Noncompositional

At the recent BLS meeting, Susan Buescher (UNM) gave a paper on the discourse function of the class of verbs known (to me) as verbs of cognition (to her as cognitive verbs). (Get the abstract here) She looked in the Santa Barbara corpus of spoken American English at the distributional properties of a set of six or so verbs like think, believe, know, wonder, and understand for some properties that might illuminate how they are used in conversation. She looked at prosodic features (like the presence of a prosodic break after the main verb), frequent subject collocations, and some sort of syntactic independence (like appearing without a complement).

It was overall a very tight analysis of the various verbs, and demonstrated various degrees of assimilation to discourse functions, like hedges, responses, and so on. She found that, for example, believe and wonder have not been appropriated for discourse functions so as to show an emergence of deviant properties. Such a thing has been argued to have happened to the verb think, where strings like I think and I don’t think are basically epistemic markers glomed onto the actual head (at least the semantic head) of the sentence. However, as I recall (I can’t seem to find her handout), she found that the combination of frequency and syntactic/prosodic independence did not reveal that think was acting in a way other than one might expect for a fully semantic-ful verb.

I wondered during the talk (and continue to wonder) exactly what properties we can look for that would reveal a change in function of some word or phrase. Certainly frequency and prosody are good clues, and I cannot refute that if these are lacking, then the case for quasi-grammaticalization is harder to make. But something tells me that the features she looked at may have been sufficient, but not necessary, conditions for demonstrating discourse function. One interesting thing to look at is the possibilities for null complementation. If we compare think with beleive and know, there are some interesting patterns:

• Who knows/thinks/believes that Superman is really Clark Kent? –I know/*think/*believe. (but I ??know/think/??believe so)
• He’ll come through for us… I *know/think/believe. (but I know it)

Of course, you could argue that the second sentence isn’t really a case of null complementation, but some sort of inverted construction (like inverted quotatives, he explained). Of course their function is rather different–they are hedges (well, know would be a sort of reverse-hedge, but it doesn’t work anyway; I’m sure works fine, though). Come to think of it, it would have been cool if she had looked at sure and certain as well. But when you’ve only got 20 minutes, you know?

Another interesting thing: Both I know and I don’t know are good in a wide variety of situations, but only I think has some special discourse status. You can’t say He’s not gonna make it…I don’t think. Well, I don’t have much else to say about it now, but I think there are many more possibilities for highlighting points of functional change in words than are usually examined (also for anyone who was at Martin Hilpert and Suzanne Kemmer’s talk at the LSA, there may well have been syntactic differences, not just frequency-collocational differences between various versions of causative make that would have demonstrated that it was actual “grammaticalization,” as Elizabeth Traugott was asking after; I suggested that changes in null complementation might be a starting point (yes, I like to pull that one out a lot – but it’s a useful diagnostic for stuff, at least in English))