Mind as Music
Cognitive neuroscience typically develops hypotheses to explain phenomena that are localized in space and time. Specific regions of the brain execute characteristic functions, whose causes and effects are prompt; determining these functions in spatial and temporal isolation is generally regarded as the first step toward understanding the coherent operation of the whole brain over time. In other words, if the task of cognitive neuroscience is to interpret the neural code, then the first step has been semantic, searching for the meanings (functions) of localized elements, prior to exploring neural syntax, the mutual constraints among elements synchronically and diachronically. While neuroscience has made great strides in discovering the functions of regions of the brain, less is known about the dynamic patterns of brain activity over time, in particular, whether regions activate in sequences that could be characterized syntactically. Researchers generally assume that neural semantics is a precondition for determining neural syntax. Furthermore, it is often assumed that the syntax of the brain is too complex for our present technology and understanding. A corollary of this view holds that functional MRI (fMRI) lacks the spatial and temporal resolution needed to identify the dynamic syntax of neural computation. This paper examines these assumptions with a novel analysis of fMRI image series, resting on the conjecture that any computational code will exhibit aggregate features that can be detected even if the meaning of the code is unknown. Specifically, computational codes will be sparse or dense in different degrees. A sparse code is one that uses only a few of the many possible patterns of activity (in the brain) or symbols (in a human-made code). Considering sparseness at different scales and as measured by different techniques, this approach clearly distinguishes two conventional coding systems, namely, language and music. Based on an analysis of 99 subjects in three different fMRI protocols, in comparison with 194 musical examples and 700 language passages, it is observed that fMRI activity is more similar to music than it is to language, as measured over single symbols, as well as symbol combinations in pairs and triples. Tools from cognitive musicology may therefore be useful in characterizing the brain as a dynamical system.