ABSTRACT
An elementary distinction that can be made in the artificial intelligence literature is between "fact-retrieval" systems and "question-answering" systems. In fact-retrieval systems, the user wishes to store vast quantities of data (e.g., indexing key words of abstracts of papers on organic chemistry) and then retrieve relevant parts of the information by sending a few key words into the system. In such systems, there is almost no inference to be done, no deduction, no subtle semantic interpretation of the question itself. The user is usually restricted to a small vocabulary of specified key words, and the syntax of his retrieval requests is usually restricted to only a few standard syntactic frames, with no possibility of complex embeddings or the like. The issue in such systems is efficiency of a given coding (keying) and organization of the information files, so that the user obtains rapid and maximum return of documents or facts relevant to his request, with a minimum of irrelevant information that must be scanned for significance. These are "fact-retrieval" or "document-retrieval" systems.
