While Signature provided some basic frequency analytics of data (such as letters, punctuation, word length, sentence length, and paragraph length), many of these measures have fallen out of fashion in current stylometric research. Rather, following John Burrows and his Delta method<\/a>, the preferred method of analysis is to use statistical frequency measures to look at n-grams.<\/p>\n In linguistics and computational statistics, n-grams are essentially numeric sets of characters or words used to measure probability.\u00a0So, for example, a word n-gram of size two (sometimes called a bigram) would take a sentence such as this quote from Roth\u2019s The Counterlife:<\/em>\u00a0\u201cHe sat down at the desk and began to read\u201d and create the following list: “he sat” “sat down” “down at” “at the” “the desk” “desk and” “and began” “began to” “to read”<\/p>\n You can also make character n-grams. Here\u2019s a list of character trigrams (3 characters) using the same sentence as above: “h e ” “e s” ” s a” “s a t” “a t ” “t d” ” d o” “d o w” “o w n” “w n ” “n a” ” a t” “a t ” “t t” ” t h” “t h e” “h e ” “e d” ” d e” “d e s” “e s k” “s k ” “k a” ” a n” “a n d” “n d ” “d b” ” b e” “b e g” “e g a” “g a n” “a n ” “n t” ” t o” “t o ” “o r” ” r e” “r e a” “e a d”<\/p>\n Essentially, you could do this manually, but it\u2019s so much easier for a computer program like JGAAP or R to do the work for you<\/a>, especially if you have a large corpus.<\/p>\n JGAAP<\/a> stands for \u201cJava Graphical Authorship Attribution Program\u201d and it was designed by Patrick Juola of Duquesne University. It is a free Java-based program for textual analysis, text categorization, and authorship attribution. It has an easy to use graphical user interface (GUI) which even includes explanations for specific analytical options.<\/p>\n However, while JGAAP can run through a number of customized stylometric analyses simultaneously, the output of data is very limited and it does not have a graphical option. The program will list the most likely candidate first and include the numerical probability data but the details of the analysis is hidden from basic users. So, JGAAP would be useful for a closed set of likely candidates with an unknown text, but the data would have to be extracted for advanced analytics or graphical representation.<\/p>\n In light of the limitations of both Signature and JGAAP, R-stylo seems like the ideal program for stylometric textual analysis. R<\/a> and RStudio<\/a> are free to download and the \u201cstylo\u201d package is easily downloadable via R\u2019s CRAN directory<\/a>. Additionally, there are a number of active scholars who use R-stylo so the support community online<\/a> is very helpful for beginners.<\/p>\n Some may balk at using R-stylo because it involves more upfront work and some familiarity with coding language. However, R-stylo\u2019s functional GUI interface, its capacity to graphically display results in a variety of different and customizable formats, and its ability to perform machine learning analytics make it the best option for stylometric analysis at this current time.<\/p>\n As an example, here is a cluster analysis of Roth\u2019s corpus according to either \u201cRoth\u201d or \u201cZuckerman\u201d novels. It’s really interesting that the only officially nonfictional entry from the corpus, Reading Myself and Others correlates so closely to The Facts: A Novelist’s Autobiography, even though this novel plays with some metafictional techniques.<\/p>\n And here is a Principal Component Analysis (PCA) of the same data. What\u2019s different here is the statistical algorithm that R-stylo uses to measure the authorship probability. Though these are both distance measures, the kind of algorithm used really affects how the data is represented. So, it\u2019s important that you really consider the best method of analysis for your specific project.\u00a0Obviously the Cluster Analysis is much easier to read without much editing, while the PCA analysis would have to be plotted using symbols, which is an additional option in R. However, I also think the messiness on this graph is good, since it shows a significant overlap between the Zuckerman and the ‘Roth’ corpora, supporting the claim that the division between these two sets of texts is certainly permeable.<\/p>\n![\"JGAAP_ROTH1\"](\"https:\/\/sites.temple.edu\/tudsc\/files\/2015\/08\/JGAAP_ROTH1.jpg\")
<\/a><\/p>\n![\"JGAAP_ROTH2\"](\"https:\/\/sites.temple.edu\/tudsc\/files\/2015\/08\/JGAAP_ROTH2.jpg\")
<\/a><\/p>\n![\"Roth_cluster_word3gram_delta_800MFW\"](\"https:\/\/sites.temple.edu\/tudsc\/files\/2015\/08\/Roth_cluster_word3gram_delta_800MFW.jpeg\")
<\/a><\/p>\n![\"Roth_PCA_delta_800MFW\"](\"https:\/\/sites.temple.edu\/tudsc\/files\/2015\/08\/Roth_PCA_delta_800MFW.jpeg\")
<\/a>
\nFinally, R-stylo has a rolling.classify option which takes a test corpus of, in this case, novels, and then reads it to determine multi-authored works. Maciej Eder most recently used this function to look at Harper Lee\u2019s newly published novel Go Set a Watchman<\/a><\/em> in order to determine the degree of Truman Capote\u2019s involvement.<\/p>\n
![\"Roth_rollingclassify_delta_800MFW\"](\"https:\/\/sites.temple.edu\/tudsc\/files\/2015\/08\/Roth_rollingclassify_delta_800MFW.jpeg\")
<\/a><\/p>\n