Discovering Semantic Classes for Urdu N-V Complex Predicates

This paper reports on an exploratory investigation as to whether classes of Urdu N-V complex predicates can be identified on the basis syntactic patterns and lexical choices associated with the N-V complex predicates. Working with data from a POS annotated corpus, we show that choices with respect to the number of arguments, case marking on subjects and which light verbs are felicitous with which nouns depend heavily on the semantics of the noun in the N-V complex predicate. This initial work represents an important step towards identifying semantic criteria relevant for complex predicate formation. Identifying the semantic criteria and being able to systematically code them in turn represents a first step towards building up a lexical resource for nouns as part of developing natural language processing tools for the underresourced South Asian language Urdu.