Writing software to analyze Biblical Greek

[Jonathan Robie]

I used the Chi-Square Contingency Table Test, after numerous consultations with a statistics professor who guided me through the statistical maze. One example of a Null Hypothesis is this: "There is no difference in dissonance between the Luke and Acts passages beyond normal, random variability."

When the Chi-Square Contingency Table Test returns a P-Value indicating a highly significant result, another intrinsic explanation for the differences in dissonance is required. (Normal, random variability is not a sufficient explanation.) By logical deduction and a review of other possible explanations, I believe it must come down to different authorship.

How did you measure normal, random variability?

A much fuller description is contained in the book...

And I'm not trying to put you on the spot ... so if you'd rather not go to the effort, that's fine.

[Patricia Walters]

I used the Chi-Square Contingency Table Test, after numerous consultations with a statistics professor who guided me through the statistical maze. One example of a Null Hypothesis is this: "There is no difference in dissonance between the Luke and Acts passages beyond normal, random variability."

When the Chi-Square Contingency Table Test returns a P-Value indicating a highly significant result, another intrinsic explanation for the differences in dissonance is required. (Normal, random variability is not a sufficient explanation.) By logical deduction and a review of other possible explanations, I believe it must come down to different authorship.

How did you measure normal, random variability?

A much fuller description is contained in the book...

And I'm not trying to put you on the spot ... so if you'd rather not go to the effort, that's fine.

I "measured" normal, random variablity by what it is not; that is, it is not abnormal. If the Chi-square Contingency Table Test returns a value greater than the range 0.05-0.01 (the range for which the statistical term "significant" applies), then the variability is abnormal and must be explained by something other than normal and random.

(Likewise, if the Chi-square Contingency Table Test returns a value greater than 0.01, then the statistical term "highly significant" cannot be applied and another explanation of the variability is required.)

I hope this helps.

[Jonathan Robie]

Thanks - I think I wasn't clear enough about my main question. You're saying that Luke and Acts are no more like each other than the similarities found in some set of texts that were written by different authors. What set of texts did you use for this? What sample did you use for comparison?

I'd assume, for instance, that the first chapter of Luke and the first chapter of Acts are much more like each other than either is to the first chapter of John. If you did clustering on some set of criteria on Luke, Acts, John, 1 John, 2 John, 3 John, and the Revelation, I would expect Luke and Acts to fall into one group, John and 1-3 John in another group, and the Revelation to fall into a third group. I think many Greek readers would intuitively group text from these books in the same way. Would your technique also see Luke and Acts as more similar to each other than either is to these other texts?

[Patricia Walters]

Thanks - I think I wasn't clear enough about my main question. You're saying that Luke and Acts are no more like each other than the similarities found in some set of texts that were written by different authors. What set of texts did you use for this? What sample did you use for comparison?

I'd assume, for instance, that the first chapter of Luke and the first chapter of Acts are much more like each other than either is to the first chapter of John. If you did clustering on some set of criteria on Luke, Acts, John, 1 John, 2 John, 3 John, and the Revelation, I would expect Luke and Acts to fall into one group, John and 1-3 John in another group, and the Revelation to fall into a third group. I think many Greek readers would intuitively group text from these books in the same way. Would your technique also see Luke and Acts as more similar to each other than either is to these other texts?

To analyze just any passage or chapter from Luke or Acts was not a rigorous enough standard for me. The question of sources looms large in Luke and obscurely in Acts. So, I spent a long time identifying those passages whose authorship is the least contested - that is, seams and summaries. I consulted a set of experts and their commentaries for the experts' opinions as to which seams and summaries were judged to be authorial. In so doing, I eliminated as much idiosyncrasy as possible in data selection. Thus, when finished, I had a set of passages from Luke and a set of passages from Acts, deemed authorial by a majority of experts; it is those I analyzed.

I did not venture into the books of 1, 2, 3 John or Revelation. It was a gigantic task to do what I did, although as a future project it might be interesting. Assuming that Luke and Acts could not be the only Greek ever written by the author/s, I was satisfied that sampling the seams and summaries offered a reliable set of data.

[moon jung]

Hi.
Patricia,

Like you, I too have an advanced computer science degree and have a great interest in biblical studies at the same time
(learned Greek and Hebrew when I was in college and continue to learn them and read texts).

I studied computational linguistics for my Ph.D at the University of Pennsylvania, and I am interested in applying techniques of corpus linguistic and discourse processing for biblical studies.

I have two projects which I would like to pursue together with other people.

(1) Search for sentence structures:

Current bible softwares, e.g. Bibleworks which I use, provides search methods where one can search for words and phrases. Bibleworks uses "regular expressions" to represent search queries. This is a "primitive" method for describing a
pattern for a sequence of strings. Instead, one could specify a sentence structure where some items are lexical terms
as a query pattern. I wished Bibleworks had provided such a feature.

(2) Computational discourse processing.
I am interested in analyzing a text above the level of individual phrases and sentences.
We can construct a "mental model or representation" of a given text or narrative starting from the existing morphological database. We can follow a standard computational linguistics where one derives "deep structures" from "surface structures" of sentences and tries to connect the current deep structure to the mental model being constructed, in terms of temporal relations, logical relations, anaphoric relations, etc. The deep structure is needed especially because
Greek allows the null subject pronoun and the null object pronoun which are bound to the prominent entities introduced in the discourse. Although this mental model cannot be constructed fully automatically,
even the partial mental model, where unresolved connections are marked "unresolved", would be a great help to biblical studies.

My language would be Python, partly because it is a full programming language rigorously defined and has a lot of
packages/modules provided by other people. Also, when processing a huge text, Python can use C++/C within it, to
speed up the computation time. I would like to apply this software to Greek texts other than NT and the related texts.

Moon Jung
Sogang University,
Seoul, Korea

[Jonathan Robie]

Hi Moon,

I'm very interested in the kinds of things you are talking about. I have a computer science degree, and I am the lead editor of the XQuery language for querying and transforming XML. Python is a great language for a lot of this, several of use Python as at least one of our languages. I tend to use XQuery quite a bit, a few weeks ago I taught for 3 day of Clifford Anderson's 2 week course in using XQuery in the digital humanities (xqueryinstitute.org), using Julius Caesar as the text. I think you will find that XQuery can do the queries you are interested in, so can emdros, which has a query language designed for corpus queries and can be called from Python.

More good news: a lot of good open data is now available, so you will have lots of data to work with. And syntax trees are available in more than one model. Here is some data you may be interested in looking at:

* The Abbott-Smith lexicon
https://github.com/translatable-exegeti ... bott-Smith

* The Dodson lexicon (for glosses)
https://github.com/biblicalhumanities/D ... ek-Lexicon

* Alpheios (alpheios.net) and Arethusa (see, e.g.http://alpheios.net/bookmarklet/odyssey.html), which support syntax tree diagramming, integrate a Greek dictionary and grammar into the web browser, and are integrating language learning tools.

* Nestle 1904 with syntax trees and morphological tags (constituent model)
https://github.com/biblicalhumanities/Nestle1904/
https://github.com/biblicalhumanities/g ... nestle1904

* SBLGNT syntax trees with morphology (constituent model)
https://github.com/biblicalhumanities/g ... ees/sblgnt

* emdros - a corpus query tool
http://emdros.org/

* Dag Haug's PROIEL syntax trees with morphology (dependency model)
http://proiel.github.io/

* Bruce Robertson's OCR work. Check out commentaries, Migne, St. John of Chrysostom, Thackaray, etc.
http://heml.mta.ca/lace/catalog

* Steve Levinsohn has made his discourse analysis available in PDFs:
http://www-01.sil.org/~levinsohns/BART.html

* Alan Bunning has the Greek textual variants up to 400 AD here:
greekcntr.org

[moon jung]

Jonathan,

wow. Thank you for the references for text resources. I was wondering how I get them.
Thanks.

Moon Jung