present data confirm this while also testifying to the fact that global melodic and harmonic context can supply the necessary disambiguating tonal information. Originally we maintained that if the strength of harmonic inference depends on the tonal clarity provided by a musical context, then differences in obviousness ratings between I and NI targets should be greatest in the consistent chord condition, in which chords explicitly reinforce various harmonies implied within a continuous melodic line (i.e., in Experiment 1). This turned out to be the case. However, statistically speaking, it appears that the presence of a continuous melodic line is fairly important. In Experiment 1, the listeners in the control condition did not differ from those in the consistent chord condition. These findings, together with those of Experiment 2, indicate that melodic context alone is highly influential in bringing about judgments that are based on harmonic implications. We emphasized that the various manipulations of harmonic and melodic context designed to affect tonal clarity rested on establishment of key and constraints of tonal grammar. To insure this, we played all melodies in the same key, or tonal frame, throughout a session. Pilot work using melodies in different keys produced greater variability and weaker implied harmony effects than reported above. Because these melodic instances, by necessity, were short relative to most musical events, it seemed clear that introduction of erratic key shifts within a session would violate the sort of normal listening experience people encounter in everyday musical settings, where longer tunes remain in the same key or modulate to a related one in predictable ways. As with tempo and meter, which also offer stable frames for musical events (Jones, Kidd, & Wetzel, 1981; Yee, Holleran, & Jones, 1994), a tonal frame cannot be haphazardly shifted about within a session without taking a toll on performance. We chose our design not only to

IMPLIED HARMONY explanation maintains that the greater frequency of the I target's pitch class (E) in the surrounding melodic context makes it less obvious than the pitch class of the NI targets. The recurrence of E is an inevitable consequence of the tonal manipulations that relied on a melodic sequence of implied chords most common to the A major scale (vs. G major) to establish high tonal clarity in A major. However, three pieces of evidence argue against this as the whole explanation of the target change effect. The first concerns the failure to find any significant I versus NI differences in Experiment 2 in the control condition, where the surrounding test measures included melodic fragments containing more instances of the I pitch than of the NI pitch. Second, if obviousness of any target pitch (S, I, or NI) decreases with an increase in the frequency of occurrence of that pitch in the context, then we would expect the I targets to be even less obvious than the S target changes because they are more common in the melodic lines of these instances than are S targets. This was not found. Finally, if this were the entire explanation for target differences in Experiment 1, then we would anticipate no Target x Context interaction in that experiment because the lower tonal clarity of the nonconsistent chord condition should have no effect on performance. This was not the case. At a larger level, however, this raises the issue of what constitutes a tonal context. Our position is that tonal context depends on key plus grammatical arrangements of tones (implied harmonies), and thus inevitably involves an element of pitch repetition. With respect to the influence of context on harmonic inferences, the pattern of results over the two experiments was rather surprising. We anticipated that harmonic context effects, taken alone, would at least rival those of melodic context, taken alone. Chord progressions alone have produced strong effects on performance in other tasks. They were used, for example, by Krumhansl and Kessler (1982) in a task in which skilled listeners rated single probe tones of various sorts to gauge a listener's developing sense of key. Such findings suggest two plausible outcomes for the present experiments: First, that chord progressions and melodic line should affect harmonic inferences equally and, second, that the combination of consistent chords with a continuous melodic line (i.e., in Experiment 1) should produce implied harmony effects substantially stronger than conditions involving either the melodic line alone (Experiment 1, control) or consistent chords alone (Experiment 2). However, insofar as judgments of I versus NI targets reflect harmonic influences, in Experiment 1 the consistent chord condition did not produce stronger evidence for such influences than did the control condition. In addition, in Experiment 2 there were virtually no reliable harmonic implication effects associated with presentations of the consistent chords alone. Perhaps the latter two outcomes explain why, in Experiment 1, combining the melodic line with reinforcing chords did not yield distinctly better performance in the consistent chord conditions than that found in control conditions, which presented only the melodic line: Consistent chords merely provided information that was redundant with the melodic line. Therefore, when chords were presented separately, as in Experiment 2, they were not especially effective at evoking harmonic implications. In any case, the findings of both experiments converge to suggest that listeners

749

weight the melody itself fairly heavily in drawing harmonic inferences from it: Without a continuous melodic line they judged both kinds of pitch changes to be equally obvious. The finding that a melodic line alone affords sufficient basis for harmonic inference is strong evidence against the assertion that a melody is perceived independently of its chordal implications (e.g., Povel & van Egmond. 1993). Instead, it confirms the time-honored music theoretic claim that a wcllformed melody can imply its own harmony. This is one of our most interesting discoveries about how context works. Previous research has not systematically compared the relative influence of chordal and melodic context (but sec Platt & Racine. 1990). It does seem somewhat surprising that even musicians do not exploit more information from purely chordal contexts: If "hearing" chordal implications is what makes the I targets seem less obvious than the NI targets, then shouldn't presenting explicit chords in the surrounding context be at least as influential as presenting the melodic line alone? At least in the present task, in which listeners knew that if a change occurred it would happen in the melodic line, it is clear that this was not the case. One interpretation of the melodic context effect focuses on the effectiveness of a continuous melodic line in guiding selective attending. When a melodic line is provided to a listener who knows that targets will always appear within that sequence, a selective attending set is encouraged in which listeners monitor only the melody (e.g., Jones & Yec. 1993; Jones, Jagacinski, Yee, Floyd, & KJapp, 1995; van Noordcn, 1975). In this case, perhaps the bitonality of nonconsistent chord conditions in Experiment 1 created a divided attention task in which listeners could not "tune out" the distracting chord conflicts. However, it would be misleading to dismiss the present findings as merely the result of a task-induced set. Such an interpretation does not do justice to the interactions of context with target change and skill observed here. Furthermore, others who have pitted instructions against structural determinants of attending in a task involving detection of small perturbations in polyrhythmic sequences found that the structural aspects of the sequences themselves had more influence on selective attending than an instructed "set" to attend in a particular way (Jones et al., 1995). There are other aspects of the structure of any multivoiced musical event that may contribute to the salience of the melodic line. When chords accompany a melody, the melodic line is typically higher in frequency and slightly louder than other voices, and this was true in the present case (Experiment 1). These physical features undoubtedly contribute to the melody capturing listeners' attention. However, if only frequency and intensity differences were responsible for these effects, then we would expect no interaction of target change with context in Experiment 1; all listeners would perform equally well in all contexts. This was not the case. There is also recent evidence that harmonically related (in-chord) pitch changes are as noticeable as harmonically unrelated ones when they occur in an upper voice that carries the melody, but they are less noticeable in the upper voice when another voice carries the melody (Palmer & Holleran, 1993). This suggests that selective attending to the melodic line may even attenuate evidence for harmonic inference.

750

S. HOLLERAN, M. JONES, AND D. BUTLER listeners' sensitivity to expected events relative to their normal discrimination levels. Thus, if in a tonal context high- and low-skill listeners give the same nonobvious rating score to a change in pitch, this score would reflect more suppression for the trained listeners relative to their normal (but higher) pitch discrimination baseline. For example, assume that Experiment 2 performance indexed pitch change discrimination levels in nontonal contexts; then high-skill listeners in this experiment, on average, were quite good at discriminating I targets (mean Experiment 2 obviousness rating of 4.52). However, in Experiment 1, these listeners scored, on average, 1.96 for I targets in the consistent chord conditions; over the two experiments then, skilled listeners showed a reduction in sensitivity to this pitch change of 2.56 points. On the other hand, low-skill listeners were less discriminating overall in Experiment 2; consequently, in Experiment 1, they showed a drop of around 1.18 for I targets in the consistent chord condition. It is primarily in this respect that it might be argued that greater reliance on a tonal scheme renders certain expected events correspondingly less obvious. Finally, skill-level manipulations and the probing of listeners' awareness of target manipulations relating to harmonic implications extend the generality of tacit knowledge to nonartificial grammars that are transformational rather than Markovian in nature. A demonstration of tacit knowledge typically involves a juxtaposition of evidence that people do reliably respond to rule violations (i.e., NI vs. I obviousness ratings), with evidence that they seem unaware of the underlying principles (questionnaire responses regarding target changes). This juxtaposition was most prominent in Experiment 1, where most listeners reliably distinguished comparisons that did not belong in the same grammatical category as their standards (NI) from comparisons that did (I), while at the same time very few could explain their performance, especially if they had little musical training: The NI comparison sequences just "didn't seem right." We think this is a fairly straightforward example of tacit knowledge of musical principles. Interestingly, even high-skill listeners, who have received training not only in attending to and identifying tonal relationships but also in labeling relevant grammatical properties, showed this effect. Although these listeners were better able to identify the changes as involving harmonic principles, less than a quarter could pinpoint the exact nature of the differences between I and NI target changes. The present data make two additional points complementary to the research on tacit knowing. The first relates to issues raised by Vokey and Brooks (1992), who maintained that people learn abstract similarities among exemplar patterns within a category rather than grammars as such. At least in musical events, it is difficult to separate abstract pattern similarities from the underlying grammar. For example, in our research there was substantial similarity between standard and comparison members of any pair because pair members differed only with respect to a single in-key pitch. Nevertheless, a single tone change (NI vs. I) can have remarkably different implied meanings in light of conventions of Western tonal harmony. This suggests that determining similarities between instances is not a simple matter. That is, by a straightforward feature account it might be maintained that

Thefindingthat listeners differentially weight various contextual determinants (melodic vs. chordal or harmonic) poses a challenge to current theories that address implied harmony (as these effects are revealed by responses to NI vs. I targets) but that fail to accommodate selective influences of context. Jones for example, maintains that a temporal context provokes an active abstraction of key specific and grammatical relationships, which direct, over time, specific attentional extrapolations (i.e., expectancies) about certain anticipated tonal relationships (e.g., I changes) and not others (e.g., NI changes; Jones, 1976, 1981, 1982, 1990). Others incorporate rule systems in which melodic and harmonic relationships will shape musical expectancies in such a way that some expectancy violations might seem more obvious than others (Carlsen, 1981; Meyer, 1973; Narmour, 1990, 1992). Tonal scheme theories, including network models, are less specific about how expectancies adapt over time to accommodate dynamically changing musical events; nevertheless, at a general level their descriptions of tonal expectancies in terms of stronger links among chords and other harmonically salient structural levels (e.g., Bharucha, 1987,1991) mean that they too offer a basis for predicting differences in obviousness ratings between NI and I targets. However, none of these approaches explains how obviousness of I versus NI targets changes as a function of the presence or absence of a continuous melodic line. This is because none currently accommodates selective influences on attending that may arise either from the task or from the relative prominence of certain-contextual relationships (e.g., melodic vs. harmonic context). Yet our data argue for the importance of these factors in explaining performance. Our third concern relates to musical skill. Musically trained listeners did not make substantially stronger harmonic inferences, if it is assumed that inference strength is gauged by differences in ratings of I versus NI targets. Furthermore, the broad outline of their performance as a function of context in both experiments was quite similar to that of less skilled listeners. It seems clear that context suppressed I target obviousness (relative to NI targets) for both skill groups. Consequently, we cannot conclude that increased training renders listeners correspondingly less sensitive to pitch changes that preserve local harmony. These findings are similar to those of DeWitt and Samuel (1990), who reported that more experienced adult listeners showed no greater tendency than inexperienced adults to restore missing tones in tonal melodies and, in some cases, seemed to be more discriminating. These results also converge with other studies that found no relationship between amount of musical training and the ability to detect in-scale versus out-of-scale pitch changes (e.g., Trainor & Trehub, 1992). However, the outcome of the our experiments appears at odds with Trainor and Trehub's main result, in which adults were less sensitive to diatonic changes than infants. A resolution of these seemingly contradictory findings is possible on the basis of the assumption that musical training not only confers knowledge of the tonal grammar but that it also heightens discriminative acuity of many structural features of an acoustic pattern (e.g., Jones et al., 1993; Spiegel & Watson, 1984). Then the presence of an implied harmony effect reflects the operation of contextually provoked tonal expectancies that suppress

IMPLIED HARMONY comparison patterns containing NI targets are more similar to their standards than those with I targets because the former represent a smaller pitch interval distance. Such an analysis, however, would predict the opposite of our findings. Here patterns judged most similar involved comparisons in which targets are defined by common tonal grammar transformations (i.e., S and I changes). It turns out that the criterion listeners adopted for categorization has little to do with the surface features of a single pitch change and everything to do with what it implies grammatically. The second point concerns the way listeners respond over time to grammatical sequences. The fact that listeners were most effective in distinguishing NI comparisons as "not right" when they could follow a continuous melodic line implies that some sequential properties (e.g., a melodic line) may have more impact on attending and hence on categorization (of NI vs. I) than others (e.g., explicit chords). In short, a theory about what constitutes abstract tacit knowledge is not sufficient to explain performance; it is also necessary to know how grammatically generated structure guides attending. References Berkowitz, S., Fontrier, R., & Kraft, L. (1986). A new approach to sight-singing. New York: Norton. Bharucha, J. J. (1987). Music cognition and perceptual facilitation: A connectionist framework. Music Perception, 5, 1-30. Bharucha, J. J. (1991). Pitch, harmony, and neural nets. In P. M. Todd & D. G. Loy (Eds.), Music and connectionism (pp. 84-99). Cambridge, MA: MIT Press. Bharucha, J. J., & Stoeckig, K. (1986). Reaction time and musical expectancy: Priming of chords. Journal of Experimental Psychology: Human Perception and Performance, 12, 403-410. Boltz, M. (1989a). Perceiving the end: Effects of tonal relationships on melodic completion. Journal of Experimental Psychology: Human Perception and Performance, 15, 749-761. Boltz, M. (1989b). Rhythm and "good endings": Effects of temporal structure on tonality judgments. Perception & Psychophysics, 46, 9-17. Boltz, M. (1991). Some structural determinants of melody recall. Memory & Cognition, 19, 239-251. Brown, H. (1988). The interplay of set content and temporal context in a functional theory of tonality perception. Music Perception, 5, 219-250. Brown, H., & Butler, D. (1981). Diatonic trichords and minimal tonal cue-cells. In Theory Only, 5, 37-55. Butler, D. (1989). Describing the perception of tonality in music: A critique of the tonal hierarchy theory and a proposal for a theory of intervallic rivalry. Music Perception, 6, 219-242. Butler, D., & Brown, H. (1984). Tonal structure versus function: Studies of the recognition of harmonic motion. Music Perception, 2, 6-24. Carlsen, J. C. (1981). Some factors which influence musical expectancy. Psychomusicology, 1, 12-29. Cohen, A. J., Trehub, S. E., & Thorpe, L. A. (1989). Effects of uncertainty on melodic information processing. Perception & Psychophysics, 46, 18-28. Cuddy, L. L., Cohen, A. J., & Miller, J. (1979). Melody recognition: The experimental application of musical rules. Canadian Journal of Psychology, 33, 148-157. Dalhaus, C. (1990). Studies on the origin of harmonic tonality (R. O. Gjerdingen, Trans.). Princeton: Princeton University Press. (Original work published 1965-1966)

751

Dowling, W. J. (1978). Scale and contour: Two components of a theory of memory for melodies. Psychological Review. 85, 341-354. DeWitt. L. A.. & Samuel, A. G. (1990). The role of knowledge-based expectations in music perception: Evidence from musical restoration. Joumalof Experimental Psychology: General, 119, 123-144. Frances, R. (1988). The perception of music (W. J. Dowling, Trans). Hillsdale, NJ: Erlbaum. (Original work published 1958) Gjerdingen, R. O. (1990). Categorization of musical patterns by self-organizing neuron-like networks. Music Perception, 8, 339-370. Gjerdingen, R. O. (1992). Revisiting Meyer's "Grammatical simplicity and relational richness." In M. R. Jones & S. Holleran (Eds). Cognitive bases of musical communication (pp. 225-243). Washington, DC: American Psychological Association. Gorfein, D. S. (Ed.). (1989). Resolving semantic ambiguity. New York: Springer. Graesser, A. C , & Bower, G. H. (Eds.). (1990). Inferences and text comprehension. San Diego, CA: Academic Press. Green, D. (1976). An introduction to hearing. New York: Erlbaum. Jones, M. R. (1974). Cognitive representations of serial patterns. In B. Kantowitz (Ed.), Human information processing: Tutorials in performance and cognition (pp. 187-229). Potomac, MD: Erlbaum. Jones, M. R. (1976). Time, our lost dimension: Toward a new theory of perception, attention, and memory. Psychological Review, S3, 323335. Jones, M. R. (1981). Music as a stimulus for psychological motion: Part I. Some determinants of expectancies. Psychomusicology, 1. 34-51. Jones, M. R. (1982). Music as a stimulus for psychological motion: Part II. An expectancy model. Psychomusicology, 2, 1-13. Jones, M. R. (1990). Learning and the development of expectancies: An interactionist approach. Psychomusicology, 9, 193-228. Jones, M. R., Jagacinski, R., Yee, W., Floyd, R., & Klapp, S. (1995). Tests of attentional flexibility in listening to polyrhythmic patterns. Journal of Experimental Psychology: Human Perception and Performance, 21, 293-307. Jones, M. R., Kidd, G., & Wetzel, R. (1981). Evidence for rhythmic attention. Journal of Experimental Psychology: Human Perception and Performance, 7, 1059-1073. Jones, M. R., & Yee, W. D. (1993). Attending to auditory events: The role of temporal organization. In S. McAdams & E. Bigand (Eds.), Thinking in sound: The psychology of human audition (pp. 69-112). London: Oxford University Press. Krumhansl, C. L. (1990). Cognitive foundations of musical pitch. New York: Oxford University Press. Krumhansl, C. L., & Kessler, E. J. (1982). Tracing the dynamic changes in perceived tonal organization in a spatial representation of musical keys. Psychological Review, 89, 334-368. Lerdahl, F. (1988). Tonal pitch space. Music Perception, 5, 315-350. Meyer, L. (1973). Explaining music. Berkeley: University of California Press. Monahan, C. B., Kendall, R. A., & Carterette, E. C. (1987). The effect of melodic and temporal contour on recognition memory for pitch change. Perception & Psychophysics, 87, 576-600. Narmour, E. (1990). The analysis and cognition of basic melodic structures: The implication-realization model. Chicago: University of Chicago Press. Narmour, E. (1992). The analysis and cognition of melodic complexity: The implication-realization model. Chicago: University of Chicago Press. Palmer, C , & Holleran, S. (1993, June). Melodic, harmonic, and frequency height influences in the perception of multi-voiced music. Paper presented at the meeting of the Society for Music Perception and Cognition, Philadelphia. Piston, W. (1987). Harmony (5th ed., M. DeVoto, ed.). New York: Norton.

752

S. HOLLERAN. M. JONES, AND D. BUTLER Journal of Experimental Psychology: Human Perception ami Performance, 15, 457-471. Spiegel, M. F., & Watson, C. S. (1984). Performance on frequencydiscrimination tasks by musicians and nonmusicians. Journal of the Acoustical Society of America. 76, 1690-1695. Tekman, H. G., & Bharucha, J. J. (1992). Time course of chord priming. Perception & Psychophvsics, 51, 33-39. Terhardt, E. (1984). The concept of musical consonance: A link between music and psychoacoustics. Music Perception, 1, 276-295. Todd, R. E., Boltz, M., & Jones, M. R. (1989). The MIDILAB music research system. Psychomusicology, 8, 17-30. Trainor, L. J., & Trehub, S. E. (1992). A comparison of infants' and adults' sensitivity to Western musical structure. Journal of Experimental Psychology: Human Perception and Performance, 18, 394-^02. vanNoorden, L. P. A. S. (1975). Temporal coherence in the perception of tone sequences. Unpublished doctoral dissertation. Technical University of Eindhoven, the Netherlands. Vokey, J. R., & Brooks, L. R. (1992). Salience of item knowledge in learning artificial grammars. Journal of Experimental Psychology: Learning, Memory, and Cognition, 18, 328-344. Wegel, R. L., &Lane, C. E. (1924). The auditory masking of one sound by another and its probable relation to the dynamics of the inner ear. Phys. Review, 23, 266-285. Yee, W., Holleran, S., & Jones, M. R. (1994). Sensitivity to event timing in regular and irregular sequences: Influences of musical skill. Perception & Psychophysics, 56, 461-471.

Platt, J. R., & Racine, R. J. (1990). Perceived pitch class of isolated musical triads. Journal ofExperimental Psychology: Human Perception and Performance, 16, 415-428. Platt, J. R., & Racine, R. J. (1994). Detection of implied harmony changes in triadic melodies. Music Perception, 11, 243-264. Povel, D. J., & van Egmond, R. (1993). The function of accompanying chords in the recognition of melodic fragments. Music Perception, 11, 101-115. Reber, A. S. (1989). Implicit learning and tacit knowledge. Journal of Experimental Psychology: General, 118, 219-235. Reber, A. S., Kassin, S. M., Lewis, S., & Cantor, G. W. (1980). On the relationship between implicit and explicit modes in the learning of a complex rule structure. Journal of Experimental Psychology: Human Learning and Memory, 6, 492-502. Reber, A. S., Walkenfeld, F. F., & Hernstadt, R. (1991). Implicit and explicit learning: Individual difference and IQ. Journal of Experimental Psychology: Learning, Memory, and Cognition, 17, 888-896. Schenker, H. (1979). Free composition (E. Oster, Trans.). New York: Longman. (Original work published 1935) Schmuckler, M. A. (1989). Expectation in music: Investigation of melodic and harmonic principles. Music Perception, 7, 109-150. Shepard, R. N. (1982). Geometrical approximations to the structure of pitch. Psychological Review, 89, 305-333. Smith, K. C , & Cuddy, L. L. (1989). Effects of metric and harmonic rhythm on the detection of pitch alterations in melodic sequences.

IMPLIED HARMONY

753

Appendix Melodic Instances (Melodies 1, 2, and 3) Used in Experiments 1 and 2 Melody 1


Test Measure 1 Test Measure 2 (unused) Test Measure 3

Consistent Chords (A Major)

\yir-til

I ' if'

I ' if-

I " Ip 1
V? I

I I* NonConststent Chords (Q Major)

if

I - If

I
Melody 2

If

if

V,

NonConsistent Chords (O Major)

r i
Melody 3
Test Measure 1 (unused) Test Measure 2

if i if
V, I

Test Measure 3

NonConststent Chords (0 Major)

Inverted chord

Received May 3, 1993 Revision received June 13, 1994 Accepted June 20, 1994

Anda mungkin juga menyukai