(07-12-2025, 09:35 AM)ThomasCoon Wrote: You are not allowed to view links. Register or Login to view.Hello again all!

I don't have a solution to the VMS and probably won't be offering anything of value here, but I wonder if words might rely on a 3-part structure, and was curious if anyone else has looked into the idea?

We know that some letter-clusters are almost always word-initial, like qok. We also know that some are almost always word-final, like aiin or dy. And others like ee very often appear in the middle.

Could all this point towards a three-slot structure, with different clusters assigned to each slot?

• Frequency-similarity correlation: High-frequency words possess more similar variants. For instance, <daiin> (863 occurrences) has 36 counterparts differing by just one glyph, while isolated words typically appear only once (Figure 3).
  
• Page-level word families: On individual folios, the most frequent tokens often differ by single glyphs. For example, on f108r, the top three words are <qokeedy>, <qokedy>, and <okedy>—each appearing 16 times.
  
• Language evolution: The gradual shift from "Currier A" to "Currier B" reflects accumulating variants rather than a sudden switch between systems. Table 2 shows how reordering sections by <chedy> frequency reveals smooth evolution rather than binary division.
  
• Self-citation mechanism: My paper proposes a generation algorithm—copying nearby words and applying small modifications—naturally produces this network structure without requiring explicit slot-based rules.
  

(07-12-2025, 10:30 PM)Torsten Wrote: You are not allowed to view links. Register or Login to view.

(07-12-2025, 09:35 AM)ThomasCoon Wrote: You are not allowed to view links. Register or Login to view.Hello again all!

I don't have a solution to the VMS and probably won't be offering anything of value here, but I wonder if words might rely on a 3-part structure, and was curious if anyone else has looked into the idea?

We know that some letter-clusters are almost always word-initial, like qok. We also know that some are almost always word-final, like aiin or dy. And others like ee very often appear in the middle.

Could all this point towards a three-slot structure, with different clusters assigned to each slot?

The Network Perspective on Voynich Word Structure

Your observation about a word structure aligns with documented positional constraints (prefixes like <qok->, middles like <ee>, and endings like <aiin> or <dy>). However, there's a complementary perspective that reveals deeper organizational principles: viewing Voynich words as nodes in a highly interconnected similarity network.

Rather than words being constructed from strict positional slots, the Voynich manuscript exhibits a continuous web of similarity relationships with remarkable properties:

Exceptional Connectivity: Network analysis shows that 84.67% of all Voynich words (6,796 out of 8,026) connect through single-glyph differences—what I call an "edit distance of 1." The longest path through this network spans 21 steps, demonstrating how thoroughly interconnected the vocabulary is (see You are not allowed to view links. Register or Login to view.).

Spatial Clustering: Words that are structurally similar tend to appear close together in the manuscript. As Figure 4 in our paper demonstrates, the average string distance between tokens increases with line distance—meaning similar words cluster on the same pages and even within nearby lines, creating localized families of related forms.

The network perspective explains several features that a rigid 3-slot framework doesn't fully capture:

• Frequency-similarity correlation: High-frequency words possess more similar variants. For instance, <daiin> (863 occurrences) has 36 counterparts differing by just one glyph, while isolated words typically appear only once (Figure 3).
  
• Page-level word families: On individual folios, the most frequent tokens often differ by single glyphs. For example, on f108r, the top three words are <qokeedy>, <qokedy>, and <okedy>—each appearing 16 times.
  
• Language evolution: The gradual shift from "Currier A" to "Currier B" reflects accumulating variants rather than a sudden switch between systems. Table 2 shows how reordering sections by <chedy> frequency reveals smooth evolution rather than binary division.
  
• Self-citation mechanism: My paper proposes a generation algorithm—copying nearby words and applying small modifications—naturally produces this network structure without requiring explicit slot-based rules.
  

Positional constraints certainly exist (certain glyphs strongly prefer word-initial or word-final positions), but the network view reveals the generative process: variants spawn from variants through incremental modifications, creating a similarity web where words are related neighbors rather than independent slot combinations. The network approach describes Voynich words less as "slot-filling" and more as organic growth through variation—each new word emerges from existing forms through small transformations, maintaining family resemblances while exploring the allowable glyph-sequence space.

A gephi project for the complete Voynich network is available at You are not allowed to view links. Register or Login to view.

(07-12-2025, 10:30 PM)Torsten Wrote: You are not allowed to view links. Register or Login to view.The Network Perspective on Voynich Word Structure