English Text Classification by Authorship and Date

We performed two experiments with statistical tech-niques for classifying documents by date and author, using large bodies of publicly-available texts. In one experiment, we produced a Markov chain of every United States Supreme Court opinion ever written, and evaluated its ability to classify American judicial opinions by decade of authorship. In the other, we ex-amined the performance of two sets of quasi-linguistic features in classifying op-ed articles from The New York Times among four authors with a support-vector machine. The results in each case were en-couraging. With the Markov chain, we could cor-rectly identify the decade of authorship of a Supreme Court opinion within one decade 85 percent of the time. With the two quasi-linguistic feature sets, we were able to measure the equivocation between pairs of authors and observe some interesting effects when more features were collected.