Answering Questions by Computer

A study of the processing of questions input to a computerized question-answering system such as the RAND Relational Data File (see RM-5085). The process consists of (1) transforming the natural-language question into a symbolic question (i.e., a certain formula of predicate calculus) and (2) generating the answer by calculating the value set of the resulting formula. This study is addressed to the second step. A key problem is the identification of "reasonable" input queries. These are characterized by introducing the concept of definite formula. A particular class of definite formulas — the proper formulas — is especially suitable for machine processing. A set of machine-recognizable sufficient conditions for their identification is given, together with rules for calculating their value sets. The definite, but improper, formulas are also studied. It is shown that definite formulas without quantifiers can be transformed into proper equivalents. For definite formulas with quantifiers, limited but useful results are obtained.