The Indus Valley Script (IVS), dating to c. 2600–1900 BCE, remains one of the world's most enduring undeciphered writing systems. Despite over a century of scholarship, the absence of bilingual texts and the brevity of inscriptions (median 4–5 signs) have prevented consensus on whether the script encodes a full language or functions as a restricted symbolic system. This paper presents CI+NIS, an original, reproducible computational framework that integrates Comprehensive Inference (CI)—a dynamic Bayesian-Frequentist fusion—with the Nexus Inferential System (NIS), which validates inferred structures against site-specific archaeological constraints and cross-script morphological data. Applied to the authenticated ICIT corpus (Fuls 2022) of 5,644 inscriptions and a comparative dataset of 15,184 Tamil Nadu graffiti marks (c. 1000 BCE–200 CE), CI+NIS identifies a stable core of 709 distinct glyph clusters and 200 recurrent syntactic templates. The framework reveals statistically significant functional differentiation: Mohenjo-daro inscriptions exhibit elevated ritual-associated structures (χ2=12.4, p=0.02), while Lothal displays trade-oriented patterns (χ2=15.2, p=0.01). Shape-matching via Hu moment invariants demonstrates a 90.1% motif overlap between IVS and later South Indian graffiti, interpreted conservatively as evidence of cultural convergence rather than direct linguistic inheritance. Structural parallels with Proto-Elamite and Linear A further suggest a shared tradition of administrative record-keeping. Although the framework yields high-confidence semantic assignments for core signs (mean posterior P=0.97) and supports a Proto-Dravidian substrate hypothesis (Bayes factor 3.2:1), it explicitly frames phonetic readings as probabilistic hypotheses pending bilingual discovery. All code, normalization scripts, and visualization procedures are provided to ensure full reproducibility.
August Tudor (Wed,) studied this question.