Solving Ancient Riddles with Neural Networks

A page of the Voynich manuscript showing a number of plant like drawings and several paragraphs in an unknown language or encrypted text
Could the Voynich Manuscript just be early world building? Image in public domain.

I, like just about everyone who has ever heard of it, have been fascinated by the Voynich manuscript for years. The idea of an eldritch textbook, written in an encrypted script and with baffling, other worldly diagrams and drawings is ripe for all manner of conspiracy and conjecture. That it is over half a century old and has managed to survive relatively intact just fuels those fires.

Personally, though, I don’t subscribe to any of the extra-terrestrial, spiritual or religious interpretations for the book. Honestly, whilst any of those would be ground breaking and revolutionary to our understanding of the universe, they also dismiss something far more interesting. The Voynich manuscripts could be some of the earliest genre fiction on the planet! That idea is genuinely more exciting to me; to find definitive proof that people 600 years ago were just as happy inventing fictional worlds, in entirety, as I am today.

Of course, ancient fiction is well documented. We have plenty of examples  from much earlier in our history, but the Voynich manuscripts are somehow more interesting then the likes of Lucian of Samosata’s True History, at least in one particular aspect. They don’t tell the tale of some great mythology or legend, at least not one which is still known, and they appear to have been created by a single author, given how consistent the writing and art style is throughout the 200+ pages. To me, that would make this veritable tome less “just a story” and more a body of work akin to Tolkien’s Middle-Earth; early proof of actual world-building, for no reason other than fun. That we may have found a 12th Century Tolkien or Martin* is far more exciting to me, personally, then any of the more grandiose theories.

With that said, it may not be that long before the Voynich manuscript finally gives up some secrets. In one of the more interesting applications of neural-network AI I’ve seen, fellows at the University of Alberta have recently been targeting the text of the manuscript. The big issue with decoding the text isn’t just the encryption; with computers we should be able to make some headway on that front. However, you can only crack an encryption system if you have some idea of what the unencrypted message will look like. For that, we need to at least know the language which was initially encrypted. Of course, if the manuscript truly was created by a 12th Century Tolkien then it may have been written in a fictional tongue, making the whole exercise fruitless. Still, this is the very mystery that Greg Kondrak, with aforementioned AI in tow, may have managed to crack.

Having trained the AI on finding lingual patterns within text (not actually deriving meaning, but recognising the mathematical models that make up a language’s semantics and syntax) it was fed the Voynich manuscript. The result: Hebrew, with a high degree of certainty, appears to be the underlying language. That’s completely amazing to me. That a computer can teach itself enough about human linguistics that it can derive the language of a block of text based on glyph placement and frequency is astonishing, but that it can then use that same logic to decode gibberish into the underlying root language is mind blowing.

Of course, the result does come with some major caveats, the first being that this is still a best guess. The AI has found a pattern, that is certain; a pattern which closely matches archaic Hebrew. But until we can decode the text and find that it makes sense in Hebrew it doesn’t get us much closer. Here, Kondrak has also made some headway, believing that the text might be using a simple alphagram system. An alphagram takes a word and reorders the letters into alphabetical order, a fairly simple form of pseudo-encryption. For example, the word “example” would become “aeelmpx”; not instantly recognisable but not the hardest riddle to solve either.

Based on that hunch, the AI has run the text through a decryption algorithm and, again, seems to have hit pay dirt. A large amount of the output words are recognisably Hebrew, with more than 80% matching known Hebrew text. Unfortunately, the AI is no good at translating ancient Hebrew, and the few sentences they have tried don’t make much sense, but it’s a promising start. Hopefully, with the help of scholars more versed in the ancient language, some light may finally be shed on just what, exactly, the Voynich manuscript is or was. And that would be pretty darn awesome!


* Or, indeed, an ancient dungeon master!

Forgotten & Surreal Instruments [#42]

Two nights ago we had the privilege of listening to the latest show put together by the Society of Strange and Ancient Instruments. Never heard of them? Well, neither had I. In fairness, had we not known one of the musicians (who, it turns out, was stepping in for another member) we still wouldn’t know about the Society and certainly wouldn’t have heard them perform.

Which is rather lucky, because both the members and the performance itself were brilliant. I had absolutely no idea what to expect and, frankly, even if I’d read a blurb or heard an explanation I doubt my expectations would have fitted the reality. In brief, the hour-or-so performance was a string quartet playing a medley of medieval and modern compositions, interspersed with readings from Sir Francis Bacon, the 16th century philosopher and naturalist. Oh, except the string instruments on which they played were not your standard violins, cellos etc. but rather the likes of the trumpet marine (one of the only ones in the world), viola bastarda (the only one in the world), gothic bray harp and utterly bizarre looking nyckelharpa (to name only a few, with each musician swapping instruments at least once throughout the performance). Layered on top of these medieval instruments were sound recordings, (occasional) electronic music and various distortions – these being the remit of the fifth musician, whom we know.

The result is a fascinating walk down both scientific and musical history. Francis Bacon has some wonderful excerpts on the nature of sound, the concept of his titular “Sound Houses” (from his New Atlantis, which I really must read) and Medieval anecdotes about noise in general. I doubt that a group of revellers truly did clap hard enough to “make the air thinner and cause the birds to fall from the sky” [paraphrased] but his observations on sound propagation are wonderfully modern and his vision for the future of music consumption is, at times, eerily prophetic. The excerpts were performed brilliantly by a live actor, lending a very clever degree of wit, movement and comedy to proceedings.

Alongside, and greatly overshadowing, the classical excerpts from Bacon was the music itself. For the most part, the musical performance was just incredibly well played Medieval fare. It’s a genre which I love, because it feels somehow incredibly alien whilst the core rhythms and structures, at times, feel almost pop like. The sound mixing was brilliant, allowing the whole medley to produce a wonderfully lyrical and complementary music whilst still permitting you to hone in on individual instruments, which you absolutely wanted to do from time-to-time. In particular, the trumpet marine was fascinating. A single stringed instrument with dozens of hidden vibration panels and sympathetic strings it could produce a bizarre array of sounds. I’m not so convinced as to the claims likening it to a trumpet, whose sound is much fuller and cleaner, but there is definitely a level of reverb and rasp which I’ve never heard from a stringed instrument before. I’m genuinely tempted to donate to their crowd funding campaign to get four of them made, just to hear what several of these instruments could produce together! Personally, though, my favourite was the weirdly altered bray harp, with small pegs fastened to the end of each string creating a very harsh, discordant noise completely antithetical to the classically perceived sounds of a harp.

Whilst there were new compositions interspersed with ancient, you really couldn’t tell them apart, and although the mixture of Medieval strings and modern audio sampling sounds odd on paper, in reality it worked extremely well. One of the last pieces performed, with the most electronic overlays, in fact felt ripe for sampling beneath a grime track, a combination I would happily pay to listen to. When paired alongside the esoteric poetry of Bacon’s prose and the atmospheric setting, an equally ancient church in the centre of a mist enshrouded town on the Moor, the result was wonderfully evocative and utterly riveting. An extremely fun way to spend the evening!