An important aspect of creative thinking in science education is the set of metaphors that students use to understand and conceptualize material. Previous research has shown that metaphorical and analogical thinking are key components in creative processes and play an integral role in academic creativity. An important part of encouraging creativity through metaphor is raising a student's awareness of the conceptual metaphors they already use. However, identifying the metaphors that students use in their learning has previously been labor intensive, requiring a great deal of time and attention from a human instructor. Previous work in computational linguistics has enabled the automatic extraction of metaphor from bodies of written text. This project improves the state of the art in computational metaphor identification (CMI) through the use of large corpora, through the employment of typed dependency parsing, and through a comparison of several selectional preference learning techniques. The result will be a technological tool kit that supports human creativity by automatically identifying metaphors in bodies of text. This research makes a significant contribution to the fields of computer science and computational linguistics by improving existing metaphor extraction techniques through the integration of current research in selectional preference learning and dependency parsing. In the area of education research, the project will evaluate the impact of CMI on both students' thinking and teachers' teaching, as well as examining the perceived utility of CMI by both groups. This research will allow these metaphors to be used to anchor instruction. This project has the potential to improve creative thinking in science instruction by helping students and their teachers perceive more effectively the ways in which students are thinking about science concepts. In addition, the research may provide insight into how this type of system could enhance creativity and learning across a wide range of other disciplines. Finally, the ability to perform computational metaphor identification may have a variety of other potential applications, from automated search and information retrieval to knowledge representation and artificial intelligence.
