Fig 1.
Four examples of historical codices.
From left to right: MS Thott.290.2° (1459; Photo: Royal Danish Library); Practica uber die grossen und manigfeltigen Coniunction der Planeten (1523; Photo: SLUB Dresden); Voynich Manuscript (Photo: Beinecke Rare Book and Manuscript Library, Yale University); MS O.2.48 (14th century; Photo: Trinity College Library, Cambridge). Just the Voynich Manuscript is considered as mysterious. (Selected by authors of this article by their discretion).
Fig 2.
Franciscus Tranchedinus, Codex Furtivae litterarum notae. Vindobonensis 2398, ca. 1450, showing the practice of using artificial, homophonic alphabet to conceal messages. Photo: Augusto Buonafalce (2008).
Table 1.
Text preprocessing table for the Takeshi Takahashi’s transliteration of the Voynich manuscript.
Fig 3.
Autocorrelation plots for an English text and its individual symbols.
Symbols with a frequency > 100 are shown. The x axis shows lags, starting at 1, from left-to-right; y axis shows the Pearson correlation; blue horizontal lines are the statistical significance levels. ‘Sig+’ stands for the percentage of ‘significant and positive’ lags, ‘Sig’ stands for ‘significant’ lags. Expected chaotical behavior of symbol repetition may be observed.
Table 2.
Summarization of ACF analysis of individual symbols of a single English text.
Fig 4.
Autocorrelation plots for the Voynich manuscript and its individual symbols.
Symbols with a frequency > 100, analogical to the previous Fig 3. The x axis shows lags, starting at 1, from left-to-right; y axis shows the Pearson correlation; blue horizontal lines are the statistical significance levels. ‘Sig+’ stands for the percentage of ‘significant and positive’ lags, ‘Sig’ stands for ‘significant’ lags. Unexpected behavior of symbol repetition may be observed.
Table 3.
Summarization of ACF analysis of individual symbols of the Voynich manuscript.
Table 4.
Description of ACF values distributions of all contained symbols for various language samples.
Fig 5.
ACF summarization data visualization.
Parallel coordinate plot (chart A) and multidimensional scaling plot (MDS; chart B) of Symbol repetition delays ACF summarization data for various language and text samples listed in Table 4. Data were scaled by min-max scaling and Euclidean distances were used for plotting the MDS. MDS explains 96.8% of the original variance. Voynich Manuscript along with the Voynich Manuscript imitation are dissimilar to the rest of the tested texts, create anti-patterns (visible in chart A) and form its own clusters (visible in chart B).
Table 5.
Enumeration of the best ligature candidates and their components.
Fig 6.
Loss scores of ligature candidates <f>, <m>, and <p>.
P stands for the percentage of finding better candidate triplet by its loss score, pos stands for the rank of the loss score, L for the overall triplet loss score ℒT. Individual rules R are plotted at axis x, probability of finding a better loss score for the given rule among any other triplets, pR is plotted at axis y. Point label contains weighted rule loss Rn and overall loss contribution of the rule in the brackets. Illustrative examples of the symbols were taken from (from left-to-right, up-to-bottom): f20v, f108v, f34v, f4v.
Fig 7.
Loss scores of ligature candidates <q>, <n>, <r>.
See Fig 6 for details. Symbol illustrations were taken from: f13v, f30v, f27v.
Fig 8.
Loss scores of ligature candidates <t>, <d>, <k>, <g>.
See Fig 6 for details. Symbol illustrations were taken from: f18v, f2r, f58v, f23v.