‘Hreran mid hondum hrimcealde sæ’: On verb attraction in Old English

Summary

The paper examines "verb attraction" in Old English, a word-order pattern where a non-finite verb leaves its expected position and joins the higher finite verb. The analysis uses Government and Binding Theory. Each verb gets its own VP and IP, and a verb can sit under V, I or C, with I placed to the right of the VP so that clause-final verbs are possible. The author draws attested sentences from the 2004 Toronto Corpus and gives tree derivations for two configurations. The first is an auxiliary plus a main verb. The second is the object-control verb hatan 'to command' plus its infinitive clause, where attraction crosses a CP boundary. In every case the derivation runs normally until one extra adjunction step at the end moves the non-finite verb next to the finite one, wherever that verb stands. The paper closes with a line from The Wanderer ("hreran mid hondum hrimcealde sæ"). It argues that attraction explains why an adverbial seems to separate the verb from its object, and that the pattern let the poet place an alliterating verb where he needed it.

Key points

• The framework is Government and Binding Theory. Old English verbs have three possible sites: V inside VP, I inside IP (head-final), and C inside CP, which is reached through topicalization.
• Every example and its identification label is taken from the Toronto Corpus as released in 2004, and the translations are the author's own.
• With an auxiliary, attraction gives main-verb-before-auxiliary order in two settings: early in the clause (examples 4–6) and clause-finally in "reversed" order (examples 7–9).
• With hatan, the finite verb can sit in V (rare, e.g. a relative clause from ChronF), in I, or in C after topicalization. The infinitive is raised out of its CP and adjoined to het in each case.
• The object of hatan is often left unexpressed. The author links this to a cultural default: commands came from people in authority, and servants or soldiers were assumed to be present to carry them out.

Interesting ideas

• I1. An odd word order is treated as an ordinary derivation plus one final movement, so a single extra step accounts for the deviation instead of a separate grammar.
• I2. Each verb in a two-verb chain keeps its own projection, complements and features, even when it sits right next to the other verb.
• I3. The verse is analysed as continuous prose without line breaks, so the meter does not distract from the syntax.
• I4. An unexpressed controllee (the person ordered) is explained by world knowledge about who carries out commands.
• I5. The movement is given a functional motive: it puts an alliterating verb exactly where the verse needs it.

For Mova

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

• I1 (tree): Repair rules could model displaced verbs in archaic or poetic English as one reversible adjunction on top of a canonical parse, so the UD tree can be restored and the change explained.
• I2 (frames): Frame extraction should take arguments from each verb's own complement structure, not from its surface neighbour, so stacked verbs like "commanded … to build" yield two linked frames.
• I3 (tokens): Poems and songs could be flattened into sentence units before tagging and parsing, with line positions kept as metadata for later stylistic use.
• I4 (world): For command verbs with no object, the world model could add a default "subordinate executor" entity, flagged as inferred, rather than leaving the role empty.
• I5 (back): The generator could treat word-order variants as options chosen for rhythm, emphasis or sound, while recording which variant was used and why.