📄 Negation and Pronoun Position as a way of investigating verb movement and the emergence of do support during Early Modern English

Summary

The paper asks whether do support in English questions was simply a consequence of losing main-verb raising (V-to-I and V-to-C movement), or whether it was partly a cause of that loss. The author collects negative interrogatives from 30 Shakespeare plays (roughly 1589–1634) using the University of Virginia Modern English Corpus search engine, filtered to drama from 1550–1650. Only questions that show either a raised main verb or inserted do, and that contain pronominal subjects and/or objects, are kept. Each example is coded by whether the pronoun stands before or after not and by whether do appears. The two strategies coexist in the same author. With a subject pronoun before negation, the sample is split almost evenly (51 of 89 use do). When not precedes the pronoun, do dominates at over 90%. All eight questions with an object pronoun before not use raising. The author argues that do arose to keep the verb from lifting an object pronoun alongside an adjacent subject pronoun, so do helped drive the loss of verb movement rather than merely follow it.

Key points

• Data: negative interrogatives from 30 Shakespeare plays, drawn from the Modern English Corpus hosted by the University of Virginia. The search was narrowed in stages: period and genre, then construction type (raising vs. do), then pronoun subjects/objects.
• Subject pronoun above not: 89 sentences, with 38 (42.69%) showing verb raising and 51 (57.30%) showing do support.
• Object pronoun above not: only 8 sentences, all with verb raising and no do.
• Not above the subject pronoun: 16 sentences, 15 (93.75%) with do. Not above the object pronoun: 25 sentences, 23 (92%) with do.
• The study builds on earlier accounts by Kroch, Han and Kroch (2000, two-step loss of movement, Penn-Helsinki corpus), Lightfoot (1993) and Macleod (2020, Hansard Corpus). Its prose summary of results partly conflicts with its own tables, so the tables should be trusted.

Interesting ideas

• I1. Cross-tabulate each example by the relative order of two small markers (negator vs. pronoun) against the competing variant (raising vs. do); a cheap positional feature can reveal what conditions a syntactic choice.
• I2. Old and new constructions coexist in one author and period, so a parser should not assume a single grammar per text; archaic main-verb inversion like "know you not" is a legitimate pattern.
• I3. Narrow a large corpus in successive filters (period, genre, construction, pronoun type) to get a small minimal-contrast sample in which only the variable of interest changes.
• I4. Report raw counts beside percentages, and check prose claims against the tables; here the narrative summary and the tables disagree, and the cells of 8 and 16 items are fragile.
• I5. The do-vs-raising choice appears to depend on the pronoun configuration, which makes it a controllable stylistic variable.

For Mova

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

• I1 (tree): Build diagnostic checks on where not (advmod) sits relative to nsubj/obj in inverted questions, and flag parses where the head verb choice conflicts with the word order.
• I2 (tokens): Make sure the tagger and lemmatizer keep a fronted lexical verb as VERB with its lemma, rather than forcing an auxiliary reading because it stands in question-initial position.
• I3 (learn): Assemble targeted evaluation slices, for example negative questions with pronoun objects, to test the parsers on rare word orders and serve as hard negative controls.
• I4 (learn): Require every evaluation report to show cell sizes and to verify textual conclusions automatically against the numbers before they enter the library.
• I5 (back): When generating archaic or Shakespearean-style questions from a meaning graph, choose between inversion and do insertion based on pronoun placement for a more period-plausible output.