A finite-state approach to linguistic constraints in Zulu morphological analysis

Conceptually, finite-state approaches are well suited to the modelling and implementation of regular linguistic (in this case morphophonological) behaviour. However, it is common knowledge that, while morphophonological phenomena are mainly regular, the development of large coverage morphological analysers for real natural languages also requires the accurate modelling of idiosyncratic behaviour.

In this paper it is firstly shown how linguistic constraints in an agglutinative language such as Zulu, can be handled in a Xerox style finite-state computational morphological analysis by means of so-called flag diacritics. Secondly, the use of multiple levels of finite-state transducers for the analysis/generation of certain forms of idiosyncratic behaviour is discussed within the context of the development of a large coverage morphological analyser for Zulu that is both efficient and accurate.