Paper: Hybrid Ways to Improve Domain Independence in an ML Dependency Parser

ACL ID D07-1120
Title Hybrid Ways to Improve Domain Independence in an ML Dependency Parser
Venue Conference on Empirical Methods in Natural Language Processing
Session Main Conference
Year 2007
Authors
  • Eckhard Bick (University of Southern Denmark, Odense Denmark)

This   paper   reports   a   hybridization   experi­ ment, where a baseline ML dependency pars­ er,   LingPars,   was   allowed   access   to   Con­ straint Grammar analyses provided by a rule­ based parser (EngGram) for the same data.  Descriptive compatibility issues and their in­ fluence on performance are discussed. The  hybrid system performed considerably better  than its ML baseline, and proved more ro­ bust than the latter in the domain adaptation  task, where it was the best­scoring system in  the open class for the chemical test data, and  the best overall system for the CHILDES test  data.