Argument structure and subcategorization
We noted two fundamental ways in which predicates may differ from each other: (i) different predicates may require different numbers of arguments; and (ii) predicates which require the same number of arguments may assign different semantic roles to those arguments. As we will see, these differences are crucial in determining the structure of the clauses in which each predicate occurs.
The ARGUMENT STRUCTURE of a predicate is a representation of the number and type of arguments it requires. The following examples show simple argument structure representations for the verbs sing, slap, love, and give:
(4) sing <agent>
slap <agent, patient>
love <experiencer, stimulus>
give <agent, theme, recipient>
In any particular sentence in which these verbs are used, each of the arguments will be associated with a specific Grammatical Relation. This is illustrated in (5), where both the semantic role and the Grammatical Relation are marked for each argument.
We can use an enriched argument structure representation to show the alignment of semantic roles with Grammatical Relations for each verb, as illustrated in (6).
Information about the set of Grammatical Relations which a particular verb assigns to its arguments is often referred to as SUBCATEGORIZATION, because it provides a way of dividing a single syntactic category (namely Verb) in to several sub-categories (those that do not take an object, those that require an object plus an oblique argument, etc.). For example, the diagram in (6d) indicates that the verb give “subcategorizes” for a subject, a direct object, and an oblique argument. A verb’s subcategorization is an important part of the information which must be found in its lexical entry. We will assume that this information is represented in roughly the form shown in (6).
We should point out that the representations in (6) are somewhat redundant, in that they contain information that is (usually) predictable. For example, a normal active transitive verb will almost always assign the SUBJ relation to its agent and the OBJ relation to its patient. Many linguists assume that a lexical entry should contain as little redundant or predictable information as possible. Now if the Grammatical Relation of each argument were fully predictable, so that it could be determined by applying a set of rules (often called LINKING rules), then we would not need to list the Grammatical Relations in the verb’s lexical entry. However, getting this approach to work properly turns out to be a fairly difficult task, and introduces more complexity than we can deal with here. For this reason, we will simply assume that all of the information contained in the representations in (6) is specified in the lexical entry for each verb.
