Tamil Brahmi inscriptions and other archaeological finds

[P_R]

http://www.hindu.com/2009/04/27/stor...2756902000.htm
Indus script does encode a language

A. Srivathsan & T.S. Subramanian
New study reported in Science shows it was no mere ‘chain of symbols’

[html:5360fa9373]<img src= "http://www.thehindu.com/2009/04/27/images/2009042756902001.jpg">
[/html:5360fa9373]
An Indus seal from Mohenjodaro.
Chennai: Computation science, information theory, and machine learning have now come to the vindication of Indus Valley scholars – providing a new type of “quantitative evidence for the existence of linguistic structure in the Indus script, complementing other arguments that have been made explicitly or implicitly in favour of the linguistic hypothesis.” This quantitative evidence comes from the results of a statistical study published online recently in the journal Science ( http://www.sciencemag.org/cgi/conten...ract/1170391v1).

Drawing from multiple disciplines, using rigorous equations, and through scientific number crunching, a team of scientists — including the well-known Indus script scholar, Iravatham Mahadevan — have demonstrated that the Indus script encodes a language and is not a mere “chain of symbols,” as an article published in 2004 claimed.

The seals and tablets of the Indus civilisation that flourished between 2500 and 1900 B.C carry examples of what has long been understood to be writing in an unknown language. Despite many attempts, the script, known for 130 years, has not been deciphered. The 2004 article, published in the Electronic Journal of Vedic Studies, challenged the idea that the Indus script encoded language and suggested that it might have been a non-linguistic symbol system like the Vinèa inscriptions of southeastern Europe and the Near Eastern emblem systems.

The new statistical study compared the pattern of symbols found on Indus Valley artifacts to five types of natural linguistic systems (the Sumerian logo-syllabic system, the Old Tamil alpha-syllabic system, the Rig Vedic Sanskrit alpha-syllabic system, English words, and English characters), four types of non-linguistic systems (including human DNA sequences and bacterial protein sequences), and the artificially created computer programming language, Fortran.

The decisive finding was that “the conditional entropy of Indus inscriptions closely matches those of linguistic systems and remains far from non-linguistic systems…The similarity in conditional entropy to Old Tamil, a Dravidian language, is especially interesting in light of the fact that many of the prominent decipherment efforts to date…have converged upon a proto-Dravidian hypothesis for the Indus script.”

The study is the collaborative work of Rajesh P.N. Rao, a University of Washington computer scientist; Nisha Yadav and Mayank N. Vahia of the Department of Astronomy & Astrophysics at the Tata Institute of Fundamental Research, Mumbai; Hrishikesh Joglekar, a software engineer from Mumbai; Ronojoy Adhikari, Faculty Fellow at the Institute of Mathematical Sciences, Chennai; and Mr. Mahadevan at the Indus Research Centre, Chennai.

Dr. Adhikari, who specialises in Novel Applications of Statistical Mechanics, has no doubt that that the Indus script was part of a structured language. Opening his Nokia mobile phone, he types the alphabets H and A one after the other. The messaging service automatically fills the next two slots with V and E. “This,” he says, “is a simple algorithm the mobile phone uses to help you complete a word quickly. It works on the principle of correlation. In English, when you use the alphabet Q, the next one that follows is often U. Every language has a probability or flexibility of what token would come after another. A token could be an alphabet or punctuation or any component of the linguistic system. We have used the idea of entropy to measure the non-randomness in a linguistic system including the Indus script.”

When Dr. Adhikari and his collaborators compared the conditional entropy of the Indus script with the conditional entropies of the various linguistic and non-linguistic systems, the results provided “quantitative evidence for the existence of linguistic structure in the Indus script.” “The Indus script,” he explains, “comes close to the entropy value of Old Tamil and lends credence to the debate that the Indus script is connected with the Dravidian language.”

The use of statistical methods is not new to research on the Indus script. The point of departure in the new study is the use of rigorous correlation techniques, a significant methodological advance.

Work on the Indus script continues. The temporal and spatial analysis of the script has been completed and awaits publication. There is scope to compare the Indus script with systems like the Chinese pictograms and the Egyptian hieroglyphics. Dr. Adhikari believes that all these efforts “are taking us closer to understanding the Indus script.”

[crajkumar_be]

PR,
Was about to post that

Decent similarities to Tamil and Sanskrit. So, where does it leave the Aryan Invasion theory and its criticisms?

[P_R]

Decent similarities to Tamil and Sanskrit.

Where does it say so ? AFAI understand they compared the symbol-sequences with several language syntaxes including Rig Vedic Sanskrit and old-Tamil. The alternate hypothesis was a comparison against certain non-linguistic sequences and it was found that it was close to the linguistic sequences. Thus the paper makes the claim that the Indus Valley sequences are likely to be representative of a language.

I am sure they must have made sure the choices of the sequences in the two sets was statistically unbiased. The proximity to old-Tamil is pretty exciting. It is also similar in Sumerian which is believed to be contemporaneous to the Indus valley civilization.

Whether the choice of nonlinguistic sequences is wide and unbiased enough is not clear. Since it was published in Science hopefully one can hope so.

For old Tamil they have used Ettuthogai, which even by the most aggressive estimates is 1500-2000 years after the Indus valley script !

Among the non linguistic sequences they have included a Fortran program for solving a particular physics problem. This confuses me a bit on how linguistic is defined. "If then else" type structure can also be called linguistic isn't it ? Perhaps podalangai can clarify.

btw, one of the authors is a software engineer from Oracle.

[crajkumar_be]

Decent similarities to Tamil and Sanskrit.

Where does it say so ? AFAI understand they compared the symbol-sequences with several language syntaxes including Rig Vedic Sanskrit and old-Tamil. The alternate hypothesis was a comparison against certain non-linguistic sequences and it was found that it was close to the linguistic sequences. Thus the paper makes the claim that the Indus Valley sequences are likely to be representative of a language.

Adha thaan naan mean pannen Doesn't mean we can conclude its similar to Tamil or Sankrit-a?
adhavadhu, randomness illama oru grammar irukkara madhiri theriyidhu nu solraanga, right?

The decisive finding was that “the conditional entropy of Indus inscriptions closely matches those of linguistic systems and remains far from non-linguistic systems…The similarity in conditional entropy to Old Tamil, a Dravidian language, is especially interesting in light of the fact that many of the prominent decipherment efforts to date…have converged upon a proto-Dravidian hypothesis for the Indus script.”

[complicateur]

Bala,
This link might provide more insight into what conditional entropy means in linguistics. Similar conditional entropy does not necessarily mean that the languages are similar. Just that they are at similar levels of sophistication. In fact the Indus Valley Civilization may not be similar to Rig Vedic Sanskrit or Thamizh. The paper is trying to refute an earlier paper's claim that the Indus valley civilization may not have been 'literate'.

[P_R]

Invented datasets.... It was pretty much the same doubt I had when I looked at the information about the datasets I used. Seemed too specific and even handpicked. The randomly generated sequence seemed odd, but I assumed 'scholars' would know what they were doing.

How they make a broadbased conclusion seemed doubtful. And the entropy measure - because even I understood it - seemed too simple to be sufficient. But all said and done the proximity to proto-Tamil than Sanskrit seems like a vengaLa kiNNam. Or its just me grasping at straws.

My lack of clarity remains regarding what is a linguistic system and what is a non-linguistic system ?

[crajkumar_be]

Thanks Compli,

Will *attempt* to understand whatever little i can

[podalangai]

Invented datasets.... It was pretty much the same doubt I had when I looked at the information about the datasets I used. Seemed too specific and even handpicked. The randomly generated sequence seemed odd, but I assumed 'scholars' would know what they were doing.

How they make a broadbased conclusion seemed doubtful. And the entropy measure - because even I understood it - seemed too simple to be sufficient. But all said and done the proximity to proto-Tamil than Sanskrit seems like a vengaLa kiNNam. Or its just me grasping at straws.

PR and others: You'll be interested in taking a look at Rahul Siddharthan's blog. Unlike Mark Liberman, he's a computer scientist, not a linguist, and he has a rather different take on the topic. Given that most of the authors were computer scientists, this gives a fairly good insight into what they thought they were doing:

http://horadecubitus.blogspot.com/20...really-do.html

The discussion in the comments is particularly interesting. The main issue that others have taken with the paper is that its null hypothesis (i.e., the artificially constructed non-linguistic sign sets) is not realistic. The comments on Rahul Siddharthan's blog, however, suggest that these were intended to represent controls, rather than a null hypothesis. This puts a rather different cast on things, though it does call into question the adequacy of their datasets. That apart, as has been pointed out on academic lists, they're lumping too many years together in one dataset.

In any event, they've said more papers using more datasets will be forthcoming and, if reports can be believed, Mark Kenoyer will be working with them on an updated paper. It'll be interesting to see how well those results match up with the preliminary analysis contained in the paper in Science.

[P_R]

Thank You podalangai. For the nth time....please post more often.

Good discussions in Rahul Siddharthan's blog. Enjoyed reading them. He kind of concedes it is "an overstatement to say the paper "proves" anything"

As the anonymous commentator in his blog put it

the heart of the matter then is the basis of the claim that the curve for any non-spoken script must lie "close" to either "Nonling Type 1" or "Nonling Type 2". Is that something generally accepted?

That is not clearly answered in the discussion there.

The same anonymous commentator ask a more vital question:

if a non-linguistic script can have an entropy value in the range of the linguistic scripts. That again is yet to be answered.

I thought Richard Sproat's comments (what does the space of non-linguistic systems look like) were also pretty much on-the-ball. But soon it descended to an appalling fistfight.

Though his attribution of motives and outright dismissal of counter-opinion is in bad taste I still think his words summarized things as they stand:

1 First we would need a much wider range of languages and
different types of writing systems.

2 Second we would need some serious corpora (not made up examples) of non-linguistic systems

In his third post on the paper Rahul Siddharthan writes thus.

Given the data for other languages in the Rao et al. paper, I would estimate P(D|HL) to be close to 1. That is, if the Indus script is a language, I would think it very likely that conditional entropies would closely resemble the data that Rao et al

I think that interpretation is right. But that significantly diminishes the 'interestingness' of Rao et al.'s paper.

[podalangai]

Though his attribution of motives and outright dismissal of counter-opinion is in bad taste I still think his words summarized things as they stand:

1 First we would need a much wider range of languages and
different types of writing systems.

2 Second we would need some serious corpora (not made up examples) of non-linguistic systems

In his third post on the paper Rahul Siddharthan writes thus.

Given the data for other languages in the Rao et al. paper, I would estimate P(D|HL) to be close to 1. That is, if the Indus script is a language, I would think it very likely that conditional entropies would closely resemble the data that Rao et al

I think that interpretation is right. But that significantly diminishes the 'interestingness' of Rao et al.'s paper.

I agree. Remember, though, that this is Rao et al's first word on the subject, not the last. Rao et al's paper is more of a "proof of concept" paper - i.e., they're trying to show that using this technique can produce interesting results. The real analysis - using the significantly larger data sets and corpora that Sproat mentions - on which they say they're working is what's going to produce the really interesting results. I think it's a positive sign that it's gotten people like Mark Kenoyer interested, and I'm quite looking forward to seeing how this pans out.