Vous êtes sur la page 1sur 32

Cognitive Psychology 44, 132 (2002) doi:10.1006/cogp.2001.0759, available online at http://www.idealibrary.

com on

The Role of Retrieval Structures in Memorizing Music


Aaron Williamon
Royal College of Music, London, United Kingdom

and Elizabeth Valentine


Royal Holloway, University of London, United Kingdom This article explores the use of structure in the encoding and retrieval of music and its relation to level of skill. Twenty-two pianists, classied into four levels of skill, were asked to learn and memorize an assigned composition by J. S. Bach (different for each level). All practice was recorded on cassette tape. At the end of the learning process, the pianists performed their assigned composition in a recital setting. The performances were subsequently evaluated by three experienced pianists according to a standardized grading system. From the cassette tapes, values for the frequency with which pianists started and stopped their practice on structural, difcult, and other bars were obtained. Starts and stops on each bar type were compared across three stages of the learning process. The analyses reveal that all pianists, regardless of level, started and stopped their practice increasingly on structural bars and decreasingly on difcult bars across the learning process. Moreover, the data indicate that starts and stops increased on structural bars and decreased on difcult bars systematically with increases in level of skill. These ndings are interpreted and discussed so as to elucidate characteristics of the retrieval structures adopted by musicians in their practice and performance and how the formation and use of retrieval structures develop as a function of expertise. Finally, the elicited values for starts on structural, difcult, and other bars are examined and discussed according to how they relate to the pianists scores on performance quality. 2002
Elsevier Science (USA)

The research presented in this article was supported by grants from the Committee of ViceChancellors and Principals of the Universities of the United Kingdom and the Rotary Foundation. The authors gratefully acknowledge the help of the participating musicians; the participating music teachers (Florence Creighton, Elaine Goodman, Heli Ignatius-Fleet, Nadia Lasserson, Murial Levin, and Danielle Salamon); Roger Chafn, Gabriela Imreh, and Carola Grindea for guidance in the early stages of this work; John Valentine for statistical advice; Susan Hallam, John Sloboda, and the referees for comments on earlier versions of this article; and Ben Chafn for computer programming assistance. Address correspondence and reprint requests to Aaron Williamon, Royal College of Music, Prince Consort Road, London SW7 2BS, United Kingdom. E-mail: awilliamon@rcm.ac.uk. 1 0010-0285/02 $35.00
2002 Elsevier Science (USA) All rights reserved.

WILLIAMON AND VALENTINE

Music is one of the richest domains for studying skilled performance. Not only must musicians recall and execute a wide range of information during performance, but they must do so in different environments, under different levels of anxiety, and generally while considering and incorporating the ideas of conductors and other ensemble members. Moreover, they must satisfy these cognitive, perceptual, motor, and social demands with novel musical insight and seemingly little technical effort. Music, therefore, provides a unique and fertile testing ground for assessing existing theories of the acquisition and performance of skilled behavior. One major feature which has been highlighted is the use of structure to organize cognition. Chase and Simons (1973a, 1973b) Chunking Theory proposes that expertise in a domain is acquired by learning a large database of chunksa chunk being a collection of information that forms a meaningful unit. These chunks are indexed by a discrimination net, an organized tree structure that enables the accessing of prescribed patterns by providing a set of instructions to the perceptual system for systematically scanning the large repertoire of patterns stored in long-term memory (LTM). The discrimination net allows rapid categorization of domain-specic patterns and accounts for the speed with which experts recognize key elements in problem situations. Chunks also give access to semantic memory consisting of productions and schemas (Simon, 1989). Each familiar chunk in long-term memory is a condition of a production that may be satised by the recognition of the perceptual pattern and evokes an action (Newell & Simon, 1972). This explains the rapid solutions typically proposed by experts and offers a theoretical account of intuition (Simon, 1986). Empirical evidence from many domains (e.g., physics, see Larkin, McDermott, Simon, & Simon, 1980; and mathematics, see Hinsley, Hayes, & Simon, 1977) reveals that experts, unlike novices, use forward search to solve problems and, therefore, make heavy use of productions based on pattern recognition. Chase and Ericssons (1982) Skilled Memory Theory, designed to address certain limitations of Chunking Theory, proposes that remarkable displays of memory result from the creation and efcient use of retrieval structures. In doing so, individuals must associate the encoded information with appropriate retrieval cues. This association permits the activation of a particular retrieval cue at a later time and, thus, partially reinstates the conditions of encoding so that the desired information can be retrieved from LTM. Only after a set of retrieval cues is organized in a stable structure is a retrieval structure formed, thereby enabling individuals to retrieve stored information efciently without lengthy search (Ericsson & Staszewski, 1989, p. 239). According to Chase and Ericsson, encoded information within the retrieval structure is organized and retrieved according to both hierarchical and serial principles. Hierarchy is the arrangement of components into various levels

RETRIEVAL STRUCTURES IN MUSIC

of complexity and importance. Serial organization refers to the sequential (i.e., linear) organization of encoded material as demonstrated in studies by Shaffer (1976) and Palmer and van de Sande (1995). Chase and Ericsson (1981) demonstrated both of these in their subject SF, who acquired exceptional digit span. At the lowest hierarchical level, SF used mnemonic associations to running times and other numerical relations in order to group digits into units. Then, SF used spatial relationships to encode digit groups into supergroups. At the time of recall, SF could easily regenerate any of the unique locations of the retrieval structure and use a given location as a cue to retrieve the corresponding digit group. Chase and Ericsson (1981) highlighted the serial principles that govern retrieval structures through a cuedrecall task. They found that SF could access digit groups when descriptions of their location within the retrieval structure were used as cues. Hierarchical organization appears to be a cognitive principle of wide generality, applying to the encoding and retrieval of both motoric and symbolic information (Johnson, 1970; Rosenbaum, 1987). Rosenbaum, Kenny, and Derr (1983) provided evidence for the hierarchical control of learned sequences of nger taps, from the pattern of interresponse times in their production. They applied their tree-traversal modelaccording to which response latencies are a function of the length of the node path to be traversedto the explanation of data from Sternberg, Monsell, Knoll, and Wright (1978). In a series of studies on the timing of brief bursts of responses in typing and speaking, these authors showed that (1) the latency of the rst response increased linearly with the length of the sequence, (2) the time to complete a sequence increased quadratically with burst length, and (3) interresponse times were longer in the middle than at the ends of a sequence. Other notable areas in which hierarchical processes have been demonstrated are language (Chomsky, 1957; Wall, 1972) and categorization (Collins & Quillian, 1969, 1970). More recently, pertinent data have come from the investigation of the effect of temporal accent structure on memory for lmed narratives. Boltz (1992) demonstrated that insertion of commercials at major episode boundaries (thus highlighting a storys underlying organization) resulted in higher recall and recognition performance and superior memory for temporal order information and details from the storys plot than did insertion of commercials within episodes (hence obscuring the storys underlying structure). Furthermore, this attentional highlighting of episode boundaries enhanced selective recall and recognition of breakpoint scenes compared with those at nonbreakpoints. These ndings suggest that episode boundaries are used as referents for attention and recall. The Skilled Memory Theory has commonly been accepted as accounting for exceptional memory (Schneider & Detweiler, 1987; Carpenter & Just, 1989; Anderson, 1990; Baddeley, 1990; Newell, 1990; Ericsson & Kintsch,

WILLIAMON AND VALENTINE

1995). Furthermore, the mechanisms based on retrieval structures have been extended beyond experts memory abilities to account for superior performance in general (Ericsson & Staszewski, 1989). Ericsson and Kintsch (1995) extended the Skilled Memory Theory into the Long-Term Working Memory (LT-WM) Theory to account for doubts in the theorys generalizability to working memory (see Baddeley, 1990; Schneider & Detweiler, 1987). They proposed that information can be encoded and retrieved through (1) a hierarchical organization of retrieval cues associated with units of encoded information (i.e., a retrieval structure, as in the top half of Fig. 1); (2) a knowledge-based, elaborated structure that permits the units of encoded information to be associated to other items in LTM or to the context (i.e., schemas and patterns, as in the lower half of Fig. 1); or (3) both. Schemas and patterns play a key role in the LT-WM Theory. Unfortunately, Ericsson and Kintsch failed to dene these terms. Gobet (1998), however, maintained that their usage seems compatible with the following denitions: A schema is a memory structure that is made both of xed patterns and of slots where variable patterns may be stored; a pattern is a conguration of parts into a coherent structure (see also Bartlett, 1932; Piaget, 1967; Rumelhart, 1980; Thorndyke & Yekovich, 1980; Alba & Hasher, 1983). According to Ericsson and Kintsch, the demands a given activity makes on working memory dictate the type of encoding used so as to attain reliable and rapid storage of and access to the information in LT-WM. For example, a musician who is performing a composition from memory will rely almost entirely on a hierarchically organized set of preformed retrieval cues to ensure that encoded information is retrieved reliably and efciently. This retrieval structure will develop throughout the course of extensive practice on the piece. Conversely, a performer who is sight reading or improvising music will continually draw from previous experience and the surrounding context but will not be able to rely on a set of preformed retrieval cues since the music has not been rehearsed prior to performance. Finally, one performing from a notated source will rely on a combination of cues, previous experience, and information drawn from the surrounding content (see Related Research in Music Cognition below for further discussion). In sum, the LT-WM Theory is consistent with the notion that superior memory performance is domain-specic and contends that the proposed increase in working memory is limited to an individuals respective domain of expertise. Also, the acquired nature of LT-WM implies that differences may exist between tasks and that individual differences in the implementation of LT-WM for a given task may potentially emerge (Ericsson & Kintsch, 1995, p. 220). Using their theoretical account, Ericsson and Kintsch explain the empirical ndings of studies on digit span memory, memory for menu orders, mental multiplication, mental abacus calculation, chess, medical expertise, and text comprehension.

RETRIEVAL STRUCTURES IN MUSIC

FIG. 1. Two different types of encoding of information stored in LT-WM. The top of the gure shows the hierarchical organization of retrieval cues associated with units of encoded information. The bottom of the gure shows the knowledge-based associations relating units of encoded information to each other along with patterns and schemas establishing an integrated memory representation of the presented information in LTM (adapted from Ericsson & Kintsch, 1995).

WILLIAMON AND VALENTINE

Two questions, however, should be addressed before aspects of the LTWM Theory, or any theory of general expertise for that matter, can be examined within the domain of music. First, is musical expertise comparable to expertise in other domains? Though the nature of performance in music may differ fundamentally from that of other domainssuch as chess, closed sport skills (e.g., gure skating, ballet, and gymnastics), and open sport skills (e.g., tennis, basketball, and eld hockey)skilled musicians share some of the same demands on skill and memory as other experts. Memory, for instance, plays an important role in closed sports because individuals must memorize sequences of movements that constitute a given performance and perform those movements with reference to dened, technical standards. Since technical skills are judged according to ideal forms, performers must make online comparisons between actual and ideal techniques. The bases for these on-line comparisons are drawn from LTM, acquired over many years of training (Allard & Starkes, 1991). Concert soloists, like experts of closed sport skills, recall series of actions that constitute a performance; refer to dened, technical standards expected by audiences and judges; and make on-line comparisons to gauge the quality of their actual performance. The second question is: Do performing musicians accumulate a vast, domain-specic knowledge base, acquired through extensive practice? Young concert artists often have 1015 concerti and six or more recital programs ready to perform by the start of their professional careers (Chafn & Imreh, 1994). This enormous amount of prepared materialnot to mention their tudesdemonstrates knowledge of such entities as scales, arpeggios, and e possession of a vast knowledge base within the music domain. Also, the 10,000 h of deliberate practice, indicated by Ericsson, Krampe, and Tesch mer (1993) as a prerequisite to reaching expert levels of musical compeRo tence, certainly qualies as extensive practice (see Williamon & Valentine, 2000, for further discussion). A multitude of additional issues must be addressed before these theories can be viewed as accurate explanations of the cognitive processes governing musical expertise. For example, do musicians index and rapidly categorize musical information into meaningful units (i.e., chunks)? Do they encode and retrieve information using retrieval structures and schemas? If so, are these structures organized according to hierarchical and serial principles? Existing research on musical skill and performance offers insight into these questions.
RELATED RESEARCH IN MUSIC COGNITION

Evidence for Chunking in Music Halpern and Bower (1982) provide evidence that musicians do, indeed, index and rapidly categorize musical information into meaningful chunks.

RETRIEVAL STRUCTURES IN MUSIC

They visually presented slides of good, bad, and random melodies to musicians and nonmusicians for 5 s each. Musicians performance on the 5-s written recall task worsened from good to bad to random melodies; however, they recalled signicantly more notes than nonmusicians in all three categories. These ndings parallel results from other studies in that highly trained individuals recalled more of the presented stimuli than those who were untrained and their ability to perform this task diminished most for random stimuli (see de Groot 1946/1978; Chase & Simon 1973a, 1973b; Allard et al., 1980; Starkes et al., 1987). They also reveal that musicians can, as other experts, encode task-relevant information rapidly and accurately, thereby suggesting that the cognitive processes governing skilled musical performance resemble, in some respects, cognitive processes in other performance domains. The results, however, contradict those of previous studies (e.g., Chase & Simon, 1973a, 1973b) in that musicians were signicantly better than nonmusicians at remembering randomly presented notes. Halpern and Bower explained this inconsistency by proposing that musicians have had 10 or more years of perceptual differentiation training with notes as well as note names to aid their encoding and recall (p. 39). They also stressed that almost any sequence or combination of notes may be judged as good music by some listeners (p. 33). Evidence for Retrieval Structures and Schemas in Music At rst glance, the nature of the cognitive mechanisms used by skilled musicians to encode and retrieve information is not clear from an inspection of existing research. This ambiguity rests on differences in the terminology used in music compared with that of other areas of skilled performance. Nevertheless, the available research provides initial evidence for the use of retrieval mechanisms during practice and performance. In a theoretical article, Clarke (1988) suggested that skilled musicians retrieve and execute compositions using hierarchically organized knowledge structures constructed from information derived from the score and projections from players stylistic knowledge. He provided examples of hypothetical knowledge structures based on a compositions formal structure. One of these postulated the formation and use of knowledge structures in memorized performances. He argued that performances of classical music from memory appear to offer the most deeply embedded generative structures (p. 3). In memorized performances, the generative structure is known entirely, or at least to a high level, in advance. This includes the highest level of global understanding (the piece as a whole) down to the lowest level (the piece as individual notes; see Fig. 2a). Therefore, it is simply unpacked during a performance (p. 5). The description of such structures bears a remarkable resemblance to that of retrieval structures by Chase and Ericsson (1982) and Ericsson and Kintsch (1995). Clarke asserted that the idea of having an entire structural representation of a composition activated during performance is unlikely,

WILLIAMON AND VALENTINE

FIG. 2. Representation of (a) a complete knowledge structure for a memorised musical performance, (b) a partially activated structure (ringed) in midphrase, and (c) a partially activated structure (ringed) approaching a phrase boundary (adapted from Clarke, 1988).

even for a piece of only moderate length. Instead, he proposed that part of the structure is activated at any one time and that the active region shifts as the performer progresses through the music. This shift occurs between regions of activated structure which vary in durational extent and generative depth. As a general rule, the depth to which the generative structure is activated is directly related to the structural signicance of phrase boundaries lying close to, or at, the players current musical location (p. 5). In the middle of a deeply embedded musical phrase, for instance, a performer may primarily be concerned with the detailed structure of connections within the

RETRIEVAL STRUCTURES IN MUSIC

phrase itself. Therefore, only a region of low-level generative connections would be active, rather than high-level structural information (see Fig. 2b). Conversely, at a phrase boundary, a performer may need to know how the previous and subsequent phrases relate to one another and to the overall structure of the piece (see Fig. 2c). At such a moment, a small area of lowlevel structural connections may be active, sufcient to specify the immediate succession of events to be played, together with a section of the higher levels of generative structure specifying larger-scale relationships (p. 4). In elaborating on how knowledge structures are formed for a memorized performance, Clarke emphasized that generative structures can be perfectly represented only when all information has become available. In the course of assembling a representation from the raw data of notation, mistaken assumptions and oversights will inevitably result in a awed structure that must be continually reassessed and retrospectively modied in the light of new evidence. Such reevaluations are jeopardized by memory limitations, and may also lead to constructive rationalizations which distort the true structure of the music (p. 6). In sum, Clarke asserted that performers retrieve and execute compositions using internal representations. He described the hierarchical nature of knowledge structures in detail and proposed how highly developed, or possibly complete, knowledge structures are used during memorized performances. Chafn and Imreh (1997) systematically observed the practice of a concert pianist (Imreh) to determine whether she used the kind of highly practiced, hierarchical retrieval structure described above to memorize and perform the third movement, Presto, of J. S. Bachs Italian Concerto. Practice for this piece was divided into 58 sessions, aggregated into three learning periods and spread over 10 months. Sessions were videotaped, and cumulative records were created showing the pianists starting and stopping points in the musicsimilar to that generated for the present study and illustrated in Fig. 3. They also examined the pianists concurrent and retrospective commentary on her practice. At the end of the 10 months, the pianist performed the Presto from memory. Chafn and Imreh conrmed Clarkes (1988) proposal that skilled musicians use hierarchical retrieval schemes to recall encoded information. Moreover, they found that the pianist organized her practice and subsequent retrieval of the Presto according to its formal structure. In her commentary in practice session 17 (of 58), she focused on sections of the formal structure in which the same theme was repeated. Specically, she spent a considerable amount of time in comparing differences between the various repetitions of the A and B themes. She then put the two themes together, remarking that I think I am going to work on these larger sections (p. 324). Chafn and Imreh used this and other similar comments as the basis for arguing that the pianist used the structure to guide the encoding and retrieval of the music,

10

WILLIAMON AND VALENTINE

FIG. 3. A sample from the cumulative records showing the segments of the music played taken from one of the middle practice sessions of a Level 4 pianist while practicing the Fugue in D Minor. The x axis represents bars of the music and the y axis depicts the cumulative number of practice segments. The beginning of each horizontal line indicates the point at which the pianist started playing the composition. The end of each line denotes the point in the music at which the pianist stopped playing. Each new line, reading from bottom to top, indicates that the pianist stopped and restarted.

suggesting that in order to monitor transitions between sections consciously, [she] needed to retrieve a conceptual representation of the next section from memory as she played (p. 325). To provide empirical support for their argument, they examined the extent to which the pianists practice was actually guided by the formal structure. They compared the number of practice segments that started and stopped at boundaries in the formal structure with the number that started and stopped at other locations. They found that starts and stops occurred more frequently at structural boundaries than in the middle of sections. They used these ndings as behavioral support for their assertion that the pianist identied the formal structure and then used it to guide the encoding and retrieval of the piece. In a follow-up study, the pianist was asked without forewarning to write out the rst page of the score from memory 27 months after her original performance, during which time she had not practiced or performed the piece for 24 months. The researchers found that accuracy of recall for notes (duration was disregarded) was signicantly better for the bars beginning each section than for bars at other locations, conrming, once again, that the hier-

RETRIEVAL STRUCTURES IN MUSIC

11

archical components of the musics formal structure formed an enduring foundation for the pianists retrieval structure.

THE AIMS OF THIS ARTICLE

Based on the work of Clarke (1988) and Chafn and Imreh (1997), one may argue that musicians do form and use hierarchically organized retrieval structures when memorizing music for performance. Indeed, Chafn and Imrehs research is the rst to demonstrate that the principles of expert memory (see Chase & Ericsson, 1981; Ericsson & Oliver, 1989) apply to concert soloists. Several remaining issues, however, must be addressed by subsequent research so that clearer insight can be gained into how these cognitive mechanisms govern musical skill. In particular, how does the formation and use of such structures change as musicians acquire greater levels of overall competence? The concert pianist in Chafn and Imrehs (1997) study used a retrieval scheme based on the musics formal structure from the early stages of her practice. Further research must explore whether these results generalize to other performers. Novice musicians may lack the required skills and experience to identify a compositions formal structure, in which case their inept domain-specic knowledge base would force them to exploit other retrieval schemes or prevent them from explicitly using any such schemes. Another important question is how do retrieval structures change across the practice process as musicians progressively learn a given composition for performance? Although Clarke asserted that performers will continually reassess and modify their representations during practice, further empirical evidence must be provided to reveal the precise ways in which these modications occur. This article aims to address these questions by examining the practice of 22 pianistsspread across four levels of skillas they prepared an assigned composition for performance. First, post performance interviews are inspected to reveal how the pianists segmented and organized their assigned composition during practice and performance. Characteristics of these segmentations are explored to address whether they were based on the formal structure or some other division of the music and whether they were hierarchically ordered. Second, empirical measures from the musicians practice are explored to determine if the pianists practice was actually guided by their segmentation of the music and to reveal possible differences in the extent of this guidance within and between ability levels. These results are discussed in terms of their contribution to furthering the understanding of how the cognitive mechanisms used in musical practice and performance develop as individuals strive to achieve musical expertise. Finally, the extent to which performers structured their practice according to hierarchical princi-

12

WILLIAMON AND VALENTINE

ples of organization is examined in relation to the quality of their resulting nal performances.
METHOD

Participants
Six piano teachers from southeast England were asked to recommend students capable of learning and performing from memory a selected piece of music suited to their level of ability. Thirty-seven pianists were recruited for the study. Of those 37, a complete set of data was collected and analyzed for 22 participants. Of the 15 pianists omitted from the analyses, 8 did not follow instructions accurately, 4 felt overwhelmed by the demands of the project, 2 did not wish to participate, and 1 withdrew for other personal reasons. Participation was strictly voluntary but encouraged by the piano teachers because the conditions of participation (described below in Procedure) were seen to contribute to students overall musicianship by providing invaluable and challenging performance experience. The participants were classied into four levels of ability based on the grading system set forth by the Associated Board of the Royal Schools of Music (see Harvey, 1994). This system contains eight grades, with Grade 1 representing the lowest level of skill and Grade 8 representing the highest. Musicians at Grade 8 are usually considered to possess high performance standards, though falling just short of expertise. The four levels span all eight grades and were stratied as follows: pianists of Grades 1 and 2 were placed in Level 1 (2 male and 3 female); of Grades 3 and 4 in Level 2 (3 male and 3 female); of Grades 5 and 6 in Level 3 (2 male and 4 female); and of Grades 7 and 8 in Level 4 (5 female). This division of the Associated Boards system was acknowledged as an acceptable stratication of ability by the six participating piano teachers, all of whom had extensive experience in preparing musicians for Associated Board grade examinations and ve of whom were, themselves, examiners for the Board. The classication system was strictly upheld, except in one instance when the pianist had never taken grade examinations. In this case, the musician was placed in the most appropriate level, as deemed by the piano teacher. Means and standard deviations for general characteristics of pianists who successfully completed the study at each level of ability, including age, years of formal training on the piano, length of time with current piano teacher, and total number of piano teachers have been reported previously (see Williamon & Valentine, 2000, p. 359). It was demonstrated that the musicians within each of these levels were sufciently comparable in terms of overall musical competence and training and that the four groups were equally representative of their intended level of skill (see pp. 362363).

The Music
The pianists were assigned one piece of music appropriate to their level of ability. All selected pieces were composed by J. S. Bach. The compositions for Levels 1 to 4 were, respectively, the Polonaise in G Minor from the Anna Magdalena Notebook (BWV Anh. 119), the Two-Part Invention in C Major (BWV 772), the Three-Part Invention in B Minor (BWV 801), and the Fugue in D Minor from the Well-Tempered Clavier I (BWV 851; Level 4 pianists also prepared the Prelude in D Minor, but only the results for the Fugue are reported in this article). The compositions were chosen with the following three criteria in mind: (1) consistency of style and composer between levels of ability, (2) position within the standard piano repertoire, and (3) relative difculty for the respective level (see Williamon & Valentine, 2000, for a discussion of how these selected works conformed to the three criteria). General characteristics of the selected compositions, including time signature, mean tempo, and the

RETRIEVAL STRUCTURES IN MUSIC

13

total number of bars and beats in each piece are listed in Table A in Appendix 1 (see Williamon, 1999, for an analysis of each compositions formal structure).

Procedure
Systematic observations of practice. The pianists were asked to record all practice for their assigned piece on cassette tape. The participants were invited to comment, either on tape or in writing, on any relevant aspect of the learning process. In addition, pianists were asked to note and describe all practice carried out away from the piano, including singing the music and analyzing the score. Participants were informed at the outset of the study that they would be required to perform the assigned piece from memory in a recital setting, attended by their teachers, parents, and fellow music students. The recitals were part of the students regular curriculum. No restrictions on the amount of time or the number of practice sessions were placed on the pianists, except for those normally afxed by themselves or their music teachers. Performance evaluations and postperformance interviews. The 22 recital performances were recorded on videotape and were evaluated by three experienced piano teachers, other than those whose students were participating in the study, two of whom were also experienced examiners for the Associated Board. Evaluations were made according to the three guiding principles set forth by the Associated Board: musical understanding, communicative ability, and technical prociency. Performers were scored for each of the above performance aspects and on overall performance quality. Performances were rated on a scale from 1 to 12, with 12 as the best rating. Results presented elsewhere (see Williamon & Valentine, 2000, pp. 363 364) reveal that scores for overall performance quality were signicantly correlated between these examiners (Evaluators 1 and 2: r .68, p .01; Evaluators 2 and 3: r .84, p .01; Evaluators 1 and 3: r .53, p .05; see Williamon & Valentine, 2000, p. 364, for a list of the mean ratings from all three evaluators on overall performance quality, musical understanding, communicative ability, and technical prociency). Following each performance, the pianists were interviewed on the practice and memorization process. All interviews were recorded on cassette tape. One set of interview questions required that participants indicate whether they had thought of their assigned composition as having component sections during both practice and performance, and if so, why and how they partitioned it. In another set of questions, they were asked to identify the bars in which difcult passages occurred in the music and explain why they were difcult. These were openended questions, not intended to lead the pianists into particular answers, such as the identication of the musics formal structure or the cataloging of difculties into specic types. Responses to questions were, in general, one to two sentences long. In addition, participants were asked to mark clearly their identied sections and difcult bars in two colors of ink, respectively, on a photocopy of their score.

Cumulative Records
The recorded practice sessions were transcribed into cumulative records for each pianist. Graphs were plotted for each practice session showing starting and stopping points for the segments of music played by each pianist. Figure 3 displays a graph taken from one of the middle practice sessions of a pianist at Level 4 practicing the Fugue in D Minor. Stutters, correcting one or two notes while continuing to play through the music, were not included in the cumulative records. All graphs were transcribed from the cassette tapes by the rst author. ve to the aims of To establish the accuracy of these transcriptions, one expert musician, na the study and unaware of the structural boundaries indicated by each pianist, transcribed one practice session for one pianist from each level of ability (the session was selected randomly from the middle of the practice process for each pianist). This amounted to 50 min of practice time and totaled 203 practice segments. The original transcriptions of these practice sessions

14

WILLIAMON AND VALENTINE

by the rst author contained 210 segments (one of the missing segments was from the practice of the Level 1 pianist, four from Level 2, and two from Level 3). The beat on which each segment started and stopped (counted cumulatively from the beginning of each composition) was obtained from the original transcription and that of the independent rater. A coefcient of repeatability (see Bland & Altman, 1996) was calculated across both transcriptions for starts and stops (with seven missing values). The resulting coefcients were, respectively, 0.34 ( w 0.12) and 0.38 ( w 0.14), thus predicting with 95% condence that, had all practice sessions been transcribed by these two raters, their transcriptions would have deviated by no more than 0.34 and 0.38 beats.

RESULTS

Segmenting the Assigned Compositions: Evidence from the Interviews All 22 participants reported segmenting their assigned piece while learning it and using this segmentation during their memorized performances. Fourteen pianists indicated that their segmentation had been pointed out and stressed by their piano teachers (Level 1: n 4 of 5; Level 2: n 5 of 6; Level 3: n 3 of 6; Level 4: n 2 of 5). These participants were students of three of the six teachers. As might be expected, their divisions of the music were highly consistent within ability levels and congruent with the musics formal structure. However, only three of these pianists (one in Level 3 and two in Level 4) reported that they were aware of this (all participants were asked explicitly if they knew of the formal structure if they had not yet volunteered the information). The remaining eight participants remarked that they had partitioned their assigned composition themselves. The reasons behind these segmentations were varied, ranging from the visual layout of the page (e.g., page breaks) to harmonic progressions (e.g., cadences). For example, one pianist in Level 1, whose segmentation was not inuenced by her teacher, reported that it was based on the change in dynamics throughout the piece, salient visual aspects of the score and repeated patterns in the music. This segmentation of the Polonaise was not the result of a formal analysis of the musics harmonic or rhythmic structure, but the sections identied by this pianist were musically sensible, coinciding exactly with the formal structure at major section boundaries. Still, not all of the pianists segmentations agreed wholly with a formal analysis of the music. One musician at Level 3 who divided the Three Part Invention into ve major sections commented that in general, the music seemed to resolve best at the end of bars that had fast notes in both hands. Although some sections identied by this pianist coincided with components of the formal structure, there were still discrepancies between the formal structure and the identied segmentation. An answer to the question of whether participants based their segmentation of their composition on the formal structure is not clear-cut from the interview data. Some of the pianists did so knowingly. Others did so unknowingly, and still others divided their composition by additional measures. Nev-

RETRIEVAL STRUCTURES IN MUSIC

15

ertheless, these interviews show that all pianists segmented their assigned piece into sections that were meaningful to each of them. Like the concert soloist in Chafn and Imrehs (1997) study, they reported using these sections to learn and recall the compositions. Therefore, the important issue is not whether the formal structure guided the rehearsal and retrieval of musical information, but whether the meaningful sections identied by each pianist guided rehearsal and retrieval. An empirical evaluation of whether this actually occurred is presented below. The interviews also provide insight into the question of whether the musicians segmentations were hierarchically organized. In many instances, the pianists reported several subsections within larger sections. An example from a Level 1 pianist demonstrates such hierarchy. She reported major sections across bars 14, 510, and 1116 and subsections at bars 79, 1314, and 1516. Similar hierarchical content was apparent in the segmentations of all pianists, especially at higher levels of ability, where the music was longer and more complex. For example, one Level 3 pianist partitioned the ThreePart Invention into four major sections that corresponded to the overall formal structure (bars 113, 1425, 2632, and 3338). Within these sections, he identied a total of seven subsections (bars 13, 46, 710, 1113, 14 16, 1719, and 2025). Since the majority of pianists in Levels 1 and 2 were inuenced by their teachers in forming their segmentations, no between-level comparisons were made with regard to the use of hierarchy. Between-level differences, however, did emerge in terms of hierarchy when the participants were asked to identify difcult bars in the music and state why they were difcult. The answers were varied, ranging from those dealing with the physical execution of the piece (e.g., large intervals, fast notes, accidentals, articulations, and tricky ngerings) to musical aspects of performance (e.g., bringing out the fugue subject and playing the piece stylistically correct). Regardless of specic classications of difculty, pianists in Levels 3 and 4 systematically identied difcult bars section by section. For example, one Level 3 pianist commented that there were no difculties in the rst section of the music. In the second section, only bar 19 was tricky in getting the hands coordinated, but the third section was most difcult, especially bars 26, 27, and 28. It took me a while to gure out how to play them, both musically and technically. Bar 36 in the last section was difcult too. In general, Level 3 and 4 pianists acknowledged that they were thinking of a particular section of the piece and then commenced to name the relevant bars within that section (the mean proportion of difcult bars spontaneously mentioned, together with their segment location, to the total number of difcult bars was 0.96 for Level 3 and 1.00 for Level 4). This was unsolicited by the interviewer and therefore suggests that, for these musicians, the identication of specic material within a composition (e.g., difcult bars) was related to how that material t into more global levels of understanding (e.g., the major sections of their segmentations).

16

WILLIAMON AND VALENTINE

Although this occurred to some extent for pianists at Levels 1 and 2, they, by and large, referred to difcult bars independently from their segmentations (the mean proportion of difcult bars spontaneously mentioned, together with their segment location, to the total number of difcult bars was 0.72 in Level 1 and 0.85 for Level 2). This implies that either (1) they did not do so because the instructions for answering did not require it or (2) the identication of local detailsuch as that required when describing difcult barswas independent from the identication of broader, more global levels of understanding. Behavioral data presented by Williamon (1999) lend support to the second of these explanations. He showed that pianists at higher levels in this sample interspersed short and long practice segments more than less skilled pianists throughout the entire learning process, suggesting that they were more exible at shifting focus between hierarchical levels of understanding. Considering that the less skilled pianists did not intersperse short and long practice segments as much and, hence, may have been less practiced at shifting focus between levels of understanding, the nding that they did not rely exclusively on global, hierarchical levels to evoke local information in their interviews is not particularly surprising. In sum, the ndings presented above reveal that participants segmented their assigned composition into various hierarchical organizations, not always coincident with the formal structure. Moreover, the data indicate that the more skilled musicians demonstrated an extended use of hierarchy when reporting information about local detail in their assigned composition (i.e., difcult bars). Nevertheless, further analyses must be performed to reveal whether empirical ndings from the recorded practice actually support the pianists claims that meaningful sections in the music guided rehearsal and retrieval. The Role of Segmentation in Practice To determine the extent to which segmentation played a role in guiding practice, bars of the assigned compositions were categorized as structural, difcult, or other. Unlike Chafn and Imrehs (1997) study, the formal structure of the music was not used as the basis of this categorization because only three pianists (one in Level 3 and two in Level 4) reported that knowledge of the formal structure inuenced their segmentation of the assigned piece. Instead, the categorization system was based on the pianists individual-specic segmentation of the music and identication of difcult bars. Bars were classied as structural if they were the rst bar in each of the identied sections and subsections. They were labeled as difcult if they had been named as such by the pianists. No differentiation was made between types of difculty. All remaining bars were placed into the other group. In four cases, two pianists in Level 2, one in Level 3, and one in Level 4, labeled one bar in their composition as both structural and difcult. In these instances, the bars were omitted from subsequent analyses. Also,

RETRIEVAL STRUCTURES IN MUSIC

17

the rst bar of each composition was excluded from all analyses. Obviously, the rst bar of a piece may play a guiding role in any hierarchical retrieval scheme; however, it was excluded because of the multitude of reasons as to why musicians may decide to start their practice at the beginning of a piece. Two of these are that musical information is organized linearly and that any attempt at a complete performance is likely to begin on the rst bar. The structural and difcult bars identied by each pianist are listed in Table B of Appendix 1. Using this classication system, the frequency with which pianists started their practice on structural, difcult, and other bars was obtained for each practice session. In order to compare these frequencies both within and between ability levels a number of difculties had rst to be overcome. In terms of within-level comparisons, the number of structural, difcult, and other bars identied by each pianist varied considerably. Consequently, the resulting frequencies may have increased or decreased based on the number of each type of bar. As for between-level comparisons, the ndings of Williamon and Valentine (2000) revealed that pianists at higher levels of ability in this sample spent more time practicing in each practice session. As a result, they may have started practice on structural, difcult, and other bars more often than pianists at lower levels of ability. Also, the number of bars in each assigned piece was different. Therefore, in the hypothetical situation that all bars were equally important in terms of encoding and retrieving musical information, the probability of the pianists starting their practice on any one bar would decrease with an increase in the length of the piece. To account for these within- and between-level inconsistencies, a score was calculated reecting the deviation between (1) the observed frequencies of starts on structural, difcult, and other bars and (2) the expected frequencies based on the number of each type of bar identied and the number of bars in the assigned piece. The equations by which these scores were calculated were derived from that used to calculate expected frequencies in the chi-squared test (see Goodman, 1957; Kendall & Stuart, 1963). The calculated valuesreferred to from hereon as s-starts for structural bars, dstarts for difcult bars, and o-starts for other barsgive an equivalent of z scores, where positive integers indicate more starts on a specic bar type than would be expected and negative integers indicate fewer starts on a specic bar type than would be expected. The equations and calculations used to obtain s-starts, d-starts, and o-starts for each pianist in each practice session are shown in Appendix 2. The means and standard deviations for s-starts, d-starts, and o-starts across all practice sessions for each level of ability are listed in the top half of Table 1. These values for the deviation of the observed from expected frequencies for each bar type (i.e., s-starts, d-starts, and o-starts) were compared using a two-factor mixed analysis of covariance (ANCOVA) with deviation as the dependent variable, bar type (i.e., structural, difcult, and other) as the

18

WILLIAMON AND VALENTINE

TABLE 1 Means and Standard Deviations (SD) for s-Starts, d-Starts, oStarts, s-Stops, d-Stops, and o-Stops for Each Level of Ability Level 1 s-starts SD d-starts SD o-starts SD s-stops SD d-stops SD o-stops SD 2.48 0.99 0.13 0.15 1.52 0.61 0.24 0.14 2.37 0.35 1.02 0.32 Level 2 6.21 1.92 1.12 0.39 2.88 0.91 1.74 0.32 1.37 0.63 0.33 0.49 Level 3 15.64 1.82 3.39 0.23 4.18 0.89 2.75 0.22 0.27 0.32 0.58 0.25 Level 4 21.97 2.19 3.33 0.41 6.43 1.02 3.45 0.28 0.92 1.26 0.72 0.31

within-subjects independent variable, and level as the between-subjects independent variable. Williamon and Valentine (2000) found that pianists at the various levels of skill in this sample differed signicantly in age; therefore, age was entered as a covariate. The analysis revealed that there was a highly signicant effect of bar type [F (2, 34) 262.25, p .001]. The values for s-starts (mean 11.58) were higher than those for d-starts (mean 1.93), which in turn were higher than those for o-starts (mean 3.75). Practice was much more likely to start on structural bars than would be expected by chance, whereas it was less likely to start on difcult or other bars than would be expected by chance. Planned comparisons revealed that sstarts were greater than d-starts [t (17) 225.16, p .001] and s-starts and d-starts combined were greater than o-starts [t (17) 373.91, p .001]. The effect of level was highly signicant [F (3, 17) 58.88, p .001]. Overall, values increased with level. However, there was also a highly signicant interaction between bar type and level [F (6, 34) 265.22, p .001]; s-starts increased as a function of level, whereas d-starts and ostarts decreased as a function of level. Planned contrasts revealed that the difference between s-starts and d-starts was greater for pianists at higher levels of ability [t (17) 263.33, p .001]. Chafn and Imreh (1997) found that their participant also stopped her practice more frequently at structural boundaries than at other locations in the music. Therefore, the deviations of the observed from expected stops on structural, difcult, and other bars were calculated to determine whether this was the case for the pianists in this study. These calculationsreferred to from hereon as s-stops for structural bars, d-stops for difcult bars, and o-stops for other barswere obtained in the same manner as shown in Appendix 2. The bottom half of Table 1 lists the means and standard

RETRIEVAL STRUCTURES IN MUSIC

19

deviations for s-stops, d-stops, and o-stops over all practice sessions for each ability level. The corresponding ANCOVA for -stops indicated a highly signicant effect of bar type [F (2, 34) 37.03, p .001]. Values for s-stops (mean 2.05) were higher than those for d-stops (mean 0.64) and o-stops (mean 0.66). Planned comparisons indicated that s-stops and d-stops combined were higher than o-stops [t (17) 62.79, p .001]. There was a signicant effect of level [F (3, 17) 3.23, p .05]; the means for Levels 1, 2, 3, and 4 were 1.21, 1.15, 1.02, and 1.08, respectively. The interaction between bar type and level was highly signicant [F (6, 34) 41.35, p .001]; the values for s-stops increased as a function of level, whereas those for d-stops decreased and those for o-stops remained fairly constant. Planned contrasts revealed that the difference between s-stops and d-stops was greater at higher levels of ability [t (17) 63.64, p .001] as was the contrast between s-stops and d-stops combined with o-stops [t (17) 25.45, p .001]. Starts and Stops on Structural Bars To explore the extent to which structural bars guided the pianists practice throughout the learning process, s-starts and s-stops were examined at three discrete stages of practice and averaged within each level of ability. Stage 1 included values for each pianists rst three practice sessions, Stage 2 included values for the middle three practice sessions, and Stage 3 included values for the last three practice sessions. Three stages, spread evenly across the practice process, were chosen to provide comparable extracts from each pianists practice. Three sessions were included in each stage to permit the maximum number of sessions per stage without exceeding the total number of sessions elicited by any participant. The mean values for s-starts at the three stages for each ability level are displayed in Fig. 4 and those for sstops are shown in Fig. 5.

FIG. 4.

Mean values for s-starts at Stages 1, 2, and 3 for each level of ability.

20

WILLIAMON AND VALENTINE

FIG. 5. Mean values for s-stops at Stages 1, 2, and 3 for each level of ability.

A two-factor mixed ANCOVA with s-starts as the dependent variable, stage as the within-subjects independent variable, level as the between-subjects independent variable, and age as the covariate revealed highly signicant effects of stage [F (2, 34) 346.70, p .001] and level [F (3, 17) 529.87, p .001] and the interaction between stage and level [F (6, 34) 114.90, p .001]. s-starts increased as a function of stage and level, and the increase across the practice process was greater for higher levels of ability (see Fig. 4). The corresponding analysis for s-stops also revealed highly signicant effects of stage [F (2, 34) 63.23, p .001] and level [F (3, 17) 857.07, p .001] and the interaction between stage and level [F (6, 34) 37.48, p .001]. s-stops increased as a function of stage and level, and the increase across the practice process was greater for higher levels of ability (see Fig. 5). Starts and Stops on Difcult Bars The analyses presented above in The Role of Segmentation in Practice reveal that difcult bars were somewhat inuential in guiding practice although not as inuential as structural bars. To explore the extent to which difcult bars guided the pianists practice throughout the learning process, d-starts and d-stops were examined at the three stages of practice. The mean values for d-starts for each ability are displayed in Fig. 6 and those for d-stops are shown in Fig. 7. The values for d-starts were analyzed by a two-factor mixed ANCOVA with d-starts as the dependent variable, stage as the within-subjects independent variable, level as the between-subjects independent variable, and age as the covariate. The analyses revealed highly signicant effects of stage [F (2, 34) 64.79, p .001] and level [F (3, 17) 62.44, p .001] and the interaction between stage and level [F (6, 34) 78.21, p .001]. d-starts decreased as a function of stage and level, the difference between Stage 1 and Stage 2 increasing as a function of level of ability (see Fig. 6). Similarly, the analyses for d-stops revealed

RETRIEVAL STRUCTURES IN MUSIC

21

FIG. 6. Mean values for d-starts at Stages 1, 2, and 3 for each level of ability.

highly signicant effects of stage [F (2, 34) 87.37, p .001] and level [F (3, 17) 72.84, p .001] and the interaction between stage and level [F (6, 34) 73.28, p .001]. d-stops decreased as a function of stage and level, the difference between Stages 1 and 2 increasing as a function of level of ability (see Fig. 7). Structural Bars and Quality of Performance Considering the extent to which the use of structural bars to guide practice has been shown to increase as a function of skill, one might predict that the sooner musicians begin using them to guide their practice on a piece the more likely they will be to produce higher quality performances. To test this prediction, partial correlations were obtained between (1) the calculated values for s-starts and s-stops at Stages 1, 2, and 3 (see Figs. 4 and 5) and (2) the mean ratings on overall performance quality, musical understanding, communicative ability, and technical prociency of the pianists performances (see Williamon & Valentine, 2000, p. 364). For these analyses, correlation coefcients were obtained across all pianists. Possible differences in

FIG. 7.

Mean values for d-stops at Stages 1, 2, and 3 for each level of ability.

22

WILLIAMON AND VALENTINE

FIG. 8. Scatterplot of s-starts in Stage 2 and mean evaluations of overall quality of performance.

the relationship between these values and quality of performance for each level of ability were controlled for by partialing out level. (Controlling for level of skill in this way amounts to an assumption that the relationship between these variables is the same for the four levels; the small sample size prevented an analysis at each level, but the scatterplot shown in Fig. 8 is suggestive of this relationship.) The resulting correlation coefcients are listed in Table 2. The analyses indicate that s-starts was signicantly correTABLE 2 Partial Correlations between s-Starts and s-Stops at Stages 1, 2, and 3 and Ratings of Pianists Performances Overall quality s-starts Stage Stage Stage s-stops Stage Stage Stage 1 2 3 1 2 3 0.39 0.45* 0.33 0.21 0.08 0.26 Musical understanding 0.44* 0.52* 0.37 0.32 0.03 0.23 Communicative ability 0.45* 0.46* 0.39 0.26 0.05 0.28 Technical prociency 0.27 0.41 0.23 0.13 0.11 0.23

* Correlation is signicant at the .05 level (two-tailed).

RETRIEVAL STRUCTURES IN MUSIC

23

lated with musical understanding and communicative ability in Stage 1 (r .44, p .05; and r .45, p .05, respectively) and overall performance quality, musical understanding, and communicative ability in Stage 2 (r .45, p .05; r .52, p .05; and r .46, p .05, respectively). No signicant correlations emerged for s-stops.
DISCUSSION

The ndings presented above indicate that the use of hierarchical structures to organize practice and to function as retrieval cues is related to level of skill as follows: (1) the overall use of structural bars in starting and stopping practice segments increased with ability level, (2) the use of structural bars in starting and stopping practice increased with stage of practice (furthermore, this increase was a function of level), and (3) the early use of structural bars to guide practice was correlated with quality of performance. These results suggest that the identication and use of musical structure in guiding practice is a salient characteristic of skill and becomes increasingly so as a function of expertise. They conrm and extend the ndings of Ericsson and Kintsch (1995), on the importance of retrieval structures in skilled performance, to the musical domain. Thus, music practice appears to parallel the development of memory structures that support the execution of skilled performance in other domains. The present results also go beyond Chafn and Imrehs (1994, 1997) case study of an expert by examining the acquisition of skill in a group of pianists at differing levels of expertise and by demonstrating a continuity between the memory strategies of experts and novices. Stage and Level The results reveal that the pianists employed structural bars to guide their practice increasingly throughout the learning process (i.e., across Stages 1, 2, and 3) and that this increase was greatest for those at higher levels of ability. In fact, the data suggest that the higher level pianists began using structural bars to guide their practice in the early stages of the learning process (Level 4 pianists elicited positive deviations from expected starts by Stage 2 and positive deviations from expected stops from Stage 1). This may have occurred because the more skilled musicians were able to recognize structural bars in those early stages and continue to use them throughout the course of practice. Why did the less skilled musicians use structural bars less in the early stages of practice? They may have needed more time to decide upon or identify structural bars, possibly resulting from a lack of experience and skill at focusing their practice in that way. Alternatively, they may have been too encumbered with the sheer difculty of physically executing the piece in these stages to make efcient use of structural bars, instead depending more

24

WILLIAMON AND VALENTINE

on the local note-to-note detail to guide their rehearsals than on more global levels of understanding. Clearly, both reasons may explain why less skilled pianists employed structural bars to guide their practice less frequently than those at higher ability levels. As for difcult bars, all of the recruited pianists focused less on difcult bars as practice progressed. Specically, they started and stopped their practice on these bars less frequently from Stage 1 to 3, suggesting that difcult bars became less inuential in directing the course of practice as the pianists approached the nal performance. Clearly, these results, when viewed in conjunction with those on structure, demonstrate that the inuence of difcult bars in directing practice was increasingly replaced by the use of structural bars to guide rehearsal (i.e., d-starts and d-stops signicantly decreased across the three stages; s-starts and s-stops signicantly increased across the three stages). The analyses of this article go beyond the ndings of existing research by exposing between-level differences in practice on difcult bars (cf. the work of Miklaszewski, 1989, who examined a pianists rst four practice sessions on Debussys Prelude Feux dArtice). In fact, the signicant main effects of level and interactions between stage and level for both d-starts and d-stops suggest that the difcult bars were not only less important for those at higher levels of skill from the onset of practice but became practically negligible as they drew nearer to the time of performance. Again, when viewed in conjunction with the ndings on structural bars, these results reveal important insight into music cognition. The highly skilled pianists used structural bars more frequently than those at lower levels to guide their practice in the early stages of the learning process and increasingly did so as practice progressed. Hence, one may argue that, although the practice of higher level pianists was inuenced by difculty to some extent ( d-stops was positive in Stage 1 for pianists at all levels of ability), they were able to set difculty aside earlier when deciding where to start and stop their practice and they increasingly chose bars of structural importance when doing so. One may also argue that the higher level pianists started and stopped their practice more on structural bars and less on difcult bars because they were working to smooth out the performance of difcult passages and integrate those passages into run-throughs of larger sections of music. Such attempts to smooth performance are similar to those observed by Miklaszewski (1989) and can be evidenced in this study by exploring the comments made by pianists during practice. One pianist in Level 3 in his second practice session, after having spent considerable time rehearsing two difcult bars, remarked that he needed to practice leading into those bars. He then started the next practice segment on one of his identied structural bars and stopped at the beginning of another structural bar. Although such comments were made by pianists in Levels 1 and 2, they were infrequent (only two comments of

RETRIEVAL STRUCTURES IN MUSIC

25

this type were made, one by a Level 1 pianist and one by a Level 2 pianist) and did not occur until the nal stage of practice. Implicit/Explicit Knowledge Considering the signicant ndings for starting and stopping practice on certain types of bars presented here, some discussion must be directed at interpreting decisions to start and stop practice. Certainly, starting practice on a specic bar may suggest intention on behalf of the musician. For example, the Level 3 pianist mentioned above started a considerable number of practice segments on two difcult bars so that he could acquire greater facility in executing them. He then intentionally started his practice on a structural bar so as to practice leading into the difcult bars. Stopping practice on a particular bar, however, does not always signify intention. Instead, stops may result from a wide range of occurrences during practicesuch as reexive reactions to errors, a desire to end a practice session, or the interruption of practice by an outside source. Still, the data concerning structural stops have been interpreted as being used to guide practice. This is supported by two points. First, the identied structural bars did not coincide with identications of difculty (in the four cases in which they did, these bars were omitted from subsequent calculations). Second, stops on structural bars increased signicantly for all pianists across the practice process. Therefore, one may argue that stops on such bars were not the result of a breakdown in the physical execution of the piece or sheer difculty but, rather, were intentionalor at least they became more so as practice progressed. On the other hand, difcult bars were extremely challenging for a variety of reasons, and unlike structural bars, stops on those bars decreased signicantly from Stage 1 to 3 for all pianists. These ndings suggest that such stops resulted from a breakdown in the physical execution and that they, consequently, decreased in number as the pianists gradually overcame that difculty. Retrieval The assumption is that using structural boundaries as starting places provides practice in their use as retrieval cues. Existing research suggests that if individuals use a retrieval scheme during performance, their retrieval may be enhanced if they use the same scheme to encode the information (Tulving & Pearlstone, 1967; Baddeley, 1990) and specically practice using it to guide retrieval (Ericsson & Kintsch, 1995). Considering that (1) the identication of structural bars was based on the pianists reports of sections in the music that were important in both practice and performance and (2) these bars were increasingly exploited across the practice process, it can be argued that this exploitation was not only important for the encoding of musical information but also for its retrieval. The ndings of Williamon (1999) support this argument. He showed that two expert pianists did, indeed, use hier-

26

WILLIAMON AND VALENTINE

archically organized retrieval structures to guide their performances of a set piece. Theoretical vs Idiosyncratic Structure The denition of structure in this study differed from that of previous work. Chafn and Imreh (1997) found that their concert soloist used the musics formal structure to guide her practice. The postperformance interviews from the present study revealed that only three of the pianists reported explicit knowledge of the formal structure. Consequently, the participants individual-specic segmentations were used as the basis for identifying structural bars. This emergence of idiosyncratic structure supports Ericsson and Kintschs (1995) prediction that individual differences in the implementation of retrieval schemes are likely to emerge in the skilled performance of a given task and be more apparent at higher levels. Future work might examine the relation of idiosyncratic to formal structure (of obvious interest to musicologists) as a function of skill and contrast use of these two types of structure as predictors of nal quality of performance. Moreover, it might also explore the generalizability of these ndings to other types of music (e.g., atonal), in which structure may not be easily distinguishable. Pedagogical Recommendations The pattern of results obtained indicates that the use of structural boundaries to organize practice and form the basis of retrieval structures is intimately bound up with level of skill, either as cause or effect. In this article, a causal relationship has been argued in that use of such strategies resulted in better quality performances. However, Hallam (1997) has suggested, on the basis of observations of student practice, that the causality may lie in the opposite direction, in that ability to use a particular practice strategy may depend on having attained a particular level of skill. Regardless, the less skilled musicians in this study seemed less able to identify structural bars and/or overcome the technical difculties of certain musical passages. Music teachers, therefore, could possibly assist their students in achieving more efcient practice and ensure that musical information is learned thoroughly by striving to help identify musically meaningful structural bars for their less skilled pupils, instructing them to use such bars to guide practice and helping them to acquire the required technical facility to overcome difculties. Similarly, teachers should further stress the importance of structural bars in guiding practice to their highly skilled students, work with them extensively to identify why certain bars are difcult, and emphasize the strategy of smoothing out performance on difcult bars by integrating them into run-throughs of larger sections of music. No doubt, many teachers do stress these points to their students. Still, an increased use of such strategies could improve the effectiveness and efciency

RETRIEVAL STRUCTURES IN MUSIC

27

of practice. Certainly, the signicant correlations between increased use of structural bars and higher quality performances in this article support this notion. Those who started their practice on structural bars in Stage 1 of the practice process received higher ratings on musical understanding and communicative ability. This positive relationship persisted in Stage 2, suggesting that the use of structural bars in practice is, indeed, an important component of high-quality performances. Summary and Conclusions The data presented in this article reveal that the pianists segmented their assigned composition into meaningful sections and reported using those sections in both practice and performance. Empirical examinations of the pianists practice conrmed the pianists reports in that they, like the concert soloist in Chafn and Imrehs (1997) study, used structural bars more than difcult and other bars to guide their practice in preparing for the required memorized performance. Despite individual differences in the pianists identication of structure, the ndings were strongest for those at higher levels of skill, increased over the practice process, and their use correlated with quality of performance. Therefore, the identication and continued use of meaningful structure in practiceregardless of what that structure may beappears to be an ability that develops with musical competence. These ndings contribute to an understanding of music cognition in general by providing evidence for the use of retrieval structures as a prominent characteristic of musical skill and by extending previous research to examine this issue further at several levels of ability. Moreover, they support the arguments of Chase and Ericsson (1982) and Ericsson and Kintsch (1995) by demonstrating that musical performers appear to implement hierarchical retrieval structures in practice so that they may use them to guide retrieval during performance.
APPENDIX 1
TABLE A General Characteristics of the Assigned Compositions Level 1 Time signature Mean tempo (beats/minute) Number of bars (without repeats) Number of beats (without repeats) 3/4 76 16 48 Level 2 4/4 70 22 88 Level 3 9/16 76 38 114 Level 4 3/4 66 44 132

28

WILLIAMON AND VALENTINE

TABLE B The Structural and Difcult Bars Identied by Each Pianist Structural bars Level 1 2 3 4 5 Level 6 7 8 9 10 11 Level 12 13 14 15 16 17 Level 18 19 20 21 22 1 5, 5, 5, 5, 5, 2 7, 7, 5, 7, 7, 7, 3 7, 14, 20, 29 14, 26, 33 6, 14, 29, 35 14, 26, 33 14, 26, 33 4, 7, 11, 14, 17, 20, 26, 33 4 21 6, 13, 17, 25, 33, 36, 39 21 13, 21, 30, 36 6, 17, 21, 36 9, 10, 11, 36, 37, 38, 43 21, 28, 30, 31, 32 9, 15, 16, 25, 26 15, 16, 37, 38, 42 16, 28, 40, 42 19, 11, 26, 13, 13, 27, 26, 13, 27, 27, 16, 28, 27, 16, 28 28, 27, 32 28, 36 27, 28 36 28, 37 15 15 7, 12, 19 15 15 15 11, 13, 14, 19, 21 5, 11, 13, 14, 21 11, 13, 14, 15, 21 3, 5, 11, 13, 14, 21 5, 11, 13, 14, 21 11, 13, 14, 21 11 11 7, 11, 13, 15 11 11 3, 3, 3, 2, 3, 6, 6, 6, 3, 6, 10 10, 15 10 6, 7, 9, 10 7, 9, 10 Difcult bars

Note. The rst bar of each piece and all bars labeled as both structural and difcult were excluded.

RETRIEVAL STRUCTURES IN MUSIC

29

APPENDIX 2 The Equations and Calculations Used to Obtain the Values of Deviation of the Observed from Expected Starts on Structural, Difcult, and Other Bars ( s-Starts, d-Starts, and o-Starts, Respectively)
Equations Measure of deviation of observed structural starts from expected structural starts s-starts fs es e s fd ed e d fo eo e o

Measure of deviation of observed difcult starts from expected difcult starts d-starts

Measure of deviation of observed other starts from expected other starts o-starts

Calculations Step 1: The proportion of structural, difcult, and other bars to the total number of bars n si number of structural bars identied by pianist i n di number of difcult bars identied by pianist i n oi number of other bars identied by pianist i N i the total number of bars (n si n di n oi ) The proportion of structural, difcult and other bars to the total number of bars: p si p di p oi n si Ni n di Ni n oi Ni (proportion structural) (proportion difficult) (proportion other)

Step 2: The number of actual starts on structural, difcult and other bars f si number of observed starts on structural bars for pianist i f di number of observed starts on difcult bars for pianist i f oi number of observed starts on other bars for pianist i M i number of total starts (f si f di f oi ) Step 3: The number of expected starts on structural, difcult, and other bars (number of expected structural starts) e si p si M i (number of expected difcult starts) e di p di M i (number of expected other starts) e oi p oi M i

30

WILLIAMON AND VALENTINE

REFERENCES
Alba, J. W., & Hasher, L. (1983). Is memory schematic? Psychological Bulletin, 93, 203 231. Allard, F., Graham, S., & Paarsalu, M. E. (1980). Perception in sport: Basketball. Journal of Sport Psychology, 2, 1421. Allard, F., & Starkes, J. L. (1991). Motor-skill experts in sports and other domains. In K. A. Ericsson & J. Smith (Eds.), Toward a general theory of expertise: Prospects and limits (pp. 126152). Cambridge, UK: Cambridge Univ. Press. Anderson, J. R. (1990). Cognitive psychology and its implications. San Francisco: Freeman. Baddeley, A. D. (1990). Human memory: Theory and practice. Boston: Allyn & Bacon. Bartlett, F. C. (1932). Remembering: A study in experimental and social psychology. Cambridge, UK: Cambridge Univ. Press. Bland, J. M., & Altman, D. G. (1996). Measurement error. British Medical Journal, 313, 744. Boltz, M. (1992). Temporal accent structure and the remembering of lmed narratives. Journal of Experimental Psychology: Human Perception and Performance, 18, 90105. Carpenter, P. A., & Just, M. A. (1989). The role of working memory in language comprehension. In D. Klahr & K. Kotovsky (Eds.), Complex information processing: The impact of Herbert A. Simon. Hillsdale, NJ: Erlbaum. Chafn, R., & Imreh, G. (1994). Memorizing for performance: A case study of expert memory. Paper presented at the Third Practical Aspects of Memory Conference, University of Maryland. Chafn, R., & Imreh, G. (1997). Pulling teeth and torture: Musical memory and problem solving. Thinking and Reasoning, 3, 315336. Chase, W. G., & Ericsson, K. A. (1981). Skilled memory. In J. R. Anderson (Ed.), Cognitive skills and their acquisition (pp. 141189). Hillsdale, NJ: Erlbaum. Chase, W. G., & Ericsson, K. A. (1982). Skill and working memory. In G. H. Bower (Ed.), The psychology of learning and motivation (Vol. 16). New York: Academic Press. Chase, W. G., & Simon, H. A. (1973a). Perception in chess. Cognitive Psychology, 4, 5581. Chase, W. G., & Simon, H. A. (1973b). The minds eye in chess. In W. G. Chase (Ed.), Visual information processing (pp. 215281). New York: Academic Press. Chomsky, N. (1957). Syntactic structures. The Hague: Mouton. Clarke, E. F. (1988). Generative principles in music performance. In J. A. Sloboda (Ed.), Generative processes in music: The psychology of performance, improvisation, and composition (pp. 126). Oxford, UK: Clarendon Press. Collins, A. M., & Quillian, M. R. (1969). Retrieval time from semantic memory. Journal of Verbal Learning and Behavior, 8, 240248. Collins, A. M., & Quillian, M. R. (1970). Does category size affect categorisation time? Journal of Verbal Learning and Behavior, 9, 432438. de Groot, A. (1946/1978). Thought and choice in chess. The Hague: Mouton. (Original work published in 1946.) ` ge, I., & El Ahmahdi, A. (1990). Mechanisms of cue extraction in musical groupings: A Delie study of Sequenza VI for viola solo by Luciano Berio. Psychology of Music, 18, 1844. Ericsson, K. A., & Kintsch, W. (1995). Long-term working memory. Psychological Review, 102, 211245. mer, C. (1993). The role of deliberate practice Ericsson, K. A., Krampe, R. T., & Tesch-Ro in the acquisition of expert performance. Psychological Review, 100, 363406.

RETRIEVAL STRUCTURES IN MUSIC

31

Ericsson, K. A., & Oliver, W. (1989). A methodology for assessing the detailed structure of memory skills. In A. M. Colley & J. R. Beech (Eds.), Acquisition and performance of cognitive skills. Chichester, UK: Wiley. Ericsson, K. A., & Staszewski, J. J. (1989). Skilled memory and expertise: Mechanisms of exceptional performance. In D. Klahr & K. Kotovsky (Eds.), Complex information processing: The impact of Herbert A. Simon. Hillsdale, NJ: Erlbaum. Gobet, F. (1998). Expert memory: A comparison of four theories. Cognition, 66, 115152. Goodman, R. (1957). Teach yourself statistics. London: English Univ. Press. Hallam, S. (1997). Approaches to instrumental music practice of experts and novices: Implications for education. In H. Jrgensen & A. C. Lehmann, (Eds.), Does practice make perfect? Current theory and research on instrumental music practice (pp. 89107). Oslo: Norges Musikhogskule. Halpern, A. R., & Bower, G. H. (1982). Musical expertise and melodic structure in memory for musical notation. American Journal of Psychology, 95, 3150. Hatano, G., & Ozawa, K. (1983). Digit memory of grand experts in abacus-derived mental calculation. Cognition, 15, 95110. Hinsley, D. A., Hayes, J. R., & Simon, H. A. (1977). From words to equations: Meaning and representation in algebra word problems. In M. A. Just & P. A. Carpenter (Eds.), Cognitive processes in comprehension. Hillsdale, NJ: Erlbaum. Hinton, G. E., McClelland, J. L., & Rumelhart, D. E. (1986). Distributed representations. In D. E. Rumelhart, J. L McClelland, & The PDP Research Group (Eds.), Parallel distributed processing: Foundations (Vol. 1). Cambridge, MA: MIT Press. Johnson, N. F. (1970). The role of chunking and organization in the process of recall. In G. H. Bower (Ed.), The psychology of learning and motivation: Advances in research and theory (Vol. IV, pp. 171247). New York: Academic Press. Kendall, M. G., & Stuart, A. (1963). The advanced theory of statistics (Vol. 1). London: Charles Grifn and Company. Kosslyn, S. M. (1980). Image and mind. Cambridge, MA: Harvard Univ. Press. Kosslyn, S. M. (1981). The medium and the message in mental imagery: A theory. Psychological Review, 88, 4466. Kosslyn, S. M. (1987). Seeing and hearing in the cerebral hemispheres: A computational approach. Psychological Review, 94, 148175. Kosslyn, S. M. (1994). Image and brain: The resolution of the imagery debate. Cambridge, MA: MIT Press. Larkin, J. H., McDermott, J., Simon, D. P., & Simon, H. A. (1980). Expert and novice performance in solving physics problems. Science, 208, 13351342. Meyer, L. B. (1956). Emotion and meaning in music. Chicago: Univ. of Chicago Press. Miklaszewski, K. (1989). A case study of a pianist preparing a musical performance. Psychology of Music, 17, 95109. Narmour, E. (1977). Beyond Schenkerism: The need for alternatives in music analysis. Chicago: Univ. of Chicago Press. Newell, A., & Simon, H. A. (1972). Human problem solving. Englewood Cliffs, NJ: Prentice Hall. Newell, A. (1990). Unied theories of cognition. Cambridge, MA: Harvard Univ. Press. Palmer, C., & van de Sande, C. (1995). Range of planning in music performance. Journal of Experimental Psychology: Human Perception and Performance, 21, 947962. Piaget, J. (1967). The childs conception of the world. Totowa, NJ: Littleeld, Adams. Pylyshyn, Z. W. (1981). The imagery debate: Analogue media versus tacit knowledge. Psychological Review, 88, 1645.

32

WILLIAMON AND VALENTINE

Pylyshyn, Z. W. (1984). Computation and cognition. Cambridge, MA: MIT Press. Rosch, E., Mervis, C. B., Gray, W. D., Johnson, D. M., & Boyes-Braem, P. (1976). Basic objects in natural categories. Cognitive Psychology, 8, 382439. Rosenbaum, D. A. (1987). Successive approximations to a model of human motor programming. In G. H. Bower (Ed.), The psychology of learning and motivation: Advances in research and theory (Vol. XXI, pp. 153182). New York: Academic Press. Rosenbaum, D. A., Kenny, S., & Derr, M. A. (1983). Hierarchical control of rapid movement sequences. Journal of Experimental Psychology: General, 113, 372393. Rumelhart, D. E. (1980). Schemata: The basic building blocks of cognition. In R. Spiro, B. Bruce, & W. Brewer (Eds.), Theoretical issues in reading comprehension. Hillsdale, NJ: Erlbaum. Schneider, W., & Detweiler, M. (1987). A connectionist/control architecture for working memory. In G. H. Bower (Ed.), The psychology of learning and motivation (Vol. 21). New York: Academic Press. Sellen, A. J., & Norman, D. A. (1992). The psychology of slips. In B. J. Baars (Ed.), Experimental slips and human error: Exploring the architecture of volition. New York: Plenum. Shaffer, L. H. (1976). Intention and performance. Psychological Review, 83, 375393. Smith, E. E., Shoben, E. J., & Rips, L. J. (1974). Structure and process in semantic memory: A featural model for semantic decisions. Psychological Review, 81, 214241. Simon, H. A. (1986). The information-processing explanation of Gestalt phenomena. Computers in Human Behavior, 2, 155241. Simon, H. A. (1989). Models of thought (Vol. 2). New Haven, CT: Yale Univ. Press. Sloboda, J. A. (1985). The musical mind: The cognitive psychology of music. Oxford, UK: Oxford Univ. Press. Sloboda, J. A., & Parker, D. H. (1985). Immediate recall of memories. In P. Howell, I. Cross, & R. West (Eds.), Musical structure and cognition (pp. 143167). London: Academic Press. Sloboda, J. A. (1991). Musical expertise. In K. A. Ericsson & J. Smith (Eds.), Toward a general theory of expertise: Prospects and limits (pp. 153171). Cambridge, UK: Cambridge Univ. Press. Starkes, J. L., Deakin, J. M., Lindley, S., & Crisp, F. (1987). Motor versus verbal recall of ballet sequences by young expert dancers. Journal of Sport Psychology, 9, 222230. Staszewski, J. J. (1988). Skilled memory and expert mental calculation. In M. T. H. Chi, R. Glaser, & M. J. Farr (Eds.), The nature of expertise. Hillsdale, NJ: Erlbaum. Sternberg, S., Monsell, S., Knoll, R. L., & Wright (1978). The latency and duration of rapid movement sequences: Comparisons of speech and typewriting. In G. E. Stelmach (Ed.), Information processing in motor control and learning (pp. 117152). New York: Academic Press. Thorndyke, P. W., & Yekovich, F. R. (1980). A critique of schema-based theories of human memory. Poetics, 9, 2349. Tulving, E., & Pearlstone, Z. (1966). Availability versus accessibility of information in memory for words. Journal of Verbal Learning and Verbal Behavior, 5, 381391. Wall, R. (1972). Introduction to mathematical linguistics. Englewood Cliffs, NJ: Prentice Hall. Williamon, A. (1999). Preparing for performance: An examination of musical practice as a function of expertise. Unpublished Ph.D. thesis, University of London. Williamon, A., & Valentine, E. (2000). Quantity and quality of musical practice as predictors of performance quality. British Journal of Psychology, 91, 353376. (Accepted March 20, 2001; published online July 11, 2001)

Vous aimerez peut-être aussi