The Voynich Ninja

(02-08-2026, 09:01 PM)Cuagga Wrote: You are not allowed to view links. Register or Login to view.I just read the article, and one thing irks me, first in terms of writing convention, implying a factually wrong statement, and then most importantly in the interpretation it supports ; here goes : 

Section 3.2 of Torsten's paper, he mentions that

Quote:Even in Quires 1 and 3, which contain exclusively Currier A bifolia and therefore cannot be affected by A/B mixing, the differential is clear (Quire 1: conjoint 0.399 against facing 0.246; Quire 3: conjoint 0.394 against facing 0.268).

Then a few paragraphs down, the notation "Quires 4-7" is used to describe quires 4 through 7, immediately before using the same notation "Quires 1-3" alongside a statement wrong for Quire 2 :

Quote:Quires 1–3, which contain exclusively Currier A bifolia, are free of this confound and still show conjoint above facing (Quire 1: 0.399 vs. 0.246; Quire 3: 0.394 vs. 0.268)

Layfield and Davis show that for Quire 2, the exact opposite statement is true, with the conjoint scores falling below both confoliate and facing, facing becoming the highest (Confoliate : 0.273 ; facing : 0.301 ; conjoint 0.206).

Which interpretation could this point support ? We have one quire where facing pages are more related than the rest ; Statistical fluke ? Distinct production history ? Methodological weakness ?

I also wonder what LSA brings to us when we compute the complete similarity matrix. What is the average similarity between Language A and Language B pages ? Between Zandbergen languages ? Between LFD hands ? Pages on distinct folios in the same quire and language ?

You're right that "Quires 1-3" is imprecise phrasing — the two quires that actually support the argument at that point are Q1 and Q3, and  I should have written it that way.

That said, at that point, I'm defending Layfield & Davis's core empirical claim (conjoint > facing manuscript-wide) against a specific alternative explanation — that their pattern could be an artifact of Currier A/B mixing within herbal quires. My argument is that Q1 and Q3 are Currier-A-pure and still show conjoint > facing, so A/B mixing isn't the explanation. That argument goes through with  Q1 and Q3 regardless of what Q2 does.

But Q2 raises a real question for their overall interpretation. Under the singulion misbinding hypothesis, conjoint bifolium partners were originally consecutive pages of the text; the manuscript-wide conjoint > facing pattern is central to their argument. Q2 shows the opposite pattern within Currier-A-pure territory (conjoint 0.206 below both confoliate 0.273 and facing 0.301). It has an easy escape (N=3 conjoint pairs, folio 12 missing, small-sample noise), but the statistical-fluke reading is easier to grant to a theory that doesn't require the pattern to be robust everywhere. Their theory does require this.

There's a deeper methodological issue that goes beyond Q2. Layfield & Davis justify excluding the herbal section from their resequencing experiment on the grounds that "each page is a distinct botanical topic" and therefore LSA-based resequencing "was not warranted" there (their paragraph 35). But they simultaneously report and rely on the herbal quire LSA differentials (their Table 8) as part of their manuscript-wide singulion argument.

The two positions are hard to hold simultaneously. If herbal pages are topically distinct enough that semantic LSA cannot detect a correct sequence there, then the observed conjoint > facing differential in the same herbal quires cannot be measuring topical semantic similarity either. Under their own topical-distinctness claim, semantic LSA has nothing to detect in herbal — yet Table 8 detects something. Either the topical-distinctness claim is overstated (in which case herbal should have been included in the resequencing experiment), or the herbal LSA differential is measuring something other than semantic similarity (in which case what?).

(02-08-2026, 10:03 PM)LisaFaginDavis Wrote: You are not allowed to view links. Register or Login to view.I think it is quite likely that the herbal pages each represent an individual topic - each is a different plant, that is - which means that the interior pages of herbal singulions would not be expected to have consistently high similarity scores when compared to one another. That is why we did not attempt to use the herbal scores to suggest a sequence for herbal singulions. The data were very clear for Quires 13 and 20, but for the first eight quires, I'm not at all sure that there IS a "correct" sequence of singulions. There is definitely more work to be done.

Lisa, thank you for engaging with this substantively — I want to work through one factual point that seems to bear on the broader question.

Your response invokes "each herbal page represents an individual topic" as the reason the herbal LSA data don't support a correct singulion sequence. The DM paper cites Layfield & Davis 2026 (the DHQ paper) for this claim: "The results for the Herbal section in (Layfield & Davis, 2026) suggested that each page is a distinct botanical topic" (DM paragraph 35).

I've gone back to the DHQ paper to check. The conclusions section (paragraph 41) applies "each page is likely a different topic" specifically to the Cosmological, Astronomical, and Astrological sections (paragraph 41.2), citing the unique diagrams and zodiac signs on each page. About the herbal section, the DHQ paper's conclusion is different: "Beginning with the first A→B transition, the scores show a marked decrease, suggesting that the A→B transition is significant" (paragraph 41.1).

The specific finding about herbal → herbal comparisons in the DHQ paper is that 77% of the manuscript's lowest scores are Herbal → Herbal comparisons at page breaks: "The overwhelming majority of the remaining scores (77% or 46) were Herbal → Herbal page comparisons, virtually all of which were near or on a page break." (paragraph 35 of the DHQ paper). That's a claim about score drops at page transitions, not about each herbal page being topically distinct.

So the DHQ paper's LSA analysis doesn't seem to establish topical distinctness for herbal pages. The topical-distinctness inference for herbal appears to rest on the illustrations (different plants each page) rather than on the LSA data — which is a defensible intuition, but it's not what the DHQ paper is being cited for in DM paragraph 35.

This matters for the current discussion because your response treats the topical-distinctness claim as the reason the herbal conjoint > facing differential doesn't support a correct singulion sequence. If the claim is an intuition from the illustrations rather than a finding from the LSA data, then it's an intuition that would predict low conjoint scores and low facing scores throughout the herbal — but the DM Table 8 shows elevated conjoint scores in most herbal quires. That empirical pattern needs an explanation, and the topical-distinctness intuition (whether derived from illustrations or from data) predicts against it, not for it.

I don't think this changes what needs to happen next — as you say, there's more work to be done. But it does mean the herbal conjoint > facing differential is left without an interpretive framework in your work so far.

Notebooks q1 and q3 are stitched in two steps using the 2+2 and 3+1 patterns, respectively. While q2 is stitched in three steps using the 1+2+1 pattern. It's more likely that it contains a "foreign" bifolio with different statistics.