Multilayered extended semantic networks as a language for meaning representation in NLP systems

Abstract

Multilayered Extended Semantic Networks (abbreviated: MultiNet) are one of the few knowledge representation paradigms along the line of Semantic Networks (abbreviated: SN) with a comprehensive, systematic, and publicly available documentation. In contrast to logically oriented meaning representation systems with their extensional interpretation, MultiNet is based on a use-theoretic operational semantics. MultiNet is distinguished from the afore-mentioned systems by fulfilling the criteria of homogeneity and cognitive adequacy. The paper describes the main features of MultiNet and the standard repertoire of representational means provided by this system. Besides of the structural information, which is manifested in the relational and functional connections between nodes of the semantic network, the conceptual representatives of MultiNet are characterized by embedding the nodes of the network into a multidimensional space of layer attributes. To warrant cognitive adequacy and universality of the knowledge representation system, every node of the SN uniquely represents a concept, while the relations between them have to be expressed by a predefined set of about 110 semantic primitive relations and functions. The knowledge representation language MultiNet has been used as an interface in several natural language processing systems. It is also suitable as an interlingua for machine translation systems.