How AI is helping us fill in the gaps in ancient languages
24-03-2022 | By Robin Mitchell
AI is an extremely powerful tool, and researchers have been using it to complete ancient transcripts that are damaged and missing information. What challenges do ancient writings present, what have researchers been able to do, and could this allow us to understand any ancient writing system?
What challenges do ancient writings present?
Human history extends for millions of years into the past, but only the past few thousand have been recorded. Those who wrote down the accounts of great kings and queens of the past believed that their words would tell great stories and achievements of their god-like deities for millennia. However, what these scholars didn’t account for was that their language may change over time or, worse, be entirely forgotten.
As a result, there are ancient languages whose messages cannot be truly known for sure. In some rare cases, there may be a piece of scripture written in multiple languages that can be used by modern archaeologists to decipher ancient languages, with one prime example being the Rosetta stone. Before discovering the Rosetta stone, the meaning of hieroglyphs from ancient Egypt could only be somewhat guessed, but the Rosetta stone included the same passage written in three languages with one, Greek, already being known. From there, both hieroglyphs and demotic could be essentially reversed engineered and understood.
Another challenge that is often presented by ancient texts is their completeness. Finding a piece of recorded history from 3,000 BC can be unbelievably challenging. Almost all recording mediums would be made from organic compounds that would break down (unless written into stone). Fragments do survive, but fragments are incomplete text whose true meaning is lost to time.
How AI is being used to recover original text from fragments
If there is one thing that is true with humans, it’s that we are very predictable. If a piece of written text isn’t about the weather, then it could be about religion, if it’s not religion, then it could be traded, and if it’s none of these, then it is probably something perverted. In fact, many ancient artefacts and drawings depict illicit acts and embarrassing body parts simply because humans are fascinated with the subject (this actually presents a strong argument that the first pictures we send into space should be of “reproductive parts” as they represent our species).
The predictability of datasets leads to patterns, and one modern invention that is excellent at recognising patterns is AI. Researchers have turned their attention to AI for reconstructing fragments of old text based on what content was written about during the time, the meaning of the fragments, and the predictability of humans.
The new tool, called Ithaca, utilises Google’s DeepMind AI that has been trained specifically on large datasets from ancient Greek inscriptions acquired by the Packard Humanities Institute in California. So far, the system has an accuracy of 62%, which is surprisingly high considering that the AI is trying to extrapolate data from virtually nothing. Furthermore, the new system can also date writing to within 30 years (remember that as time progresses, style and spelling changes). If the system is used in conjunction with humans, the overall accuracy of recovering text jumps to 72%, which can help gain valuable insight into the past.
Could AI be used to decipher languages of the past?
Considering that humans are predictable, as are their languages, it is possible that a future AI could be fed an unknown language of the past with some example text and then decode the language. This would potentially revolutionise the field of archaeology as any inscription could be understood and therefore give great insight to the people who lived thousands of years ago.
Whether or not this is possible is not clear as humans have been struggling to decipher many ancient languages for hundreds of years. Furthermore, other considerations need to be made, such as imagery that sounds like a piece of text which may change the meaning of a piece of text. To make matters more complex, some languages may descend from other lost ancient languages meaning that they have nothing in common with already understood languages.
AI is an exciting tool, and its use in the field of languages raises some fascinating opportunities.