Groundbreaking AI project translates 5k-year-old cuneiform | Hacker News Reader

Groundbreaking AI project translates 5k-year-old cuneiform

timesofisrael.com

43 pointsisaacfrond18 comments

I wish more research would dig into the ancient Persian tablets as well. There’s just so much mystery around Persia, and unfortunately, in many contexts (movies, books, websites, or discussions in general), people act as if Persia never existed. This is despite the fact that Persians had such an undeniable impact on human history. Cyrus the great, for example, was the one who liberated Jews in Babylon and wrote the first human right cylinder.

At best, Persians are portrayed as bar bars or black mercenaries (absolutely nothing wrong with being black. But Persians were definitely not black).

> and wrote the first human right cylinder.

This interpretation of the Cyrus Cylinder is highly contested https://en.wikipedia.org/wiki/Cyrus_Cylinder

From Wikpedia: The interpretation of the Cylinder as a "charter of human rights" has been described by various historians as "rather anachronistic" and tendentious. It has been dismissed as a "misunderstanding" and characterized as political propaganda devised by the Pahlavi regime. The German historian Josef Wiesehöfer comments that the portrayal of Cyrus as a champion of human rights is as illusory as the image of the "humane and enlightened Shah of Persia". D. Fairchild Ruggles and Helaine Silverman describe the Shah's aim as being to legitimise the Iranian nation and his own regime, and to counter the growing influence of Islamic fundamentalism by creating an alternative narrative rooted in the ancient Persian past.

Unfortunately, academia is just as much a slave to geopolitics as governments, media, tech, sports, etc. Hopefully things will get better in the future.

From the byline:

“Accuracy is debatable”

Another case of AI hype and clickbait headlines meshing nicely.

> According to the study, the neural machine translation provided a BLEU4 score of 36.52 for cuneiform to English, and a score of 37.47 for transliterated cuneiform to English. BLEU4 scores are from 0 to 100, with 0 being the lowest and 100 being a perfect translation, which even a human translator could not achieve. Around 37 is considered fairly good for an early-stage translation model, explained Gutherz.

> Gutherz said that Google Translate, a privately-funded commercial tool that has been in existence for over a decade, would get a BLEU4 score of about 60 translating from Spanish to English.

It looks more like an early stage of a significant AI accomplishment to me. The glass is half full.

Because there's such a large corpus of 5000 year old cuneiforms waiting to be added to the training set? Ok, that sounds a bit snarky, but how would you expect to improve certainty when there is so little data?

There actually are...

"There are far more of these tablets in existence than I could have imagined. In fact, between half a million and two million cuneiform tablets are estimated to have been excavated in modern times, of which only approximately 30,000–100,000 have been read or published."

https://library.stanford.edu/blogs/special-collections-unbou....

I feel like this must be a really rewarding time to be alive for diligent archivists and demotivated researchers alike. I hope the revelation that proper cataloging could mean that it's (once again?) possible for single researchers or small teams to make discoveries and then immediately see their findings reverberate in a virtual instant throughout their field (as updated data sets are used to re-calibrate models) could serve as an incentive for better data collection and management too.

And those are just the tablets we've found. It's quite likely there are millions of other tablets just sitting in caves and and old, buried cellars waiting to be excavated. There are nowhere near enough Assyriologists in the world to even think about handling how many tablets we have access to, let alone the new ones we will find in the future, so machine translation will be a huge boon to the field, even if the translations are a little bit iffy.

Right now, Assyriologists will prioritize - if the tablets come from, say, someone's old garage, they are unlikely to be translated. If it comes from a palace or a temple, then the odds are better. But imagine the situation where someone ended up with a copy of the Declaration of Independence in their garage.

Seems like it's only a matter of time before this hits an inflection point and an explosion of insights from ancient cultures emerges. Exciting. Many long-held beliefs will be upturned. Many people will be digging their heels in...

> Many long-held beliefs will be upturned. Many people will be digging their heels in...

Yeah, according to the AI, the tablets say the people of Akkad had a great interest in 21st century culture, and they worshipped Harry Styles as a literal god.

It's true, Watermelon Sugar is clearly about the economic prosperity that followed Sargon's unification of the Babylonians.

Even if inaccurate, if it produces something that looks interesting, then a human can go back and verify it. Maybe this is best as bulk processing that gets it close, and then can be refined, verified, etc...

I see this as a potential step toward a "universal translator" type of project that can take any unknown language and translate it, with some level of accuracy, to a known language. Such early stage projects will be of dubious accuracy, but once the methods are improved, this could take over machine translation in general.

>Some of the translated tablets have information that is still relevant today. “If he cleans his garments, his days will be long,” according to one Akkadian scribe more than 3,000 years ago.

Ah, yeah... super relevant lol This must be why some people wear the same clothes every day. Faster days!

I assumed that it meant that keeping clean would keep you healthy.

Ah, that makes sense. My bad I just read it as an irrelevant phrase lol

And if the tablets are this confusing when the translation is vetted by experienced experts, adding a choppy translation software to the mix will make it a nightmare

Reply on news.ycombinator.com