Time to retire Currier languages?

[MarcoP]

An advantage of the Currier A/B split is that it is easy to understand and to verify: pages where CUVA bigram SO (EVA:cho) is more frequent than ED are A, the others are B. This is cruder than Currier's analysis, which acknowledged that, when also looking at other bigrams, there are a few hard to classify pages (mostly astro/cosmological). But I would say that these plots show two visibly distinct clusters.
The left graph is labelled by scribe, the right by folio. Colours correspond to sections, as in the graph at the bottom of You are not allowed to view links. Register or Login to view.. 

   

What I am greatly puzzled about is the variability inside both Currier A and Currier B. Since Scribe 1 is basically the same as Currier A, the variability is still there is one splits the manuscript by Scribe. In other words, we cannot say "bigram distributions are different because different scribes preferred different spellings".
The problem with Scribe 1 is particularly clear on the  SO vs EO plot, where the Pharma pages (yellow squares) and some of the HerbalA (i.e. Scribe1-Herbal) pages bound near the Pharma pages appear to belong to the same cluster as the other scribes, while most Herbal A pages form a separated cluster.

   

This can also been observed in the top 20 words from Herbal A and Pharma. The list for Pharma includes four 'eo' words (three of them in the top 10), while no 'eo' words appear in the HerbalA list.


╔══════════╤═════╤═╤════════╤═════╗
║ Herbal_A │ ....│ │ Pharma │ ....║
╠══════════╪═════╪═╪════════╪═════╣
║.daiin....│ 4 9 │ │ daiin..│ 4 1 ║
╟──────────┼─────┼─┼────────┼─────╢
║.chol.....│ 2 6 │ │ chol...│ 1 8 ║
╟──────────┼─────┼─┼────────┼─────╢
║.chor.....│ 1 8 │ │ cheol *│ 1 5 ║
╟──────────┼─────┼─┼────────┼─────╢
║.s........│ 1 4 │ │ okeol *│ 1 4 ║
╟──────────┼─────┼─┼────────┼─────╢
║.shol.....│ 1 3 │ │ ol.....│ 1 4 ║
╟──────────┼─────┼─┼────────┼─────╢
║.cthy.....│ 1 1 │ │ or.....│ 1 3 ║
╟──────────┼─────┼─┼────────┼─────╢
║.chy......│ 1 1 │ │ qokeol*│ 1 1 ║
╟──────────┼─────┼─┼────────┼─────╢
║.sho......│ 1 0 │ │ aiin...│ 1 0 ║
╟──────────┼─────┼─┼────────┼─────╢
║.dain.....│ 1 0 │ │ okeey..│ 1 0 ║
╟──────────┼─────┼─┼────────┼─────╢
║.dy.......│ 0 9 │ │ qokol..│ 1 0 ║
╟──────────┼─────┼─┼────────┼─────╢
║.dar......│ 0 8 │ │ qokeey.│ 1 0 ║
╟──────────┼─────┼─┼────────┼─────╢
║.shor.....│ 0 7 │ │ dal....│ 1 0 ║
╟──────────┼─────┼─┼────────┼─────╢
║.chey.....│ 0 7 │ │ chor...│ 1 0 ║
╟──────────┼─────┼─┼────────┼─────╢
║.or.......│ 0 6 │ │ chey...│ 0 9 ║
╟──────────┼─────┼─┼────────┼─────╢
║.qotchy...│ 0 6 │ │ dar....│ 0 9 ║
╟──────────┼─────┼─┼────────┼─────╢
║.shy......│ 0 6 │ │ cheor *│ 0 9 ║
╟──────────┼─────┼─┼────────┼─────╢
║.cthol....│ 0 6 │ │ s......│ 0 9 ║
╟──────────┼─────┼─┼────────┼─────╢
║.ol.......│ 0 5 │ │ shey...│ 0 8 ║
╟──────────┼─────┼─┼────────┼─────╢
║.qokchy...│ 0 5 │ │ dol....│ 0 8 ║
╟──────────┼─────┼─┼────────┼─────╢
║.cthor....│ 0 5 │ │ okol...│ 0 7 ║
╚══════════╧═════╧═╧════════╧═════╝

Lately I have been asking myself: is it possible to understand if the changes in bigram distributions are a continuum or a set of discrete steps? My impression is that a continuum could be a better explanation, but I think that quantitative methods should be able to provide a more objective answer.

In conclusion, the simple split between two languages is certainly inaccurate (and Currier already knew it had limits). At a certain point, it should be replaced with something better but I don't think we have a sound alternative yet. Working with scribes offers a finer granularity, but the variability within Scribe1's corpus is an issue.