📄 Arguments or macroroles? : Two functional approaches to Old English quirky case

Summary

The paper asks how Old English "quirky" case should be explained: accusatives, genitives and datives that break the usual pattern. It compares two functional theories. In Functional Grammar (FG), argument structure rests on semantic valency, and case is treated as a matter of surface expression. In Role and Reference Grammar (RRG), argument structure rests on macroroles (ACTOR, UNDERGOER), which are partly fixed by syntax, so case becomes central to how semantics is linked to syntax. Using examples from the Helsinki Corpus, the authors show that Van Valin's RRG case rules cover most Old English clauses. They also argue that quirky case is more widespread than Roberts (1995) claimed. Their main proposal is that quirky case does not belong to particular intransitive state verbs. Instead, it marks constructions that depart from both the transitive and the intransitive prototype, and such constructions also allow more case options. Prototypicality is defined by macrorole alternation that depends on telicity. They add a Principle of Lexical Template Instantiation, which favours fully specified argument mappings. This ties marked morphology to argument structure that is less than prototypical.

Key points

• RRG case rules for accusative languages (Van Valin 1991):
  • the highest-ranking macrorole takes NOM;
  • the other macrorole takes ACC;
  • non-macrorole core arguments take DAT by default.
• In a ditransitive with sellan 'give', the recipient is denied UNDERGOER status and therefore surfaces in the dative. A supporting fact is that Old English had no passive with the recipient as subject.
• The examples are drawn from the Helsinki Corpus, specifically its diachronic section (Rissanen & Ihalainen 1984), and a few are cited from Bosworth & Toller's dictionary (BT).
• The four construction types:
  • Prototypical transitive: active accomplishment with creation/consumption verbs (writan 'write' + a bounded object).
  • Less-prototypical transitive: the activity use of those same verbs.
  • Prototypical intransitive: active accomplishment with motion verbs (faran, swimman + a goal phrase).
  • Less-prototypical intransitive: the activity use of motion verbs.
  • State and causative-state verbs are non-prototypical either way.
• The Principle of Lexical Template Instantiation says that templates tend to map maximal implementations onto syntax, so that semantic participants line up one-to-one with constituents as far as possible. It is meant to work together with, and apply before, the Completeness Constraint of Van Valin & LaPolla (1997).

Interesting ideas

• I1. Treat case marking as the output of a small ranked algorithm over generalized roles rather than a per-verb list. Deviations then become measurable signals instead of noise.
• I2. Whether a construction is transitive depends on the clause, not just the verb. The same verb (drink, write) is two-role with a bounded, fully referential object and one-role without it.
• I3. Telicity, contributed by the object or by a goal phrase, defines how prototypical a construction is. That gives a cheap syntactic cue for aspect.
• I4. Marked morphosyntax is predicted to co-occur with greater variability, meaning more alternative case choices, in non-prototypical constructions.
• I5. A maximal-instantiation principle says to prefer analyses in which every semantic participant has its own syntactic constituent.

For Mova

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

• I1 (frames): Map UD obj/iobj/obl onto ACTOR/UNDERGOER through a ranking rule, and log the cases where VerbNet roles and that rule disagree as explainable exceptions.
• I2 (frames): Decide UNDERGOER status per clause from object referentiality (determiner, count/mass), so that "was drinking beer" and "drank a pint of beer" get different frame completeness.
• I3 (world): Mark events whose frames include a bounded object or a to-goal as completed or culminated, which helps the event-order and state-change logic in the world model.
• I4 (learn): Expect witness parsers to disagree more on state and experiencer verbs, and use non-prototypicality as a prior for flagging silver-data trees that need review.
• I5 (back): When generating text from the meaning graph, prefer surface forms that realize every frame participant explicitly, and only reduce arguments when there is a stated reason.