Geoff Pullum has a nice discussion of here of various issues concerning 'that'-clauses (forgive me Geoff!) and their ilk. [Personally, I prefer to use the 'that'-clause terminology (parochial though it be) because it is simple--no descriptivists here.] Unfortunately, he dances around the hard issue about how syntax and semantics interact. Consider his example:
There's an arrogance in the scientific community that they know better than the average American.According to Geoff, the 'that'-clause here specifies the content of some attitude invoked in the proposition expressed by this sentence. That seems basically right. At the same time, he maintains that the 'that'-clause is not a complement of (i.e., syntactic argument of) "arrogance", or anything else. That too seems right.
But there is pretty clearly a tension in these two views. In saying that the 'that'-clause functions semantically to specify the content of some attitude, it looks like Geoff is committed to the view that the semantic value of the 'that'-clause is the (semantic) argument of some property or relation. This naturally suggests the view that although "arrogance" is unary, its semantic value (being arrogant) is binary, i.e., a two-place relation. But that seems quite wrong; arrogance is a property not a relation. What relation then is the semantic value of the 'that'-clause a relation of, if any?
It seems to me we have two choices. (1) We can try to articulate a view according to which the fact that the semantic value of the 'that'-clause specifies the content of the community's arrogance, it isn't a semantic argument to any relation. (2) We can try to articulate a view according to which there is some unarticulated relation around which is involved in the semantic interpretation of the sentence.
In my view, (2) is preferable. In particular, I find the original sentence to be synonymous with the following:
There's an arrogance in the scientific community's belief that they know better than the average American.If so, we are back at an earlier debate about whether or not we should account for this additional semantic richness by positing an richer underlying LF or simply a more complex semantic mapping from surface structure to meaning.