Lines interrupted by drawings

[MarcoP]

Marco, while you're right that words starting [che, she] or [yk, yt] behave differently after mid-breaks, they still have the same preference. Both [che, she] are less common after mid-breaks and after line breaks, while both [yk, yt] are more common. They're simply not responding to the same degree, though they are showing the same preference. Naturally, it depends on the statistics whether the smaller changes are robust enough to draw evidence from.

Hi Emma, thank you for your comment!
I agree with your observations. In general, the direction of change with respect to space-breaks is the same for image and line-breaks. Among prefixes, qo- and cho- may be the most notable exceptions.

I, too, find the [qo] statistics interesting here. Note that [qo] is much lower after mid-breaks while [ot] is much higher. I know we've discussed in the past the relation between [t] and [k], especially in the presence of [q].

I do hope that these comparisons can help us understand more about qo-. As you say, ot- is much higher for image-breaks than for space-breaks; ok- is also higher: the two differences add up to about 4%, while the qo- difference is 11%. It seems possible that also yt- and yk- (8% difference total) are involved in what is happening here.

Interestingly, [sa] is more common after both line breaks and mid-breaks, yet words starting [a] are only missing after line breaks and not mid-breaks. I had thought that [sa] was replacing [a] at the start of lines, yet clearly must rethink this idea.

If I understand correctly, your idea seems to work for line-breaks, but not for image breaks. I find it difficult to understand if one should conclude that the idea is not correct, or that line-breaks and image-breaks are significantly different phenomena. Even if directions of change are largely consistent, the prefix histogram  seems to me to suggest that the second option is worth considering.

I collected some more data, this time focussing on last-first combinations. Also in this case, I only considered pages containing at least one image-break.
   

The graph illustrates last-first character combinations across the different kinds of break. For instance, -y.o- is based on a word ending by -y followed by a word starting by o- (e.g. qopchy.otshol). The combinations included are those that are expected to be more frequent across image-breaks: since the set of image-breaks  is limited, measuring combinations that are expected to be small is difficult. The total number of couples in each set is:

• space-break 11050
  
• line-break 1868
  
• image-break 751
  

The plotted measure is the difference between the actual number of occurrences for a combination and the expected count (based on the frequency of the suffix and prefix), normalized by the total number of couples in each set. When the suffix of a word and the prefix of the following word are perfectly independent, the measure is 0. From the graph, it is clear that space-breaks produce word couples that differ from the expected much more than the other two cases.
If one adds up the absolute values of the normalized differences, the resulting totals for these combinations are:

• space-break 18.0%
  
• line-break 7.1%
  
• image-break 7.6%
  

Considering that line-breaks and image-breaks are about one order of magnitude fewer than space-breaks, there could be a greater influence of accidental / random deviation in the smaller sets: yet these sets appear to deviate much less than space-separated couples.
In other words, when word A is separated from word B by a space, there is some kind of "interference" between the two words which results in predictable suffix.prefix combinations (the phenomenon we discussed in You are not allowed to view links. Register or Login to view.). This kind of "interference" does not happen, or is much weaker, across line-breaks and image-breaks.

In the paper, we also discuss how last-first dependence is stronger in Currier B than in Currier A (as already noted by Currier himself). I am now thinking of comparing image-breaks with Currier A, in order to see if any strong parallel can be found: the relative rarity of EVA:q is another trait in common between the two sets.