📄 Distributional Properties and Endocentricity of English Gerunds

Summary

The paper addresses a long-standing puzzle about English gerund phrases. Outside the phrase they pattern like nouns, but inside they often behave like verbs. This makes it hard to respect endocentricity, the principle that a phrase takes its category from its head. Kim argues that there are two distinct gerund types, sorted by their internal dependents:

• Nominal gerunds take adjectives, determiners and of-PPs. They are treated as ordinary derived nouns produced by a lexical rule.
• Verbal gerunds take adverbs, direct NP objects, not and auxiliary have. They remain verbs.

The work is a theoretical HPSG analysis with no experiments. Its evidence is constructed examples plus a few attested sentences from the Cobuild corpus. Those sentences show that prepositions can take verbal complements, such as if-clauses and wh-clauses, so a verbal gerund after a preposition is not anomalous. Accusative-subject gerunds are analysed as object control by a preposition or verb, parallel to want him to apply. Possessive subjects are treated as specifiers. The payoff is an account that keeps endocentricity without inventing a hybrid category. It also links gerunds to to-infinitives.

Key points

• The paper uses a two-way taxonomy with clear diagnostics:
  • Nominal gerunds take an adjective, a determiner (including no) and an of-PP, and allow only a genitive subject.
  • Verbal gerunds take an adverb, an NP object, not and have, and allow a genitive or accusative subject.
• A key minimal pair: Brown's deft painting of his wife and Brown's deftly painting his wife are fine, while the crossed variants are ungrammatical.
• Critique of Pullum (1991, GPSG): the analysis forces VFORM onto the head of an NP. It covers only one gerund subtype and leaves control via AGR unexplained.
• Critique of Malouf (1996, HPSG): it creates a new mixed category gerund. That category overgenerates, wrongly allowing Brown's deft painting his wife.
• Proposed lexical entries:
  • A preposition or object-control verb can select an NP plus a gerundive VP, as in proud of his daughter winning the prize.
  • A lexical rule licenses verbal gerunds with an optional possessive specifier.

Interesting ideas

• I1. Decide whether an -ing form is a noun or a verb from its dependents (adjective/determiner/of-PP versus adverb/NP object/not/have), not from the form alone.
• I2. Let one surface form carry two separate lexical entries, a derived noun and a verb, instead of inventing a hybrid category, as many zero-derived pairs like praise already do.
• I3. Treat X winning after a preposition or verb as control, so the accusative NP is the understood subject of the gerund, mirroring want him to apply.
• I4. A nominal gerund inherits the argument structure of its root verb, with the of-PP realising the verb's object.
• I5. The paper's starred minimal pairs work as natural negative controls, sentences a correct grammar must reject or analyse differently.

For Mova

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

• I1 (tokens): the tagger could add a post-check that reassigns -ing forms to NOUN or VERB (VerbForm=Ger) based on the dependents it sees.
• I2 (tree): keep gerunds strictly NOUN or VERB in treebank conversion, so downstream rules never meet a mixed category.
• I3 (frames): when a gerund follows proud of / remember + NP, fill the gerund's agent role from that NP rather than from the main-clause subject.
• I4 (frames): map painting of his wife onto the VerbNet frame of paint, with the of-object as theme, so nominalised events still produce who-did-what-to-whom frames.
• I5 (learn): build a small suite of gerund minimal pairs to test whether the tagger and parser are sensitive to adjective versus adverb and of-PP versus NP object.