Trailfinder: A Case Study in Extracting Spatial Information Using Deep Language Processing

Abstract

The present paper reports on an end-to-end application using a deep processing grammar to extract spatial and temporal information of prepositional and adverbial expressions from running text. The extraction process is based on the full understanding of the input text. It is represented in a formalism standard for unification-based grammars and with a language-independent vocabulary as far as spatiotemporal information is concerned. The latter feature in principle allows portability of the extraction algorithm across languages and applications, as long as the domain is kept constant. The present application is called ’Trailfinder’, and supports web-queries about information concerning mountain hikes. A standard hike-description is parsed by an HPSG-based grammar augmented by Minimal Recursion Semantics (’MRS’; (Copestake 2002)). To represent domainspecific meaning concerning location and direction, we enrich MRS structures with featurebased interlingua specifications. Utilizing the ’Heart of Gold’ (HoG) technology developed as part of the Deep Thought project, and conversion algorithms employing XML sheets, these specifications are mapped to the query interface language.