agnel.io

in build historical linguistics

Reading the Indus script with computers

Statistical and AI tests of short Indus inscriptions against candidate languages, with falsifiable results.

inscriptions measured
~2,500
candidate languages tested
4
claims of decipherment
0
Stack
  • Python
  • Statistics
  • LLMs

Nobody can read the Indus script. The inscriptions are short, most about five signs long, there is no bilingual text to work from, and the language behind them is unknown.

This project does not claim to crack it. It measures the statistical patterns in about 2,500 inscriptions and tests how well they fit four candidates: Old Tamil, Burushaski, Elamite and an Elamo-Dravidian model, with Vedic Sanskrit as a control.

The output is a reproducible pipeline and a ranked set of hypotheses that new evidence could prove wrong, with a clear line where measurement ends and guessing begins.

vtx · type help, Tab to complete, Esc to close