A practical text search system for unindexed data

A hardware/software system for searching unindexed documents or records has been developed, based on finite state automata (FSA's). Disk data are streamed through a set of FSA's. The FSA control tables represent a large number of simultaneous query terms. Special circuit features are used to reduce the memory explosion to manageable proportions. Query terms may be fixed words or character groups, numeric ranges, words containing fixed- or variable-length "don't cares", or contiguous word phrases. Following term recognition, query resolution circuits apply AND, OR, NOT, and proximity relationships to determine document hits. A 340-million-character disk can be searched in less than five minutes. For large data bases, disks may be parallelled to avoid increasing search time.