Tree Distance in Answer Retrieval and Parser Evaluation
Martin Emms · 2005
The use of syntactic tree-distance as a surrogate for semantic distance in an answer retrieval task is investigated.The feasibility of this is confirmed by showing that retrieval performance increases with parse quality, and an application of this to parser evaluation is discussed.Variant definitions of tree-distance involving parameters such as whole vs sub-tree, node weighting, wild-card trees and lexical emphasis are compared with each other and with sub-string distance.