SESHAT — Quantum-Assisted Decipherment of Undeciphered Scripts

Seshat — Egyptian goddess of writing, knowledge, and measurement. Counterpart of Thoth. Keeper of the House of Books.

---

Research Question

Can quantum-inspired combinatorial search, constrained by known cross-script correspondences, narrow the phoneme-assignment space of Linear A sufficiently to identify its linguistic affiliation?

Methodology

Linear A corpus (Younger database)
    +
Linear B corpus (DĀMOS, Oslo) + known A↔B sign correspondences
    ↓
seshat-core [Rust]
Statistical analysis: sign frequency, bigrams, positional patterns
    ↓
seshat-anneal [C++/CUDA]
QUBO formulation with Linear B anchor constraints
Simulated Annealing baseline vs Quantum-Inspired Annealing (TFI model)
    ↓
seshat-analysis [Python/JAX]
Cross-linguistic LM trained on Linear B, Luwian, Hurrian
Scoring and ranking of candidate phoneme assignments
    ↓
seshat-viz [Rust/egui]
Interactive sign table, probability heatmaps, annealing convergence
    ↓
paper/ [LaTeX]

Repository Structure

seshat-core/        Rust — corpus parser, sign inventory, bigram matrices
seshat-anneal/      C++/CUDA — SA and QA-sim annealing engines
seshat-analysis/    Python/JAX — LM validation, statistical analysis
seshat-viz/         Rust/egui — interactive visualization
data/
  linear_a/         Linear A tablets (Younger database)
  linear_b/         Linear B corpus (DĀMOS)
  correspondences/  Known A↔B sign mappings
  comparative/      Reference corpora: Luwian, Hurrian, Mycenaean Greek
paper/              LaTeX manuscript
scripts/            Data acquisition utilities

Data Sources

• Linear A: John Younger's Linear A Database (University of Kansas) https://people.ku.edu/~jyounger/LinearA/
• Linear B: DĀMOS Database of Mycenaean at Oslo https://damos.hf.uio.no/
• Luwian/Hurrian: CLARIN-D / CMAWL corpus

Status

• Phase 1: Corpus ingestion & statistical baseline (seshat-core)
• Phase 2: QUBO formulation & annealing engine (seshat-anneal) - SA: 100% anchor recovery (51/51), perplexity 80.2 on GORILA corpus - QA-sim: matches SA quality, 8.76× wall-clock overhead vs SA
• Phase 3: Cross-linguistic LM validation (seshat-analysis) - 3-gram Kneser-Ney trained on Linear B, Luwian, Hurrian — perplexity 8.32 - AB8 (a) identified with probability 0.79; ambiguous signs narrowed to 2–5 candidates
• Phase 4: Interactive visualization (seshat-viz) — in progress
• Phase 5: Paper — complete, pending arXiv endorsement

Authors

Antonio Zambudio Rodriguez