After having the annotators work on a small sample of popular music, we all met together last week to compare our annotations and see what issues came up. Encouragingly, the annotated boundaries were once again in high agreement, and mostly the letters (which indicate musical similarity) were consistently applied. The greatest variations were in the instrumentation and function layers, so we spent the bulk of the meeting discussing those.
By the end, we had more or less agreed on a restricted function vocabulary that we would stick to for the popular music. You can download a PDF of the list here. It includes the usual suspects (intro, verse, chorus, bridge, outro) as well as a number of other terms, all grouped into categories. That way, if the large vocabulary poses a problem for any application of this data, the terms in the same category may all be collapsed.
The central challenge in building such a vocabulary is to strike a balance between generality and specificity. A lack of either presents a danger: without generality, it becomes hard to compare the analyses to each other in order to draw conclusions; without specificity, the resulting analyses risk being meaninglessly general. In constructing a vocabulary we are sticking our necks out (in being so bold as to propose a theory of formal function adequate to describe all popular music), at the same that we're hedging our bets (by dividing the terms into categories, which may be collapsed into a more restricted set as necessary).
The annotators are all advised that if, in the course of their work, they begin to sense that a new function label is deserved, we encouraged them to contact us and let us know. That hasn't happened yet, but perhaps it will once we delve into jazz and classical music, where the whole notion of dividing music into discrete sections is often more contentious.