Unconventional Methods in Voynich Manuscript Analysis

UNCONVENTIONAL METHODS IN VOYNICH MANUSCRIPT ANALYSIS Ivan Zelinka1,  , Oldrich Zmeskal2, Leah Windsor3, Zhiqiang Cai3 1Department of Computer Science, Faculty of Electrical Engineering and Computer Science Vˇ SB-TU, Czech Republic 2Institute of Physical and Applied Chemistry, Faculty of Chemistry, Brno University of Technology, Czech Republic 3Institute for Intelligent Systems, University of Memphis, USA ivan.zelink[email protected]  , zmesk[email protected], [email protected], [email protected] Abstract This paper discusses the possible use of unconventional algorithms on analysis and categorization of the unknown text, including documents written in unknown languages. Scholars have identified about ten famous manuscripts, mostly encrypted or written in the unknown language. The most famous is the Voynich manuscript, an illustrated codex hand-written in an unknown language or writing system. Using carbon-dating methods, the researchers determined its age as the early 15th century (between 1404-1438). Many professional and amateur cryptographers have studied the Voynich manuscript, and none has deciphered its meaning as yet, including American and British code-breakers and cryptologists. While there exist many hypotheses about the meaning and structure of the document, they have yet to be confirmed empirically. In this paper, we discuss two different kinds of unconventional approaches for how to handle manuscripts with unidentified writing systems and determine whether its properties are characterized by a natural language, or is only historical fake text. Keywords: analysis, complex network, fractal geometry, text, Voynich manuscript. Received: 10 May 2019 Accepted: 1 June 2019 Published: 24 June 2019 1 Introduction This paper discusses the possible use of various unconventional algorithms and techniques on analysis and categorization of the unknown text. We selected the Voynich Manuscript (hereafter, VM) as the experimental text and quantity of interest, as it is the most famous illustration of a hand-written codex in an unknown writing system (amongst the such books as the Book of Soyga, Codex Seraphinianus, Hypnerotomachia Poliphili, the Oera Linda Book, the Ripley Scrolls, the Smithfield Decretals, the Rohonc Codex, the Red Book, Prodigiorum Ac Ostentorum Chronicon). The timeline of the VM and its researching can be found in various research reports and publications as for example in [4], [7], [11], [13] or well written chronology in [33]. The VM is full of symbols that were partially interpreted by [30], and it is claimed that various plants, animals, a mineral, and cities were deciphered. These claims, however, have not been verified. Other scholars have also examined the codex’s properties, including [7], [12], [25], [18], [21] [24], [26] or [15]. Unaffiliated researchers and hobbyists have also pursued investigations into the Voynich Manuscript. The VM studies are also available on the web as the web documents and the web blogs (not a scientific document) as for example [18], [28] or [31]. In this paper, we do not attempt to translate or decipher the codex. The aim of this paper is to identify statistically and computationally meaningful attributes of the text when analyzed by means of two unique methods. The structure of the paper is as follows. First, we provide a brief overview of VM. Next we describe our experimental design: an algorithm for fractal geometry ([10], by SW HarFA [20]). Using this methodology, we analyze the VM alongside the Ernst Hemingway book The Old Man and the Sea, and the Bible. We then describe the fractal properties of language and the complex networks of topology, including visualizations of the VM, the Bible, and The Old Man and the Sea, translated into 120 languages1,2. The aim of this workflow is to investigate how much VM differs from the books written in natural language. In the second experimental section, we perform an analysis based on the text conversion into the complex-like network ([19], [9], Fig. 10 and Fig. 11), where each vertex represents one word from the manuscript. Also randomly generated text is used for such analysis. Our analyses demonstrate that VM is likely not a historical fake, but rather that it is written in natural, yet unknown/undecrypted/undecoded, language. 1https://www.bible.com/languages 2https://ebible.org/download.php https://doi.org/10.13164/mendel.2019.1.001 ISSN: 1803-3814 (Printed), 2571-3701 (Online) MENDEL — Soft Computing Journal, Volume 25, No.1, June 2019, Brno, Czech RepublicX 1 2 Voynich Manuscript The Voynich manuscript [11], Fig. 1, is a mysterious illustrated book, written in an unknown language. According to the radiocarbon dating (C14) it was written sometime between 1404-1438. The book is named after Wilfrid Michael Voynich (Wojnicz, Vojnic), an American merchant of Polish-Lithuanian origin who acquired the manuscript in Italy in 1912. At present, the text is owned by Yale University under the designation MS408. The manuscript is the subject of interest of many scholars of various disciplines, especially cryptologists. It is also connected with the Czech environment, as the first known owners were scholars from Prague, and it is even possible that it was written there. The book probably had about 272 pages in 17 sheets, but only 240 were preserved. The text is written on parchment with unknown lettering (20-30 characters with simple punctuation) in an unknown language. The manuscript consists of over 170,000 characters with spaces dividing the text into about 35,000 groups of varying length (accepted as words, or discrete lexical items) while 8114 of those words are considered unique ”word types”. In the entire book the characters and words distribution seems to be affected rules and indicates the existence of spelling and grammar. Statistical analysis, as in [1], [14], of the text suggests that Zipf’s law applies to the language used, the average length of each word is roughly the same as Latin and English [32]. Importantly, the language used in Voynich’s manuscript is quite different from the European languages. However, some signs suggest similarities with Semitic [or Indo-European?] languages, although some scholars have suggested that the manuscript is of Latin American origin [29]. Several words in the manuscript are also written in Latin, in a language reminiscent of ancient languages around the Mediterranean Sea, although it is possible that the Latin passages were added later. According to the human-generated classifications of the illustrations, the book is divided into several parts; it is likely that it is a tractate from several areas: •Botanical part. The illustrations contain pictures of common European plants. However, most of them are difficult to identify; some have not yet been identified. •Astronomical part. Illustrations include astronomical diagrams, astrological diagrams, and symbols, sketches of the zodiac. In some diagrams, the names of the months of the year are Latin, in Romanesque, perhaps in Catalan or Occitan. •Biological part. The illustrations show mostly miniatures of naked women, bathers in strange formations, some of which resemble body organs. Some women have crowns. •Cosmological part. Illustrations remind remotely of maps of a strange landscape or cosmological sketch. There are castles, perhaps volcanoes. •Pharmacological part. Illustrations show parts of plants (roots, leaves). Perhaps it is a pharmaceutical prescription, which may also suggest that the text is divided into short paragraphs in this section. •Recipes. Without illustrations, the text is divided into short paragraphs, separated by blooms or stars. The text has not yet been decrypted, and its history and provenance are very fuzzy. One way to determine the time of the book’s creation is according to the style of hairstyles, clothing, and castles shown in the illustrations. A more scientific approach using radiocarbon dating methods was also used. The period of creation is estimated to be between 1404 and 1438. Most studies of the Voynich Manuscript have relied on qualitative analysis, using the illustrations as benchmarks to demarcate the chapters and contents. Given recent advancements in computational linguistics methodology, we are now able to explore this document using rigorous, statistical approaches. 3 Fractal Geometry and the Text We now turn to the empirical analysis of the VM, using fractals. Fractals are geometric objects whose basic structure, fragmented or irregular, is repeated at multiple scales. The term, coming from Latin, was proposed by the mathematician Benoit B. Mandelbrot in the 70s in 20th. century, to describe the quality of these fractured geometrical objects. Fractal geometry describes self-similar objects, a characteristic feature of complexity and complex dynamics. Self-similarity is a phenomenon that appears in the dynamics of many natural, artificial or mathematical systems [16]. Self-Similarity is the property of a geometric object that contains a part which is identical, or very similar, to the geometric structure of the whole object. In other words, a subset of the parent object is similar to the parent object. Mandelbrot’s book, The Fractal Geometry of Nature [16], became a bestseller and described how to construct very complicated structures using simple principles. However, the history of this discipline also has deeper roots, starting in the 17th century with recursive self-similarity stated by the philosopher and mathematician Gottfried Unconventional Methods in Voynich Manuscript Analysis MENDEL — Soft Computing Journal, Volume 25, No.1, June 2019, Brno, Czech RepublicX 2 (a) (b) (c) Figure 1: A few page samples from Voynich manuscript. Leibniz. Then, in the 19th century, the mathematical study of continuous, but not differentiable functions, became increasingly rigorous. It was Karl Weierstrass who first described a function that was continuous but without derivative at any of its points. Relevant researchers who also made contributions within this field were Georg Cantor, Gaston Julia, Henri Poincare and others [8]. Nowadays, fractals have several applications like data compression, antenna systems, modeling natural patterns, computer graphics, digital art, etc [3], [16]. Self-similarity, as a typical attribute of fractals, can be shown graphically on two classic fractal objects: a snowflake and fern in Fig. 2. Upon zooming in on any part of the object, its structure will resemble the basic object again. There are many algorithms for constructing fractals, as for example: •Iterated Function System (IFS) uses geometric replacement rules, that may be deterministic or stochastic [6]. •Escape-time fractals use a recurrence relation at each point in a certain space [3]. •Strange attractors use the solutions of a system of differential equations that exhibit chaos [8] and have a fractal dimension [8]. •Lindenmayer systems (L-systems) use string rewriting [23] and a formal grammar to resemble branching patterns present in many biological structures like plants, cells from the immune system, etc [23], [3]. The most known algorithms are the Iterated Function System (IFS) and the Escape-time algorithms that present a tight relationship [6] between them. An essential characteristic of affine transformations is that we need a few numbers for a perfect description of an object. Each fractal has its unique description via a few numbers (coefficients of affine transformations [6]), and a slight change over them also changes the shape of the fractal. On the other hand, Escape-time algorithms work differently as it is based on the idea of convergent and divergent trajectories over a search space. They take a point from such search space and calculate how many iterations are needed to leave a certain predefined area. The number of iterations is then used to determine the colour for such point [16]. Thus, IFS produce black-white pictures while TEA colored one. Because fractal presence and attributes have been observed and discovered in any natural language as reported in [10], we employ this methodological approach for analyzing the VM in a similar way as reported in pioneer work [9] by L. Hrebicek, then in [17], [27], [5], [2]. 4 Complex networks The complex network is in fact a graph, i.e. a structure, that consist of nodes (e.g. vertices) and edges (connections amongst them). The topology is usually specified regarding regular lattices or complex networks [19], which in recent years has become one of the best-developed areas of complex systems theory. The structure of complex networks can be observed in many systems. Large-scale networks, exhibiting complex patterns of interaction among vertices exist in both natures and man-made systems like genetics, ecological ecosystems, economics, sociology, scientific publications, the Internet, power grids, among many others [19]. I. Zelinka et al. MENDEL — Soft Computing Journal, Volume 25, No.1, June 2019, Brno, Czech RepublicX 3 (a) (b) Figure 2: Fractal samples, the snowflake, and the fern. The name complex networks [19] comes from the fact that they exhibit substantial and non-trivial topological features, with patterns of connection between vertices that are neither purely regular nor purely random. Among many studies, two well-known classes of complex networks are the scale-free networks and the small-world networks (see examples in Fig. 10). Research in the field of complex networks has become interdisciplinary and had joined researchers from many scientific fields and perspectives. Complex networks are very relevant in the scope of this article, so we recommend to read its introduction in [19]. Complex networks’ connection with analyzing text and literature was also introduced in [9] by L. Hrebicek, the Czech scientist and expert on the Turkish language. This work and idea have inspired us to use the complex network approach on the Voynich manuscript. The complex network visualization also inspired by ideas in [9] was used to convert the Voynich manuscript into a visual form for further analysis, see Fig. 10 and Fig. 11. 5 Experiment design For the VM analysis was used fractal geometry and very basic complex network analysis. Experiment conditions and parameter setting is described here. 5.1 Data The data used for our experiments were based on Hemingway’s novel The Old Man and the Sea (120 language mutations), the Bible book in English (at the moment we are expanding those experiments for Bible in 150 languages) and computer translation of the Voynich manuscript 3. As the Voynich text was used the computer translation known as the Currier, while other existing as Benninett, FSG, Frogguy, Eva, Takahashi EVA, ... has not been used here. These computer translation - character sets differ in symbols that interpret individual letters from VM. Used data must be preprocesed. Original source files contained symbols and letters that were not a part of the text, so firstly before experiments all those objects were remover till only words from own text remains, see Fig. 3. During the analysis we found, as the curiosity, the most frequently used words, see Fig. 4. 5.2 Used software and hardware For reported calculations, algorithms and visualizations have been used standard PC with an Intel Core I7 processor and 8Gb RAM with SW Mathematica R v. 11.0. Additionally for the fractal analysis was used SW HarFA [20] to calculate the fractal dimension of the text. 3http://www.voynich.nu Unconventional Methods in Voynich Manuscript Analysis MENDEL — Soft Computing Journal, Volume 25, No.1, June 2019, Brno, Czech RepublicX 4 FGAG2, GDAE, AR, GHAM, SOE, SORG, 0D0RC2, GDOR, SOE8G, 2ORG, DZAR, ORG, DAIR, THAM, SOR, AR, HZAR, HZAR, 8ALA, 2GAIIR, SCDG, OR, GDAM, SO8, HZOARG, HZC2, 8ARAM, 2G, 0OM, OHCCG, OHCAR, ROEOHG, HZAAR, 8AM, ODAM, OR, ODAL, 2AIRG, TCAR, HZAM, PZAR, FZAM, G8ARAISG, O8AR, SG, SOE, ⋯33 571 ⋯, 4ODA2, AECCGTG, O4ODAM, AE, SCG, 4ODAR, ODARAE, ODCG, SPZCG, OHCCG, OODAR, ODG8G, O2AM, SDG, 4ORAM, TDZCG, 4ODCG, EDCTG, ODCCG, ODAE, TC8, DAEG, 2GDAR, AM, OEDACG, 8AM, OEG, 0A2OR, SCG, 4ODCG, ODCOEAL, TCG, 4OE, OR, TCCG, 4OR, ARAK, OE, EDAL, 2O8AE, TAE, THZG, TDZG, 4OE, AM, ARG large output show less show more show all set size limit... (a) FGAG2, GDAE, AR, GHAM, SOE, SORG, 0D0RC2, GDOR, SOE8G, 2ORG, DZAR, ORG, DAIR, THAM, SOR, AR, HZAR, HZAR, 8ALA, 2GAIIR, SCDG, OR, GDAM, SO8, HZOARG, HZC2, 8ARAM, 2G, 0OM, OHCCG, OHCAR, ROEOHG, HZAAR, 8AM, ODAM, OR, ODAL, 2AIRG, TCAR, HZAM, PZAR, FZAM, G8ARAISG, O8AR, SG, SOE, ⋯33571 ⋯, 4ODA2, AECCGTG, O4ODAM, AE, SCG, 4ODAR, ODARAE, ODCG, SPZCG, OHCCG, OODAR, ODG8G, O2AM, SDG, 4ORAM, TDZCG, 4ODCG, EDCTG, ODCCG, ODAE, TC8, DAEG, 2GDAR, AM, OEDACG, 8AM, OEG, 0A2OR, SCG, 4ODCG, ODCOEAL, TCG, 4OE, OR, TCCG, 4OR, ARAK, OE, EDAL, 2O8AE, TAE, THZG, TDZG, 4OE, AM, ARG large output show less show more show all set size limit... (b) Figure 3: The Voynich data preprocessing, a) original data, b) the same data under font, a screenshots from the Mathematica R software. 8AM 4 ODAM TC8G SC8G 4 ODCC8G 4 ODCCG 4 ODC8G TOE TCG SCG OHAM OE ODAM 4 ODAE AM 4 OHAM 8AR TDZG 8AE TCCG 4 OHC8G 4 ODAR ODCCG OHC8G 2AM SOE TCOE TOR SCCG ODCC8G 4 ODG OHCCG 4 ODCG ETC8G ODC8G OHAE OR HZG 4 OE 4 OHCC8G OHCC8G OHAR 8G T8G SCOE THZG ODAE SCC8G 4 OHTG AR 4 ODT8G OHG 4 OHAR SDZG 4 ODTG 4 OHG 4 OHAE 4 ODOE GHAM HZCG TG 4 ODAN 8AK SG OEDAM DCC8G TCO8G ODAR AE HZOE TO8G RAM EDAM SO8G TCC8G DAM SOR OHOE ORAM HAM OEAM 8OE ODCOE TCOR TCDG 8AIR ODG 8AN O8AM 2OE HOE 2AR TAR SO TCO 2AE AK 2 8OR R (a) the and you of to They your He His daughters daughter Abraham Joseph was that LORD Then years replied because Jacob my said father with will him when them have brother had brothers God Abram place from there every were sons me So named This gave their Now may Isaac not for In Laban before wife land give came son she earth Esau after But her down these day who Out what went are man well took men one has by up saw on as be a Am go We all Let is old Us do it at I If (b) Figure 4: The most frequent words in Voynich manuscript (8am in Voynich computer code) and the Bible (The). The function WordCloud has been used in Mathematica R . I. Zelinka et al. MENDEL — Soft Computing Journal, Volume 25, No.1, June 2019, Brno, Czech RepublicX 5 The words in the text has been taken as graph nodes and its ordering edges of that (a) The words in the text has been taken as (b) Figure 5: The principle of the text-graph conversion. 5.3 The fractal analysis A central question about the Voynich manuscript is its authenticity - whether or not it is a fake. The veracity of the Voynich Manuscript has been hotly debated, i.e., whether it is a randomly generated nonsense strings, or whether it is written in some natural language. For our analysis, it is not relevant if the VM is open or encrypted. To find the scientifically supported answer and investigate this question, we use fractal analysis and complex network visualization on the text of the VM. For analysis of the Voynich manuscript, we used Mathematica R software and HarFA [20]. The Mathematica R  software was used for the visualization of our data and also for all computation and analysis joined with compex network approach. The HarFA has been used for fractal analysis. In the fractal analysis, we calculated the fractal dimension of the Voynich manuscript and the Hemingway’s novel The Old Man and the Sea, across 120 languages. In the second experiment, we also use the Bible in English translations. 5.4 The complex network visualization In the case of a complex network, visualization was used only Voynich manuscript itself and compared with randomly generated text and its visualization as the network, to show that the fake text, based on random word writing, has different network structure and network parameters. Thus among the many of our various experiments, fractal analysis and complex network analysis and visualization are reported here. The conversion of the VM into a complex network has been done in a simple manner. The words in the text have been taken as the graph nodes and its ordering as the edges of that graph. Simply, if the text is The words in the text have been taken as the graph nodes and its ordering as the edges of that graph, then it is taken as the The→words→in→the→text→have→been→taken→as→the→graph→nodes→and→its→ordering→as→the→edges →of→that→graph. The symbol →is taken as the oriented edge from Node1to Node2. The graph then looks as on Fig. 5. Such kind of visualization can capture important text features in a complex network structure. For example at Fig. 6 is visible that 110th word in the text is repeated, so there are two the same words. Because of the limited number of words in the VM, the line of the graph starts to create loops – nodes that become a local or global centre of the network. For example, the English Bible translation is one of the dominant nodes representing word the). An example of the VM network growth is in Fig. 7 and of The Bible at Fig. 8. The word the is in a class of lexical items called function words [22]. Function words are the glue that binds content words together in sentences. These include articles, conjunctions, temporal and causal connectives, and prepositions. We may hypothesize that the most frequently occurring words in the VM - as in other natural languages - are function words. The same has been done for the randomly selected words in order to simulate random-like-fake text. Two cases of randomly generated text were used. In Case 1, 500 words were randomly taken and converted to a network. The result is in Fig. 12. In the Case 2, to keep the experimental conditions as precise as possible, we generated a set of 7391 unique words (Set1), and then randomly created a set of 33663 words (Set2) from Set1. Then the same text-complex network was generated. This part was repeated 100×, and networks generated using this method are shown in Fig. 11. To compare VM with a well-translated book, we also selected The Bible. The Bible network is shown in Fig. 8. Unconventional Methods in Voynich Manuscript Analysis MENDEL — Soft Computing Journal, Volume 25, No.1, June 2019, Brno, Czech RepublicX 6 RSCG 4 OE TCG OE DCC8 2OE Figure 6: Word repetition. The color represent different degree centrality. In order to compare properly in some global view the complex networks attributes from random, VM, or Bible networks, we generated and used the following Algorithm 1. Input: TEXT: the text (Bible or Voynich); SSW: the max size of the sliding window; RW: the range of the sliding window; y= 10, x = 1: the counters; while y < SSW do •TEXTy= Take first ywords from position x; foreach TEXTyand x < RW do •CNy= convert TEXTyinto a complex network as on Fig. 10 ; •Calculate degree centrality of the CNy; •BDC = Select the biggest degree centrality from CNy; •Plot BCD as color on coordinates (x,y) as in Fig. 13; •x++; end •y++; •x = 1; end Algorithm 1: General building pseudocode for the text degree centrality map as on Fig. 13. The Algorithm 1 simply construct networks from the text of different length (see Fig. 13 y-axis from 10 to 200 words); for each text window size (string) is that done for positions from x=1 to 1000. For each network we calculated degree centrality 4and represented by color in the figure, which can be considered as the map of the degree centrality of the text over the specific text range. The number of word strings processed for those maps at Fig. 13 was 200,000 (200 ×1000) for each and 5,000,000 (1000 ×5000) for each Fig. 15. A 10,400,000 word sets from VM was processed to get degree centrality maps at Fig. 13 and 15. 4https://en.wikipedia.org/wiki/Centrality I. Zelinka et al. MENDEL — Soft Computing Journal, Volume 25, No.1, June 2019, Brno, Czech RepublicX 7 FGAG2 GDAE AR GHAM SOE SORG 0D0RC2 GDOR SOE8G 2ORG DZAR ORG DAIR THAM SOR HZAR 8ALA 2GAIIR SCDG OR (a) FGAG2 GDAE AR GHAM SOE SORG 0D0RC2 GDOR SOE8G 2ORG DZAR ORG DAIR THAM SOR HZAR 8ALA 2GAIIR SCDG OR GDAM SO8 HZOARG HZC2 8ARAM 2G 0OM OHCCG OHCAR ROEOHG HZAAR 8AM ODAM ODAL 2AIRG TCAR HZAM PZAR FZAM G8ARAISG O8AR SG PZOG OG8AR S 2 FZOAM SO8ARG GSCG SO8G ODTO G OHTOE TOHZG O2TG 8AN TOR DO2 FZOE HCO8G G8AN PZC2AM OE2 PZCG GHAN SOSG PZO8AEC2 ODSO DSOG OHAIRN OHCOE SO8AN 2DZCG SOG DZCG DO8AM PZG PZ08AIIE2 HZCG SC OE8AN 8 OM TOE O8AM TO8AN T8G ODON 80R HZG DO8 SDZCG DZOR (b) Figure 7: The growth of the Voynich network. The color represent different degree centrality. Unconventional Methods in Voynich Manuscript Analysis MENDEL — Soft Computing Journal, Volume 25, No.1, June 2019, Brno, Czech RepublicX 8 days later while they were still in pain two of Jacobs sons (Dinahs brothers Simeon 00 61 00 64 n Levi) took their swords went into the unsuspecting city slaughtered every male They killed Hamor his son Shechem with Dinah out Shechems house away other came upon slaughter looted because sister had been defiled flocks herds donkeys everything else or field carried off all possessions women children plundered houses Then Jacob said to Levi You have brought trouble me by making repugnant Canaanites Perizzites people this land We are few number if unite against attack I my household will be destroyed But answered Should he treated our like a prostitute God Arise go up Bethel Words visualized=171 From possition=22773 Neighborhood highlited=0 Figure 8: The Bible network, first 500 words - an example. 1.70 1.75 1.80 1.85 1.90 1.95 1.2 1.4 1.6 1.8 2.0 Fractal measure 1 Fractal measure 2 Figure 9: The Fractal analysis of the Voynich manuscript (red dots) vs. Hemmingway’s The Old Man and the Sea, across the 120 language mutations. On the xand yaxes are different fractal dimensions of the text. I. Zelinka et al. MENDEL — Soft Computing Journal, Volume 25, No.1, June 2019, Brno, Czech RepublicX 9