Requirements are discussed for an information retrieval language that enables users to employ natural language sentences in interaction with computer-stored files. Anticipated modes of operation of the system are outlined. These are: the search mode, the dictionary mode, the tables mode, and the statistical mode. Analysis of sample sentences parsed manually indicate that sentences can be machine analyzed by about 40 strings, many of which are combinations of others. The sentences produced about 130 distinct words excluding all index terms. It is suggested that, by properly classifying verbs with their objects, a semantic interpreter can be developed. (Author)