Dating Ancient Manuscripts with Help from Modern Software

Paleographers perform an important initial step in the textual criticism of the Bible, from studying the development of ancient Greek handwriting to determining the date and provenance of undated manuscripts. One maxim of textual criticism is that, all things being equal, the earliest readings are preferred. However, ancient literary manuscripts often were not dated by their scribes, so how are we to determine the date of undated biblical manuscripts?

Traditional dating methods

Aside from using modern scientific methods that may damage a manuscript, sometimes there are clues that we can use to determine a date. For example:

1. Some manuscripts have a colophon. This is a note appended to the end of the manuscript by a scribe. It may provide a date and other information, yet sometimes colophons have been forged, even in antiquity, to make a manuscript appear older.
2. The archeological context may produce a relative date for dating some manuscripts.
3. Sometimes a documentary manuscript like a deed or will was reused, usually by writing on the opposite side of the document that is dated, which also produces a relative date.
4. Some manuscripts have none of these “helps” for dating. In that case, paleographers traditionally date a manuscript by its writing style. This is the most common method.

In this last procedure, paleographers classify the handwriting of dated manuscripts into various handwriting styles, which have been ordered chronologically by century. Then, to determine a date for an undated manuscript, they compare the script of the undated manuscript to these chronologically ordered styles.

Having begun in the eighteenth century, the field of paleography is relatively young, and yet a surprising number of issues have developed with this methodology. For example, the definitions of the various handwriting styles can lack clarity, since the same adjectives, like upright and round, are used to describe multiple styles. Furthermore, scribes employed multiple styles during the same time period, so a clear linear progression from one style to the other does not always occur. These two issues alone can make a comparison of parallel writing styles difficult, so scholars spend many years developing their expertise. Thus, paleography is often considered more of an art than a science.

Traces of subjectivity   

Over the years, some manuscripts have been wrongly dated. Ernest Colwell highlighted one case: “There is in Leningrad one folio of a Greek Gospel for which [Caspar Rene] Gregory accepted the date AD 1247. Yet he assigned the manuscript on Mount Sinai from which it was taken to the fourteenth century.”¹Ernest C. Colwell, Studies in Methodology in Textual Criticism of the New Testament, ed. Bruce M. Metzger, NTTS 9 (Leiden: Brill, 1969), 125. Pasquale Orsini confirmed the difficulty in dating manuscripts using the examples of P.Kellis Lit. II 97 and the Codex Tchacos, for which paleographic dating contradicts the scientific data, and even the historical and archaeological contexts relating to these manuscripts.²Pasquale Orsini, Studies on Greek and Coptic Majuscule Scripts and Books, vol. 15 of Studies in Manuscript Cultures, eds. Michael Friedrich, Harunaga Issacson, and Jorg B. Quenzer (Berlin: de Gruyter, 2019), xiii–xiv. In a recent example, Georgi Parpulov redated the important New Testament manuscript GA 35 from the eleventh century to the fourteenth century. This manuscript is considered a prime representative of the Byzantine text and so was the base text for The Gospel According to John in the Byzantine Tradition published in 2007.

Perhaps problems, such as these, stem from the subjective nature of the methodology traditionally employed. Brent Nongbri has highlighted several issues with the use of handwriting styles, such as circularity and potential bias toward early manuscripts. In his study of manuscript P75, Nongbri cautioned against dating papyri too early. He was concerned that most parallels chosen for dating P75 were themselves paleographically dated, and therefore, they were not independent witnesses for a second-century date for P75.³Brent Nongbri, “Reconsidering the Place of Papyrus Bodmer XIV–XV (P75) in the Textual Criticism of the New Testament,” JBL 135 (2016): 408.

In another case, according to Nongbri, leading scholars use the widely accepted, second-century dating of manuscript P52 in assessing the date of the Gospel of John, yet Nongbri maintained that P52 has the same attributes as third-century papyri, which should broaden the acceptable date range for this manuscript.⁴Brent Nongbri, “The Use and Abuse of P52: Papyrological Pitfalls in the Dating of the Fourth Gospel,” HTR 98 (2005): 23–32. Not all have accepted Nongbri’s results, but his work is a reminder that paleographically dating manuscripts has been subject to controversy and debate among scholars regarding the accuracy actually obtained by this process.

Using computational modeling

The need for a more objective approach is apparent. Models are key to discovering patterns, testing predictions, communicating explanations from the known to the unknown, and ultimately, gaining objectivity. Furthermore, employing a graphical modeling and analysis tool, like decision tree ensembling, allows a user to see the results of an analysis easily. Another benefit of a decision tree model is that it is appropriate to use when a series of decisions need to be made to determine an answer, which corresponds nicely to the procedure for dating a manuscript. Therefore, the purpose of my research is to determine whether computational modeling may be used to provide a more objective basis for dating ancient Greek manuscripts.

The research for this project proceeded in three broad steps. First, a group, or domain, of similar manuscripts was chosen, which included Greek minuscule literary manuscripts for the initial project. In other words, manuscripts selected to be modeled had to be written in the Greek language in the minuscule hand, which is a script having lower-case letters that are sometimes connected. Furthermore, the contents of the manuscripts had to be literary in nature rather than the every-day type of documents known as documentary. In addition, only internally dated manuscripts were used, rather than manuscripts dated paleographically to a century or range of centuries. These manuscripts were chosen from the collections at Oxford’s Bodleian Library, the Cambridge University Library, the BNF in Paris, and the University of Michigan’s Harlan Hatcher Graduate Library. Additional manuscripts for the model were selected from the online collection at the Center for the Study of New Testament Manuscripts.

Second, the structure of the model had to be devised and the manuscripts analyzed against it. To devise the model, secondary literature on paleography at the Tyndale House Library in Cambridge was reviewed to accumulate a list of over 100 attributes of Greek minuscule literary manuscripts. These attributes are “omnitextual,” meaning they are not only about the shapes of letters and ligatures, but also include codicological and orthographical attributes, such as the material of the manuscript and the shapes of breathing marks.

In order to work with the attributes programmatically, the attributes are also discrete, or granular, rather than a combination of various characteristics like handwriting styles would be. The chosen dated manuscripts were compared against this list of attributes to record evidence for the existence of these attributes in each century in which the manuscripts were written. Thus, the corresponding attributes were noted for each dated manuscript in order to build a model of the Greek minuscule literary domain. This procedure is called training the model.

Third, a second set of dated Greek minuscule literary manuscripts was selected in order to test, or validate, the accuracy of the model, which has been named Omnitext™. In blind tests, these manuscripts were compared to the model and computations made for each century in which a corresponding attribute was found. This procedure is called ensembling.

Attribute	9th c.	10th	11th	12th	13th	14th	15th	16th	17th
	✔	✔	✔	✔	✔	✔	✔	✔	✔
					✔	✔	✔	✔	✔

For example, if a matching attribute from a manuscript appeared in the model throughout the minuscule period, from the ninth through the seventeenth centuries, every century receives a point, also called a vote. However, if the next matching attribute only appeared in the model from the thirteenth through the seventeenth centuries, then only those centuries receive a vote. In this example, the thirteenth through seventeenth centuries rise in importance, and the ninth through twelfth centuries fall in importance. The matching of corresponding attributes continues for approximately 100 attributes, and the plausible date further narrows.

When all attributes have been examined, the votes for each century are aggregated to produce a secure date based on the evidence in the model. The century with the most votes is the predicted “winner.” If there is a tie, a date-range between the centuries is predicted. For each test, the model predicted the correct date, either to the century or to a range of centuries including the correct date. The test results indicate that a decision tree model using ensembling can provide a more objective means to determine a secure date range for those manuscripts that are undated.

The future

From Omnitext, which is software with a patent pending with international license based on this methodology, three reports may be obtained about the manuscript examined. First, and most importantly, is the date prediction. Second is a report of the most impactful attributes for dating the manuscript, in other words, the attributes that helped narrow the date range. Third is a report of the dated manuscript in the model that most closely matches the undated manuscript that was compared to the model, based on their corresponding attributes. Thus, Omnitext is providing evidence for paleographical decisions and, in this way, commending the field as more of a science than an art.

Recently, Omnitext has been integrated with the New Testament Virtual Manuscript Room at the Institute for New Testament Textual Research at the University of Münster. The goal is to include other researchers in the paleographical analyses of manuscripts by comparing manuscripts to the model. This procedure is called a correlation.

Recently, a graduate student has tested the process and students at Birmingham Theological Seminary, Samford University, and New Orleans Baptist Theological Seminary are making Omnitext correlations. If you would like a demonstration of the process or want to analyze a manuscript against the Omnitext model (or have your students do so), please see Omnitext.org and contact Dr. Pat Sanders. With the development of this new tool, a more secure and objective date range based on evidence may be obtained in dating Greek minuscule manuscripts.

Notes

• 1
  Ernest C. Colwell, Studies in Methodology in Textual Criticism of the New Testament, ed. Bruce M. Metzger, NTTS 9 (Leiden: Brill, 1969), 125.
• 2
  Pasquale Orsini, Studies on Greek and Coptic Majuscule Scripts and Books, vol. 15 of Studies in Manuscript Cultures, eds. Michael Friedrich, Harunaga Issacson, and Jorg B. Quenzer (Berlin: de Gruyter, 2019), xiii–xiv.
• 3
  Brent Nongbri, “Reconsidering the Place of Papyrus Bodmer XIV–XV (P75) in the Textual Criticism of the New Testament,” JBL 135 (2016): 408.
• 4
  Brent Nongbri, “The Use and Abuse of P52: Papyrological Pitfalls in the Dating of the Fourth Gospel,” HTR 98 (2005): 23–32.

Pat Sanders

Pat Sanders (PhD, New Orleans Baptist Theological Seminary) is a lecturer at Samford University as well as core faculty and director of the Research and Writing Center at Birmingham Theological Seminar. Her research interests include manuscripts, paleography, textual criticism, and digital humanities. The Omnitext™ software is based on her PhD dissertation.