📄 A Greek–Semitic bilingual (?) with a sacred regulation from Dreros

Summary

The paper re-examines a lost archaic inscription from Dreros in eastern Crete. It was first edited by H. Van Effenterre, and only his drawing survives. The authors propose that the stone carries a bilingual text. Line 1 (Text A) is a right-to-left letter string, τυπρμηριηια, which they read as Eteocretan. Lines 2–3 (Text B) are a Greek sacral rule about an oath and something that must be pure.

Earlier attempts forced line 1 into Greek ("hypermeridia", a sacrificial thigh portion), but that required inserting letters and altering a clearly read one. The authors refuse such emendations. Instead they transliterate the consonantal string into Hebrew characters and segment it into North-West Semitic words: (y)tw pr mhr yhyh, roughly "with a bullock, let it be ready." The verb form yhyh would parallel Greek γένοιτο, as Cyrus H. Gordon had earlier argued. Supporting cues include vowel letters used as matres lectionis, a vertical word divider, and the writing direction. If confirmed, the reading would revive the idea that Eteocretan was Semitic.

Key points

• Find details:
  • The stone was discovered in 1936 by P. Demargne and H. Van Effenterre.
  • It lay 3–4 m deep, reused in a Hellenistic cistern near the temple of Apollo Delphinios.
  • It was later lost in WWII bombing while held at the Neapolis museum.
• Object details:
  • The stone was a grey schist block of 99 × 23 × 23 cm.
  • It carried three lines in the archaic Cretan alphabet.
  • It is dated to roughly the second half of the 6th century BC.
• Rejected Greek reading: Van Effenterre (1989), following Duhoux, read τ(α) υπ<ε>ρμηρίδια. This needs two inserted letters plus turning H into Δ, which the authors judge too many interventions for a fragment.
• Proposed Semitic segmentation:
  • tw: preposition "with," compared with Phoenician forms and the name Ithobalos.
  • pr: bull or calf, including sacrificial calves in Exodus 24:5, rendered μοσχάρια in the LXX.
  • mhr: "ready, apt."
  • yhyh: 3rd person masculine singular prefix form of hyh, "be/become."
• Corroborating features:
  • The text is mainly consonantal.
  • The letters υ, ι, α serve as vowel letters.
  • A vertical-bar separator is also known in Phoenician epigraphy.
  • The text runs right to left.

Interesting ideas

• I1. Treat the attested characters as reliable and penalize any interpretation by how many corrections it needs; a hypothesis requiring three edits to short, clearly read input should lose to one requiring none.
• I2. Use the understood half of a bilingual text as an anchor for the opaque half, matching a functional slot such as the optative verb γένοιτο to a candidate word (yhyh) in the other language.
• I3. Build a case from several independent weak cues (vowel-letter usage, separator, writing direction, lexical parallels) rather than one decisive feature.
• I4. Segment an unbroken letter string into word candidates and keep only the split that composes into a grammatically sensible clause.
• I5. Re-encode a sequence in a representation suited to the hypothesized structure (here, a consonantal script) so that patterns become visible.

For Mova

How these ideas could serve Mova, our small open language model — its files, stages and tests:

• I1 (tokens): Give spelling/normalization repair an explicit edit budget, and prefer analyses that explain the raw token string without "fixing" it.
• I2 (back): When translating through the meta-world, align the frame slots of a known-language sentence to constrain the reading of a poorly covered language.
• I3 (learn): Mirror this in silver-data acceptance by requiring agreement of independent witnesses, each weak alone, before trusting a label.
• I4 (tree): For ambiguous tokenization (hyphenated or run-together forms), score candidate splits by whether they yield a well-formed UD tree, not by lexicon lookup alone.
• I5 (tokens): Alternative normalized views of a token, such as stripped affixes or phonetic keys, can expose lemma matches that the surface spelling hides.