Ancient data, modern math and the hunt for lost cities of the Bronze Age | Hacker News Reader

Ancient data, modern math and the hunt for lost cities of the Bronze Age

washingtonpost.com

125 pointsFtuuky21 comments

Cryptography and data mining/science (ML/AI) have their roots in the study of Epigraphy https://www.google.com/search?q=epigraphy Combined, they enable the process of discovery, new hypothesis generation which in turn can lead to new inventions and cures e.g. extending human lifespan which also happens to be the only way humans will make it to other habitable planets http://www.eso.org/public/news/eso1736/

> extending human lifespan which also happens to be the only way humans will make it to other habitable planets

[citation needed]

What about hibernation technology? Digital upload? FTL travel?

I know the last two seem vastly more fantastical than the first, or than age extension, but it seems odd to announce flatly that something is The One And Only Way.

Got another way?

Wow, so interesting. TIL that we have 10s of thousands of 4,000 year old clay tablets. What a data set!

This was the only downer of the article:

> As a final check, the authors ran the model against the location of known ancient cities to see whether its results matched the actual archaeological record.

> One (sic) two out of three of the known cities they tested against, the model nailed it. But it whiffed on the third.

2/3 is a bit troubling.

As well known, 2/3 ain't bad.

I think of it this way:

1. Suppose we tested on 1 datum and got 0/1. Not good.

2. How about just 2? 1/2. Hmm.. coin flip.

3. Ok, let’s do 3. 2/3. Well, better than half but it’s easy to get 2 heads on 3 coin flips.

4. ....

1/2 being equivalent to a coin flip is only true if we're looking at a binary value. In reality, much fewer than one in two coin flips would accurately pinpoint the location of an ancient city. Even if it only worked half the time, that would still be much better than random chance.

BTW, I believe the parent comment that you're replying to is a lazy joke referencing an old '70s song, not a serious argument.

Ah I missed it!

You make a good point.

> BTW, I believe the parent comment that you're replying to is a lazy joke referencing an old '70s song, not a serious argument.

Or your belief is a lazy attempt at understanding it, and you missed the double entendre even though you agreed with the point...

we have 10s of thousands of 4,000 year old clay tablets. What a data set!

Year 6017: we have 10s of thousands of 4,000 year old https://en.wikipedia.org/wiki/Floppy_disk . Hhmmmm...

Floppies from the 1980s/1990s are already magnetically degraded to the point of unusable, in most cases. (Source: Recent 2017 attempt to archive remnant data from period storage media before donating hardware to a museum)

That was my (sarcasm) point...unless we change, they'll find NOTHING.

Oh yeah. Excuse me for not winging your digit. Anyway, we concur.

Mesopotamian studies is really interesting. There's a lot of intact records from archives that were found in the basements or lower levels of buildings that were destroyed in fires -- clay tablets only break under the most intense heat.

Most are administrative records, but as TFA shows even these can be incredibly interesting. Only a small percentage have been translated -- there's not that many people with the diligence to learn cuneiform! But if you want to go to Oxford and spend 5-10 years learning ancient dialects and writing systems, it could be quite rewarding.

> But if you want to go to Oxford and spend 5-10 years learning ancient dialects and writing systems, it could be quite rewarding.

What of an online course?

If you know of one, post it in the comments.

The primary issue though is that the source material isn't digitized yet.

As it happens, whilst the sources mentioned in the article (Old Assyrian archival texts excavated in Kaneš, Turkey) are unfortunately not yet available online, there are a substantial number of cuneiform texts available in interactive, scholarly-edited forms on the Open Richly Annotated Cuneiform Corpus: http://oracc.museum.upenn.edu/projectlist.html

Admittedly there are swathes of cuneiform tablets which aren't available online yet (for example the OA texts, Old Babylonian letters, many tablets that haven't even been entered into museum catalogues yet) but the online corpus presents some very rich material.

There is also a Github repository https://github.com/oracc

(NB I am not affiliated with Oracc but I do use it heavily for research and am a big fan!)

This is a good site to visit for those interested in learning more about cuneiform script and the Akkadian language (an ancient language related to Arabic, Aramaic, and Hebrew).

http://oracc.museum.upenn.edu/saao/knpp/cuneiformrevealed/in...

The site itself is a public outreach site about divination during the late Assyrian empire (c. 680 - 620 BC), so there's a lot of interesting information on there.

You'll need to install a Unicode cuneiform font if one isn't already present on your system. Unhelpfully, Unicode doesn't distinguish between cuneiform sign forms from 2000 BC and those from c.800 BC, which are graphically simplified and easier to learn for a beginner (though less aesthetically pleasing). The 800 BC signs (called the 'Neo-Assyrian' script) are usually those taught in a first-year Akkadian class.

The 'CuneiformNA' font is available for download here: http://oracc.museum.upenn.edu/doc/help/visitingoracc/fonts/

Also on that page is a CuneiformOB (Old Babylonian) font, for the 2000BC sign forms.

Alternatively, Google provide a cuneiform font as part of their Noto project. These are also Old Babylonian sign forms:

https://www.google.com/get/noto/#sans-xsux

If you'd like to do some machine learning projects on it, here's where to get the data:

Machine assisted translation of cuneiform texts

https://cdli.ucla.edu/?q=news/machine-assisted-translation-c...

Are the models using logistic regression? I’ve seen logistic regression models end up in 60-70% accuracy band while deep neural network models end up in 80-90% with such data sets

If you have a look at the paper (figure 12), although the actual location of some cities lie outside the predicted confidence regions, the model gives a pretty good hint of where they are.

It is easy to see how the study could be followed adding other factors such as geographical data, or the goods that are traded between different regions. But still, this simple model gives a very good estimation.

I may be wrong, archaeology is not my field, but I'd guess 2/3 is a high probability by their standards. I do not think they usually have so many chances when they start digging.

Reply on news.ycombinator.com