Syntactic pattern matching with GraphSpider and MPL

We present MPL (Metapattern Language), a new formalism for defining patterns over dependency-parsed text, and GraphSpider, a matching engine for extracting dependency subgraphs which match against MPL patterns. Using a regexp-like syntax, MPL allows the definition of subgraphs matching user-specified patterns which can be constrained by word or word class, part-of speech tag, dependency type and direction, and presence of named variables in particular locations. Although MPL and Graph- Spider are general-purpose, we developed a set of patterns to capture biomolecular interactions which achieved very high precision results (92.6% at 31.2% recall) on the LLL Challenge corpus. MPL specifications and pattern sets, and the GraphSpider software, are available on SourceForge: http://graphspider.sf.net/