Computational linguists
sign statistics, language models, stylometry
Undeciphered scripts and disputed texts can be approached through sign statistics, language models and stylometry. These problems need computational linguists who can set up tests that a proposed reading must pass.
12 problems need computational linguists
- 01When was Beowulf composed?
Train a dating model only on independently dated Old English texts, treat copying as explicit noise, and test it on held-out texts preserved in late copies.
Chronology · Europe · 680–1020also needs philologists - 02What new evidence can fix when and why five historically extinct species disappeared?
Build name-variant searches and HTR models that find these species in Chronicling America, VOC transcriptions and Icelandic periodicals, and report recall against hand-checked samples.
Chronology · Global · 1500–1945also needs archivists, geneticists, data scientists, philologists, archaeologists and GIS specialists - 03Which lost works of Galen survive unrecognised in Arabic and Syriac?
Model how Hunayn-school translators rendered Greek terms, using the Glossarium Graeco-Arabicum, and apply it to OpenITI texts to flag passages matching no extant Greek work.
Texts & authorship · Middle East · 160–1270also needs archivists and philologists - 04Which of the journeys in Ibn Baṭṭūṭa's Riḥla did he actually make?
Run passim-style text-reuse detection of the Riḥla against OpenITI, report precision and recall on the known Ibn Jubayr borrowings, then list every other candidate source passage.
Texts & authorship · Global · 1325–1355also needs philologists and GIS specialists - 05Does the Indus script encode language, or a non-linguistic sign system?
Build comparison corpora of known scripts and non-linguistic systems trimmed to Indus text lengths, extending Sproat (2014), and test measures beyond entropy such as sign-position regularities.
Languages & scripts · South Asia · 2600–1900 BCEalso needs archaeologists and data scientists - 06Do khipus record more than numbers, and can their non-numerical signs be read?
Using the Open Khipu Repository schema, write rival readings of colour, ply and attachment as models that predict non-numerical features of held-out cord groups.
Languages & scripts · Americas · 1400–1800also needs archaeologists and data scientists - 07What language does Linear A record, and can any proposed family beat chance?
Write the Anatolian, Semitic and isolate hypotheses as explicit generative models and score them on held-out texts against shuffled sign values and unrelated control languages.
Languages & scripts · Mediterranean · 1800–1450 BCEalso needs philologists - 08What texts did Newton copy in the alchemical manuscripts whose sources are unidentified?
Build alignment searches over the Newton TEI, EEBO-TCP and EMLAP that tolerate translation, alchemical symbols and spelling variation, and report ranked candidates with their null results.
Texts & authorship · Europe · 1660–1700also needs philologists, archivists and conservation scientists - 09What do the thousands of unread Shang oracle-bone graphs mean?
Build held-out tests in which a proposed reading must predict an unread graph's slot, collocations and parallel charges in inscriptions withheld from the analysis.
Languages & scripts · East Asia · 1250–1046 BCEalso needs philologists and archivists - 10Does Proto-Elamite record language, and is it the ancestor of Linear Elamite?
Train sign-distribution models on proto-cuneiform, where logograms and phonetic signs are known, then test whether the residual Proto-Elamite signs contain a comparable phonetic subset.
Languages & scripts · Middle East · 3300–1880 BCEalso needs philologists - 11Was rongorongo an independent invention of writing?
Compare rongorongo sign sequences against written languages and non-linguistic symbol systems, and test proposed readings by predicting held-out passages under alternative sign segmentations.
Languages & scripts · Pacific · 1400–1870also needs archaeologists and conservation scientists - 12Who supplied the medical knowledge that Hans Sloane's catalogues attribute to unnamed people?
Classify the attribution of every Vegetable Substances entry against the manuscript facsimiles, flagging composite entries and 'from the same' chains, and publish coding rules with a hand-checked sample.
Science & knowledge · Global · 1680–1753also needs archivists, philologists and data scientists
Can you help?
Propose an approach or leave a comment on any problem above, or write to say what you could contribute.
Drafted by Claude Opus 5.5