A Simile Recognition System using a Commonsense Sensory Association Method

We propose a method for comprehending the significance of metaphorical expressions by using an intuitive sensory association method implemented on computer. The metaphors we address in this work are similes. When simile expressions are used in human conversation, it appears that the listener uses intuitive sensory associations, cultivated through experience, to recall the characteristics of the subject and the predicate and comprehend the meaning of the metaphor by replacing the predicate with another appropriate word to describe the subject. A sensory association method has been proposed that is capable of clarifying these sorts of intuitive sensory relationships between nouns and their characteristics. The sensory association method outputs the sensations and impressions that humans naturally feel in response to a given noun. In this work, we construct a simile comprehension system based on the sensory association method and seek to use it to get computers to understand similes. In this paper, we define comprehension of a simile—such as “cheeks like apples”—as to convert the simile into the phrase “red cheeks.” The capacity to perform this conversion demonstrates the computer's understanding that the two expressions are synonymous. The results of our tests indicated an accuracy of 65.7%; thus, by introducing a sensory association method we were able to exceed the accuracy achieved in a previous study.