another problem benefitting from a large data set: parser optimization. by looking at the frequency distribution of markup constructs, you can infer parser optimizations.