I think, this would be a nice machine learning research example (i.e determining an XML Schema, given sample XML documents). I wish this could be computed by machine learning techniques. Incidentally these days I'm studying an online Machine Learning course from Coursera (Stanford University). They primarily talk about following three methods to solve machine learning problems: Linear (and non linear) regression, Logistic regression & neural networks. From what I've learnt from this course so far, it seems we can't solve the XML Schema problem via the mentioned known machine learning methods (its not trivial to map XML document information; by string values, or hashes to a machine learning model, either three I've mentioned).