30-10-2022, 01:26 AM
(This post was last modified: 30-10-2022, 10:56 AM by Torsten.
Edit Reason: small change
)
Stolfi suggested a similar structure for Voynich words (see You are not allowed to view links. Register or Login to view.). See also Stolfis three layer model (see You are not allowed to view links. Register or Login to view.)
However, such models do have in common that they also generate numerous words which do not exist in the Voynich text. This happens since the number of different glyph combinations is much more restricted than these models suggest. For instance the glyph after EVA-q is most likely EVA-o (in 97.5% of the cases). And the resulting sequence EVA-"qo" is most likely followed by EVA-k (in 59% of the cases) or another gallow glyph (25 %). There are always similar restrictions in place. For instance a group of EVA-e/ee/eee is either followed by EVA-d (34%), EVA-y (27%), EVA-o (23%), etc. The glyph before EVA-i/ii/iii is in 94% of the cases EVA-a etc. (Note: The empty slot beside EVA-q and the glyph combinations used in your table, like EVA-ar/al/ol or EVA-e/ee/eee, are probably chosen too describe such restrictions.)
Therefore it is possible to improve such a model by allowing only certain glyph combinations after each glyph. This results in a tree-like-model with a new branch for every newly added glyph. Each path within this model stands for a valid Voynich word. In this model one branch starting from the node EVA-"q-o-k" would for instance describe the probability for EVA-"q-o-k-ee", whereas a different branch would stand for EVA-"qo-k-o", and a third branch would describe the probability for EVA-"qo-k-y", etc. This way this model also covers the highly restricted number of different glyph combinations.
The reason for this behavior is the network character of the Voynich text (see You are not allowed to view links. Register or Login to view.). One result of the network character is that "high frequency tokens also tend to have high numbers of similar words" [see Timm & Schinner 2019, p. 6]. This means for every frequently used word it is possible to find some similar words.
For instance beside "daiin" (863 times) there is also "aiin" (469), "dain" (211), "saiin" (144), ...
beside "saiin" (144) there is also "aiin" (469), "sain" (68), "saiir" (6), ...
beside "aiin" (469) exists also "ain" (89), "kaiin" (65), "chaiin" (45), ...
...
beside "chol" (396) exists also "chor" (219), "cho" (68), "chal" (48), ...
beside "cho" (68) exists also "chy" (155), ...
beside "chy" (155) exists also "chey" (311), "chdy" (150), ...
beside "chey" (311) there is also "chedy" (511), "cheey" (174), ...
beside "chedy" (511) there is also "chdy" (150), "cheedy" (95), "cheody" (89), "kedy" (44), ...
beside "kedy" (44) there is also "okedy" (118), ...
beside "okedy" (118) there is also "qokedy" (272), "otedy" (155), ...
beside "qokedy" (272) exists "qokeedy" (305), "okedy" (118), "qotedy" (91) etc.
This way the words of the Voynich text result in a single network, connecting 6796 out of 8026 words (=84.67 %). The longest path within this network has a length of 21 steps, substantiating its surprisingly high connectivity [see Timm & Schinner 2019, p. 4]. For the core network see table 1 in Timm & Schinner 2019, p. 7. Side note: The more restricted tree model would only represent a different view for the Voynich network.
BTW: I have tried to describe some core parts of the network by using a 2D-grid (see
network_grid.txt (Size: 21.7 KB / Downloads: 26)
). Please note: since the network is in fact multidimensional it is possible that similar words, like "qokeey" and "qokeedy", do occur in different parts of the grid.
However, such models do have in common that they also generate numerous words which do not exist in the Voynich text. This happens since the number of different glyph combinations is much more restricted than these models suggest. For instance the glyph after EVA-q is most likely EVA-o (in 97.5% of the cases). And the resulting sequence EVA-"qo" is most likely followed by EVA-k (in 59% of the cases) or another gallow glyph (25 %). There are always similar restrictions in place. For instance a group of EVA-e/ee/eee is either followed by EVA-d (34%), EVA-y (27%), EVA-o (23%), etc. The glyph before EVA-i/ii/iii is in 94% of the cases EVA-a etc. (Note: The empty slot beside EVA-q and the glyph combinations used in your table, like EVA-ar/al/ol or EVA-e/ee/eee, are probably chosen too describe such restrictions.)
Therefore it is possible to improve such a model by allowing only certain glyph combinations after each glyph. This results in a tree-like-model with a new branch for every newly added glyph. Each path within this model stands for a valid Voynich word. In this model one branch starting from the node EVA-"q-o-k" would for instance describe the probability for EVA-"q-o-k-ee", whereas a different branch would stand for EVA-"qo-k-o", and a third branch would describe the probability for EVA-"qo-k-y", etc. This way this model also covers the highly restricted number of different glyph combinations.
The reason for this behavior is the network character of the Voynich text (see You are not allowed to view links. Register or Login to view.). One result of the network character is that "high frequency tokens also tend to have high numbers of similar words" [see Timm & Schinner 2019, p. 6]. This means for every frequently used word it is possible to find some similar words.
For instance beside "daiin" (863 times) there is also "aiin" (469), "dain" (211), "saiin" (144), ...
beside "saiin" (144) there is also "aiin" (469), "sain" (68), "saiir" (6), ...
beside "aiin" (469) exists also "ain" (89), "kaiin" (65), "chaiin" (45), ...
...
beside "chol" (396) exists also "chor" (219), "cho" (68), "chal" (48), ...
beside "cho" (68) exists also "chy" (155), ...
beside "chy" (155) exists also "chey" (311), "chdy" (150), ...
beside "chey" (311) there is also "chedy" (511), "cheey" (174), ...
beside "chedy" (511) there is also "chdy" (150), "cheedy" (95), "cheody" (89), "kedy" (44), ...
beside "kedy" (44) there is also "okedy" (118), ...
beside "okedy" (118) there is also "qokedy" (272), "otedy" (155), ...
beside "qokedy" (272) exists "qokeedy" (305), "okedy" (118), "qotedy" (91) etc.
This way the words of the Voynich text result in a single network, connecting 6796 out of 8026 words (=84.67 %). The longest path within this network has a length of 21 steps, substantiating its surprisingly high connectivity [see Timm & Schinner 2019, p. 4]. For the core network see table 1 in Timm & Schinner 2019, p. 7. Side note: The more restricted tree model would only represent a different view for the Voynich network.
BTW: I have tried to describe some core parts of the network by using a 2D-grid (see
network_grid.txt (Size: 21.7 KB / Downloads: 26)
). Please note: since the network is in fact multidimensional it is possible that similar words, like "qokeey" and "qokeedy", do occur in different parts of the grid.
![[+]](https://www.voynich.ninja/images/collapse_collapsed.png)