Construction of a Corpus-Based Metaphor Generation Support System Built on Japanese Literature

Metaphor generation is influenced by emotional or sensuous knowledge structure. Hence, there is a possibility that it might be difficult to select an adequate vehicle in metaphor generation. The purpose of this research is to construct a metaphor database based on a literature corpus and use the database to build a metaphor generation support system. At first, sentences including metaphors were extracted from a literature corpus. The extracted sentences were analyzed using dependency parsing in order to construct a metaphor database for the metaphor generation system. The system outputs candidate vehicles from a given topic and its expressed features by searching the sentences including the topic or the features in the database. Furthermore, an experiment was conducted to evaluate the usability of the system. In the experiment, participants were asked to generate a sentence including a metaphor from a shown image with or without the system. Third-party evaluation was conducted to evaluate the metaphors generated in the experiment. The results seem to suggest the efficiency of the system.