ABSTRACT
The model we are proposing is a spreading activation model combined with a mechanism for carrying out symbolic computations on representations that have become available through activation. Symbolic computations are required for the modeling of the parsing process, which we conceive of as consisting of three different but related stages. The first stage, segmentation, concerns the mapping of the speech input onto form-based access representations of full as well as bound forms (affixes, bound stems). Prosodic information, resyllabification, stress shifts, tone sandhi, and other phonological mutations may complicate this mapping operation. The second stage, licensing, involves checking whether representations that have become co-active can be integrated on the basis of their subcategorization properties. The third stage, combination, deals with the computation of the lexical representation of the complex word from the lexical (syntactic and semantic) representations of its constituents, given that this integration has been licensed.
