Relational Data File

An overview of the prototype Relational Data File, a computer databank distinguished by the storage of data in a series of separate "atoms" of one binary relational sentence each, and the ability to retrieve data not explicitly stored that can be inferred from stored data. The prototype RDF contains 70,000 sentences extracted from scientific and technical source items. Each sentence is encoded in four 36-bit computer words, with synonyms automatically translated to standard form and ambiguities flagged for manual correction. New entries in the synonymy/ambiguity dictionary update all stored data. Each accepted sentence is quadruplicated and stored in four subfiles by each of the key elements. A string dictionary decodes entries for output. A special retrieval language, INFEREX, allows retrieval through any inference scheme that can be expressed in the predicate calculus and set theory. The RDF structure seems well suited for very large databases where new types of data are continually being added and other types removed and where data are incomplete (as inrapidly changing fields), redundant, or unpredictable.