Sequence Memory in Music Performance

How do people remember and produce complex sequences like music or speech? Music provides an example of excellent sequence memory under fast performance conditions; novices as well as skilled musicians can perform memorized music rapidly, without making mistakes. In addition, musical pitches repeat often within a melodic sequence in different orders, yet people do not confuse the sequential ordering; temporal properties of musical pitches aid sequence memory. I describe a contextual model of sequence memory that is sensitive to the rate at which musical sequences are produced and to individual differences among performers. Age and musical experience differentiate adults’ and children’s memory for musical sequences during performance. Performers’ memory for the sequential structure of one melody transfers or generalizes to other melodies in terms of the sequence of pitch events, their temporal properties, and their movements. Motion-analysis techniques provide further views of the time course of the cognitive processes that make sequence memory for music so accurate. KEYWORDS—sequence learning; memory retrieval; motor learning; music performance. Music is one of the most complex sequential behaviors that people produce. People of varying musical experience can hum tunes lasting several minutes, with little practice. How are they able to do this? Musical sequences contain pitch events that must be produced in a particular order; like speech, the meaning will change if the units (tones) are reordered. Yet musicians can produce long melodies from memory with few errors (Finney & Palmer, 2003). Musical sequences are complex on many dimensions: Musicians must remember the patterning of finger, hand, or foot movements, as well as the patterning of pitches and durations. The pitch sequence is important, but it is not the only important dimension. For example, the first five notes in ‘‘Mary Had a Little Lamb’’ and ‘‘The First Noel’’ are identical, yet most listeners do not confuse these two melodies. That is because the time between tone onsets (forming musical durations) differs between them. Thus, the timing of music is necessary for distinguishing musical sequences in memory. Musical sequences can also be remembered in terms of how they are performed. For example, trombonists, clarinetists, and guitarists perform the same melody with different sequences of finger, arm, and hand movements. Which sequential aspects (pitch, motor, timing, etc.) are most important for performers? Are the sequence dimensions represented in memory independently? These questions are the focus of research on typing, handwriting, sports, and other sequential behaviors (cf. Schmidt & Lee, 1998); only music performance requires a hierarchy of prespecified times for when each sequence event must be produced (rhythms), and thus offers a good testing ground for understanding the time course of sequence memory. MEMORY FOR MOTOR AND PITCH SEQUENCES One approach is to examine how sequence memory transfers or generalizes from one performance to another. In an experiment using a transfer paradigm, pianists practiced one melody and then, as quickly as possible, performed a second melody that was the same as or different from the first melody in terms of either the motor sequence (the sequence of hand and finger movements) or the pitch sequence (e.g., A B C E). Comparisons of the amount of time the musicians took to perform the second melody relative to the first melody indicated both motor and auditory transfer of the melodic information (Palmer & Meyer, 2000). The more similar the two melodies were in the required hand and finger movements or in their pitch sequence (melodic-contour and interval information), the greater the transfer—that is, the faster they could perform the second melody. Most importantly, the two types of information had independent effects on the results. Novice child pianists were more dependent on motor information (hand and finger positions) than were skilled pianists; skilled pianists relied more on the sequence of pitches than on Address correspondence to Caroline Palmer, Department of Psychology, McGill University, 1205 Dr. Penfield Ave, Montreal QC H3A 1B1, Canada; e-mail: [email protected]. CURRENT DIRECTIONS IN PSYCHOLOGICAL SCIENCE Volume 14—Number 5 247 Copyright r 2005 American Psychological Society the sequence of motor movements (Palmer & Meyer, 2000). Thus, performers preserved pitchand motor-sequence information independently, and the importance of the two sequence dimensions changed with age and experience. Are motor sequences encoded as memory for when events should be produced? Motor-control research has focused on whether sequences of effector movements (limb movements) are encoded separately from timing information about the movements (Schmidt, 1975). For example, pianists may encode a musical sequence in terms of finger orderings, or in terms of the arrival times for each keypress. When the rhythmic sequence (patterning of keypress times) and the motor sequence (patterning of fingers to be used) were manipulated separately between two melodies that pianists performed in a transfer-oflearning task, the finger sequence transferred between melodies independently of the rhythmic sequence; the more similar the sequences were in their patterns of musical durations or finger movements, the more quickly pianists were able to perform the second melody (Meyer & Palmer, 2003). This research suggests that motor movements are not preserved in memory simply as a pattern of temporal information for when to move effectors, but instead as a sequence of finger orderings that is remembered independently of their timing. These findings extend what we know about effector-independent sequence representations from tasks like handwriting or typing to music performance, in which events must be performed not only in a certain order but also at precise times. INDIVIDUAL DIFFERENCES IN SEQUENCE MEMORY How do individual differences influence performers’ sequence memory? Music performance is a skill in which individual abilities differ widely, compared with other skills like handwriting or talking. Highben and Palmer (2004) compared performers’ memory for auditory information (the sequence of sounded pitches) with motor information (the sequence of finger movements) by removing auditory or motor feedback during performance. The feedback that pianists received during normal practice was replaced in the experiment with instructions to imagine the missing feedback: how the piece sounded or how the finger movements felt. After performers practiced with a musical score under reduced-feedback conditions, pianists performed from memory under normal-feedback conditions (i.e., with all feedback). Independent tests of auditory-imagery ability and motor-imagery ability were collected as well. Performance from memory showed significant effects of removing motor or auditory feedback during practice; pianists’ memory for the music, based on pitch errors, was worst when both types of feedback were unavailable during learning. There were important individual differences in how each type of feedback affected performance: Pianists who scored high on the aural-imagery test performed best from memory following the removal of auditory feedback during practice, compared with pianists who scored low. Musicians’ aural-imagery skills were predictive of how much they relied on auditory feedback during learning to perform a novel melody. Removing motor-feedback did not differentiate pianists; all pianists scored high on motorimagery skills. These findings confirm that an accurate auditory image is important for successful performance from memory (McPherson, 1997), and individual differences exist in the extent to which memory for musical sequences is encoded in motor movements and in auditory images (Highben & Palmer, 2004). ROLE OF WORKING MEMORY How much of a musical sequence is available in memory during performance? Despite anecdotal accounts such as that of Mozart’s ability to remember an entire choral chant after one hearing, most psychological studies indicate that there are strict limits on the amount of sequence information available in memory during performance. Serial-ordering errors (sequence events that are produced in the wrong order, such as ABCD being reordered as ADCB) often tell us which sequence events are active in memory at a given time. For example, a speech error such as ‘‘He brought the store’’ instead of the intended ‘‘He brought the book to the store’’ indicates that ‘‘store’’ was available three words earlier than its intended sequence location. Drake and Palmer (2000) found that child and adult musicians’ pitch-ordering errors tended to arise from sequence events within a range of 3 to 4 pitches, with older musicians’ errors reflecting events farther away than younger musicians. One correlate of age that may account for these differences is working memory: a temporary store of information necessary for ongoing complex tasks. Palmer and Pfordresher’s (2003) measures of pitch-ordering errors in music performance also spanned 3 to 4 events on average; in performances at faster tempi, the range was smaller, and at slower tempi, the range was larger. Furthermore, two pitches tended to be substituted for each other if they shared the same stress or metrical accent, analogous to similarity-based principles that affect speech errors and other memory lapses (Dell, Burger, & Svec, 1997). Palmer and Pfordresher (2003) proposed a quantitative model of sequential-memory retrieval that predicts which sequence events musicians can remember during performance. The model’s predictions are shown in Figure 1; the black bars indicate the model’s predictions for how active or accessible each tone is at the time at which the performer is producing the circled tone. The model is based on two common memory processes: interference and decay. The graded descent in event strength indicates that nearby events are more accessible than events farther away from the tone currently being produced. The rate of decrease is related to tempo: the faster the performer plays, the less time available for retrieval and the steeper the graded descent. The model also predicts that the greater the performer’s working-memory capacity, the less steep the graded descent in memory strength. The peaks within that graded descent, shown in Figure 1, indicate that 248 Volume 14—Number 5 Sequence Memory in Music Performance