I have been awarded a research-leave grant from the Arts and Humanities Research Council (AHRC) to complete a project during 2007-8 on the automatic derivation of harmonic-contrapuntal structure from a musical score, in the manner of a Schenkerian reduction. Since Michael Kassler's work in 1967, there has been periodic interest in the implementation of Schenkerian theory by computer, but the question of whether this is practical remains open. I hope to finally answer this question of practicality, and (if the answer is positive) to create prototype software to derive structural analyses from MIDI-like information.
There has been good progress to date, resulting in demonstration software to produce a reduction matrix embodying possible reductions for spans of music consisting of a few bars. Other software has shown proof-of-concept for a mechanism for attaching a goodness score to reductions, allowing the 'best' reduction to be extracted from the matrix in dynamic-programming style and allowing the distibution of scores among all possibilities to be estimated. Some progress has been made towards defining a metric to find the 'best' analysis, on the basis of comparisons of a sample of possible (but not necessarily 'good') analyses with existing analyses made by experts. So far, this has been based only on a small number of very short analyses, because the computational demands for longer analyses are currently too great.