The Indus "script"?

[Steve Farmer]

Simon Brodbeck writes, on the "Indus script" question:

> I hope it will not be out of place here to say how little I am  
> looking forward to the apparently forthcoming treatment of this  
> issue in this forum.
>
> I have no doubt that if any present parties have anything  
> substantial to add to the debate, we shall see it in print before  
> long. In the meantime, I wonder how much is to be gained by  
> repeating, in more or less rhetorical variants, contributions that  
> have already been made in print.
>
Sorry for the delayed response, and just hit "delete", Simon. :^)  
I'll try to make this my only post, unless someone responds to the  
specific evidence I discuss below.

I agree that this isn't the ideal forum for this discussion. That  
would require the contributions of archaeologists, comparative  
historians, and computational linguists, few of whom are on the List.  
There is nothing rhetorical in what I say below, nor am I simply  
repeating points already made in print. Two major points:

1. My last post on this List was actually nine years ago, and the  
only reason I posted last week was to respond to a pop news story  
sent to the List by an Indologist on the paper by Rao et al. --  
actually the worst of the stories published so far, entitled  
"Artificial Intelligence Cracks 4000-Year-Old Mystery." When  
something like that is posted without comment on a research List,  
supposedly overturning the work of you and your colleagues, most  
people would feel constrained to respond. :^)

Now that the statistical flaws in Rao et al. have been thoroughly  
discussed by well-known computational linguists including Mark  
Liberman and Fernando Pereira (neither of whom I know), there isn't  
much to add beyond what they say here:

http://languagelog.ldc.upenn.edu/nll/?p=1374 (updated since it first  
appeared)
http://earningmyturns.blogspot.com/2009/04/falling-for-magic- 
formula.html

The bottom line is a critical point that is rarely mentioned  
anywhere, let alone on the Indology List: no single statistical  
measure can distinguish speech-encoding systems (writing) from symbol  
systems that encode semantic data of other types. (See here  
especially Pereira's comments in the second link above.) The only  
reason that the data of Rao et al. look superficially convincing to  
non-specialists is because of the huge gap they show in their data  
(see their chart in the first link above) between clustered groups of  
Indus signs and a carefully selected group of natural languages and  
two wholly *invented* data sets of supposedly "representative  
examples" of nonlinguistic symbols -- neither of which are anything  
like anything found in the ancient world. Rao et al. calculated *no*  
values at all for ancient nonlinguistic symbol systems beyond those  
from the Indus -- despite what they suggest in the main body of their  
paper. In other words, they slipped one by the Science reviewers.

If you don't think this paper will affect future work in Indus  
studies by people, including researchers, many of whom get their  
views of research outside their specialized fields from outside  
reports and newspapers, think again.

The question of whether Indus symbols did or not encode writing may  
seem a trivial Glasperlenspiel to those whose work doesn't involve  
studies of the impact of literacy on premodern civilizations. It is  
anything but an empty scholastic game: writing is an enabling  
technology that when it existed fundamentally transformed ancient  
civilizations. And surrounding the Indus -- among the BMAC, in the  
cities of the Gulf, and in the urban centers of the SE Iranian  
plateau (e.g., at Jiroft, whose supposed "writing" has been shown in  
the past year to consist of modern forgeries)  -- we find no writing  
either.

The significance of this for Indus studies and ancient studies in  
general will be discussed at length in Kyoto at the end of this  
month, by me and Michael Witzel, by perhaps the leading specialist on  
Iranian and Gulf archaeology, Dan Potts, and others. Nothing old or  
rhetorical about that discussion either.

2. I also feel constrained to point out that the summaries Parpola  
gave on this List in the wake of the discussion of Rao et al. --  
which didn't directly pertain to our views of Parpola's work -- have  
little to do with our real views. E.g., he summarizes and disposes of  
what he claims as one of the arguments of Farmer, Sproat, and Witzel  
this way:

> All literary civilizations produced longer texts but there are none  
> from the Indus Valley — hence the Indus script is no writing  
> system: Farmer and his colleagues reject the much repeated early  
> assumption that longer texts may have been written on birch bark,  
> palm leaves, parchment, wood, or cotton cloth, any of which would  
> have perished in the course of ages as suggested by Sir John  
> Marshall in 1931 (I, 39). Farmer and his colleagues are ready to  
> believe the Indus script thesis only if an Indus text at least 50  
> signs long is found.
>
> *But* even though Farmer and his colleagues speak as if our present  
> corpus of texts was everything there ever existed, this is not the  
> case. More than 2100 Indus texts come from Mohenjo-daro alone, and  
> yet less than one tenth of that single city has been excavated.  
> Farmer and his colleagues do not know what has existed and what may  
> be found in the remaining parts of the city, even if it is likely  
> that only imperishable material of the kinds already available  
> continue to be found. The Rongo-Rongo tablets of Easter Island are  
> much longer than 50 signs. But does this make it certain that they  
> represent writing in the strict sense?
>

Nothing here represents any of our nuanced views. We certainly never  
speak "as if our present corpus of texts was everything there ever  
existed." Nor would we claim that every symbol string in the world  
over 50 signs long is "writing." Richard Sproat is in fact an expert  
on Rongo-Rongo, and we could point of course to manuscript length  
mnemonic or "prompt" texts among the Mixtecs, Aztecs, or (in Asia)  
the Naxi, of which I have indeed talked about at length in lectures  
attended by Parpola in 2005 (Kyoto) and 2007 (at Stanford). But Indus  
symbols don't belong to this subtype of nonlinguistic symbols, as we  
argue -- nonlinguistic like linguistic signs come in a surprisingly  
wide variety of 'flavors'. We in fact say something like that on the  
first page of our 2004 paper.

In any event, our arguments about "text" length are considerably more  
sophisticated than the way they are summarized above.

The "lost manuscript" thesis, which was accepted with little  
discussion from Hunter 1929 to Marshall 1931 to Parpola 1994, was  
introduced to explain away the obvious lack of anything remotely  
looking like a text in the Indus Valley. The reason the argument  
fails was something we first noted: no premodern civilization is  
known that wrote long texts on perishable materials but failed to  
leave *obvious* and *abundant* examples of such texts behind on  
durable materials as well. We're not talking here about obscurely  
buried archives.

If in a Gedankenexperiment you took away all perishable materials of  
any type from China, India, Central Asia, Iran, Anatolia, Greece,  
Italy, Mesopotamia, Mesoamerica, etc., it wouldn't change one bit our  
understanding of which civilizations in those regions were literate.  
This would be due to the massive numbers of long texts, often  
numbering in the hundreds of thousands, that the literate  
civilizations of these regions left behind in obvious places.

In 2009, five years after our paper was published, we no longer find  
speculation about the literary "activities of the Harappan scribes  
and scholars", which Parpola 1994 compared "on the analogy of  
'empires' comparable to the Indus Civilization" to Aztec  
"writing" (Parpola 1994: 54). (Actually, Aztec texts aren't really  
"writing" at all in the technical sense, as noted above.)  In his  
recent post, Asko tells us instead that the "type of texts I expect  
to be lost is exemplified by the Mycenaean Linear B tablets, i.e.  
economic accounts, not literature, which was probably handed down  
orally...."

This doesn't make speculation about lost Indus archives more  
credible, due to the massive scale of the excavations conducted since  
the 1920s. Compare here -- we haven't written about this anywhere --  
with finds of Linear B, of which we have many thousands of long  
texts. The first Minoan site ever excavated, at Knossos, quickly  
turned up no less than 4300 Linear B texts (plus of course seals,  
etc.). Nearly *all* of these texts are far longer than what is by far  
the longest "Indus text" on a single surface, consisting of 17 high- 
frequency but non-repeating signs on a square about 1 inch square.

After the find of 4300 long Linear B texts at Knossos, another 1000  
or so showed up at Pylos. Then 300 or so from Thebes. And now we have  
perhaps another 300 so far from other sites. Others turn up every  
year. These are not obscure finds.

Archaeological science rarely proves anything outright, but  
eventually hypotheses based on speculation concerning *possible*  
finds introduced to "save" a thesis, but that never materialize are  
eventually quietly abandoned -- especially when that speculation  
conflicts with what is commonly known from cross-cultural  
archaeological studies from many other parts of the world. (India may  
be "different", as Indologists often say, but not *that* different.)

The Indus left thousands of short symbol strings behind on many types  
of materials, not just "seals" -- pots, potsherds, metal plates,  
weapons, molded terracotta tablets, incised shells, cones and rods,  
etc. -- the same kinds of materials on which other civilizations  
(including those that routinely wrote on perishable materials) left  
thousands of *long* texts behind. This is especially true of  
potsherds, which were among the most prevalent writing materials in  
the ancient world, since they were far cheaper than rather expensive  
perishable materials (including cloth). You have to have an  
explanation for that, and of course for all the missing texts, and  
without one the traditional "Indus script" thesis quite frankly isn't  
credible.

That's it for me, as noted above, unless someone wants to discuss the  
evidence discussed here. All this will be picked up in Kyoto later  
this month, so I'm fine in leaving it here. The big issue to the  
minds of me and my colleagues -- Michael Witzel, Richard Sproat, the  
archaeologist Dorian Fuller, others -- isn't the now standard  
question that we introduced five years ago ("Is it a writing system  
or not?"), but about all the exciting new research avenues that are  
opened up as soon as you recognize the abundant evidence that says it  
wasn't.

Best wishes,
S. Farmer