9/18/23 Speaking to the Ancient Past

The English expression "language barrier" reveals part of our mental conception, at least in English, toward other languages. "Barrier" implies a wall, a fortress, built around each language and harboring secrets within. We each live within our own fortresses of our native language or languages, and if we dare to go searching out for other one, we run up against a wall. The knowledge, methods, beliefs, and ideas of that language are all inaccessible to us, initially. 

Of course, language barriers and boundaries exist not only between different modern places, but also between our modern time and the past. Since 1945, for example, the languages spoken in North and South Korea have divergedsignificantly, thanks to decisions of the North Korean government and its isolation from English or other influence. These systematic changes have constructed a barrier, one that stands not only between the two countries but between both countries and engaging with their past. For another example, cuneiform is the earliest known writing system, used and adapted to write in several ancient languages, like Akkadian, Hittite, Elamite, and Urartian which were used in the modern day Middle East. Archaeologists have dug up over half a million cuneiform tablets, but less than 20% of them have actually been read. They still fall beyond the barrier.

So researchers have access to this massive wealth of early human writing, but we can't actually read what humans 5000 years ago thought important enough to write down. Why not?  Knowing what people in the past wrote down is worthwhile for academics in many fields, not just archaeology, but also history, anthropology, linguistics, and more. But the painstaking process of reading a cuneiform tablet involves copying down the glyphs by hand, transliterating all of them individually and then translating that into a modern language, typically English. And, not only is the process itself long, but to even begin translating a tablet requires years of specialized study of ancient languages, not exactly a lucrative financial choice.

But the number of translated tablets is likely about to soar upwards. Over the summer, a team of archaeologists and computer scientists in Israel created an AI-based tool that can almost instantly transliterate and translate Akkadian cuneiform writing into English. While not perfect, the translations score highly on the machine translation evaluation scale, and one of the aforementioned human scholars can iron out the errors to make the translation presentable.

This type of headline, and not just language related, is constantly in the news lately. "Researchers develop new AI to accomplish X task in Y field," almost copy-pasted, to the point where the term AI itself blurs and seems to lose meaning. But in this case, the artificial intelligence, built using large language model tools like neural networks, is giving us what is essentially a time machine. 

This time machine is significant not only because it will facilitate scholarship, but because it cuts down on the amount of know-how and labor required to understand these texts. In the same way that instantaneous machine translation allows us to communicate with people in other languages who otherwise would have been in fortresses beyond our reach, this tool allows ordinary people, people who don't have the means or time to dedicate years of their life to learning an ancient language, to hear what the past has to say. 

But how exactly is the past getting to us? The research was conducted in Israel, but the AI's default is to translate into English. And when the researchers were painstakingly translating by hand, they translated into English as well. Even though translation between English and other world languages is simple with machine translators, the default to English translation shows its hegemony as a lingua franca in most academic fields. And anything translated from the cuneiform to another language would have to first go through English. In this letter to the editor in the Lancet medical journal, an editor of another journal argues for the adoption of English-language publishing across the board, even going as far as to say that languages compete in a Darwinian process of evolution, with English being the current 'most fit' — ignoring the value of research and knowledge produced in a diverse set of languages, and the knowledge produced when languages interact, as well as the legacy of colonization that led to English hegemony. 

So the past is speaking to us — through a machine, and primarily in English. Thanks to an artificial intelligence, we can read the words of the people who lived thousands of years ago and first began to write things down in clay. Many students choose to study languages because of the joy of making connections with people across the world, people of different backgrounds and lives. We wouldn't choose to dedicate years of our life to understanding a language that died 2000 years ago. But there is a sense of wonder in the ability to understand humans, our ancient ancestors, even if they can't speak back to us. And really, when you scour through your language textbooks, study for vocab quizzes, give presentations in class, you're doing the same kind of painstaking work of these archaeologists and computer scientists — tearing down the language barrier, one brick at a time, so you can speak to the people beyond it.

For more, read the peer-reviewed paper about the new Akkadian translation tool, and read about how some researchers think that cave art is linked to the origins of language. Maybe one day AI will be able to 'translate' the etchings of those even more ancient humans.