[Article] New Paper: Subtle Signs of Scribal Intent...

[asteckley]

I think the importance of last words on lines being shorter on average needs to be emphasized, because any theory regarding how the text was generated needs to explain it. Consider something like Rugg's proposed grill method -- I'm not sure why there would be anything special about last words on a line if that was the mechanism.

Actually, I can imagine some version of the grill method that might produce anomalies in the length or choice for the first or last word on a line. I believe Rugg's  main result was in showing that a simulation-device (like one inspired from the Cardan grille) could produce various structure and do so rather easily, but that specific device wasn't required -- it could have been any number of different variants of the the one he showed. 
So the rules of the device might have had some relation to lines of script.  

But it is a greater stretch to imagine how or why such a device could produce weirdness related to the drawing intrusions.

On the other hand, for both scenarios -- that of using some simulation device, or of actual meaningful language, it is plausible that the scribe could simply depart temporarily from any rules or precise meaning that might otherwise be dictated, and just choose a word that is shorter or longer for whatever reason.  But one would think that there would be less flexibility to do that in the case of writing meaningful script (because he still has to maintain the conveyance of some prescribed meaning).

If it's really the case that average word length shifts based on word position in the line (correcting for the effect of the result regarding words-before-drawing-elements from asteckley's paper), that's potentially a whole other world of hurt for everybody's hypotheses.

That is true. I know those the statistical tests are robust and reliable and based on solid assumptions, and I know that the "data is the data" and the results have to be accepted for what they say.  Nevertheless, even I still find them surprising and hard to accept. Why should the drawings have any such effects on the token choices?  

(One can expect, by the way, that given the probabilistic nature of the statistical tests, there will be some small number of false-positives that sneak through. For example, a p-value test using a threshold of 0.01 indicates a 1% probability that the observed data cannot be explained by random sampling effects. But that means one can expect that about one test in 100 will pass the test when it really shouldn't.  Now I didn't discuss how any of that affect the results in the paper because there just wasn't enough room --the paper was limited to 10 pages-- and adding that extra layer of complexity in the overall description of the study results didn't seem worth while. But suffice to say, the small number of expected false positives does not change the overall results.)