AI Takes On Dead Languages: It Finds the Patterns, but Meaning Is Still the Scholar's Job

Neural networks are already helping historians read texts that have stayed incomprehensible for millennia. But for all the power of these algorithms, deciphering a script still requires a human: AI is excellent at finding regularities in symbols, but only a linguist can give them meaning.

What a neural network can do in linguistics

The main advantage neural networks have over a researcher is the speed and scale of statistical analysis. In a few hours, an algorithm can process a corpus that would take a scholar years: it counts symbol frequencies, builds co-occurrence matrices, and identifies structural patterns.

One of the most telling examples is Ithaca, a project developed by a DeepMind team together with historians and published in Nature in 2022. The system restored damaged fragments of ancient Greek inscriptions. The algorithm's accuracy was about 62%, while specialists working without AI picked the correct option in only 25% of cases. When historians worked in tandem with the neural network, the result rose to 72%.

Ithaca worked well precisely because a large corpus of deciphered texts already existed for ancient Greek. The algorithm drew on thousands of inscriptions that scholars had studied before, and used that foundation to make predictions for unknown fragments.

• Ithaca (DeepMind, Nature, 2022): 62% accuracy in restoring inscriptions
• Historians without AI: about 25% correct answers
• Historians + Ithaca: 72%
• The largest undeciphered writing systems: Linear A (Crete), Proto-Elamite (Iran), Rongorongo (Easter Island), the Indus script (Mohenjo-Daro)

Why can't we do without a linguist?

AI can describe the structure of a text, but it can't link that structure to meaning. This is especially true where there is no "anchor": a parallel text like the Rosetta Stone for Egyptian hieroglyphs. For Linear A, Proto-Elamite, or Rongorongo, no such entry point exists, and without one, decipherment turns into guesswork.

An algorithm will detect that certain symbols appear next to each other more often than others, but without cultural context it can't put words to them. The linguist brings what the data lacks: knowledge of language typology, an understanding of how trade records differ from religious texts, and an intuition for grammatical categories. None of this can be learned directly from a corpus.

"AI is fantastically good at pattern recognition, but human intuition remains key."

That is what researchers in computational linguistics interviewed by Ars Technica emphasize.

How scholars' work is changing

The "neural network + linguist" pairing doesn't replace the scholar; it changes the pace of the work. The algorithm takes over the routine: frequency analysis, restoring missing symbols, and finding structural matches in parallel corpora. The specialist focuses on interpretation, the stage that requires an understanding of the world, not just of the text.

Some researchers compare this shift to how MRI changed diagnostics: the doctor still makes the diagnosis, but can see details that were previously out of reach. Teams that once spent months hand-coding corpora now get initial patterns within days. Several dozen undeciphered scripts remain in the world, and some of them, specialists believe, could be read in the coming years thanks to this partnership.

What this means

The use of AI in linguistics shows how neural networks amplify expertise rather than displace it. Where parallel texts and enough data exist, the algorithm works as an accelerator. Where cultural context is opaque or data is scarce, the scholar still has the last word.