Year: 2007
Type: conference-paper
Abstract: We propose dynamically selecting n for nbest outputs returned from a dialog system module.We define a selection criterion based on maximum drop among probabilities, and demonstrate its theoretical properties.Applying this method to a dialog-act detection module, we show consistent higher performance of this method relative to all other n-best methods with fixed n.The performance metric we use is based on ROC area.
Authors: Junling Hu, Fabrizio Morbini, Fuliang Weng, Xue Liu
Institutions: Robert Bosch (United States), McGill University
Language: English
Topic: Speech and dialogue systems
Subfield: Artificial Intelligence
Field: Computer Science
Domain: Physical Sciences
Open Access status: hybrid
OpenAlex ID: W53604603