Editing Info:1d5d8211-a482-402f-8cd2-b99a6cb3d1a5 (section)

=== What we know ===

The Indus script comprises approximately 400–600 distinct signs, depending on analytical decisions about variants and allographs. This range places it between a purely logographic system (which would require thousands of signs) and an alphabetic one (which would need only 20–40). Most scholars interpret the sign count as consistent with a logo-syllabic system — one combining logograms (signs representing whole words) with syllabic signs representing phonetic values — though this interpretation is contested.

The inscriptions are predominantly short. The average inscription length is approximately 4.6 signs, and the longest known text contains only 17 signs. This extreme brevity constrains all analytical approaches, as statistical methods designed for longer texts lose reliability with such small samples. The shortness of the texts has also been used by some scholars (notably Farmer, Sproat, and Witzel 2004) to argue that the system is not linguistic at all but rather a collection of non-linguistic symbols analogous to heraldic devices or potter's marks.

The writing direction has been established as predominantly right-to-left, based on evidence of sign compression on the left side of seals (where writers running out of space compress the final signs) and from seal impressions that reverse the orientation. Some texts appear boustrophedon (alternating direction).

Asko Parpola's Dravidian hypothesis, developed over more than four decades and synthesised in his 1994 monograph ''Deciphering the Indus Script'', proposes that the underlying language belongs to the Dravidian family, based on the geographic distribution of Dravidian languages in South Asia and on proposed rebus readings of individual signs. For example, Parpola interprets a frequently occurring "fish" sign as representing the Dravidian word ''mīn'' ("fish"), which is homophonous with ''mīn'' ("star") in several Dravidian languages, thus encoding an astronomical reference. While influential, this hypothesis remains unverified because no proposed readings have been confirmed through independent means.

Rajesh Rao and colleagues (2009) published a landmark paper in ''Science'' demonstrating that the conditional entropy of the Indus sign sequences falls within the range observed for natural languages, and outside the range for non-linguistic symbol systems (such as DNA sequences or Fortran code). This entropic evidence supports the linguistic hypothesis but has been challenged by Farmer and Sproat, who argue that the method cannot reliably distinguish language from structured non-linguistic systems of comparable complexity.

More recently, Daggumati and Revesz (2021) published a method for identifying allographs (variant forms of the same sign) in the Indus script by data-mining positional distributions within inscriptions. Their analysis reduced the effective sign list, potentially making decipherment more tractable. They also applied convolutional neural networks to compare Indus signs with those of other Bronze Age scripts (Linear A, Cretan Hieroglyphic), identifying structural similarities that may indicate shared typological features.