Voynich and Padua | The Boundary Grammar Problem

One of the oldest strong results in Voynich text analysis is also one of the easiest to underuse.

The line matters.

Prescott Currier argued in 1976, from large textual samples, that the Voynich line behaves like a functional entity.

Certain symbols or groups are strongly sensitive to where a word occurs in the line.

Some forms are rare at line beginnings.

Some occur disproportionately at line ends.

Repeated sequences in Currier’s large samples did not simply run across the line boundary as though line wrapping were arbitrary.

This means a line break may be part of Voynich syntax. If so, every transcription that treats the text as ordinary prose with accidental wrapping throws away information.

Quick Read

Currier’s 1976 analysis found strong positional effects at line beginnings and endings.
He concluded that the line is a functional entity, while explicitly saying he did not know what that function was.
The result does not prove poetry, cipher blocks, tables, chanting, recipes or any other particular genre.
It does show that physical layout and textual behaviour interact.
Paragraph starts may introduce another boundary state, often associated with visually distinctive initial forms.
Boundary effects may reflect syntax, formatting rules, encryption procedures, abbreviation, filler behaviour, scribal convention or technical record structure.
For the Padua hypothesis, boundary grammar is useful only if comparable positional systems can be shown in Paduan technical manuscripts better than in matched Italian and Central-European controls.

A Line Is Not Necessarily a Sentence

The first correction is conceptual.

If the line is functional, that does not mean it is a grammatical sentence.

A line could correspond to:

a complete instruction;
a metrical or mnemonic unit;
a record field;
a cipher block;
a tabular row rendered without visible columns;
a calculation step;
a scribal layout unit.

The evidence tells us that the line boundary changes textual probability.

It does not yet tell us why.

Currier’s Boundary Observation

Currier reported several position-sensitive phenomena.

Some character groups were unexpectedly rare at the beginning of lines.

Some symbols showed a strong preference for line-final position.

In large samples from the herbal and biological material, repeated sequences did not continue from one line across the boundary into the next.

He therefore concluded that line position governed aspects of token formation or selection.

That is an unusually strong structural statement for an undeciphered text.

Line Start Is a Distinct State

The first word of a line is not merely a normal word that happened to land beside the margin.

Its morphology can differ.

This creates several hypothesis classes.

Syntax Hypothesis

Line openings correspond to grammatical beginnings and therefore prefer particular forms.

Record Hypothesis

Each line is a record with predictable opening fields.

Cipher-State Hypothesis

Encoding resets or changes at the beginning of each line.

Scribal-Convention Hypothesis

Certain expanded or ornamental forms are preferred at visible boundaries.

Each predicts different behaviour elsewhere.

Line End Is a Different State Again

Line endings are especially interesting because they interact with physical space.

Currier noted groups that appeared to fill or complete lines and a symbol strongly associated with line-final words.

This raises the classic filler problem.

Are some final forms semantically empty additions used to reach the margin?

Possibly.

But filler is not the only explanation.

line-final morphology;
punctuation-like closure;
checksum or cipher termination;
record-field completion;
abbreviation expanded only at endings.

The important fact is positional dependence.

Paragraph Start Adds Another Boundary Layer

Voynich paragraphs often begin with visually prominent forms, including gallows-like characters or expanded combinations.

This suggests that paragraph start may be a separate state from ordinary line start.

Possible functions include:

topic initiation;
entry heading;
capitalisation-like signalling;
encryption reset;
structural marker;
scribal emphasis.

The correct test is comparative.

Does the first token of a paragraph behave like an intensified line-initial token, or does it belong to a distinct morphological population?

Page Start May Be Yet Another State

Technical manuscripts often treat a page opening differently from continuation text.

Headings, rubrication, larger initials and new entry sequences can cluster there.

Voynich gives us an opportunity to test nested boundaries:

page start;
paragraph start;
line start;
middle;
line end;
paragraph end.

If each state has different token probabilities, the text may behave more like a structured record system than unconstrained prose.

Image Boundaries Can Create Text Boundaries

The problem becomes more complex on illustrated pages.

Text frequently flows around drawings.

A line may end because the picture interrupts it rather than because the linguistic unit is complete.

This gives us an excellent discriminator.

If line-end morphology remains strong even when a line is shortened unpredictably by an image, then line position itself may drive the effect.

If the effect changes around drawings, layout planning and text generation may interact more deeply.

Boundary Grammar Can Reveal Production Sequence

Codicological reconstruction suggests drawing outlines were generally placed before text.

That means the scribe often knew the remaining writing space.

If token choice responds to available line length, the textual system may actively adapt to page geometry.

This is stronger than simple word wrapping.

the page may constrain token generation itself.

The “No Repeat Across Line Boundary” Result Is Particularly Important

Currier emphasised that repeated sequences in his large herbal and biological samples did not simply continue over the end of a line into the start of the next.

If ordinary prose were merely wrapped to fit available width, there would be no obvious reason for repetition to respect the physical line so strongly.

This is one reason line grammar deserves its own research programme.

Boundary Effects Are Not Semantics

We must keep the firewall clear.

A line-initial form is not therefore “start”.

A line-final form is not therefore punctuation.

A paragraph-initial gallows is not therefore a title.

Position gives functional constraints before it gives word meaning.

Currier A and B Must Be Tested Separately

Currier’s original work already noted that some positional restrictions were especially pronounced in his Language B material.

This means boundary grammar may vary by textual population.

Possible explanations include:

different language or dialect;
different encoded source material;
different scribal implementation;
different page function;
different production phase.

Boundary effects should therefore be crossed with hand, bifolio and section rather than averaged across the whole manuscript.

The Hand Problem

Lisa Fagin Davis’s five-scribe model gives a second axis.

If the same boundary rule appears across several scribes, it likely belongs to the shared writing system rather than one person’s habit.

If each hand implements boundaries differently, scribal agency becomes more important.

This is a public, testable bridge between palaeography and text statistics.

What Would a Technical Record System Look Like?

If each line behaves like a record, we might expect stable positional slots.

opening class;
middle payload;
closing class;
optional quantity or category fields.

The slots need not correspond to modern columns.

They can be expressed sequentially.

This model is especially relevant to recipe, astronomical and catalogue hypotheses because those genres often use repeated field order.

What Would Ordinary Prose Look Like?

A prose model should tolerate words moving across line boundaries with less impact on their form.

Paragraph boundaries would matter more than arbitrary line breaks.

If Voynich repeatedly violates those expectations, an ordinary prose model becomes less attractive.

This does not identify the alternative.

It narrows it.

The Padua Question Comes Later

Boundary grammar by itself cannot localise Padua.

Many medieval technical traditions use structured lines, tables and abbreviated entries.

The Padua hypothesis only gains if a recovered boundary operation matches a Paduan technical genre or scribal practice better than the matched controls.

For example:

Paduan astronomical canons;
medical consilia;
receptaria;
university quaestiones;
local tabular practices.

Until then, boundary grammar is manuscript evidence, not geographic evidence.

Public-Safe Boundary Test

Separate page, paragraph and line boundaries.
Measure token classes by positional state.
Repeat separately for Currier A/B and each scribal hand.
Control for shortened lines caused by illustrations.
Test whether line-final forms correlate with available remaining space.
Test whether paragraph starts are simply stronger line starts or a different class.
Compare with structured technical manuscripts from competing regions.
Do not assign word meanings from position alone.

What Boundary Grammar Can Legitimately Support

Voynich token behaviour depends in part on line position.
The physical line is likely a functional entity.
Layout may participate in the writing system.
Boundary behaviour can distinguish competing genre and encoding models.

What It Cannot Support

that lines are sentences;
that line-final material is meaningless filler;
that paragraph-initial glyphs have a known semantic value;
that positional effects prove a cipher;
that any boundary system is specifically Paduan.

Research Sources

Continue the Series

The Final Idea

The manuscript may be telling us something important before any word is deciphered.

Where a Voynich token sits may be part of what it is allowed to be. If that is true, the page is not a passive container for text. The page is part of the grammar.

Voynich and Padua | The Boundary Grammar Problem: Why Line Starts, Line Ends and Paragraph Starts Behave Differently