1975 年 2 巻 2 号 p. 33-44
A computer program called POET generated “poems” through a system which incorporated (1) transitional dependencies of words, (2) transformational structures of language, and (3) semantic, or imagery, aspects in the decision about a general theme of a poem. It was assumed that each sentence in a “poem” contains a single “thematic word” as the main subject. Thematic words were selected on the basis of the similarities of words to the major thematic word (which was to be given by the experimenter) in an image space defined according to Osgood's Semantic Differential Space. The other words in the individual sentences were assumed to be generated on the basis of both transition probabilities of word associations under the given theme and choice probabilities of transformational rules in language habit.
Some of the “poems” generated by the system were introduced and discussed in relations to psychological processes of sentence generation.