In this paper we suggest an approach to combine principles and techniques from traditional and connectionist paradigms in order to overcome the specific problems of traditional and connectionist parsers and to simultaneously preserve their advantages. The described approach - which is deeply hybrid - is based on the idea of a dynamic generation of a neural network according to the rules of a context-free grammar and driven by the input. The neural network represents the derived parse-tree or chart of the input structure, and is involved actively in the parsing process by activation flow. The parsing is truly incremental, hypothesizes expected structures, and gives estimates for the 'probailities' of the expected structures. The hybrid parsing method has been used successfully in three different natural language processing systems (PAPADEUS, INKAS, and INKOPA).
CITATION STYLE
Kemke, C. (1996). A hybrid approach to natural language parsing. In Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics) (Vol. 1112 LNCS, pp. 875–880). Springer Verlag. https://doi.org/10.1007/3-540-61510-5_147
Mendeley helps you to discover research relevant for your work.