A coincidence detection network for spatio-temporal coding: application to nominal composition

摘要

This paper describes an architecture for semantic interpretation and classification of nominal phrases. The network, based on the coincidence detection of incoming events, shows an ability to extract relevant linguistic features. Through a combinatorial study, we justify that the topology is rich enough to extract the desired feature associations. It is designed to handle and to combine several linguistic problems. Two particular applications are reported which account for a computational approach of compositionality with implicit knowledge representation: 1. The interpretation of locative noun phrases 'cooking+Prsp+NP' is inferred from a set of associations of prepositions and noun phrases. 2. The semantics of 'Noun+a+Noun' compounds is linked to the acceptabilities of linguistic tests.