📄 Linguistic change: the grammatical environment of participial non-finite clauses in Old English and in present-day English

Summary

The paper asks whether a new, more compact construction spreads through a language according to how grammatically complex its surroundings are. It compares present-participle non-finite clauses (PNF) with adjectival or adverbial dependent finite clauses (DC) that could replace them. Following Natural Phonology terminology, PNF is treated as the "weakened" variant (cheaper for the speaker, harder for the hearer) and DC as the "strengthened" one. Old English is taken as the early stage of the change, because the participle's verbal government was only then emerging, and present-day English as the later stage. Matched basic and control samples from both periods were tagged for markedness-based parameters, such as tense, mood, number, transitivity and modality, and for extra parameters motivated by pidgin and creole studies. In Old English, PNF clearly favoured unmarked, simple environments. By present-day English, PNF environments had gained marked values while DC environments had lost them. This supports the claim that weakened variants move from simple contexts into complex ones, and it offers frequency evidence for deciding which grammatical values count as "complex".

Key points

• The research programme began in 1986 at Ljubljana and Maribor under Janez Orešnik. Earlier tested pairs were Old English expanded vs non-expanded tenses (1990) and ge-verbs vs unprefixed verbs (1994), and each reached roughly 75% correct predictions.
• Old English data came from Sweet's edition of the Orosius and Skeat's editions of the Gospels of Mark and John. The first basic sample (B1) held all 114 PNF instances from the Skeat material. The study used three basic/control sample pairs in total, with B2/B3 and C2/C3 drawn from present-day English.
• Markedness followed Mayerthaler's Natural Morphology. Non-affirmative modality, non-present tense, non-indicative mood, non-singular number and transitivity were treated as "complex".
• The added parameters covered the number of arguments, person, animacy, the interaction of stativity and tense, and the syntactic role (subject, direct object, indirect object, adjunct) that the PNF/DC subject's referent plays in the matrix clause.
• The mean count of marked values was highest for Old English DC and lowest for Old English PNF. Over time it rose for PNF and fell for DC.

Interesting ideas

• I1. Variant pairs that carry the same meaning can be ranked by speaker economy versus hearer transparency, as in "old men and women" versus "old men and old women", so the choice between them becomes a tradeoff rather than a matter of style.
• I2. Each participial clause is aligned with a finite dependent clause that paraphrases it, which makes the clause's implicit subject and its adverbial relation (when, as, because) explicit.
• I3. The paper records which matrix role, mostly subject or direct object, supplies the referent for the participle's understood subject. This yields a usable prior for resolving controllers.
• I4. Clause complexity is scored by simply counting how many parameters take marked values, which gives a transparent and additive difficulty measure.
• I5. Every hypothesis is checked against a control sample containing the competing variant, and the share of confirmed predictions is reported honestly instead of claiming a clean win.

For Mova

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

• I1 (back): The generator could choose a participial clause in simple contexts and switch to an explicit finite clause when the surrounding clause carries many marked features, which would keep output decodable.
• I2 (frames): Participial adjuncts could be expanded into finite-clause frames, with the hidden subject filled from the matrix clause, giving every reading an inspectable who-did-what.
• I3 (frames): Coreference for PNF subjects could try the matrix subject first, then the direct object, and only then the less likely indirect object or adjunct roles.
• I4 (answer): Test items or story sentences could be graded by their marked-value count, so that reader performance can be broken down by grammatical complexity.
• I5 (learn): Silver-data or repair-rule claims could be evaluated against a matched control construction and reported as a prediction hit rate, in the spirit of negative controls.