AI Deciphers Ancient Greek Linear A Script: A Milestone in Interdisciplinary Research

This article introduces a graduation project that uses machine learning (ML) and natural language processing (NLP) technologies to attempt to decipher the Linear A script, known as "Europe's oldest unsolved mystery". The project demonstrates the application potential of AI in archaeology and historical linguistics, as well as the methodological value of interdisciplinary research, providing a new example of the integration of humanities and technology.

Linear A is a syllabic script used by the Minoan civilization on Crete from 1800 to 1450 BCE, and it is the predecessor of the already deciphered Linear B. Its decipherment faces three major obstacles:

• Limited corpus: The number of existing inscriptions is limited, and most are short administrative records
• Unclear language affiliation: Scholars cannot determine the language it records
• Lack of bilingual comparison: There is no Rosetta Stone-like bilingual text, making traditional cryptography methods difficult to work

The project adopts a computational archaeology paradigm, with core technologies including ML (character recognition, pattern discovery), NLP (structural analysis, contextual relationships), and statistical modeling (quantification of character co-occurrence). The research steps are:

1. Data collection and digitization
2. Extraction of features such as character morphology and strokes
3. Unsupervised learning to discover character clustering patterns
4. Comparison with writing systems like Linear B
5. Hypothesis generation based on statistical significance

AI application scenarios in the project:

• Character recognition: Using CNN or Transformer to process ancient scripts with large morphological variations and high noise
• Sequence modeling: Using RNN/Transformer to learn syllable combination patterns
• Transfer learning: Using pre-trained Linear B models to transfer to Linear A analysis, solving the small sample problem

Key challenges and solutions addressed by the project:

• Small sample learning: Expanding samples through data augmentation (rotation, scaling), semi-supervised learning, and active learning
• Domain knowledge integration: Incorporating linguistic rules and historical context (knowledge graphs, rule-guided models)
• Interpretability: Using attention visualization and feature importance analysis to ensure verifiable conclusions

Two-way value of interdisciplinary research:

1. Technology empowers humanities: AI processes large-scale data to discover subtle patterns that are hard for humans to detect
2. Humanities feed back to technology: Ancient script research promotes innovation in small-sample and noise-robust algorithms This project represents a new paradigm in digital humanities, where scholars will transform into a combination of data scientists and domain experts

Future directions:

• Technical deepening: Multimodal learning (integrating images, text, and archaeological context), generative models (generating Linear A characteristic texts), crowdsourcing collaboration platforms
• Application expansion: Extending the methodology to deciphering unsolved ancient scripts such as the Indus Valley script and the Rongorongo script of Easter Island