This model looks very interesting. How does it compare to heuristic-based approaches like Defuddle and Readability?