In this Master's thesis, the nature and domain of verb subcategorization is explored. This exploration results in an implementation of the theory developed. This implementation serves as an example of the way subcategorization should be handled in an NLP system.

In chapter 1, it is concluded that each main verb subcategorizes for a (maximal) number of arguments, including adjunct-arguments. It is argued that arguments are an appropriate abstraction to use for subcategorized elements. The arguments have to be specified in terms of syntactic categories (category selection). The category selection of those arguments is determined by the main verb itself.
Auxiliaries and copula are treated differently.

In the domain of subcategorization, there are properties and structures which have to be clarified, such as: the position and word order of complements, passive voice, ergativity, reflexivity, optionality, selection of complementizer, provisional phrases, sentential subject. These properties and structures are derived from other basic properties and structures, in order to avoid redundancies in the lexicon.

Next, in chapters 2 and 3, three linguistic frameworks (LFG, HPSG, GB) and two existing implementations (AMAZON/CASUS and EUROTRA) are discussed, in particular with respect to processing subcategorization.

Finally, in chapter 4, an implementation of the theory developed above, in the framework of DCG, is discussed.

It is concluded, that subcategorization frames are indispensible in NLP systems, but that they should be reduced to abstract schemata, from which more specific features can be derived automatically.