On Gahl and Garnsey on Grammar and Usage

1. Gguu and kgku

My paper ‘Grammar is grammar and usage is usage’ (Newmeyer 2003; henceforth GGUU) defends the classical ‘Saussurean’ position that grammatical knowledge forms a distinct cognitive system. This system is called upon in language use, and in turn is in part shaped by use, but generalizations about usage lie outside of grammar per se. In the course of GGUU, I provided arguments against stochastic approaches to grammar, that is, approaches in which grammatical constructs (rules, structures, or whatever) are linked to the probability that that construct might be called upon in actual speech. One particularly compelling argument for my position was put forward originally in Lavandera 1978. Lavandera pointed out that syntactic variants often (indeed, typically) differ in meaning. Hence the probabilities assigned are in principle just as likely to be a function of the meaning that one wants to convey than of some inherent property of the syntactic construct itself.

‘Knowledge of grammar, knowledge of usage: Syntactic probabilities affect pronunciation variation’ (Gahl & Garnsey 2004; henceforth KGKU), is largely devoted to attempting to undermine the Lavandera-GGUU position by providing examples of pronunciation variation in English, including the frequency of /t,d/-deletion, that seem to be sensitive to probabilities attached to syntactic structures, where those probabilities are not a simple function of the meaning conveyed by the structures. KGKU concludes that such a result is ‘consistent with the notion that knowledge of grammar includes knowledge of probabilities of syntactic structures, and that this knowledge affects language production’ (p. 748).

I argue below that KGKU is not successful in dissociating probabilities from meaning (broadly defined). However, it raises a number of theoretically important issues and, in fact, does point the way toward how stochastic grammar might, in principle, find support.

2. Verb bias and plausibility

Two constructs are central to KGKU: verb bias and plausibility. The former is the probability of a particular syntactic structure, given a particular verb. Consider, for example, the verbs confirm and believe (KGKU, pp. 755–56). As 1a,b and 2a,b illustrate, both can occur with simple direct objects (DO) and with sentential complements (SC).

(1)

a. The CIA director confirmed the rumor once it had spread widely.

b. The CIA director confirmed the rumor should have been stopped sooner.

(2)

a. The job applicant believed the interviewer when she discussed things with her.

b. The job applicant believed the interviewer had been dishonest with her.

Confirm, since it is used more frequently with a DO complement than with an SC complement, is called a DO-bias verb, whereas believe, given that it is used more frequently with an SC complement than with a DO complement, is an SC-bias verb.

The (semantic) plausibility of a sentence is derived from the judgments by subjects in experiments on how likely an event might be to happen in real life. KGKU assumes, following Trueswell 1996 and Garnsey et al. 1997, that verb bias and plausibility are independent of each other. For example, the verb regret has neither a strong DO-bias nor a strong SC-bias, yet the sentence The senior senator regretted the decision is judged much more plausible than the sentence The senior senator regretted the reporter (Garnsey et al. 1997:90).

In a nutshell, KGKU demonstrates that for sentences of equal plausibility, there are more probability-related differences in pronunciation variation observed when sentence type and verb bias match (i.e. DO sentences with DO-bias verbs, and SC sentences with SC-bias verbs) than in conditions where they do not match. To be more specific:

Our central result was that the contextual probability of a syntactic pattern, given a particular verb, affected the pronunciation of words in those structures. The distribution of /t,d/ -deletion and the duration measurements bore out our specific predictions: (i) verbs were more likely to undergo /t,d/-deletion in high-probability syntactic contexts than in low-probability contexts; (ii) clause boundaries that had a low probability, given a verb’s bias, affected verb and pause durations to a greater extent than did clause boundaries that had a...