A SIMD-PARALLEL MODEL OF HIERARCHICAL, RULE-BASED PATTERN-RECOGNITION, WITH APPLICATION TO THE OPTICAL INTERPRETATION OF NATURAL-LANGUAGE

This paper presents a parallel model of parsing and semantic operations suitable for natural language (NL) understanding that is designed for implementation on synchronous SIMD architectures and optical processors. We transform the adjacency matrix of the graph of a chart parse, derived from a bottom-up chart parser, into an image that can be operated upon, in parallel, by low-level image processing functions available on SIMD architectures. We show how such transformations can be applied to semantic operations to yield SIMD-parallel graph inference. Theory is expressed in terms of image algebra (I4), a rigorous, concise, inherently parallel notation that unifies the design, analysis, and implementation of image processing algorithms. IA has been implemented on numerous parallel architectures, and IA preprocessors, interpreters, and class libraries are available for the C-++, FORTRAN, and Ada languages. We discuss practical issues of implementing our rule-based system on electro-optical architectures, which exhibit high parallelism and throughput.