Search

The decomposition of long complex English sentences into shorter kernel-like constituent sentences (microsentences) has often been suggested as an avenue toward conducting automatic retrieval of natural language messages. To explore the prospects of such c step, the authors attempted in 1963 to prepare a general program for deriving microsentences from longer sentences that had been syntactically analyzed by the Harvard Multipath Analysis Program. The basic idea was to extract the subject, verb and object (if any) of each clause and to reassemble these materials into a grammatical microsentence. A program is described in this paper, which was designed to operate on the tree structure output of the analyzer, and the microsentences that were produced are exhibited. The authors conclude that while microsentences of the quality achieved do not open up immediate prospects for improving the performance of automatic message retrieval systems, they may have practical value in man-machine systems using human monitors to select the preferred syntactic interpretation of a sentence. [ABSTRACT FROM AUTHOR]