A Quantitative Confirmation of the Currier Language Distinction
For Voynich manuscript researchers, this provides quantitative evidence supporting a long-standing qualitative distinction.
The paper tests whether Currier's A/B language distinction in the Voynich manuscript reflects a genuine structural property, finding that an unsupervised model recovers the split with ARI=0.383 and a supervised classifier predicts labels with 89.2% accuracy.
We present a quantitative analysis of character-pair substitution ratios in the Voynich manuscript, testing whether Currier's A/B language distinction (1976) reflects a genuine structural property of the text. A Beta-Binomial mixture model applied to raw character counts without access to labels recovers the Currier split with ARI = 0.383. A supervised Beta-Binomial classifier trained on a subset of folios predicts the A/B identity of held-out folios at 89.2% accuracy. The character pairs separate into three functional regimes that constrain any theory of the Voynich writing system.