February 06, 2005

Syntactic and notional number

In reference to my posts on one of the SAT's "sentence error" questions (here and here), Maryellen MacDonald sent email with a sketch of relevant psychological research about number in nouns, verbs and pronouns. This led me to wonder about the difference between underpants and underwear. I'm not referring to Ken Keeler's theory that "the word 'underpants' is 20% funnier than the word 'underwear'" -- as far as I know, this plausible hypothesis still awaits psycholinguistic validation. Rather, Maryellen's note suggested an insight into the contrast between "these underwear" and "these committee".

Maryellen wrote:

The question of why verbs and pronouns tend to vary in the extent to which they agree with collective nouns is a topic of some interest in psycholinguistics. Probably the most relevant paper is this one:

Bock, J. K. Nicol, J. & Cutting, J. C. (1999). The ties that bind: Creating number agreement in speech. Journal of Memory and Language.40, 330-346.

The claim goes roughly as follows: collective nouns are syntactically singular but semantically (or "notionally" in this literature) plural. Subject-verb agreement is largely a syntactic process, described as percolation of number features over a phrase structure during utterance planning. There's little or no influence from the meaning (or notional plurality) of the words. Antecedent-pronoun agreement, however, is more influenced by semantics. For most nouns, there are similar patterns of agreement for both verbs and pronouns, but in the case of collectives, there's a conflict in number between syntactic number and notional number, and thus the verbs and the pronouns have different distributions of agreement patterns with collectives.

I've described this approach briefly enough that I probably haven't made it sound much better than a redescription of the data, but in fact there's a very large agreement literature in language production, and Bock and colleagues' views capture quite a lot of what turns out to be very complex patterns of greater or lesser tendencies to agree across different situations. Still, my own views lean more toward a constraint satisfaction view of subject-verb agreement, rather than the more purely syntactic view. A paper about collectives from this perspective, is

Haskell, T.R. & MacDonald, M.C. (2003). Conflicting cues and competition in subject-verb agreement. Journal of Memory and Language, 48, 760-778.

We didn't discuss verb vs. pronoun agreement in this paper, but Bock's idea about greater weight to the notional plurality in pronoun agreement could be incorporated. Of course there needs to be a story (on anyone's account) for why pronoun agreement is more semantically-influenced than is verb agreement, and I at least don't know of a satisfying answer as of yet.

Here's the abstract from Bock, Nicol and Cutting 1999:

Coherence in language relies in part on basic devices like number agreement. To assess meaning-based (notional) versus form-based (morphological) control of number agreement, we examined how speakers created number agreement for collective nouns, which can carry conflicting notional and morphological number. The agreement targets were verbs and two types of pronouns, produced in the course of a sentence-completion task. Comparisons of the verbs and pronouns indicated that verbs tended to reflect the morphological number of the collective controller, whereas pronouns were more likely to reflect the notional number. This argues that the number features of pronouns may be retrieved under control from the speaker's meaning, while the number features of verbs are more likely to be retrieved under control from the utterance's form. The implication is that the retrieval of words during language production is influenced by two distinct types of information, consistent with an inflectional account of agreement.

The experiments described in this paper also connect with the recent discussion here about the "agreement with nearest" phenomenon, because the materials used include subject phrases like

The actor(s) in the soap opera(s)
The cast in the soap opera(s)

The theory in Bock, Nicol & Cutting -- that verb agreement is basically syntactic, while pronouns are "notional" or semantic -- suggests that pluralia tantum should be considered both syntactically and "notionally" plural, since pants, scissors, clothes etc. usually take both plural verbs and plural pronouns. I would say "these pants are missing their price tag", not "*this pants is missing its price tag". But what about the contrast between underpants and underwear? These seem pretty much the same to me "notionally", but they generally take different pronouns as well as different verbs, at least according to my intuitions:

If your underpants are (*is) dirty, you should wash them (*it).
If your underwear (*are) is dirty, your should wash (*them) it.

Being an empirical sort of person, I did a few quick web searches to check my intuitions on this point. Most of the examples of "underwear are" or "underpants is" are irrelevant:

Marbles in my Underpants is definitely a much darker book than The Soap Lady, but in many ways it's almost a companion piece.
Demonic Underpants is the single most important development in the history of music since ANkST & ANkHS.
Whether Mr Morrison insists his mother wears racing driver's underpants is a matter that still requires clarification.
The silk fibres of pure silk in the silk long underwear are a natural, breathable insulator, so you won't overheat.

However, there are plenty of examples suggesting that "notional" number (and/or random fluctuations) play a role here as well:

Bravado Underwear are non-returnable and non-exchangeable, for hygienic reasons.
The Rugged Bear Girls Long Underwear are flame retardant, 100% Cotton and machine washable.
I drove home thinking about how these underwear would feel under my cargo shorts.
Sexy silk underwear are great at the gym.
This underpants is made of Polartec® Power Dry®, a patented fabric with two unique surfaces.
Yeah, well short underpants is better than no underpants at all, Chewie.
Dude. Girls' underpants is where Cooties COME FROM!
MAYOR: (Nick Maloney) Aaargh! Oh, my God! Get this underpants off me! [Burst of theme music]. ANNOUNCER: Don't miss Attack of the Killer Italian Y-Fronts.

Note that there is variation not only in verb agreement, but also in agreement with demonstratives -- this vs. these. This is an interesting difference between collective nouns and pluralia tantum. As far as I can tell, no competent speaker of English is ever seriously tempted to use these committee or these family as ad sensum plurals, while it's relatively easy to create contexts in which these underwear or this underpants are natural outcomes.

The complexity of interactions between form and context in this area (even leaving out undergarments) motivate the framework proposed in Haskell and MacDonald 2003, whose abstract explains:

Traditional theories of agreement production assume that verb agreement is an essentially syntactic process.
However, recent work shows that agreement is subject to a variety of influences both syntactic and non-syntactic, which raises the question of how these different sources of information are integrated during agreement production. We propose an account of agreement production in which several information sources contribute activation to singular and plural verb forms. Conflict between cues leads to competition which can in turn magnify the influence of subtle cues. Three fragment completion experiments tested key predictions of this constraint satisfaction approach. Experiment 1 demonstrated competition effects on verb choice and sentence initiation latencies. Experiments 2 and 3 demonstrated that conflicts between semantic and grammatical cues allow morphological regularity to exert a small but detectable effect on agreement. These results suggest that the constraint-satisfaction framework may provide a productive approach for understanding agreement production.

Haskell and MacDonald also have some interesting new things to say about the "agreement with nearest" issues:

One important issue in the study of agreement production is how local nouns come to influence agreement. In the traditional approach to agreement, attraction errors occur when features originating on the local noun are mistakenly allowed to percolate to the top of the subject noun phrase (Vigliocco & Nicol, 1998). In the constraint satisfaction approach, however, factors should influence agreement to the extent they exhibit a reliable correlation with verb marking. For correct agreement, local noun number is largely independent of verb marking, which would seem to predict that local noun number should play no role. However, there is an interesting subset of cases for which local noun number, rather than head noun number, appears to govern agreement: compare a bunch of marbles were rolling around the floor to a bunch of sand was blowing around. In such expressions, the head noun (which is often a collective) acts much like a quantifier (Michaux, 1992).

Thus, this particular construction (a <singularnoun> of <pluralnoun>, hereafter the 'a number of' construction) is correlated with the use of a plural verb. In the constraint satisfaction framework, the more similar a given construction is to this one, the more strongly the plural verb form will be activated. For a highly dissimilar phrase such as my winter jacket, the plural verb form will not be activated at all, and plural verbs should be produced very rarely. In contrast, a more similar construction such as the key to the cabinets will result in the plural form being slightly activated, and plural verb forms should occasionally be produced—which is in fact the case (Bock & Miller, 1991). Greater similarity to the 'a number of' construction should make the production of plural verbs even more likely. Note that this proposal essentially amounts to the claim that distributional information, which has been shown to play a prominent role in language comprehension, also has an impact in production as well. This possibility was considered in more detail by Thornton, Haskell, & MacDonald (2001).

Read the whole thing!


Posted by Mark Liberman at February 6, 2005 06:59 AM