AI and deciphering Linear A and Etruscan: Can machines crack undeciphered languages?

Overview

The Conversation examines the role of artificial intelligence in deciphering undeciphered languages such as Linear A and Etruscan. It argues that AI acts as a rapid assistant to human researchers, but cannot replace the essential anchors and rigorous review historically required for decipherment.

• AI serves as a fast research assistant, testing hypotheses across thousands of signs and patterns
• An anchor such as a bilingual text or known language family remains essential
• Cross-lingual transfer can reveal patterns from known languages to unknown ones
• Independent expert scrutiny and peer review are necessary to verify any AI-assisted claims

Author: The Conversation

Background and the decipherment problem

The article discusses how deciphering ancient languages has always depended on an anchor, typically a bilingual inscription or a known linguistic relative. Linear A, the script of Bronze Age Crete, is described as a language isolate with no confirmed link to any known language, while Etruscan also lacks a full decipherment. This absence of anchors makes decipherment uniquely difficult and raises questions about the usefulness of AI in such work. AI can test hypotheses at scale, comparing sound patterns and sign sequences across large corpora and even predicting missing characters in fragmented inscriptions. Yet AI cannot conjure meaning out of nothing and must operate within the frame of an anchor—something to ground statistical correlations in a real linguistic relationship.

The AI as a research assistant

The piece clarifies what AI actually contributes. The breakthrough case described involved a self-taught AI engineer who, starting from a Semitic-root hypothesis, used AI-generated scripts to test the pattern against Linear A inscriptions. The process did not yield an automatic translation; rather, it enabled rapid cross-checking of a human hypothesis against thousands of characters, a task that would take humans years. In this sense, AI is a research assistant that accelerates pattern testing and data synthesis, rather than a source of independent linguistic insight.

The need for a genuine anchor

A central claim is that statistical pattern matching alone cannot manufacture meaning. Without an anchor such as a known language family or bilingual text, AI-driven approaches risk mistaking coincidence for pattern. Linear A and Etruscan remain hard precisely because their anchors are missing or incomplete, and no amount of computation can create one from scratch.

Cross-lingual transfer and its limits

The article notes that AI models trained on known language families can sometimes infer patterns in related unknown languages, a cross-lingual transfer. Examples include scripts like Ugaritic where the language family is known. However, even with such transfers, AI cannot derive semantics without an anchor. The conversation emphasizes that AI’s strength lies in identifying patterns and testing hypotheses rather than deriving meaning or translations.

Evidence, verification, and future prospects

The piece cautions that evidence in this field is scant and hard to verify. Linear A’s surviving corpus is small, making independent verification challenging. Claims in this space rely heavily on expert scrutiny and peer review rather than numerical confidence scores. Nevertheless, AI is portrayed as a powerful accelerant for decipherment, enabling more researchers to engage with these problems than institutional resources would allow. The ultimate breakthrough will still require a robust comparative anchor and rigorous human evaluation, not AI alone.

Conclusion

The article concludes that AI’s role in deciphering languages like Linear A and Etruscan is best understood as a fast assistant to human researchers. It speeds up pattern testing, reduces manual cross-referencing, and expands participation, but it does not replace the need for an anchor and careful human review to separate meaningful signals from coincidence.