Would be interesting to run this algorithm on published scientific papers (another group recently claimed a different algorithm as being able to detect schizophrenia purely from text) and speech from politicians.