This article describes the operation of two new tests of authorship and offers some results. Both tests rely on controlled contrasts of word-frequency and both exclude the very common words, which have been put to such good use in recent years. One test treats of words used with some consistency by a target-author but more sporadically by others. The second treats of words used sporadically by the target-author but not by most others. (The inclusion of words that some other authors use avoids the strict constraint that has impoverished this form of evidence.) In suitable cases, both tests prove very accurate. The fact that evidence of authorship can be detected in these three distinct frequency-strata helps to explain why such tests should work at all and so encourages the development of even better ones.
Literary and Linguistic Computing Vol. 22, Issue 1, p. 27-47