You should submit your critique to the magazine, or write your own article, and see what the response is.
From what I can tell from the article, they did do more than "just" count words and sentence length.
I think you misunderstood me. I was talking about the feature set and not the analysis of the data.
The article you linked to performed PCA on the number occurrences of function words in text blocks.
The program you linked to counted word occurrences and sentence length.
On the other hand the use of Markov chains to capture sentence structure has been kicking about since Shannon mentioned it in passing in 'A Mathematical Theory of Communication'. It is a sufficient statistic to capture convincingly the voice of a person*, and it and its cousins such as string kernels are used in contemporary textural analysis. I'm not aware that the methods have spread to author identification however and it would be nice to see that they had.
*Complexity wrote a bot which mimicked Interesting Ian with it.
