Certain connectionist models attempt to explain global context effects in tonal music in terms of activation spreading through a schematic knowledge set (e.g. Bigand, Madurell, Tillmann & Pineau, 1999). In doing so, they represent tonal structures as vectors, with events weighted according to their serial order. Accordingly, the issue of determining global context effects is simply one of computing the degree of association between upcoming events and the resultant vector representing the passage thus far. Music theory, by contrast, represents tonal structures as syntactic trees, with the determination of global context effects not only involving serial order information, but also making significant reference to the geometric structure of those trees (e.g. Lerdahl & Jackendoff, 1983). Thus, spreading-activation models, if they can adequately account for global context effects, offer the prospect of eliminating one component of current tonal theory.
This paper will examine the theoretical assumptions underlying both connectionist and music-theoretic approaches to global aspects of tonal structure. In particular, it will offer a critical evaluation of spreading-activation accounts of global context effects. Various structural phenomena will be considered that implicate the existence of syntactic relations in tonal music. Following this, some experimental data will be presented that indicate the cognitive reality of these syntactic relations. It will be argued that spreading-activation models are incapable of handling such phenomena because their representational systems are insufficiently rich and their processing mechanisms insufficiently powerful. The failure of spreading-activation models to account for global context effects suggests that the processing mechanisms employed in tonal cognition involve more powerful computational architectures.