This fourth annual report summarizes the investigation of (1) a "sentence dictionary" and (2) a "word government dictionary" for use in automatic abstracting and extracting systems. The theory behind the sentence dictionary and its compilation is that a separation of significant from nonsignificant sentences can be accomplished on the basis of syntactic sentence types. In the initial experiments dealing with the implementation of the sentence dictionary, parts-of-speech were used to designate the syntactic classification of sentences. As this classification proved too discriminative, a new system for syntactic classification was generated in which the part-of-speech strings were replaced by a phrase structure representation of the sentences, which were assigned automatically by a parsing program. The degree of complexity needed for the phrase structure is currently being determined by experiment. The compilation of the word government dictionary, also discussed in this report, was done on the assumption that the set of government words is a small subset of all English words and that this word government information would be useful in computational linguistics and information retrieval. Such investigation has led to the conclusion that a word's government pattern is often linked to its semantic meaning. (RM)