Paper: Introduction To The INLG'06 Special Session On Sharing Data And Comparative Evaluation

ACL ID W06-1418
Title Introduction To The INLG'06 Special Session On Sharing Data And Comparative Evaluation
Venue International Conference on Natural Language Generation
Session Main Conference
Year 2006
Authors

NLG was not inherently different from NLU at all, in fact the output representations used in the latter were just as much there by gentle(wo)man’s agreement as any common inputs to NLG would be. The strong NLG traditions of user-oriented and task-based evaluations using human evaluators would be part of the evaluation paradigm in shared-task evaluations, while parallel research might look at—but not impose—bespoke automatic methods for NLG. Money would be needed for data resource creation, but not necessarily for anything else; evidence that this was possible could be found in successful and vibrant shared-task initiatives run on a shoe-string, such as CoNLL and SENSEVAL. VAL. The community would create its own forum for reviewing, updating and adding tasks and evaluation methods. NL...