Pseudo-Origen's Homily on Jeremiah & the Words of the Savior

[Peter Kirby]

Homily 1 on Jeremiah. This text was identified as (more like) Origen's (than anyone else).

The "fragments from the catena" of "on Jeremiah" (which were closest to Clement of Alexandria) begin here:

https://books.google.com/books?id=UFInXpKNapEC&pg=PA280

BINGO!

[Peter Kirby]

DING DING DING DING DING DING DING!

[Ben C. Smith]

Wow! Peter, I am impressed. You may really be on to something here.

[Stephan Huller]

Yes I second that. Breakthrough technology. I am working with documentary filmmakers on a number of projects. Could be very interesting. First though, do write a paper and submit it to a journal. Breakthrough stuff

[Stephan Huller]

Also it would be great to compare Methodius with Clement. I've always wondered if they were one and the same

[Tenorikuma]

May I also suggest comparing Acts with the Pastoral epistles.

[Ben C. Smith]

Methodologically speaking, Peter, is it more secure that Origen is not the author than that Clement is? IOW, is it possible that Clement was selected simply because the real author was not on your list, and Clement happened to be closest of those who were? Is that how it works?

Ben.

[Peter Kirby]

Methodologically speaking, Peter, is it more secure that Origen is not the author than that Clement is?

Sort of.

Yes, there is more probability behind the conclusion that Origen was not the author ... although that doesn't imply that "Clement is the author" is improbable, less probable than not (that may be obvious, but I think I should spell that out).

IOW, is it possible that Clement was selected simply because the real author was not on your list, and Clement happened to be closest of those who were? Is that how it works?

That is possible, sure.

The answer as to how 'really' probable it is that Clement is the author requires both stylometric and non-stylometric analysis.

If you asked me how to set that up, I would say that the 'non-stylometric analysis' should be used to create the prior probability that Clement is the author.

Then the stylometric analysis should be used to calculate the consequent probabilities, based on the stylometry, that Clement is or is not the author. This should be done on the basis of empirical/experimental data about the reliability and rate of accuracy of this stylometric method when working with samples thus and thus big and with such and such number of controls, with this number of matches ('N' out of 'T') to the candidate author, etc.

It must be kept in mind that Clement was picked as the most probable author twice with 3000+ word samples. This isn't a large number of times, but it is greater than once. Conclusions based on more than one reading have much higher probability of being correct (than those based on just one).

Also, there was a relatively high absolute level of correspondence with Clement of Alexandria (p-values > 0.1 with text length > 3000), well within range of plausible authorship. So it's not just the best of a bad bunch.

I'm not trying to oversell the conclusion that Clement is the author (and I also don't want to paint myself into a corner), but it does deserve further analysis, including further stylometric analysis. It is definitely not insignificant that Clement was considered the closest match twice. It would be even more significant, for example, if the text were sliced into three smaller samples and if Clement were still the closest match three times. Or if the number of controls were increased. And so on. Or if scholarly analysis might vindicate such a conclusion.

[Ben C. Smith]

It must be kept in mind that Clement was picked as the most probable author twice with 3000+ word samples. This isn't a large number of times, but it is greater than once. Conclusions based on more than one reading have much higher probability of being correct (than those based on just one).

Also, there was a relatively high absolute level of correspondence with Clement of Alexandria (p-values > 0.1 with text length > 3000), well within range of plausible authorship. So it's not just the best of a bad bunch.

That clarifies things. There is an absolute element involved as well as a relative element. Thanks.

Ben.

[Peter Kirby]

You may really be on to something here.

Breakthrough technology.

Thank you! I wish I could claim that I invented this, but the basic idea of fitting frequently-used words to a normal distribution in reasonably-large samples and comparing authorship on the basis of several of them has been around a while (and used to good effect before). Only a couple things seem new to me:

1) The particular weighted-average method of combining Z-scores in order to produce an aggregated p-value (aggregated over all the words). This formula is different from what I've seen.

2) A public interface for the program. Most stylometry programs never leave the researcher's close circle. Among other things, this makes it hard to repeat the experiment.

I'm not sure if that's enough to publish a paper. Maybe the third part is:

3) Actual application of this method to a wide array of patristic texts (something made possible by the TLG).

Stylometry is (or, seems to be) mistrusted and/or misunderstood by most biblical scholars. Perhaps a sober assessment of a bunch of church fathers could help spread the word about it being neither unreliable in general nor necessarily inapplicable to real situations.

I actually believe that there are better methods out there than what I'm using now, so there's that fact too.

Thanks for the words of encouragement.