At McGill this summer, SALAMI isn't the only major annotation-creation project that is taking place: The McGill Billboard Project is led by professor Jonathan Wild, and its goal is to assemble upwards of 1,500 chord annotations of a representative sample of the roughly 20,000 songs that have appeared on the Billboard Top 100 chart since the listing was introduced in 1958. With this data, the project seeks to chart the emergence and rise to popularity of different familiar chord progressions.

Besides sharing some human resources and technological efforts (witness the parallel websites), the two projects have begun to tack a similar course in the way annotations are created. The Billboard annotators reported that in producing chord labels (a difficult task that requires expert ear training), they found that the best workflow saw them first transcribing all the unique chord progressions they heard, and then producing a full transcription by analyzing the large scale structure of the piece (identifying the verse, chorus, and so forth), and identifying each with a set of chord sequences, copy-pasting them wherever the section repeated. In effect, they were doing much of the work that SALAMI annotators do.

Therefore, in cooperation with our colleagues at the University of Southampton, the Billboard workflow is going to be split up in a way that makes their job easier, and that gives SALAMI more data too! Billboard annotators will now create a list of chord progressions associated with major sections of each song, and these partial transcriptions will then be used by SALAMI annotators at Southampton to produce annotations that do double duty for SALAMI and Billboard. The Southampton SALAMI workflow will look no different from the McGill one, except that instead of using lowercase letters to indicate subsections of larger phrases, they will use the strings of chord labels provided by the Billboard transcribers. The chord progressions can automatically be converted to lowercase letters to create SALAMI annotations, or all the other information can be discarded to leave time-aligned chord annotations for Billboard.

By doubling some of this effort, we can compare this slightly modified Southampton workflow to the standard McGill workflow, which will allow us to ask some interesting questions afterwards. It will allow us, for instance, to compare within-institution consistency to between-institution consistency, to test how well the annotation approach has been communicated between teams. It may also be eye-opening to compare the usage of prime symbols at McGill to the chord progressions used at Southampton: how varied does a single chord progression have to be to earn a prime symbol?

Next Post Previous Post