Wednesday, April 13, 2016

Handwriting Analysis and Dating the Bible

Ostraca from Arad dating ca. 600 B.C. (source)
There’s currently a story making the rounds on the major news sites (NY Times, Guardian, etc.). The headlines are, as headlines often are, “exciting”:
  • “New Evidence on When Bible Was Written: Ancient Shopping Lists”
  • “The Bible Is Really Old, Handwriting Analysis Reveals”
  • “Parts of Bible may have been written earlier than expected, archaeologists say”
The study these stories are referencing was published Monday in the Proceedings of the National Academy of Sciences. What the article actually does is try to show that there were at least six literate people at the desert fortress of Arad ca. 600 B.C. It takes a bit more work to get from there to an earlier-than-expected date for the Bible obviously. Here is how the paper summarizes its own significance:
Scholars debate whether the first major phase of compilation of biblical texts took place before or after the destruction of Jerusalem in 586 BCE. Proliferation of literacy is considered a precondition for the creation of such texts. Ancient inscriptions provide important evidence of the proliferation of literacy. This paper focuses on 16 ink inscriptions found in the desert fortress of Arad, written ca. 600 BCE. By using novel image processing and machine learning algorithms we deduce the presence of at least six authors in this corpus. This indicates a high degree of literacy in the Judahite administrative apparatus and provides a possible stage setting for compilation of biblical texts. After the kingdom’s demise, a similar literacy level reemerges only ca. 200 BCE.
I’m certainly not qualified to comment on ancient Israelite literacy rates so I will leave that to others. But what’s much more interesting about this paper, in my opinion, is the use of computer handwriting analysis to detect multiple scribes at work on ostraca. Here’s the description of the method:
Our algorithmic sequence consisted of three consecutive stages, operating on digital images of the ostraca (see Supporting Information). All of the stages are fully automatic, with the exception of the first, which is a semiautomatic step.
  1. Restoring characters
  2. Extraction of characters’ features, describing their different aspects (e.g., angles between strokes and character profiles), and measuring the similarity (“distances”) between the characters’ feature vectors.
  3. Testing the null hypothesis H0 (for each pair of ostraca), that two given inscriptions were written by the same author. A corresponding P value (P) is deduced, leveraging the data from the previous step. If P ≤ 0.2, we reject H0 and accept the competing hypothesis of two different authors; otherwise, we remain undecided. [Note the assumption favoring different hands.]
 The important thing is that the computer software has to have “binary” images in step two. This seems to mean that the computer needs a crisp black and white image with just the letter and nothing else. But that means reconstructing letters from pottery that’s worn, dirty, scraped, etc. (see photo below). That’s the stage I have some doubt about since the handwriting algorithms work by measuring distances between vectors (see step 2 above). Depending on how sensitive those measurements are, the reconstruction process could make a big difference.

Visual representation of the process of restoring lost letters.

The authors did test the method on 18 modern Hebrew writers and it worked in about 98% of the cases. That’s impressive. But then they didn’t have to reconstruct their handwriting either since they had nice, clean scans of alphabets the 18 people wrote for them.

In the end, the analysis found at least six different hands at work in the 16 inscriptions used in the study. Thus (a) literacy rates were high and thus (b) the Old Testament was compiled in the 7th rather than the 2nd century B.C. That at least explains the headlines. The article actually doesn’t spend much time getting us from a to b. (Remember, it is a science journal.)

Still, I did find the handwriting analysis itself most interesting. I’d like to see the method applied to something like the correctors in Sinaiticus and see what it comes up with.


  1. Peter,
    I agree with you that the handwriting analysis is the most interesting item. It would indeed be interesting to apply this technique to NT texts, not only for correctors but where entries are present in the margin to ascertain if from original hand. Of course, all this presupposes that the method is further tested and confirmed.

  2. I'm not sure I buy the idea that computers make better judgements on this sort of thing than humans do.

    1. Better comparison: humans using computers vs. humans not using computers.

  3. Interesting discussion here too:

  4. On what basis did they decide the probability threshold of (IIRC) 0.20?

    1. I couldn't figure that out either. 0.20 may be the standard.

  5. More commentary on this study from Alan Millard and Walt Kaiser via Christianity Today.