Figure 2 The timing profile of an attempt to imitate an inverted version of one of the four melodies used in the imitation experiment (see text). Note the ‘cross over’ halfway through the performance, where the performer, having followed the inversion target quite closely, reverts to the timing profile of the original performance target.

One final feature of the data throws further light on the performers’ strategies in attempting to imitate the transformed versions of the melodies. If the data are analysed for learning effects across the three attempts at each version of each tune, it emerges that only the more moderate transformations (the two translations) of the two longer melodies (32 and 36 notes long) show any significant improvement over the three attempts. Again, further work is needed to establish the reliability of this pattern of results and to provide a more secure basis for interpreting it, but a plausible explanation might hinge on the distinction between an analytic versus a wholistic approach to the task. For the shorter melodies (10 and 15 notes long), it is possible for the performers to hold the melody in immediate memory in a wholistic fashion for the short amount of time from first hearing the melody to finishing their imitation attempt (about 5 to 10 seconds). This has a kind of ‘all-or-nothing’ effect on their imitation attempts: either the whole thing is grasped reasonably accurately and then imitated quite successfully, or else very little is grasped and the imitation is poor. There is little scope for improving over the three trials because the basis of the imitation attempt is not a structural representation from which the imitation is regenerated, but is more like a perceptual ‘snapshot’ which is simply reactivated. For the longer melodies, however, this kind of immediate memory strategy is simply impossible, since the melody lasts too long and contains too much information to be held in an extended echoic manner. The only option is to form some kind of more abstract representations. The results show that the original and inverted versions do not improve over successive attempts-but possibly for quite opposite reasons: the well

structured characteristics of the original mean that it is readily understood and represented by the performers, leaving little opportunity for improvement, whereas the inversion is so peculiar that it very largely defies being represented at all. The two translations, being less radical as disruptions of the relation between structure and expression, give the performers enough that can be grasped to lay down the basic outlines of a representation on the first attempt, but which is improved upon in the subsequent repetitions. The two translations improve because they lie between the “all” of the original version, the conventionally of which makes it easy to represent on first hearing, and the “nothing” of the inversion, the bizarreness of which makes it almost impossible to represent in the course of three hearings.