Entropic Evidence for Linguistic Structure in the Indus Script

April 23, 2009 5:05 PM

Scholars at odds over mysterious Indus script. The Indus script is the collection of symbols found on artifacts from the Harappan civilization, which flourished in what is now eastern Pakistan and western India between 2,600 and 1,900 B.C. A new analysis using pattern-analyzing software suggests that the script may constitute a genuine written language. [Via]
posted by homunculus (20 comments total)

Previously.
posted by homunculus at 5:06 PM on April 23, 2009

*hugs homunculus just ever so slightly longer than is mandated by the awesomeness of this post.*

If I had a choice for a superpower, it would be the ability to travel back in time and sit undetectable while observing the amazing advances of life from random molecules interacting to humans developing language. Also, a lot more.
posted by Science! at 5:28 PM on April 23, 2009

Looks intriguing, but I can't read the actual paper and the Wired piece is breathless PR. I'll wait to see what the skeptics say. In that previous thread, I said "barring some unexpected discovery we're unlikely ever to know whether it represents a language or not, let alone what language it might be"; this may or may not be that unexpected discovery. In any case, thanks for the post!
posted by languagehat at 6:01 PM on April 23, 2009

Yes, great post. I'm always awed that such ancient mysteries not only exist, but are sometimes understood. I'll be curious to see how this one unravels (or not) over time.
posted by Dee Xtrovert at 6:18 PM on April 23, 2009

I have no idea how they get from 'pattern' to 'this word means cow'. I am not a linguist. Languagehat, do you know what process linguists take?
posted by kldickson at 6:44 PM on April 23, 2009

I have no idea how they get from 'pattern' to 'this word means cow'.

I don't think that they do get from "pattern" to "cow". All that the new research does is say that the grammar patterns in the Indus script seems to fall within the parameters of where actual, spoken languages fall. Therefore, it's probably the script of an actual ancient language.

Some scholars have had a "so what" reaction. So what that it's probably from a spoken language - doesn't help us to decipher what the symbols really mean.
posted by gemmy at 6:54 PM on April 23, 2009

languagehat, you aren't missing much. The paper is extremely short: not quite a page worth of text, one column of footnotes, and two small charts. There isn't much contained in it that isn't in the Wired and New Scientist articles.
posted by jedicus at 7:03 PM on April 23, 2009

||H♅| <- First Post! Sorry, logomanic episode
posted by BrotherCaine at 7:25 PM on April 23, 2009

My first thoughts were: "Awesome! I hope someone incorporates this into some kind of mystery novel where they go all Indiana Jones breaking the secret code!"

Clearly, I've been reading too many fluff mysteries of the "manuscript lost to history containing HIDDEN SEKRITS" variety. Still, that would be a pretty awesome plot point.
posted by grapefruitmoon at 7:47 PM on April 23, 2009

I'm sceptical. This is a paper by (mostly) non-linguists published in a non-linguistics publication. The argument could be made that they're simply dealing with 'patterns' common to all languages, and so no particular knowledge is needed. Languages, and therefore the scripts which encode language, are not random, this is perfectly true. However,

When aimed at language, this statistical technique comes up with a measure for the "orderedness" of words, letters or characters – from totally ordered to utterly random.

We simply don't know what kind of information these symbols stand for - whether sounds, words, grammatical markers, and so on - or even whether the they encode information in different ways depending on the symbol or its position. Moreover, we're not even sure what is two symbols or two variants of one symbol, making statistical studies potentially useless. But this paper doesn't even attempt to look at any of this, they only look at 'orderedness' when compared to 'randomness', as the rebuttal points out. Needless to say, we already know the symbols are ordered, as is true of most things humans do deliberately.

I read Farmer's main paper a few years ago, and liked it a lot, but I remember there was also a good rebuttal of some of its main points too. The main problem I (amateur that I am) have with the argument that Indus symbols don't encode language is that it makes them illiterate. We know the Indus civilisation was sophisticated and had substantial long term contacts with literate societies, but didn't directly borrow writing. This make sense if they already had writing, but requires explanation or acknowledgement otherwise.


I have no idea how they get from 'pattern' to 'this word means cow'. I am not a linguist. Languagehat, do you know what process linguists take?

Eh, the problem here is really about the script: is it a script? what kind of script is it? how does it work? We're not even close to knowing how to understand the first thing about the language, as we don't even know what we're looking at. 'Decipherment' is always somewhat of a catchall, ranging from known languages in unknown scripts (easy-ish once you have an opening), through unknown languages in known scripts (slow slog building up of vocabulary and understanding), to unknown languages in unknown scripts (hard to impossible). Some researchers take external data from potentially cognate languages or structurally similar scripts in order to turn 'unknown' to 'known' and make the job easier. But as this post makes clear, the idea that there is a language beneath these symbols, or even that the symbols have the kind of order to encode any information - linguistic or otherwise - is up for debate.
posted by Sova at 7:50 PM on April 23, 2009 [3 favorites]

So... if I understand it right, the NewScientist article is saying that Rao and his colleagues studied the conditional entropy within a bunch of the little sequences on these tablets and stamps, and found that the symbols within each sequence had about the same amount of randomness as a language. Okay.

Link number two (totally awesome post, btw) says that they've found thousands of these stamp seals, and each one has an animal and a sequence of symbols on it. Could each stamp represent a single word in the Indus language? I'm picturing an ancient scribe who stamps sentences into clay or mud using multiple seals, instead of custom carving out each seal for a single purpose, like the Mesopotamians apparently did. Wouldn't that effect the way that the conditional entropy is calculated?
posted by Kevin Street at 9:07 PM on April 23, 2009

I guess what I'm trying to say is, wouldn't the conditional entropy of a random collection of words (or Indus seals) be different from the conditional entropy of an actual document in a language? Context, word order, sentence construction and so on, isn't that where the non-randomness of a language comes from?
posted by Kevin Street at 9:32 PM on April 23, 2009

I'm waiting for them to tackle Linear A.
posted by Araucaria at 10:41 PM on April 23, 2009

Yeah, suppose the people who think that the script is symbols for some non-liguistic purpose are correct. I expect that the iconography of a god, or mnemonic signs for recipes, or indeed any sort of communication are going to exhibit non-random frequency and clustering.

Eg, think of the visual grammar of pictogram signs at airports. Arrows are always at the extreme left or right of a sign. They are orientated to one of eight compass points. A man sign (for male toilet) is usually in close proximity to a female sign, sometimes there is a wheelchair.

So the presence of order doesn't in itself prove that these signs represent a script for a human language. It seems to me that there is a missing step of showing what patterns are exhibited in writing systems but not in other kinds of collections of signs.
posted by i_am_joe's_spleen at 10:46 PM on April 23, 2009

My layman's perspective is that of course it is a language, like, duh. Why would ancient people devote resources, time, effort, into making meaningless random scribbles?
posted by Meatbomb at 11:48 PM on April 23, 2009

More insomniac ramblings...

The harappa.com website mentioned in homunculus's "Previously" has a lot more information. Apparently there's plain rectangular seals with symbols but no animals, and square seals like the ones pictured in the NewScientist story that come with animals. So right there you've two different kinds of objects.

The square seals often (but not always) have unicorns, and the unicorns are always shown with a sacred object. And there's a boss on the back of them where a cord can be run through. So the square seals might have been worn around a person's neck. (An ancient nametag?)

And apparently they've found both Indus seals in Mesopotamia, and seals with Mesopotamian letters on them in the Indus ruins. So there was trade and/or communication going on.
posted by Kevin Street at 12:43 AM on April 24, 2009

Thanks, Sova; I'm sticking with my original skepticism.
posted by languagehat at 6:18 AM on April 24, 2009

"Thanks, Sova; I'm sticking with my original skepticism"

This agglomeration of bizarre symbols constitutes the entire "languagehat text" - an artifact and mystery as old as time. Discovered in a dank corner of the ancient website "metafilter" - and whose carbon dating suggests that it was composed as far back as yesterday - the pictograms hint at the possibility of an actual human language being regularly used by the primitive language-hat species of the distant past.

If - as some scientists suggests - the comment above was actually used to communicate some crude "meaning," it would revolutionise our understanding of protolinguistics. Yet many are not convinced: the idea that primitive homonid such as languagehat could create a working communicative system that we would recognise as a language represents a paradigm shift in our understanding, and would mean that we have to actually listen to him when he criticises many of our most cherised AskMe answers.
posted by the quidnunc kid at 3:53 PM on April 24, 2009 [1 favorite]

The harappa.com website mentioned in homunculus's "Previously" has a lot more information. Apparently there's plain rectangular seals with symbols but no animals, and square seals like the ones pictured in the NewScientist story that come with animals. So right there you've two different kinds of objects.

The square seals often (but not always) have unicorns, and the unicorns are always shown with a sacred object. And there's a boss on the back of them where a cord can be run through. So the square seals might have been worn around a person's neck. (An ancient nametag?)

Indus symbols are recognisably found on a variety or objects, though mainly seals. Similar graffito marks on pottery also exist, as does a visually related repertoire of painting. 'Texts' can be found alone, or with pictures, or indeed with geometric marks not usually considered to be writing themselves. The 'unicorn' you mention is the side view of a long-horned bull, and the 'sacred object' is just another way of say 'we don't know'. Impressions from seals have been found in warehouses, and so obviously fulfilled the function of creating identifying tags for goods - a common use for seals elsewhere. But despite all this, there is still no definite way of knowing whether or not they encoded language.

A good moderate description of the Indus symbols and their context can be found in the works of Parpola. He considers it to be a script, most likely logo-syllabic, and believes that the underlying language must be some kind of Dravidian. He even goes so far as to proffer a very limited number of 'readings' for various sign combinations. But I have to stress, as above, that everything he says is tentative, and I think Farmer more or less shit in his bed.

And apparently they've found both Indus seals in Mesopotamia, and seals with Mesopotamian letters on them in the Indus ruins. So there was trade and/or communication going on.

Trade went on for centuries, and there was even a settled colony of traders in Lagash (southern Iraq today). There is no doubt whatsoever that the Indus people knew about writing, hence my qualms about making them illiterate. The Indus seals found in Mesopotamia (supposedly) have different symbol patterns/frequencies from those found in the Indus valley, and have been proposed to belong to Indus traders using Sumerian/Akkadian names (it was very common back then to translate your name word for word into another language).

Oh, one last thing, if you want a great description of how decipherment takes place The Decipherment of Linear B by John Chadwick is well worth reading.
posted by Sova at 6:03 PM on April 24, 2009

One of the authors of the Science paper explains and defends his work:
The Indus 'non-script' issue is a non-issue.
posted by Idle Curiosity at 1:21 AM on May 4, 2009