AI-generated conjecture · a wager, not a finding
← All conjectures · Literature & poetics
The dictionary is a graveyard with citations for headstones
Status is derived only from the shepherd-authored triage/prediction data above -- community submissions and claims are a separate overlay and can never change it (see the participation panel below).
Claim (verbatim)
The electronic Dictionary of the Irish Language (eDIL, dil.ie), which digitises and revises the Royal Irish Academy's great Dictionary, is built entirely from citations: every headword is a lemma illustrated by quotations from the medieval corpus, each tagged with the source it was drawn from. A great many of those sources are the Old Irish glossaries - Sanas Cormaic (Cormac's Glossary), O'Mulconry's Glossary, O'Davoren's Glossary - which were themselves antiquarian salvage operations, compiled to explain words already obscure to medieval readers and citing texts that were rare or dying even then. Where a headword's only attestations are glossary citations, the dictionary is functioning as a fossil bed: the word, and often the lost text it was quarried from, survives nowhere but in the lexicographer's specimen jar. The loss mechanism is that a lexicon outlives its literature - a glossator's quotation can be the last trace of a poem or a legal tract that otherwise perished entirely. Restated: a substantial minority of eDIL headwords are attested only from the glossary tradition, and each such lemma is the sole surviving witness to a scrap of vanished Irish text. The dictionary is not a summary of the corpus; in its rarest entries it is the corpus's only grave-marker.
Prediction clause (verbatim)
Prediction: among eDIL headwords, a substantial minority - provisionally set at 5 percent or more - will have a citation set drawn entirely from the Old Irish glossary tradition (Sanas Cormaic, O'Mulconry's Glossary, O'Davoren's Glossary and their kin), marking them as glossary-only survivals (primary clause: a glossary-only headword share of at least 0.05; the verdict follows it). This operation is not yet built: resolving it requires filtering every eDIL headword by the source-sigla of all its citations and classing each siglum as glossary versus running-text, a bulk query eDIL does not currently expose; the primary clause therefore voids until a citation-source-provenance index exists.
Kill-dataset (verbatim)
Kill (not yet built): eDIL (dil.ie) headword-and-citation data with every citation's source-siglum classified as glossary or running-text - count the headwords whose entire citation set is glossary-sourced. Needs a citation-source-provenance index over eDIL that the public interface does not yet provide in bulk.
Nobody has run this test. The kill-data is named above. If you can run it — or you know the paper that already settles it — claim the kill or submit the prior scholarship. Kills and prior scholarship are credited here, by name, as they come in.
Provenance
Run: Fresh agent generation · model: claude-fable-5
Fresh blind generation by claude-fable-5, 2026-07-17, North Atlantic vernacular wave (Old Norse / medieval Irish / medieval Welsh) instrument-anchored on the Skaldic Project (skaldic.org), CELT (celt.ucc.ie) with eDIL (dil.ie), and the printed censuses (Mac Cana 1980; Bromwich TYP 3rd ed. 2006; Thurneysen 1921): every Kill names a real open corpus or printed edition and a countable operation on a self-index of loss (catalogue-to-corpus survival fractions, patron-era gradients, citation-only census of lost poems/codices, in-codex presence rate, bipartite list-vs-shelf mismatch, corroboration asymmetry, single-witness-passage counts), thresholds far from 1 with coverage guards and in-prediction disambiguation. Deliberately DISJOINT from the owned European ground: w01 no.23 'Prose is the ark' owns the skaldic.org stanza-embedding-in-prose fraction, w07 no.5 owns lausavisur legal distribution, w07 no.26 owns kenning-innovation decay, w07 no.37 owns the Mabinogi-as-legal-casebook coding, w07 no.38 owns the composition-vs-manuscript date-gap (uses skaldic.org only as one of four corpora), w07 no.47 owns bishops'-saga miracle registers, and w07 no.36 owns Irish scribal-marginalia seams - none of my operations reuse these; item 4 (treatise-only drapur) is adjacent to w01 no.23's poetics-treatise secondary clause but counts whole-poem transmission channel, not the metrical-anomaly rate, seam flagged. CONFIDENCE FLAGS (named facts kept out of load-bearing thresholds where soft): Skaldatal poet count '~140' is MODERATE (counts vary); the exact number of Snorra-Edda-only eddic compositions is MODERATE (threshold set at a robust >=3, with Grottasongr and Heimdalargaldr as certain); Mac Cana tale-list total '~200' is MODERATE (threshold set as a fraction, not a count); Cin Dromma Snechtai reconstructed contents are MODERATE (scholarly reconstructions differ - Thurneysen vs Carney/McCone); Bromwich TYP triad count '~90+' and survival rate are MODERATE; Ifor Williams's '12' historical Taliesin poems is the standard figure but approximate; White/Red Book single-witness-passage count '>=20' is MODERATE on exact number, robust in direction. Dropped candidates: (a) a lausavisur-only survival-channel item - dropped as duplicating w01 no.23; (b) a kenning-family type-token item - dropped as duplicating w07 no.26; (c) a Codex Regius great-lacuna prose-recovery item - dropped as leaning on the w01 no.23 prose-ark mechanism; (d) a second glossary hapax-rate item (Sanas Cormaic alone) - merged into the single eDIL fossil-bed item to avoid redundancy; (e) an Irish scribal-marginalia item - dropped as owned by w07 no.36. Item 9 (eDIL glossary-only headwords) is flagged not-yet-built: it needs a citation-source-provenance index over eDIL that the public interface does not expose in bulk.
Novelty / leakage triage
anticipated in the literature — this exact test has never been run
The fossil-bed phenomenon is thoroughly documented: Russell's and Moran's work on the Irish glossaries treats them precisely as salvage of otherwise-lost text, the Early Irish Glossaries Database supplies the glossary side in structured form, and eDIL's entries visibly carry glossary-only attestations. But the load-bearing 5-percent share requires classifying the source-sigla of every citation of every headword, an index the public eDIL interface does not expose in bulk - the wave itself flags the operation as not yet built. No published count of glossary-only lemmata as a share of the dictionary exists to leak the answer in either direction. Adjacent, pending the citation-provenance instrument.
- P. Russell, 'The Sounds of a Silence: The Growth of Cormac's Glossary', Cambridge Medieval Celtic Studies 15 (1988)
- P. Moran (ed. and trans.), De origine scoticae linguae (O'Mulconry's Glossary) (Turnhout, 2019)
- Early Irish Glossaries Database (P. Russell, S. Arbuthnot, P. Moran, University of Cambridge, www.asnc.cam.ac.uk/irishglossaries/)
- eDIL: electronic Dictionary of the Irish Language (dil.ie)
Predictions
No prediction registered yet.
Weigh in
No community feedback yet.
New here? Create an account first
Create an account or sign in and your feedback is tied to you — you can track it, get replies, and claim this conjecture so others know you’re working on it. Prefer not to? Just leave your take below as a guest — only the name you type is shown.
Add your take
Posted immediately (spam is removed). Community feedback is never an adjudicated verdict and never changes this conjecture's triage label or status above.