Carys J. Craig & Guillaume Laroche, Out of Tune: Why Copyright Law Needs Music Lessons. Well worth reading in full. Here’s some choice bits:
[Lucy Pollard-Gott] found that, as listeners became more familiar with a given theme and listened to varied versions of that theme, both musicians and non-musicians were more able to identify elements of “theme structure” in variations. In plain language, the better someone knows a musical theme, and in a context where she is asked to compare that theme to another, the more likely it is she will draw a link between the two themes and deem them to be related, even when the two themes are somewhat dissimilar yet loosely share some common musical features.
This finding has tremendous implications for the lay listener test. First, it suggests that the recognition of similarity is an acquired skill, not a stable binary yes/no response. Rather, no can become yes over time and repeated listenings, to a point where the two themes need not be particularly alike in order for connections to be drawn between them. Second, it suggests this process is unidirectional; while no can become yes over time, yes cannot become no. Once points of similarity are drawn, a listener cannot go back to a state of mind in which those connections do not exist. … It is all too simple to create the conditions that favour a finding of recognizable similarity.
Thus, the question of “recognizability” of one work in another is not has objective as the lay listener test purports to be; quite the contrary, one can train people to hear connections between melodies, given sufficient time. This does not bode well for composers falsely accused of infringement where there is merely coincidental similarity, even where there are notable differences in the musical themes or expressive details that the composer might point to as evidence of independent creation.