📄 Embedded topicalisation in Old English: Does it exist?

Summary

The paper challenges a widely held claim in Old English syntax: that topicalisation, meaning the fronting of an object or PP to the start of a clause, never happens inside subordinate clauses. Earlier work either banned it outright (Pintzuck, Stockwell & Minkova) or accepted it without counting cases (Allen).

The author first contrasts Present-Day English, where embedded topicalisation is ungrammatical, with German, using an informal grammaticality survey. He then runs CorpusSearch 2 queries over the YCOE, a syntactically annotated collection of Old English prose produced at York, Toronto and Helsinki. The texts cover narrative, argumentative and technical prose, and include both native OE works and translations from Latin.

The search targets four embedded orders: OSV, XSV, OVS and XVS. Each hit is coded for:

• pronominal vs. full-NP object
• full-NP vs. impersonal 'man' subject
• nominal vs. adverbial clause

All four orders are attested, including some with heavy, relative-clause-bearing objects. Fronted objects are mostly pronouns. Inverted clauses always have full-NP subjects, which points to information-structure motives. The result also undermines a theory of the later XV→SV change that relies on nothing being extractable from the VP in subordinate clauses.

Key points

• Data and tool: ten prose texts drawn from the 2003 YCOE (the syntactically annotated Old English prose corpus from York, Toronto and Helsinki), searched with CorpusSearch 2.
  • Texts: Chronicle A/E, Orosius, Bede, Ælfric's Lives of Saints and Old Testament, Cura Pastoralis Preface, Boethius, Herbarium, Medicina de Quadrupedibus.
  • The Cura Pastoralis Preface yielded no hits.
• German survey: every informant accepted canonical SOV. For embedded OSV, only four accepted it fully, ten accepted it in colloquial contexts only, and most rejected it. Nobody accepted embedded OVS.
• Topicalisation with inversion: embedded OVS/XVS is rarer than OSV but occurs in five of the ten texts, so it is not an isolated quirk.
• OVS profile: all fronted objects are pronouns and all subjects are full NPs. Most clauses are adverbial.
• XSV profile: only two cases have the impersonal 'man' as subject.

Interesting ideas

• I1. Test the "never" claim directly: a categorical ban is checked by searching a parsed corpus for the supposedly impossible configuration and counting hits across genres, rather than trusting the handbooks.
• I2. Code every hit on a few fixed variables: each match is tagged for pronoun vs. NP topic, subject type and clause type, which exposes which configurations really drive a pattern.
• I3. Read word order as given vs. new information: fronted pronouns carry known information, while heavy subjects pushed rightward after inversion carry new information. Word order thus signals information status.
• I4. Balance the sample by genre and origin: texts are chosen across genres and split into native vs. translated sources, because both can shift syntax.
• I5. Judge acceptability on a graded scale: the German survey allows a "fine but only colloquially" answer instead of a yes/no verdict.

For Mova

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

• I1 (learn): Treat "rule X never fires" assumptions in the parser or the repair rules as hypotheses. Mine silver treebanks for counterexamples before hard-coding a constraint.
• I2 (tree): When auditing non-canonical orders (fronted objects, inversion) in gparse output, break errors down by argument weight and clause type to find where parsing fails.
• I3 (frames): Use fronting and inversion as cues for topic vs. new information when building frames. This can help coreference prefer fronted pronouns as links to earlier discourse.
• I4 (learn): Build evaluation sets stratified by genre and by original vs. translated text, so that parser accuracy is not inflated by one register.
• I5 (answer): Let readers flag constructions as marked or colloquial rather than simply correct or wrong, giving graded and explainable confidence on unusual sentences.