Complex predicates (CPs) are a highly productive predicational phenomenon in Hindi and Urdu and present a challenge for deep syntactic parsing. For CPs, a combination of a noun and light verb express a single event. The combinatorial preferences of nouns with one (or more) light verb is useful for predicting an instance of a CP. In this paper, we present a semi-automatic method to obtain noun groups based on their co-occurrences with light verbs. These noun groups represent the likelihood of a particular noun-verb combination in a large corpus. Finally, in order to encode this in an LFG grammar, we propose linking nouns with templates that describe preferable combinations with light verbs.