Linear A Computational Analysis

Computational structural analysis of the Minoan Linear A script (~1850-1450 BCE). This project applies information-theoretic, morphological, and cross-domain methods to extract linguistic structure from the undeciphered corpus without requiring a bilingual text.

What This Is

Linear A is one of the last major undeciphered writing systems. These scripts analyze ~1,720 known inscriptions using computational methods to:

• Test proposed grammatical rules against the libation formula corpus (41 variants)
• Measure information-theoretic properties (Shannon entropy, bigram predictability)
• Compare the Minoan sound system against candidate language families
• Apply statistical controls (permutation testing, ablation, perturbation analysis)
• Evaluate cross-domain convergence across linguistic, archaeological, and genetic evidence
• Validate methodology using Linear B (Mycenaean Greek) as a negative control

Key Findings

• 6/6 morphological agreement rules confirmed with zero exceptions across the libation formula corpus
• Shannon entropy of 4.70 bits/sign — squarely in the natural language range
• 9/9 administrative structural features match other Bronze Age accounting systems
• Three-way vowel convergence: Linear A corpus (a=43%, i=21%, u=18%), Beekes' pre-Greek substrate reconstruction (/a/, /i/, /u/), and Hurrian (3 vowels: a, i, u) all independently point to the same phonological system
• Hurro-Urartian shows strongest overall fit at 77.5% across 8 dimensions, with a 37-point gap over the next candidate (Semitic at 40.3%). This is a composite score — see Calibration & Limitations below for what it does and doesn't mean.
• Linear B negative control validates methodology: Mycenaean Greek scores 30.8% — a known IE language is correctly identified as dissimilar to Linear A
• Bootstrap confidence: P(Hurro-Urartian = #1) = 100.0% across 10,000 resampled iterations. Note: with 7 candidates and 8 correlated dimensions, this is expected for any clear leader — it confirms consistency, not certainty.
• Corpus-wide analysis: 1,720 inscriptions processed — 31.8% of all words identified with ~25 vocabulary items, 5 administrative document types classified, arithmetic totals verified on multiple tablets
• -RO nominalizer rule confirmed: KU-RO (total), KI-RO (deficit), SA-RO (liquid-allocation) — productive suffix turning roots into abstract nouns
• Balance-sheet accounting identified: SA-RA₂ (allocation) sections followed by KI-RO (deficit) sections on same tablets, with deficit amounts consistently less than allocations — tracking commitments vs. deliveries

Visualizations

The control validation script generates 5 publication-quality figures:

Scoring Heatmap

Multi-Dimensional Radar Chart

Bootstrap Score Distributions

Language Family Similarity Space (MDS)

Bronze Age Mediterranean Geographic Map

Scripts

Script	Description
LINEAR_A_CROSS_CULTURAL_STRUCTURE.py	NEW — Cross-cultural structural analysis using 3,000+ ancient texts across 6 civilizations (Sumerian, Egyptian, Hittite, Hebrew, Greek, Vedic). Universal libation formula invariants, evolution timeline 3000-500 BCE, structural parse of the libation formula, offering-term frequency validation
LINEAR_A_CORPUS_ANALYSIS.py	Full corpus statistical analysis: 1,720 inscriptions, word frequency tables, commodity co-occurrence, 5 document types identified, -RO nominalizer grammar rule, 31.8% corpus readability, arithmetic-verified tablet readings
LINEAR_A_SUBSTRATE_PREDICTIONS.py	Beekes pre-Greek substrate test: 10 prediction entries, three-way phonological convergence (substrate × Linear A × Hurrian), semantic domain overlap analysis, triangulation methodology
LINEAR_A_CONTROL_VALIDATION.py	Linear B negative control, expanded 38-item vocabulary, 5 visualizations, source documentation, honest assessment
LINEAR_A_STRUCTURAL_ANALYSIS.py	Core analysis: libation formula corpus, morphological rule testing, information theory, sign frequency, co-occurrence networks
LINEAR_A_PHONOLOGICAL_ANALYSIS.py	Sound system analysis: phonotactics, vowel harmony, consonant clusters, syllable structure, rhythm patterns
LINEAR_A_HURRIAN_COMPARISON.py	Systematic Hurrian grammar comparison: case system mapping, vocabulary, verbal morphology
LINEAR_A_ENHANCED_COMPARISON.py	Extended comparison with pre-Greek substrate, Urartian three-way analysis, Van Soesbergen readings
LINEAR_A_STATISTICAL_CONTROLS.py	Reviewer-proof controls: baseline distributions across 6 language families, ablation, permutation testing, sign-reading perturbation
LINEAR_A_CROSS_DOMAIN_CONVERGENCE.py	Bayesian convergence across 8 independent evidence domains (linguistic, archaeogenetic, maritime trade, material culture, religious iconography, etc.)
LINEAR_A_TRANSLATION_ATTEMPT.md	Full writeup: established facts, analysis results, proposed translations, confidence levels, honest assessment of unknowns
LINEAR_A_SYNTHESIS_REPORT.md	Comprehensive synthesis: 30+ vocabulary items, 5 document types, 8 grammar rules, religious formula structure, corpus statistics, honest limitations
PROCESS_DOCUMENTATION.md	How this analysis was produced: 5-stage pipeline, knowledge base construction, cross-cultural methodology, reproducibility notes

Running

The core analysis scripts use only the Python 3 standard library. The control validation script requires matplotlib and numpy for visualizations.

# Core analysis (no dependencies)
python3 LINEAR_A_STRUCTURAL_ANALYSIS.py
python3 LINEAR_A_STATISTICAL_CONTROLS.py
python3 LINEAR_A_CROSS_DOMAIN_CONVERGENCE.py

# Control validation with visualizations (requires matplotlib, numpy)
pip install matplotlib numpy
python3 LINEAR_A_CONTROL_VALIDATION.py

Methodology

Confidence protocol: Every claim is tagged [HIGH], [MEDIUM], [LOW], or [SPECULATIVE].

The approach treats Linear A as a systems analysis problem rather than a purely linguistic one:

1. Structural analysis — Variant comparison across 41 libation formula inscriptions reveals grammatical agreement rules
2. Administrative pattern matching — Cross-cultural comparison with Sumerian, Egyptian, and Linear B accounting systems
3. Statistical controls — Every proposed pattern is tested against null hypotheses and competing language families
4. Cross-domain convergence — Independent evidence streams (DNA, trade networks, material culture) evaluated via Bayesian updating
5. Negative control validation — Linear B (deciphered Mycenaean Greek) run through the same pipeline to verify methodology doesn't produce false positives
6. Substrate prediction testing — Beekes' ~800 pre-Greek substrate words (independently reconstructed from Greek-internal evidence) used as an external prediction corpus for proposed Linear A readings — three-way convergence with Hurrian phonology
7. Cross-cultural structural analysis — 3,000+ ancient texts from 6 civilizations (Sumerian, Egyptian, Hittite, Hebrew, Greek, Vedic) analyzed for universal ritual formula invariants. Zero-exception rule identified: every libation formula across 2,500 years contains agent + deity + action. Applied to constrain Linear A libation formula structure.
8. Corpus-wide computational analysis — All 1,720 inscriptions processed through statistical pattern matching: commodity co-occurrence, word frequency, document type classification, arithmetic verification of totals

Calibration & Limitations

What the composite score means: The 77.5% is an average across 8 dimensions (vowels, structural features, case system, vocabulary, geography, timeline, scholarly support, religious parallels). It is not a probability. Different weighting schemes produce different numbers, but Hurro-Urartian remains #1 under all tested configurations including linguistic-only scoring with cultural dimensions removed entirely (see LINEAR_A_STATISTICAL_CONTROLS.py — Test 6).

Existing controls (run these scripts to see full output):

• Ablation: Remove any single dimension — Hurro-Urartian stays #1 in 8/8 tests
• Permutation: 10,000 bootstrap iterations with dimension resampling — 100% win rate
• Perturbation: Randomly flip 10-30% of feature values — Hurro-Urartian still #1 in >90% of trials at 30% noise
• Lexical control: Linear A vocabulary scores significantly above random pseudo-lexicons (p < 0.01)
• Cultural weight removal: Scoring on linguistic dimensions only (no geography/timeline/culture) — ranking unchanged

Bootstrap caveat: P(#1)=100% across 10,000 iterations sounds definitive but shouldn't be overread. With only 7 candidate families and 8 partially correlated dimensions, any clear leader will dominate bootstrap resampling. It confirms internal consistency of the result, not that the result is necessarily correct.

Cross-domain independence: The Bayesian convergence analysis treats evidence streams (linguistic, genetic, trade, iconography) as independent, but they are partially correlated — geographic proximity drives both trade contact and genetic admixture. The posterior probability should be read as "consistent across domains" rather than a precise Bayesian probability.

Candidate set size: We compare against 6 realistic candidate families for Linear A. This is small, but it reflects genuine archaeological and geographic constraints — you can't meaningfully compare Linear A to language families with no Bronze Age Mediterranean presence. The tradeoff is acknowledged.

The Circular Reasoning Problem (acknowledged)

Using Linear B phonetic values to read Linear A introduces a Greek phonological filter. Our strongest evidence comes from morphological and structural patterns (agglutination, case agreement, word order) which are less affected by specific phonetic value assignments than vocabulary comparisons.

Data Sources

Primary Corpus

• SigLA Database (Salgarella & Castellan, 2020) — ~3,000 signs from ~400 inscriptions
• GORILA Corpus (Godart & Olivier, 1976-1985) — 5 volumes, ~1,427 inscriptions, ~7,362 sign tokens
• Younger, J.G. — Linear A Texts in Phonetic Transcription (University of Kansas)

Reference Grammars

• Wegner (2007) & Wilhelm (1989) — Hurrian
• Salvini (2008) & Diakonoff (1971) — Urartian
• Hoffner & Melchert (2008) — Hittite
• Huehnergard (2011) — Akkadian
• Allen (2014) — Egyptian
• Rix (2004) & Bonfante (2002) — Etruscan
• Hewitt (1995) — Kartvelian
• Ventris & Chadwick (1956, 1973) — Mycenaean Greek

Pre-Greek Substrate

• Beekes (2010) "Etymological Dictionary of Greek" — 700+ non-IE words identified
• Beekes (2014) "Pre-Greek: Phonology, Morphology, Lexicon"

Vocabulary Comparison

• 38 items across 7 language families (6 competing + 1 control)
• Categories: established Linear A words, morphological suffixes, pre-Greek substrate, administrative terms, libation formula elements

Support

This is a side project built out of genuine curiosity about ancient writing systems. If you find it useful or interesting and want to support continued development, you can buy me a coffee:

No pressure at all — the code is free and always will be.

License

MIT