Learning distributed representations of conceptual knowledge

摘要

Summary form only given, as follows. The authors argue thatdistributed representations must satisfy five criteria in order to serveas an adequate foundation for constructing and manipulating conceptualknowledge. These criteria are: automaticity, portability, structureencoding, semantic microcontent, and convergence. In the authors'approach, distributed representations of semantic relations (i.e.propositions) are formed by recirculating the hidden layer in recurrentparallel distributed processing networks. The authors' experiments showthat the resulting distributed semantic representations (DSRs) satisfyall of the above criteria. They believe that DSRs can help supply animportant building block in developing more complex connectionistarchitectures for higher level inferencing, as is required in naturallanguage processing