SIMD Parallelism for Symbol Mapping

Variable-length character strings and other symbolic objects form the major data items in Artificial Intelligence and related applications. The rapid manipulation of such symbols is greatly facilitated if they are first mapped into a simple internal identifier space. We describe SIMD-parallel associative hardware which achieves this mapping. Our hardware unit, known as the lexical token converter, is part of a novel architecture for knowledge-based systems. This architecture uses the concept of active memory to achieve highspeed manipulation of symbolic structures. We describe this concept and, in particular, its conventions for information representation.