Text patterns support a different way to assemble the Voynich manuscript

The Voynich manuscript’s reading order may be part of its mystery. Colin Layfield and Lisa Fagin Davis combined analysis of the book’s physical structure with statistical comparisons of its text, arguing that its folded sheets once functioned as separate units. Binding those units into nested gatherings could have disrupted relationships between passages and illustrations.[1]

Their Digital Medievalist study proposes a “singulion” structure: individual folded sheets, or bifolia, read in sequence rather than nested inside a conventional quire. The contribution is a reconstruction hypothesis about how the book was used. It supplies no translation of the unknown script, but it offers a different arrangement against which to test its patterns.[1]

Start with the object, not the alphabet

Medieval books were physical engineering projects. Parchment sheets were folded and stacked; sheets nested inside one another in gatherings; writing and images were laid out with the anticipated binding in mind. A conventional reconstruction of a book’s contents assumes those construction choices are understood. The Voynich manuscript, held at Yale University, has long displayed features that make a straightforward reconstruction difficult, including changes in hands and unusual relationships among illustrations and facing pages.[1]

Davis and Layfield proposed that some of those anomalies make better sense if sheets were originally separate units. Under that arrangement, the writing associated with one leaf need not continue naturally onto the next leaf in its present binding. An illustration apparently straddling an awkward gutter, or a change in scribal style at an unexpected point, can then become evidence about assembly rather than a sign that the original author was inexplicably inconsistent.[1]

But a manuscript cannot be rearranged on intuition alone. Physical signs such as staining, sewing holes, parchment structure and writing hands need to be evaluated together. A proposed sequence that improves a statistical score but contradicts the material evidence would not be a convincing reconstruction. The researchers have described further work on stains and other physical markers to test their idea.[1]

Statistics without translation

The computational method used in this work is latent semantic analysis, or LSA, a technique developed decades before modern large language models. It uses patterns of symbol co-occurrence across a text collection to represent relative relationships between passages. The method can compare passages even when their meaning is unknown; it does not need to identify the manuscript’s language or derive a grammar first.[1]

The researchers compared text at the scale of pages, leaves, bifolia and quires, and used known medieval works as reference cases. The question was whether patterns of similarity fit one physical arrangement better than another. This is a way to test continuity without first assigning meanings to the manuscript’s symbols.[1]

Reconstructing the reading unit

A folded sheet can preserve a local relationship even if its position in the bound book changes. That makes the proposed unit useful: scholars can examine whether handwriting, illustrations and text patterns agree at the same scale. A strong reconstruction would make several kinds of evidence fit together, rather than merely improve one statistical score.[1]

The statistical method also depends on assumptions. Similar symbol distributions may indicate related material, but a writer can change topics deliberately, and repeated symbols need not carry the same role as words in a known language. Physical corroboration is therefore part of the reconstruction, not an optional addition to it.[1]

The stakes extend beyond this peculiar book. Digitization presents historical material as neat sequences of images. Search tools then treat those sequences as given. If the sequence is a later editorial decision, a computational method may find regularities in an artifact's reconstruction rather than in the original work. Scholars of damaged archives, reconstructed papyri and fragmentary books face similar problems.[1]

The proposal shifts the point of attack from decoding a symbol to recovering a reading unit. If the folded sheets preserve the book’s original organisation more faithfully than its current binding, future language analysis will have a better object to work on.[1]

Report an error or suggest a correction