In my paper and in this forum I believe I have further developed the work of Timm & Schinner and provided explanations for the following:
Dense local edit-distance connectivity
However because my paper, my posts and my generator do not even TRY to explain:
...some responses have declared the work a failure.
I did not set out to produce a perfect reconstruction of the Voynich Manuscript. I set out to investigate whether a constrained copy/mutate system with limited working rules could reproduce a reasonable proportion of the manuscript's statistical and structural behavior in a very human doable format. At this point I believe the answer to that question is yes, since much of the criticism now being directed at this work concerns aspects of the Voynich which the current model was never intended to reproduce.
It seems that no matter how many times I state that this is a statistical model, I keep getting dragged into debates about why its visual appeal is lacking.
Therefore, unless future criticism addresses the actual statistical and generative scope of this work - namely the copy/mutate + ledger model and constrained copy/mutate generation in general - rather than demanding full visual, paleographic, or semantic reconstruction, I am unlikely to spend much more time responding to those objections.
While I do greatly appreciate feedback, I am primarily interested in criticism directed at the actual claims and scope of the model, rather than features that are well beyond that scope. I also welcome suggestions on how the system could be improved in the future to better reproduce the manuscript visually, but I will not consider the current model a failure simply because it does not yet do so.
Dense local edit-distance connectivity
- Why so many Voynich words differ by only one mutation (ED1).
- Why families cluster locally.
- Words being derived from nearby prior words rather than independently invented.
- Local propagation of forms through insertion/deletion/substitution.
- Why some glyph combinations are common and others effectively impossible.
- Legal transition structure within tokens.
- Why forms like daiin/daiir/dair/etc. behave as mutation neighborhoods.
- Persistence and expansion of lexical cores.
- Different behaviors at line start, paragraph start, internal positions, etc.
- Gallows concentration patterns.
- Why gallows cluster at paragraph and line starts.
- Why they behave differently from ordinary glyphs.
- Why source relationships collapse strongly at sheet/quire level rather than purely page-to-page.
- Why many Scribe 1 pages reduce to a very small dominant source pool.
- Why neighboring folios share mutation neighborhoods and recurring cores.
- Different lexical environments and mutation ecologies between early and later manuscript regions.
- Statistical shifts across manuscript regions without requiring a language change.
- Approximate Voynich-like token length behavior.
- Non-random frequency falloff emerging from recursive reuse and mutation.
- How a copy/mutate system alone can generate a vocabulary size comparable to the Voynich.
- How a copy/mutate system can generate hapax token counts comparable to the Voynich.
- Why the text is repetitive but not trivially repetitive.
- A practical workflow a medieval scribe could actually execute repeatedly with limited tools and memory requirements.
- Why some forms appear isolated or weakly connected after many mutation generations.
- Why certain high-frequency forms remain stable over long spans.
- How language-like statistics can emerge without underlying semantic language encoding.
- The text is structured, but the structure may arise from constrained generation rather than natural language.
- Recursive reuse naturally creates apparent grammatical consistency.
- Mutation constrained by adjacency legality prevents explosive randomness.
- Because production continually feeds on prior output.
- Recursive expansion from limited initial material.
- Statistical resemblance does not require translation or plaintext recovery.
However because my paper, my posts and my generator do not even TRY to explain:
- Exact CLS ordering behavior
- Stroke-level paleography
- Scribal motor habits
- Full glyph-class asymmetry
- Exact entropy profile reproduction
- Semantic meaning
- Encoding/decoding of plaintext
- Perfect Voynich reproduction
- Every positional phenomenon
- Every rare glyph behavior
- Exact Currier separation
- Illustration-text relationships
...some responses have declared the work a failure.
I did not set out to produce a perfect reconstruction of the Voynich Manuscript. I set out to investigate whether a constrained copy/mutate system with limited working rules could reproduce a reasonable proportion of the manuscript's statistical and structural behavior in a very human doable format. At this point I believe the answer to that question is yes, since much of the criticism now being directed at this work concerns aspects of the Voynich which the current model was never intended to reproduce.
It seems that no matter how many times I state that this is a statistical model, I keep getting dragged into debates about why its visual appeal is lacking.
Therefore, unless future criticism addresses the actual statistical and generative scope of this work - namely the copy/mutate + ledger model and constrained copy/mutate generation in general - rather than demanding full visual, paleographic, or semantic reconstruction, I am unlikely to spend much more time responding to those objections.
While I do greatly appreciate feedback, I am primarily interested in criticism directed at the actual claims and scope of the model, rather than features that are well beyond that scope. I also welcome suggestions on how the system could be improved in the future to better reproduce the manuscript visually, but I will not consider the current model a failure simply because it does not yet do so.
![[+]](https://www.voynich.ninja/images/collapse_collapsed.png)