Indus Script - Decipherability Question

Decipherability Question

In a 2004 article, Farmer, Sproat, and Witzel presented a number of arguments in support of their thesis that the Indus script is nonlinguistic, principal among them being the extreme brevity of the inscriptions, the existence of too many rare signs increasing over the 700-year period of the Mature Harappan civilization, and the lack of random-looking sign repetition typical for representations of actual spoken language (whether syllabic-based or letter-based), as seen, for example, in Egyptian cartouches.

Asko Parpola, reviewing the Farmer, Sproat, and Witzel thesis in 2005, states that their arguments "can be easily controverted". He cites the presence of a large number of rare signs in Chinese, and emphasizes that there is "little reason for sign repetition in short seal texts written in an early logo-syllabic script". Revisiting the question in a 2007 lecture, Parpola takes on each of the 10 main arguments of Farmer et al., presenting counterarguments for each. He states that "even short noun phrases and incomplete sentences qualify as full writing if the script uses the rebus principle to phonetize some of its signs".

A computational study conducted by a joint Indo-US team led by Rajesh P N Rao of the University of Washington, consisting of Iravatham Mahadevan and others from the Tata Institute of Fundamental Research and the Institute of Mathematical Sciences, was published in April 2009 in Science. They conclude that "given the prior evidence for syntactic structure in the Indus script, (their) results increase the probability that the script represents language". Farmer, Sproat, and Witzel have disputed this finding, pointing out that Rao et al. did not actually compare the Indus signs with "real-world non-linguistic systems" but rather with "two wholly artificial systems invented by the authors". In response, Rao et al. point out that the two artificial systems "simply represent controls, necessary in any scientific investigation, to delineate the limits of what is possible." They state that real-world non-linguistic systems were indeed included in their analysis ("DNA and protein sequences, FORTRAN computer code"). Farmer et al. have also compared a non-linguistic system (medieval heraldic signs) with natural languages using Rao et al.'s method and conclude that the method cannot distinguish linguistic systems from non-linguistic ones. Rao et al. have clarified that their method is inductive, not deductive as presumed by Farmer et al., and their result, together with other known attributes of the script, increases the evidence that the script is linguistic, though it does not prove it. In a follow-up study published in IEEE Computer, Rao et al. present data which strengthen their original conditional entropy result, which involved analysis of pairs of symbols. They show that the Indus script is similar to linguistic systems in terms of block entropies, involving sequences up to 6 symbols in length.

A discussion of the linguistic versus nonlinguistic question by Sproat, Rao, and others was published in the journal Computational Linguistics in December 2010.