The Weekly Mystery: AI Attempts to Read the Voynich Manuscript
This manuscript from Middle Ages has not been decyphered since 400 years, and modern AI models can not understand it as well! How is that possible?
For this week's brain-teaser, let me introduce you to the Voynich Manuscript. This book has been researched for almost 400 years by linguists, historians, cryptologists, and mystery amateurs, but nobody was able to read even a single word. The decyphering of this fascinating manuscript is ongoing, and today we would address the question if AI, capable of finding patterns invisible to the human eye, could help us understand this mystery.
The Voynich Manuscript is a handwritten book that has been carbon-dated to the early 15th century, however, neither the author nor the country of origin could have been determined. It features illustrations that are unique to this manuscript, and its text, written by five different scribes, was written in an unknown language and cipher that has yet to be translated. Even the content and purpose of the manuscript are a mystery, with theories ranging from a secret code to encrypted information about alchemy, medicine, or astronomy.
The decoding attempts have been many, starting from the moment the manuscript was discovered in the 17th century. Some have proposed that it’s a simple substitution cipher, while others believe it’s a more complex code that includes symbols representing words or ideas rather than individual letters. Some researchers have suggested that the manuscript is a work of fiction, a hoax, or a ciphertext created as part of a secret society. There is no definitive answer, and the mystery surrounding the Voynich Manuscript continues to capture the imagination of people around the world. Excellent online resources, such as www.voynich.nu, provide valuable information and starting points for your own research.
Mathematical and AI approaches have been used in attempts to unravel the mystery, including:
- Statistical analysis: Using frequency analysis, n-grams, and entropy analysis to determine if the text follows any recognizable patterns.
- Image processing: Using computer vision techniques to identify patterns in the manuscript’s illustrations.
- Machine learning: Training algorithms on historical texts to identify patterns in the Voynich Manuscript, and comparing them to other known languages.
- Cipher-breaking techniques: Using various cipher-breaking techniques such as frequency analysis, brute-force methods, and genetic algorithms to try and crack the code.
The student research project from the University of Adelaide in 2014 attempted to decipher the manuscript using computational analysis techniques such as Word Recurrence Intervals and N-gram relationships. Additionally, supervised learning algorithms such as Support Vector Machines (SVM) and Multiple Discriminant Analysis (MDA) were applied, however, the efforts were unsuccessful.
In the more recent developments, the International Conference on the Voynich Manuscript 2022 that took place at the University of Malta also introduced some new machine-learning approaches for decyphering the manuscript. The topics varied from the identification of parts of the text written by different scribes; to the classification of probable languages used; to the discussion of whether the text was written in the real language or just gibberish.
Despite these modern attempts, the contents of the Voynich Manuscript remain a mystery, and its true meaning and purpose remain unknown.
More statistical approaches to solving the mystery of the Voynich Manuscript are summarised in the excellent blog Computational Attacks on the Voynich Manuscript.
Resources:
You can view the Manuscript online at https://collections.library.yale.edu/catalog/2002046.
If you are interested to find out more resources about the Manuscript, please refer to the bibliography at http://www.voynich.nu/refs.html#abbott.
International Conference on the Voynich Manuscript 2022: https://www.um.edu.mt/__data/assets/pdf_file/0019/503335/Conferencebookofabstracts.pdf
