At CICM 2008 (Workshop DML), Stephen Watt presented his work on analyzing the frequency of symbols, that would be an interesting infrastructure for further cop-based analysis.
See Michael’s blog and the DML Proceedings.
Another talk (Workshop MathUI) was on his handwriting recognition of mathematical notations: Presenting his Representation Approach. See paper
The challenge is that there is no fixed dictionary. But maybe CoPs provide some restrictions of potential parsing results? Or is frequency a better approach?