📄 The Three Hosts of Doomsday in Celtic and Old English

Summary

Breeze takes up a penitential commonplace found in late Old English homilies and confessional texts. It holds that being shamed before a single confessor now is preferable to being exposed at Judgement before the three hosts of heaven, earth and hell. Malcolm Godden had treated this as a distinctively Anglo-Saxon motif. C. D. Wright had already pointed to Irish parallels, and Breeze sharpens that case. He first catalogues the Old English attestations by function. He then brings in an earlier witness Wright did not use: the unedited final section of the Hiberno-Latin Liber de numeris (mid-eighth century), whose manuscripts he traces across South Germany, the Rhineland and the Loire. Drawing on Bouhot's study, he also argues that the supposedly Bonifatian homily is a later Carolingian compilation. That homily shares wording with the Hiberno-Latin Catachesis Celtica. The conclusion is that the motif was most likely coined by an Irishman, or less likely by a Continental scholar steeped in Hiberno-Latin. This matters because it moves a supposedly native English theme into the Irish-influenced learning of the Continent.

Key points

• The paper inventories Old English witnesses, labelled (a)–(t). Fifteen use the shame formula, and none survives in a manuscript much earlier than about 975. Seven give it as advice (a–g), six as a warning (h–m, mostly Ælfric), and two passages (Vercelli VIII, Christ III) appear to have dropped or replaced the hosts. At least five more mention the Three Hosts without any shame.
• Liber de numeris is not the same work as the seventh-century Liber numerorum. It is a Hiberno-Latin guide to themes countable from one to eight, probably produced at Salzburg among the scholars around St Vergil (d. 784). Its oldest copy is Colmar MS 39 (c. 775, from Murbach), and Breeze cites its Three Hosts statement from London, BL, MS Royal 5 E. VI.
• The motif's first Anglo-Latin use is on the Continent, not in England: a letter by Alcuin to Irish monks, dated between c. 792 and 804.
• The sermon credited to Boniface survives early only in Paris, B. N., MS lat. 10,741. Bouhot treats it as a preacher's handbook that may be little older than 850, and its phrasing is nearly identical to Catachesis Celtica (Vatican, Reg. lat. 49, of Breton provenance).
• The Three Hosts theme persists in Welsh vernacular texts to the fourteenth century and in Irish ones to the seventeenth.

Interesting ideas

• I1. Breeze sorts every occurrence of one formula by the job it does (counsel, threat, or bare mention), and notes copies where one element seems to have been cut or swapped; this separates a motif's content from its pragmatic use.
• I2. A text is dated by what travels with it: a baptism tract of 811/812 bound alongside the sermons gives a lower bound for the collection.
• I3. Near-verbatim overlap between two texts is treated as evidence of dependence, strong enough to overturn an authorship attribution.
• I4. The claim that a motif is unique to one tradition rested partly on incomplete coverage: the Patrologia Latina edition was based on an incomplete and unreliable late copy, and the key passage sat in an unedited part. Checking the full transmission overturned it.
• I5. Liber de numeris organises doctrine as numbered sets (six ages, seven gifts, eight sins), a compact enumerative scheme for storing knowledge.

For Mova

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

• I1 (frames): prag could tag a recurring proposition with distinct speech-act labels (advice vs. warning) depending on context, rather than treating identical content as one act.
• I2 (world): event ordering could use co-occurrence anchors, so that a dated companion fact bounds when an undated entity or state can have arisen.
• I3 (learn): the same reasoning can flag near-duplicate passages between silver training data and evaluation sets, and trace which source a witness parser's output actually derives from.
• I4 (learn): evaluations claiming a phenomenon is absent should be run against complete sources and paired with negative controls, not judged from a partial sample.
• I5 (world): the knowledge graph could store explicitly counted groupings as first-class sets with cardinality, so readers can check "three hosts" or "seven gifts" against their listed members.