More comment from Steve Farmer on a further story (posted on the Ma'at History board by Rich.)

*****************


General permission is given to repost the following on other Lists.

Benjamin Fleming writes (see full posts at the end with links)

> Here we go again. Just in case you did not get enough the last time
> around, here is a report about (presumably) *another* study from Rao
> and associated to be (or already, I can't tell) published in the
> Proceedings of the National Academy of Sciences this week.

The Science Daily article tells us, gullibly:

> The new study looks for mathematical patterns in the sequence of
> symbols. Calculations show that the order of symbols is meaningful;
> taking one symbol from a sequence found on an artifact and changing
> its position produces a new sequence that has a much lower
> probability of belonging to the hypothetical language.

Indeed, here we go again. And it is no longer even faintly
intellectually interesting. Doesn't anyone ever pass these papers on
to a computational linguist for review?

Well, only to repeat things we've said hundreds of times now, long
before these articles appeared: The fact that there is "order" or
"meaning" in Indus symbols has been known since the 1920s, many
decades before there were computers. You can show that in a dozen
different ways: looking at positional frequencies, conditional
entropy, Markov processes (apparently what they used here, according
to the news story), Zipfian distributions, a zillion other ways --
including just looking at a random bunch of symbol chains with the
naked eye.

That's how difficult it is to show there is order in the symbols.

In the mid to late 1960s what was known since the 20s using the naked
eye was "discovered" again using the "magic of computers" by the
Finns and Soviets. A whole string of other writers have used
computers many times since to prove the same obvious point.

The joke is that there is "order" and "meaning" of some sort in
virtually all symbol strings. Finding that has absolutely nothing to
do with whether the symbols are linguistic or not.

Richard Sproat is in fact giving the keynote address at a big computational
linguistics conference in Singapore three days from now in which this
new claim will surely become a laughingstock. (Neither Rao nor anyone else
in his group is a computational linguist.) Here is a link to that
conference:

[conferences.inf.ed.ac.uk]

Here is an abstract of Richard's kickoff speech, which will in fact
deal in part with this issue:

[conferences.inf.ed.ac.uk]

As we wrote in 2004 in "Collapse of the Indus-Script Thesis," long
before Rao et al. showed up:

"Statistical regularities in sign positions show
up in nearly all symbol systems, not just those that encode speech."

We then demonstrated that statistically using medieval heraldic signs
and gave detailed critiques of the long string of writers (none of
whom Rao et al. ever cites, interestingly enough) who have made the
same claims for some fifty years or more. You'll find these
discussions at length in our paper:

[www.safarmer.com]

I discussed this issue again in Kyoto at the end of May 2009, again
showing statistical evidence, in front of a large forum of Indus
researchers. Rao et al. didn't have a single supporter, and a number
of researchers there told me that they had opposed the publication in
Science of their earlier paper.

Sproat and I also discussed this issue in a ltechnical paper
published in 2006 in a volume from the CSLI Studies in
Computational Linguistics, at Stanford University, writing
in part:

[www.safarmer.com] (page 10):

> Non-linguistic sign systems often display levels of formal
> structure no less extreme than those seen in linguistic systems:
> witness the complex syntactic structures in mathematical
> expressions, or the recurrent sign groups that regularly show up in
> non-linguistic sign systems in the ancient Near East (Farmer et
> al., 2004 [where we *show* and *discuss* visual examples]).

In this case, we were in fact criticizing computational work done in
the the 1960s (!) by Koskenniemi (one of the Finns) that is
*far* more sophisticated than anything Rao et al. are doing
nearly 50 years later. The fact that the Finns and Soviets were
taking the same approach (which was soon shown to be a
dead end) nearly a half century ago has somehow slipped
everyone's attention. There was no indication in the earlier
Rao study that they are even aware of this work.

When Michael Witzel, Richard Sproat, and I publish our Kyoto paper,
later this year (we're all busy now) we'll be sure to discuss this
again in detail -- and all the other crazy stuff published to
attempt to uphold the old "script" thesis over the past five
years. Here's our abstract to our paper:

[www.safarmer.com]

"The collapse of the Indus-script thesis, five years later:
Massive non-literate urban civilizations of ancient Eurasia"

In prelude:

You can use the same methods to show there is order in Mesopotamian
kudurru symbols, clusters of freeway signs, Vinca symbols, Iroquois
wampum, Naxi mnemonic symbols, Aztec or Mixtec "picture writing,"
series of horoscopal signs, alchemical symbols, medieval heraldic
signs, boy scout or military medals, etc. etc. There is no end to the
list.

Back in 2007 one of Rao's collaborators (Mayank Vahia, to whom I'm
copying this paper, and who I believe belongs to our List) sent me a
similar paper and I gently informed him that this had nothing to do
with language and that the Soviets and Finns had done the same thing
in the 60s. That didn't stop them from pursuing the issue, however,
since they were getting funding (in Chennai) from a new institute
founded right after we published our paper whose main purpose
is to oppose our views (well funded too, from interesting sources).

Mayank, are you willing to come on the List -- you can bring Rao with
you if you want -- to discuss what significance you think are found in these
findings?

What is clear is that this has nothing to do with "science" but with
a propaganda war, which is an issue we also explicitly discussed in
Kyoto. That doesn't change the data, and this article of Rao's will
be trashed as thoroughly as his article in Science was: not everyone
reading those magazines is ignorant of computational linguistics.

Well, we'll have to waste a lot more time dealing with this new
nonsense, I guess. In the long run, this probably works to our
benefit, since this paper won't fly any further than the last Rao
effort. I'm sure it will be discussed in Singapore, where in his
Keynote Address, as noted above, Richard will be discussing these
issues in front of people who actually specialize in the field. (I'm
copying this message to Richard also: maybe he can find a few minutes
to say something about Rao's newest.)

Big question: will the reporters check this out any more thoroughly
this time? I talked to one reporter this morning who is writing a
story....

Cheers,
Steve

**********

Hermione
Director/Moderator - The Hall of Ma'at

Rules and Guidelines

hallofmaatforum@proton.me



Edited 1 time(s). Last edit at 03/24/2022 09:21AM by Hermione.