Consider the following sentence:
1) Mary killed the guy.
This sentence comprises of the subject Mary, the predicate kill and the direct object the guy. The subject is realized as an NP, the predicate is realized as a verb and the object is realized as an NP as well. Note what happens if we leave out either the subject or the object:
2) * killed the guy 3) * Mary killed
In both cases we get an ungrammatical sentence. Why is this? Let's take a look at the meaning of kill. We know that killing is an activity with two participants: the one who kills and the one who is killed. Admittedly, sometimes the killer and the killee are one and the same person, but, linguistically, they are usually spelled out by means of different expressions as in John killed himself as opposed to *John killed John (where both Johns refer to the same person). It is thus not unreasonable to state that, semantically, the predicate has a designated number of logical elements or participants that it requires. We will call these minimally required elements arguments.
Predicates take different numbers of arguments; in natural languages it is usually assumed that the number of arguments a predicate can take is at most three or four (note that this number refers to the syntactic categories that are actually realized; it does not refer to the participants involved in an activity denoted by the verb). Consider the next sentences where each Predicate is written in boldface and each argument is given a number:
4) [Henry]1 snored 5) [The girl]1 kissed [the zebra]2 6) [John]1 sent [a parcel]2 [to his family]3 7) [John]1 bet [Brian]2 [five bugs]3 [that he will lose the match]4
The predicate snore in 4) needs at least and at most one argument. For this reason, it can be called a one-place predicate (or monadic predicate). Kiss in 5) wants two arguments, which is why it can be called a two-place or dyadic predicate. 6) and 7) follow the same pattern. The number of arguments a predicate takes is referred to as a predicate's valence or its arity (in some textbooks, you will find the terms un-ary predicate, bi-n-ary predicate instead of one-place etc., but those are just terminological differences). Within GB, the valence of a predicate is part of its lexical information, and it is usually called Argument Structure.
As noted before in brackets, it is necessary to make a distinction between arguments that are realized by syntactic categories and such that are not. Take, for example, the verb meet. We know that the verb denotes an activity that involves at least two participants, say John and Mary:
8) John meets Mary.
However, saying that meet requires two NPs won't get us very far since 9) is possible as well:
9) The kids met.
Here, we just have one NP. Meet licences this argument since the NP is plural and thus there are semantically at least two participants involved in the action denoted by the verb. So it is important to keep in mind that argument structure, semantically, tells us something about the number of arguments a verb wants. This requirement can correlate with the syntactically realized number of caterories, but it need not.
Note that the arguments of a predicate are not necessarily NPs: in 6) the indirect object is required by the verb send, and it is realized as a PP. And the last argument the predicate bet takes in 7) is a finite sentence. It thus makes sense to distinguish three different levels of description of a sentence:
|Functional Level||Subject||Predicator||Direct Object|
|Form Level||[S[NP N]||[VP V||[NP Det N]]]|