This SGER addresses the problem of metaphoric inference in Natural Language Understanding (NLU). Language is fundamentally a creative act and NLU systems require extensive semantic knowledge and sophisticated inference methods that are capable of dealing with figurative language. A prototypical example of the power of figurative language in everyday discourse is the ubiquitous use of metaphor to perform most linguistic functions including predication, modification, and reference.
To date, no comprehensive computational system addressing all the aspects of metaphor has been implemented or even designed. Our project seeks to remedy this situation by addressing key challenges in building a scalable, open source, model of metaphor. Building on previous modeling work and leveraging results from Cognitive Linguistics, we are exploring techniques to a) design and populate a machine readable metaphor ontology, b) analyze the metaphoric encoding of crucial discourse information, including event structure and communicative intent, and c) use machine learning algorithms for metaphor recognition from textual sources.
On a practical scale, this project provides a machine readable ontology of metaphors, allowing programmatic access to the relational structure of the data, enhancing reuse, and fostering incremental development. Coupled with the metaphor recognition tools, this takes us part way (a companion requirement is a corpus annotation effort) toward creating an open and scalable metaphor resource. From the scientific side, this project creates a better understanding of the form and content of metaphoric inference and the first computational framework to explore the automatic processing of this information from text languages and media.
