Does the Indus script encode language, or a non-linguistic sign system?

If the Indus signs are writing, the Harappan civilization was literate and every proposed decipherment (Dravidian, Indo-Aryan, Munda or other) is a legitimate project; if not, it becomes the main case of a large urban society administered without writing, which changes comparative accounts of states and bureaucracy. The answer also governs how seals, tags and pot inscriptions are read as evidence for trade, taxation and identity across a region of perhaps a million square kilometres.

Inscriptions average about five signs and the longest have only a few dozen, so statistical signatures that distinguish languages from other systems are weak at this length. Farmer, Sproat and Witzel (2004) argued the brevity and sign-repetition patterns are incompatible with writing; Rao et al. (2009) answered that conditional entropy places the corpus within the range of linguistic scripts; Sproat (2010, 2014) showed that several non-linguistic systems also fall in that range. The corpora used by different teams differ in sign lists (roughly 400 to 700 signs) and inclusion rules, and the most complete database (ICIT) requires permission to access.

A pre-registered battery of measures, fixed before the Indus data are scored, classifies a large set of known writing systems and non-linguistic systems (heraldry, mason's marks, deity symbols, administrative codes) correctly when all are trimmed to the Indus length distribution, and then assigns the Indus corpus to one class with stated error rates that are robust to choice of sign list.

• Assemble an open, versioned Indus corpus reconciling the Mahadevan, CISI and ICIT sign lists, and build comparison corpora of known scripts and non-linguistic systems sampled to match Indus text lengths, object types and sign inventories, extending Sproat's 2014 dataset.
• Fix the classifier and thresholds in a public pre-registration, then report results across alternative sign segmentations.
• Progress depends on negotiated open access to ICIT and to the CISI photographic archive, and on recording unpublished excavated material from Pakistan and India.

Drafted by Claude Opus 5.5, 2026-10-08

1. 1977
   Iravatham Mahadevan, The Indus Script: Texts, Concordance and Tables, Memoirs of the Archaeological Survey of India 77. New Delhi: Archaeological Survey of India.Computer-generated concordance of the texts then known, with sign frequencies, positions and site distributions; still the most widely used reference sign list.
2. 1987
   Jagat Pati Joshi and Asko Parpola (eds.), Corpus of Indus Seals and Inscriptions, vol. 1: Collections in India, Annales Academiae Scientiarum Fennicae / Memoirs of the Archaeological Survey of India 86. Helsinki: Suomalainen Tiedeakatemia.First volume of the photographic corpus (CISI); later volumes cover Pakistan (1991) and new material and collections elsewhere (2010 onward).
3. 1994
   Asko Parpola, Deciphering the Indus Script, Cambridge: Cambridge University Press.Synthesis arguing that the script is logo-syllabic writing of a Dravidian language, with rebus-based readings of selected signs.
4. 2004
   Steve Farmer, Richard Sproat and Michael Witzel, The Collapse of the Indus-Script Thesis: The Myth of a Literate Harappan Civilization, Electronic Journal of Vedic Studies 11 (2): 19–57.Argued from text brevity, sign repetition and the rarity of singleton signs that the Indus signs are non-linguistic symbols.
5. 2009
   Rajesh P. N. Rao, Nisha Yadav, Mayank N. Vahia, Hrishikesh Joglekar, R. Adhikari and Iravatham Mahadevan, Entropic Evidence for Linguistic Structure in the Indus Script, Science 324 (5931): 1165.Showed that the conditional entropy of Indus sign sequences falls between rigid and random systems, in the range of known languages.
6. 2010
   Rob Lee, Philip Jonathan and Pauline Ziman, Pictish Symbols Revealed as a Written Language through Application of Shannon Entropy, Proceedings of the Royal Society A 466 (2121): 2545–2560.Applied a related entropy method to Pictish stones; frequently cited in the Indus debate as an example of how such measures can classify non-linguistic systems as writing.
7. 2010
   Nisha Yadav, Hrishikesh Joglekar, Rajesh P. N. Rao, Mayank N. Vahia, Ronojoy Adhikari and Iravatham Mahadevan, Statistical Analysis of the Indus Script Using n-Grams, PLoS ONE 5 (3): e9506.n-gram models found strong directional and positional structure consistent with a formal language, while stating that this does not by itself prove the encoding of speech.
8. 2010
   Richard Sproat, Ancient Symbols, Computational Linguistics, and the Reviewing Practices of the General Science Journals, Computational Linguistics 36 (3): 585–594.Criticized the entropy argument and showed that non-linguistic systems can produce similar values; Rao et al. replied in the following issue.
9. 2014
   Richard Sproat, A Statistical Comparison of Written Language and Nonlinguistic Symbol Systems, Language 90 (2): 457–481.Compiled corpora of non-linguistic systems and found that no single published measure cleanly separates them from writing; the closest existing template for a controlled test.
10. 2015
    Bryan K. Wells and Andreas Fuls, The Archaeology and Epigraphy of Indus Writing, Oxford: Archaeopress.Presents the ICIT database with Fuls's appendices on automated segmentation, positional analysis and classification of undeciphered writing systems.
11. 2023
    Andreas Fuls, A Catalog of Indus Signs, Mathematica Epigraphica 4. Berlin: self-published.Sign-by-sign frequencies and positional behaviour keyed to his Corpus of Indus Inscriptions (2022, over 5,500 texts), concluding the script is logo-syllabic with about 700 signs.
12. 2023
    Bahata Ansumali Mukhopadhyay, Semantic scope of Indus inscriptions comprising taxation, trade and craft licensing, commodity control and access control: archaeological and script-internal evidence, Humanities and Social Sciences Communications 10.Argues from object context and sign structure that inscriptions served administrative functions, a reading compatible with logographic writing.
13. 2026
    Ashish Nair, How Non-Linguistic Is the Indus Sign System? A Synthetic-Baseline Scorecard, arXiv preprint 2604.17828.Scores 1,916 deduplicated inscriptions against simulated heraldic and administrative baselines and seven real non-linguistic corpora, finding that none reproduces the Indus profile; not yet peer reviewed.

• Finnish Academy of Science and Letters, Helsinki, with the Archaeological Survey of India and Department of Archaeology and Museums, Pakistanpartly digitized Corpus of Indus Seals and Inscriptions (CISI), vols. 1 (1987), 2 (1991), 3.1 (2010) and later partsPhotographic corpus of seals and inscribed objects arranged by collection; published in print, with the underlying photographs not openly available online.
• Interactive Corpus of Indus Texts (ICIT), Technische Universität Berlindigitized Relational database of inscribed objects, texts and sign list compiled by Bryan K. Wells and Andreas FulsOver 4,500 inscribed objects and 5,500 texts with analysis tools; access is by request to the administrator.
• Harappa.com (Jonathan Mark Kenoyer, Richard H. Meadow and colleagues)partly digitized Images and reports from the Harappa Archaeological Research Project, including excavated seals, tablets and inscribed potteryOpen image collection with stratigraphic context for many Harappa inscriptions, useful for dating and object-type controls.
• National Museum, New Delhipartly digitized Harappan Gallery: seals and sealings from Mohenjo-daro, Harappa, Lothal and other sites allocated to IndiaMajor holding of Indus seals; images available through CISI vol. 1 and partly online, with study access by application.
• Archaeological Survey of India / Department of Archaeology and Museums, Pakistannot digitized Excavated material from Mohenjo-daro, Harappa, Dholavira, Kalibangan, Rakhigarhi and other sites, including unpublished inscribed objectsRecent excavations have added inscriptions not yet in the corpora; recording them requires collaboration with the excavating authorities.

Compiled by Claude Opus 5.5, 2026-10-08