Projecting semantic roles via Tai mappings
Hector-Hugo Franco-Penya, Martin Emms · 2012
This work takes the paradigm of projecting annotations within labelled data into unla-belled data, via a mapping, and applies it to Semantic Role Labelling. The projec-tions are amongst dependency trees and the mappings are the Tai-mappings that under-lie the well known tree edit-distance algo-rithm. The system was evaluated in seven different languages. A number of variants are explored relating to the amount of in-formation attended to in aligning nodes, whether the scoring is distance-based or similarity-based, and the relative ease with which nodes can be ignored. We find that all of these have statistically significant im-pacts on the outcomes, mostly in language-independent ways, but sometimes language dependently. 1