andrewcriddle

In a recent thread on the Authorship of the Pastorals, I mentioned Anthony Kenny's "A Stylometric Study of the New Testament (or via: amazon.co.uk)"

I have since reread it Kenny argues that his stylistic arguments tend on the whole to support Pauline authorship of all the Pastorals. However his results seem IMHO to indicate a somewhat different conclusion.

His analysis shows the large stylistic difference of Titus from the typical Paulines and that 1 Timothy is closer to Titus than to a typical Pauline.

However 2 Timothy is more different from Titus than from any other Pauline and slightly closer to Philippians than to 1 Timothy. (1 Timothy is closest to 2 Timothy then to Titus).

The most obvious explanation of these facts taken on their own is that 2 Timothy is by Paul 1 Timothy is Post-Pauline with very close imitation of 2 Timothy and Titus is by the same author as 1 Timothy with less imitation of 2 Timothy.

This evidence obviously cannot be taken on its own but other evidence in the Pastorals of post-Pauline authorship such as the more developed church order is found more clearly in Titus and 1 Timothy than in 2 Timothy.

Hence there seems at least a plausible case that 2 Timothy unlike the other Pastorals was written by Paul himself.

Andrew Criddle

Julian

Could you talk a bit about what stylometric factors he uses and his analytical methodology? I have been dealing a lot with this issue (stylometric measurement and comparisons) myself lately ( http://julian.textcrit.com ) but am not ready to publish any results. I would be in a position, however, to comment with some authority on his methods and results.

Is the work you are referring to a paper or a book?

Julian

S.C.Carlson

Raymond Brown in his Introduction (or via: amazon.co.uk) indicates that there is increasing support for that idea.

Toto

more on stylometrics in this thread - Inquiry on synoptic statistics

TedM

Interesting stuff Andrew. 2 Tim seems most like Paul to me but that's my ameteur opinion. It's good to know that sometimes the experts agree with such 'gut' feelings!

When discussing hepaxes or whatever they are called several months back, and the Pastorals in particular, I recall reading that many stylistic differences can sometimes occur within just a few verses--perhaps indicating interpolation or just the uniqueness of the verses. I'm curious whether Kenny's research/conclusions revealed a concentration of such differences in only certain sections/verses of the Pastorals. IOW does his work suggest that some sections in each of them seem quite Pauline while others don't?

ted

Julian

I took a look at the Table of Contents and the Index over Amazon. Let me first say that to charge $168 for a 160 page booklet is downright criminal. Especially considering that the book hopelessly outdated and, based on what I can gather from the index, completely inadequate.

First of all, word frequency, word length, sentence length and so forth, represented as a document-term matrix will only get you so far, especially in light of the brevity of the pastorals. Short texts do not yield reliable results even when sophisticated and modern techniques are applied. This has been amply illustrated in [Stamatatos01] despite so-called text-length independent paradigms [Yule44] and [Honore79], which really aren't [Tweedie98].

[Stamatos01] used modern Greek texts and achieved good results but only on longer texts and the pastorals hardly qualifies with 1 Timothy = 1591 words, 2 Timothy = 1238 words and Titus = 659. They also a applied a chunker to the text, something which I am sure wasn't available to Kenny, seeing how they are still working on a dependency corpus for modern Greek [Prokopidis05], not to mention Koine where no work has been done in favor of X-bar and SFG approaches.

Minimum edit distances work well on synoptic issues [Andre98] but is hardly a stylometric issue. Inverse Document Frequency, Poisson deviations, Cluster analysis, Singular Value Decomposition, Principle Component Analysis, Latent Semantic Analysis/Indexing (probabilistic or otherwise) and many other approaches have been applied and some work well but generally under conditions more favorable than those that exist in the GNT. Of course, none of these methods were available to Kenny in 1986. The federalist papers have been a subject of study, to find out who the real authors were, and the first attempt at this was Mosteller and Wallace, 1964, and a later paper [Fung03] uses Feature Selection via Concave Minimization and Support Vector Machines, yet another method unavailable to Kenny. There are lots of other examples.

I find it hard to believe that there would be anything of value in Kenny's book but I could be wrong on that score, having not read it (nor will I at that price tag) but I suspect that the newer methodologies is where we need to go solve some of these issues. But even with all this fancy new math it is doubtful that we will ever get an answer much better than 'likely' or 'not likely.'

Anyways, I am working on applying many of the fancier techniques which have proven themselves effective but I am still a few months away from results I can back up mathematically.

Julian

A small selection of papers for those who might be interested, most (all?) should be available online:

[Stamatos01] : Computer-Based Authorship Attribution Without Lexical Measures
[Yule44] : The Statistical Study of Literary Vocabulary
[Honore79] : Some Simple Measures of Richness of Vocabulary
[Tweedie98] : How variable may a Constant be? Measures of Lexical Richness in Perspective.
[Prokopidis05] : Theoretical and Practical Issues in the Construction of a Greek Dependency Corpus.
[Andre98] : Comparison of literary texts using biological sequence comparisons and structured document capabilities
[Fung03] : The Disputed Federalist Papers: SVM Feature Selection via Concave Minimization.

andrewcriddle

Hi Julian

Could you clarify something ?

Are you saying that it is impossible in principle to come to a firm decision on stylistic grounds alone as to the authorship of works as short as the Pastorals ? Or are you saying that the actual stylistic differences claimed between the Pastorals and Paul's undisputed writings are too small given the brevity of these writings to allow a firm decision ?

Andrew Criddle

Julian

It is impossible to come to a firm decision. There, I've said. It is possible to establish, however, some reliable percentage as to similarity. It is also possible to come up with a p-value given that this lends itself well to a simple null hypothesis. Stylometric measurements can give us a lot of information but I doubt that we will ever get above the 80% mark. When the text is short, around 1000 words or less, then that probability seems to get closer to somewhere between 50-60%. I suspect that once a whole battery of the modern techniques are applied we can do better, possibly much better, than that.

So I guess what I am saying is that stylistic grounds form our only means of determining authorship, short of some archaeological discovery that sheds light on a topic (a signed and notarized confession of the author(s) ). I am also saying that stylometric measurements based solely on word counts/length and that sort of thing, are insufficient to get very close to our goal, although they certainly help.

For example I have received good results based on word counts and a Pearson chi-square test. I mention this because it was in the index of Kenny's book. But I get even better results when generating 30 different tests and subjecting them to a PCA. You see what I am getting at? The word count issues are only a primitive first step in a long march and I doubt that long march will get us much closer than, say, 80%. But that is 80% that is mathematically provable and that's a lot better than what we can say now which is mostly conjecture and consensus.

Julian