Paper: Cross-lingual Model Transfer Using Feature Representation Projection

ACL ID P14-2095
Title Cross-lingual Model Transfer Using Feature Representation Projection
Venue Annual Meeting of the Association of Computational Linguistics
Session Main Conference
Year 2014
Authors

We propose a novel approach to cross- lingual model transfer based on feature representation projection. First, a com- pact feature representation relevant for the task in question is constructed for either language independently and then the map- ping between the two representations is determined using parallel data. The tar- get instance can then be mapped into the source-side feature representation us- ing the derived mapping and handled di- rectly by the source-side model. This ap- proach displays competitive performance on model transfer for semantic role label- ing when compared to direct model trans- fer and annotation projection and suggests interesting directions for further research.