Implicit Memory Relations Between Contemporary Musical Materials

The perception of similarity between materials from a nontonal musical work entitled The Angel of Death by Roger Reynolds was examined in three experiments. The implicit memory relation between themes and their electroacoustic transformations was tested with two recognition paradigms. The first experiment studied the priming effect in which earlier presentation of an original theme facilitates memory coding of its transformed version. The second experiment investigated the extent to which the themes could refresh the memory trace of a previously presented transformation. The results of Experiment 1 revealed no priming effect, i.e. previous exposure to themes does not enhance recognition of transformations. In Experiment 2, there was a partial significant facilitating effect of the theme-transformation relation for the recognition task, almost certainly depending on the musical density of transformation used. In a third experiment, musical density was equalized between electroacoustic transformations to test once again the refreshing effect of themes on transformed versions. No significant facilitating effect was observed, suggesting that musical cues of similarity between themes and transformations were hardly perceived. These results are discussed with respect to the memory for surface features in nontonal music.

Many studies (Krumhansl & Castellano, 1983; Shepard & Jordan, 1984) have demonstrated that musical schemata in long-term memory influence the way musical events are encoded and memorized. Such cognitive schemata correspond to the knowledge acquired from repeated exposure to musical pieces belonging to a given culture. Indeed, simple exposure to tonal music would be sufficient for a listener to acquire implicit knowledge concerning the regularities of this musical system (Tillmann, Bharucha & Bigand, 2000). Implicit learning allows us to acquire highly complex information without conscious and controlled processing. Listeners generate expectations about what incoming events are likely to be encountered and when they are likely to do so. The way these expectations are more or less fulfilled by the composer contributes to the perception of musical expression (Meyer, 1956; Jackendoff, 1991). One way of generating these expectations consists of varying and developing thematic materials. In Western music, development and variation are usually primed by the exposition of the themes, but they can also prime the exposition of the themes in their literal form. The memory for these related auditory events based on a certain degree of similarity between various aspects of the materials and structure guides listeners' attention and contributes to the cognitive integration of the piece.

According to Gernsbacher (1990), development and variation in music may act in a similar way to language by contributing to the integration of all the sections of a discourse into a whole. In their text comprehension model, Kintsch & Van Dijk (1975, 1978) demonstrated that memory traces of stimuli are achieved by focusing attention on a selection of key elements of the perceived structure. In music, such key elements could correspond to what McAdams (1989) called form-bearing dimensions. They refer to certain categories of auditory attributes such as rhythm, duration, pitch, and timbre to which listeners are sensitive and from which they can distinguish different levels of relations among musical events. It has been pointed out by McAdams & Matzkin (2003) that the most crucial link between two different musical patterns is the degree of similarity (i.e. a theme followed by its variation). Thus, the comprehension of musical form would depend on the perception of invariant cues between musical events. A number of studies have addressed the question of similarity in music (Dowling, 1972; Krumhansl, Sandell & Sergeant, 1987; Pitt & Monahan, 1987; Bartlett & Dowling, 1988; Matzkin, 2001; McAdams & Matzkin, 2003), and demonstrated that listeners are able to extract prominent cues from the musical surface in order to get a coherent representation of the musical structure.

Recent work by Deliège (2001) concerned the implicit memorization of prominent cues from music. She developed the notion of imprint formation to designate the processes involved in music perception and more specifically to describe the perceptual consequence of the insistent use of the same cue in more or less varied forms within a given musical piece. According to her, variations of a given musical cue would not be memorized in all their diversity and complexity. Memorization is based on the simplification and reduction of inputs to integrate them into a network of knowledge. Given this rule of information coding in memory, the concept of imprint implies the retention of critical information about a collection of more or less similar presentations of a given cue. This notion of imprint is based on the model of prototype according to which a number of items possess a set of properties originating from a basic model. The formation of an imprint would largely depend on the features memorized from previous listening. It would be determined by the content of the schematic knowledge in implicit memory. As outlined in Berz (1995), tonal structure is certainly an organizational element that influences the memorization of musical events. Some researchers have argued that the memorization of a simple tonal melody merely involves the elaboration of a mental model in which few cues from the musical surface are needed (cf. Sloboda & Parker, 1985). In other words, for a tonal piece, the automatic activation of musical structure kept in memory would suffice to convey relevant cues for recognition. A number of studies used melodic contour and pitch height as the most prominent cues for the processing of musical information. For instance, Dowling & Fujitani (1981) demonstrated that contour information was relatively easy to encode. Dewitt & Crowder (1986) revealed that contour and intervallic content could be easily retained over time.

With an unconventional musical structure, it seems more difficult for listeners to situate musical information within some kind of previously memorized organizational framework. Without previous knowledge of a new musical structure, more surface features would probably be needed to identify similarities between two musical events. In other words, memorization of atonal musical sequences probably first requires the encoding of musical surface features rather than the internal representations of a learned stylistic norm. Concerning the surface features of atonal music, listeners' abilities to encode and memorize them have been clearly demonstrated (Deliège, 1989, 2001). Moreover, Lerdahl (1989) argued that listeners should be able both to encode surface characteristics of atonal music and to elaborate more abstract representations of its structure. One can hypothesize that surface features would require the processing of sound events, while the elaboration of abstract representations would require the integration of the relation between musical events and the hierarchical structure of the piece.

In a study intended to investigate memorization ability for surface features and the more abstract representation of these features in atonal music, Krumhansl (1991) assessed the effect of pitch change on musically trained listeners' recognition memory for different sequences from a contemporary piece by O. Messiaen. Her goal was to test recognition memory performance with an unfamiliar musical style for which listeners did not possess schematic knowledge. In this study, the second half of the piece was transformed in different ways in order to change the pitch independently of other musical dimensions in the piece. Some transformations were constructed to preserve the melodic contour and the rhythmic framework, whereas others modified the specific coupling between values of pitch, duration, and dynamics. The task consisted of presenting the original first half of the piece followed either by one of the several transformations of materials from the first part (transformed version) or by original excerpts from the first (original old version) or second (original new version) parts of the piece. For each pair of sequences (original plus one of the transformed or old/new original versions), listeners were asked to judge whether or not the latter sequence came from the original part of the piece. The repeated listening effect of the original part of the piece on recognition performance was also investigated. Krumhansl (1991) supposed that (a) if the original old version was accurately recognized, this should indicate that listeners memorized the surface features of music; (b) if the original new version was well recognized it should signify that listeners were able to elaborate and memorize a more abstract representation of surface features in order to generalize them to another part of the piece. Finally, she hypothesized that the recognition performance for the different modifications of the piece's pitch structure should indicate the listener's sensitivity to different kinds of surface regularities. First of all, her findings revealed that repeated exposure to the original first part of the piece had no effect on memory performance, suggesting that the recognition process was not based on memory trace reinforcement, but rather on a rapid comparison between perceived immediate musical cues. The results also clearly showed that listeners both encode and abstract the surface features of the music. Furthermore, transformations preserving the correlation between pitch height and duration of the original versions were highly recognized, whereas those not respecting this relation were not recognized as belonging to the original version. This suggests that the pitch-height/duration relation, and more generally the melodic contour, constitute a salient feature for extracting similarity cues. In order to test the influence of previous musical knowledge on memory recognition processing, Krumhansl (1991) carried out the same task with professional musicians to whom the musical structure of the piece was described. Interestingly, findings revealed that recognition performance was strongly similar to that observed for nonprofessional listeners unfamiliar with the piece's structure. Such results surprisingly reveal that familiarity with the structure of the piece does not change the musical criteria used to judge similarity between musical sequences. More generally, these data demonstrate that listeners possess a long-term memory capacity to encode, maintain, and elaborate abstract representations of surface features within a musical piece, whatever the level of familiarity with its structure.

In the present study, the recognition of perceptual invariance between original and transformed versions from The Angel of Death by Roger Reynolds was studied through implicit memorization of similarity relations. This study specifically concerns the elaboration of implicit musical association between excerpts from an unconventional style of music. We assume that similarity relationships between thematic cells and their transformations create associative links between different sections of the piece. Our aim was to investigate whether these links result in memory priming and refreshing effects.

The specificity of the relation between themes and variations in The Angel of Death is that some variations in the electroacoustic part of the piece consist of a systematic segmentation of original thematic materials according to a particular transformation algorithm. The composer has created two kinds of algorithms: SPIRLZ and SPLITZ. Both transformations, illustrated in Figure 1, keep the nature of original sound material but impose different changes on their temporal structure. Memory for similarity cues between themes and their electroacoustic transformations certainly requires the ability to abstract common characteristics shared between them.

In the SPIRLZ algorithm, the material is considered as a series of contiguous segments. The algorithm rearranges these segments in time, starting somewhere in the middle and then progressively concatenating segments taken alternately from before and after those already used. It performs this operation in several cycles and the size of the segments and the starting segment can be changed to create an accelerating or decelerating impression. The composer conceives of the operation as a sort of spiraling through the original material.

The SPLITZ algorithm is a transformation that also consists of a segmentation of the original excerpt. The even- and odd-numbered segments are separated spatially (into right and left channels in our case) and the odd ones are played in chronological order, while the even ones are played in reverse order. Different amplitude envelopes are also applied to the even and odd segments. The relative durations of the segments are defined according to a number series derived from one of those used in The Angel of Death. This transformation gives the impression of a very irregular sound with many silences from which a multiply reflecting hall-mirror impression can result.

According to the composer, both algorithms cut the original excerpts into constituent segments in order to rearrange them in a particular way that modulates their original structure. This modulation affects the time domain of the musical structure by imposing a new formal idea on the original source. Through such transformations, the composer algorithmically applied fragmenting identities to original excerpts from the thematic materials of the piece, similarly to transformations applied in the making of the computer layer of the piece at a later stage in the compositional process.

With such materials, the question is whether the similarity of the materials in the themes and variations creates implicit knowledge about their relationships. First, we wondered if listening to the original themes could facilitate recognition of their algorithmically transformed versions. To test this question, we first carried out a priming experiment. Two other experiments were conducted in order to test the refreshing effect (re-exposition) of the transformed versions on their subsequent recognition.

The priming effect refers to the increase in accuracy of recognition that occurs as a consequence of the prior exposure to musical information shared by original and transformed versions, without any intention or task-related motivation. In this experiment, the effect of the previous presentation of original excerpts (prime) was tested on the recognition of their subsequently memorized transformations.

Thirty participants performed the experiment. All of them were unfamiliar with the musical piece. Participants were recruited without regard to musical training because the study initially aimed to test implicit memory for unusual musical events. As a result, subjects had different levels of musical experience: 17 were musically untrained and 13 had from 4 to 20 years of instrumental practice. Each participant was randomly allocated to one of two experimental conditions.

Eight orchestral thematic excerpts were used as primes. Their duration varied from 2.4 to 30.7 s. For each prime, one transformed version was constructed (four SPLITZ and four SPIRLZ). Six transformations of other thematic excerpts were also created as foils. They were selected from thematic materials so as to be not too distinct but sufficiently different from the target excerpts. The duration of the transformations was constant (24 s). At the request of the composer, a small amount of reverb was added to the transformed sound files to avoid reverberation discontinuities at the segment boundaries.

The experiment was run on a Macintosh computer with the PsyScope software (Cohen, McWhinney, Flatt & Provost, 1993). Musical excerpts were presented over a Sennheiser HD 520 II headphone.

The experiment consisted of three phases: 1) exposure to primes, 2) exposure to targets, and 3) target recognition (Figure 2). In the first phase, four thematic excerpts were presented in random order one after the other. These four excerpts were picked from among eight possible thematic primes according to the experimental condition (C1 versus C2, see Figure 2) being carried out by the listener. During the exposure of primes, the participants were asked to simply listen to them once each. They were only asked to be attentive to each one. In the second phase, eight electroacoustic transformations were presented. Half of them (either four SPLITZ or four SPIRLZ as function of the excerpts presented in phase 1) were the related transformed versions of the four primes. Before listening to the set, participants were warned that they had to memorize each electroacoustic excerpt. Finally, in the recognition phase, the same eight transformations were presented in random order along with six new transformations that had no relation to the primes (i.e. they were transformations of other thematic materials from the piece). Participants had to indicate whether the transformation had been heard previously during the memory phase by pressing either the "yes” or "no” key on the computer keyboard. They were asked to rate their degree of certainty on a 6-point scale displayed on the computer screen.

Two groups of participants performed the experiment, each being primed by a different half of the set of thematic excerpts. Figure 2 illustrates the three phases of the experiment and the relation between the musical stimuli. It shows that all participants heard the same transformations in the memory and recognition phases but were not primed by the same thematic excerpts. This experiment lasted about 20 minutes.

The assumption was that the four transformations previously primed by their original themes would be easier to recognize than the unprimed ones, because their coding in memory would be facilitated by previous exposure to the primes. Moreover, we expected that listeners would easily reject the electroacoustic transformations that had not been heard previously.

Recognition performance was estimated by calculating the mean correct responses weighted by the rated degree of certainty on a 6-point scale. For instance, if a participant gave a correct response with a certainty rating of 4, the score was 4. If an erroneous response was given with a certainty of 5, the score was "5.

First of all, the mean correct rejection rate was calculated for new transformations not previously heard. The mean correct rejection value reached 88% showing the listeners' ability to distinguish previously heard transformations from those that had not been presented previously. Moreover, the mean certainty rating for correct rejections calculated for each new transformation gave values varying between 2.3 and 4.8 (mean value = 4). Such results indicate that listeners did not respond at random and easily distinguished between electroacoustic transformations that were previously memorized or not.

A repeated-measures analysis of variance (ANOVA) with between-subjects factor Condition (C1 versus C2), and within-subject factor Priming (primed versus not primed) was performed on recognition scores. Surprisingly, this analysis did not show a significant effect of the priming relation on recognition performance for either transformation set. This result suggests that the previous presentation of the original themes did not facilitate the recognition of their memorized transformed versions. It appears that similarity between thematic excerpts and transformations does not result in a priming effect.

The high mean rate of correct rejection for new transformations indicates that listeners can encode and store sufficient prominent musical cues to distinguish different electroacoustic transformations belonging to the same musical piece but coming from different sources. However, it seems that the memorization of transformations did not benefit from the prior presentation of the thematic units to which these transformations were applied. In other words, the results could suggest that the coding in memory of transformations was not enhanced by the previous presentation of the related musical events.

In Experiment 2, we investigated the potential effect of the relation between thematic materials andtransformations with a slight change in procedure. This time, the transformations to be memorized were presented first. Half of them were then refreshed in memory by the presentation of the four themes, and the recognition test still occurred at the end.

A memory-refreshing paradigm provides a new way to examine the extent to which the exposure to a theme allows the memory traces of related transformed versions to be "revived" and reinforced for their subsequent recognition among new items.

Fifty-eight new participants performed this second experiment. They had never heard the musical materials from The Angel of Death before. Twenty-three had no musical training and 35 had instrumental practice (3-20 years). As in Experiment 1, each participant was randomly allocated to one of two experimental conditions.

The stimuli and equipment were identical to those of Experiment 1.

In the first phase, participants were presented with eight electroacoustic transformations for which they gave aesthetic judgments. In the second (refreshing) phase they again made an aesthetic judgment for four thematic excerpts related to four of the previously heard transformations. In the last (recognition) phase, they had to recognize the transformations presented in the first phase. The aesthetic judgment was a pretext to encode each excerpt more deeply in memory. The three phases of the refreshing task are illustrated in Figure 3. We assumed that refreshed transformations would be better recognized than those that were not refreshed and that the new transformations would be rejected.

As in Experiment 1, the mean correct rejection of previously unheard transformations was very high (85%) and the mean certainty rating reached 4 (varying between 1.0 and 5.8). Once again, listeners showed a good ability to distinguish already memorized transformations from other new electroacoustic transformations.

The critical point of this study was to assess whether the refreshed transformations (primed in Experiment 1) were easier to recognize than the others. The ANOVA with between-subjects factor Condition (group C1 vs. group C2), performed on recognition scores of transformed versions refreshed either in Condition 1 (SPIRLZ transformations) or Condition 2 (SPLITZ transformations) revealed, as displayed in Figure 4, a significant interaction of Condition X Refreshed transformation (F(1,56) = 6.64; p = .0126). Two other ANOVAs with between-subjects factor Refreshing transformation were separately performed on recognition scores for SPIRLZ and SPLITZ transformations. These analyses show that the four SPIRLZ transformations were recognized more often when they were refreshed than when they were not (F(1,56) = 4,53; p= .038). However, the results do not show a significant effect of Condition on recognition scores for SPLITZ transformations. This finding raises the question of the equivalence of the two refreshing conditions based on different kinds of electroacoustic transformations.

In the recognition phase, as has already been shown in Experiment 1, listeners easily managed to reject the new transformations not previously heard above chance level. This correct rejection score strongly suggests that the to-be-memorized transformations are well encoded and that memory traces are maintained over time to govern the recognition process.

Moreover, our results indicate that electroacoustic transformations are better recognized when they are refreshed by their related original themes. This supports the idea that the memory traces of the transformations can be properly elaborated as well as reinforced by intervening exposure to the original version. However, findings reveal an asymmetry of refreshing condition, suggesting a disparity between the refreshing effects of the two transformed versions. Indeed, the type of algorithm likely influences the perception of its structure and content. This raises questions about the specificity of each transformation algorithm and the ease with which it can be encoded and memorized. As explained above, SPLITZ transformations give the impression of a very irregular sound sequence with many silences. Such silences may make the encoding of the transformations' global shape more difficult, thus weakening their memory traces and consequently making identification of similarity between transformed and original versions more problematic. This illustrates and underlines the difference in sound density between SPIRLZ and SPLITZ transformations that may have an effect on the perception of similarity between materials.

In order to control for this factor, we conducted a third experiment with a new set of transformations derived from the previous, but for which we harmonized the musical density. Moreover, to suppress the confounded factors of groups and kind of transformation, we created additional stimuli in order to counterbalance transformed versions within refreshing conditions. We expected that sound density harmonization would render the SPLITZ transformations as easily encodable and memorizable as SPIRLZ transformations. In this third experiment, new listeners were asked to perform exactly the same refreshing task described previously.

In this third experiment, we used exactly the same refreshing paradigm as before and changed our stimuli by generating a new set of SPIRLZ and SPLITZ transformations of The Angel of Death thematic materials with similar sound event densities. For compositional reasons, all of these new stimuli came from the piano versions of the thematic materials. For each original piano theme, two electroacoustic transformations (one SPIRLZ and one SPLITZ) were applied. This design allowed us to examine the differential effect of the transformation algorithm (SPIRLZ versus SPLITZ).

Fifty-seven participants performed the experiment. They were not familiar with Reynolds' music. Thirty-four had no musical training and 23 had instrumental practice (3-20 years). Each participant was randomly allocated to one of four experimental conditions of the experiment.

As mentioned above, new sets of SPIRLZ and SPLITZ transformations were designed to have equivalent event densities. Each of the eight original excerpts, with durations of about 18 s, was transformed with both algorithms. Sixteen transformations were then constructed: half with SPIRLZ and half with SPLITZ. Four groups of listeners were formed in order to counterbalance the refreshed and unrefreshed transformations as a function of the kind of electroacoustic transformation (SPIRLZ versus SPLITZ). Finally, given the additional conditions, another set of new transformations was constructed to obtain an equal number of never-heard and already-heard transformations.

The equipment and procedure were identical to those of Experiment 2. The hypotheses were also similar: we expected that refreshed transformations would be better recognized than those that were not refreshed, whereas the new transformations would be rejected.

The mean correct rejection rate and mean certainty rating were 78% and 3.4 (range: 1.0-5.6), respectively. Once again, the results show that listeners succeeded in performing above chance. This indicates that in spite of the similarity between the transformed versions' temporal structures, listeners succeeded in distinguishing new transformations from those they had already heard.

A repeated-measures ANOVA with between-subjects factor Condition (the two halves of the sets of transformations for both SPLITZ and SPIRLZ algorithms) and within-subject factor Refreshing (refreshed versus unrefreshed) were performed on recognition scores. Results showed that whatever the kind of electroacoustic transformation and the nature of the original excerpts, there was no significant effect of refreshing on the recognition scores. No facilitation function for subsequent memory processing was provided by the relationship between original excerpts and transformations.

Despite the harmonization of sound density between electroacoustic transformations, no significant effect of refreshing context was demonstrated. This result suggests that sound density was not the crucial factor in recognition performance in the preceding experiment. Because the refreshing phase seems to have no influence on subsequent recognition, one wonders whether listeners are able to perceive the similarity relations between original themes and their transformed versions. When asking them about the similarity between certain transformations, it appears that most of the listeners had no conscious awareness of such a relationship or encountered difficulty in distinguishing the specificity of the transformed versions, because only the difference between the two categories of algorithms was prominent enough to be really perceived. That is, they only noted differences between SPLITZ and SPIRLZ transformations. Finally, it appears that the reinforcement of memory traces is not obvious and likely depends on additional variables.

In this study, we assumed that similarity relations between themes and transformations would result in priming and/or refreshing effects. Thematic excerpts from of The Angel of Death and their respective transformation through Reynolds' SPLITZ and SPIRLZ algorithms were presented to participants. In Experiment 1, we expected that thematic units would prime the processing of the respective transform, which would result in better recognition performance of these primed transformations. In Experiments 2 and 3, we assumed that recognition of to-be-memorized transformations would benefit from the subsequent presentation of the thematic cells from which the transformations were derived.

Regarding the overall findings, it appears that they fail to give clear evidence of memory trace reinforcement for transformations previously primed or subsequently refreshed by their related original themes. On the one hand, these findings would suggest that participants did not perceive the similarities between themes and their SPLITZ and SPIRLZ transformations. On the other hand, it may also be that participants actually perceived these associations, but that they did not result in either priming or refreshing effects.

Several points may explain this weak effect of similarity on priming and refreshing. The first concerns the nature of the memory test we used to investigate implicit memory for similarity cues between transformations and original themes. Implicit and explicit memories are considered to refer to distinct retrieval processes and different procedures of testing. In explicit memory tasks, the measures are generally achieved by free or cued recall paradigms or recognition tests, and participants are informed that their memory will be tested. This explicit processing involves controlled retrieval effort, intention and awareness on the part of participants. Inversely, the implicit memory task does not use explicit instructions and aims to avoid participants developing memory retrieval strategies and to ensure the involvement of automatic and unaware memory processes.

It is likely that the explicit method used in the current experiments to investigate implicit elaboration in memory engages two levels of processing that compete to govern the behavioral responses. Indeed, explicit memory recognition might conceal implicit processing because listeners were informed that they had to recognize previously heard transformations among many others, leading them to focus their attention on certain musical features. This goal-oriented activity requires encoding and recovering strategies for retrieval that may prevent the implicit processes to be captured. Further studies with indirect measures are needed to measure the implicit retrieval of similarity.

Another question concerns the interaction between short- and long-term memory processes. Gerrig & McKoon (2001) used the concept of resonance to design the memory process by which inputs stored in working memory (i.e. perceptual cues) interact with those contained in long-term memory. Such interaction would determine the accessibility of information over time and would depend on the degree of association between pieces of information. This theory emphasizes the role of previous knowledge for information integration. In our study, the musical events were unconventional and did not necessarily match with preexisting knowledge schemata in long-term memory. Therefore, the salient musical cues extracted from transformations in working memory would perhaps not match sufficiently with representations in long-term memory for recognition. This raises the question of memory trace stability for this kind of musical event. Accordingly, our findings could be explained by the difficulty in maintaining dynamic traces of transformations because of their unconventional and complex structure, which was unlikely to be related to musical schemata previously stored in memory. In comparison with tonal musical materials, the transformational algorithms are defined by unusual attributes like the sound texture and the effects of stretching or condensing to which they give rise. This underlines the strong influence of previous knowledge on the integration of new musical structures in memory.

Furthermore, the relationship between original excerpts and transformations is based on the communality of pitch height and texture, while they differ in their duration structures and pitch contours. There is strong agreement that melodic contour and rhythmic configuration constitute prominent features for the perception of similarity between musical events (cf. Matzkin, 2001). It may be that transformations do not preserve the essential surface features required for the long-term memory integration, thus limiting capacities for storage. As a result, the prominent musical features might be temporarily maintained in working memory, but, without sufficient strength of association with previous schematic knowledge, such memory traces would be more volatile. Further, the interpolation of unrelated transformations in the recognition phase could suffice to disrupt musical information held in working memory store. This interference could be important especially because transformations that served as foils had a musical structure that was similar (i.e. produced with the same transformation algorithms) as the targets.

As McAdams and Matzkin (2003) emphasized, the feasible range of variation in cognition for the memorization of similarities is limited and has to take psychological constraints into account. One of these constraints could be a memory-based context dependency related to the local versus global musical context in which recognition processes are involved. Reynolds has pointed out that "musical form requires considerable duration in which to make its optimum impact, and this is even more true when the stylistic, the musico-linguistic terrain one is inhabiting is unfamiliar." [Cf. The time line of the compositional process and Computer transformations]. This emphasizes the necessity of taking into account the global context of music perception in order to measure to a finer degree the impact of the previously heard transformations on recognition. This supports the important role of the global musical form in the memorization process and underlines the fact that familiarization with musical context seems to be determinant in the imprint formation for the recognition of related musical events. Deliège (2001) demonstrated that the time over which listeners were exposed to a given piece strongly influenced their recognition performance. Thus, the more often they heard the piece, the better their recognition was. This suggests that the rehearing of a given musical event would have a clear effect on the sedimentation of the abstracted cues in memory, leading to a better quality of memory trace.

In conclusion, the explanatory framework above raises the crucial question of the psychological role that algorithms used for transformations might play in Reynolds' music. It may be that listeners need more exposure to their features in order to elaborate more stable memory traces. From an empirical standpoint, this motivates further testing of memory for similarity in a repeated global presentation context using indirect measures for a better evaluation of the cognitive system's plasticity for new musical events.

This project was funded by the Cognitique program of the French Ministry of Research. The generation of the SPLITZ and SPIRLZ transforms was made possible with the generous technical help of Benjamin Carson who resurrected at Ircam the algorithms, originally programmed at UCSD, and of Roger Reynolds who chose the parameters for all of the stimuli used.

• Bartlett, J. C. & W. J. Dowling (1988). Scale structure and similarity of melodies. Music Perception, 5, 286-313.
• Berz, W. L. (1995). Working memory in music: A theoretical model. Music Perception, 12, 353-364.
• Cohen, J. D., MacWhinney, B., Flatt, M., & Provost, J. (1993). PsyScope: An interactive graphic system for designing and controlling experiments in the psychology laboratory using Macintosh computers. Behavior Research Methods, Instruments and Computers, 25(2), 257-271.
• Deliège, I. (1989). A perceptual approach to contemporary musical forms. Contemporary Music Review, 4, 213-230.
• Deliège, I. (2001). Prototype effects in music listening: An empirical approach to the notion of Imprint. Music Perception, 18, 371-407.
• Dewitt, L. A., & Crowder R. G. (1986). Recognition of novel melodies after brief delays. Music Perception, 3, 259-274.
• Dowling, W. J. & Fujitani, D. S. (1971). Contour, interval, and pitch recognition in memory for melodies. Journal of the Acoustical Society of America, 49, 524-531.
• Dowling, W. J. (1972). Recognition of melodic transformations: inversion, retrograde, and retrograde inversion. Perception and Psychophysics, 12, 417-421.
• Dowling, W.J., and Fujitani, D.S. (1971).Contour, interval, and pitch recognition in memory for melodies. Journal of the Acoustical Society of America, 49, 524-531.
• Gernsbacher, M. A. (1990). Language Comprehension as Structure Building. Hillsdale, N.J., Lawrence Erlbaum Associates.
• Gerrig, R. J. & McKoon, G. (2001). Memory processes and experiential continuity. Psychological Science, 12, 81-85.
• Jackendoff, R. (1991). Musical parsing and musical affect, Music Perception, 9, 199-230.
• Kintsch, W., & Van Dijk, T.A. (1975). Comment on se rappelle et on résume des histoires. Langages, 40, 98-116.
• Kintsch, W., & Van Dijk, T.A. (1978). Toward a model of text comprehension and production. Psychological Review, 85 (5), 363-394.
• Krumhansl, C. L. & Castellano, M. A. (1983). Dynamic processes in music perception. Memory and Cognition, 11, 325-334.
• Krumhansl, C. L. (1991). Memory for musical surface. Memory and Cognition, 19, 401-411.
• Krumhansl, C. L., G. J. Sandell & D. C. Sergeant (1987). The perception of tone hierarchies and mirror forms in twelve-tone serial music. Music Perception, 5, 31-78.
• Lerdahl, F. (1989). Atonal prolongational structure. Contemporary Music Review, 4(1), 65-88.
• Matzkin, D. 2001. Perception de similarité de mélodies tonales et non tonales: Etude pluridisciplinaire. Unpublished PhD thesis, Ecole des Hautes Etudes en Sciences Sociales, Paris.
• McAdams, S. & Matzkin, D. (2003). The roots of musical variation in perceptual similarity and invariance. In I. Peretz & R. Zatorre (Eds.), The Cognitive Neuroscience of Music (pp. 79-94). Oxford: Oxford University Press.
• McAdams, S. (1989). Psychological constraints on form-bearing dimensions in music. Contemporary Music Review, 4(1), 181-198.
• Meyer, L. B. (1956). Emotion and Meaning in Music. Chicago: University of Chicago Press, Chicago.
• Pitt, M. A. & Monahan, C. B. (1987). The perceived similarity of auditory polyrhythms. Perception and Psychophysics, 41, 534-546.
• Shepard, R. N. & Jordan, D. (1984). Auditory illusions demonstrating that tones are assimilated to an internalized musical scale. Science, 266, 1333-1334.
• Sloboda, J. A. & Parker, D. H. H. (1985). Immediate recall of melodies. In P. Howell, I. Cross, & R. West (Eds.), Musical Structure and Cognition. London: Academic Press.
• Tillmann, B., Bharucha, J., & Bigand, E. (2000). Implicit learning of tonality: A self-organizing approach. Psychological Review, 107, 885-913.