tailieunhanh - Báo cáo khoa học: "Recognizing Authority in Dialogue with an Integer Linear Programming Constrained Model"

We present a novel computational formulation of speaker authority in discourse. This notion, which focuses on how speakers position themselves relative to each other in discourse, is first developed into a reliable coding scheme ( agreement between human annotators). We also provide a computational model for automatically annotating text using this coding scheme, using supervised learning enhanced by constraints implemented with Integer Linear Programming. | Recognizing Authority in Dialogue with an Integer Linear Programming Constrained Model Elijah Mayfield Language Technologies Institute Carnegie Mellon University Pittsburgh PA 15213 elijah@ Carolyn Penstein Rose Language Technologies Institute Carnegie Mellon University Pittsburgh PA 15213 cprose@ Abstract We present a novel computational formulation of speaker authority in discourse. This notion which focuses on how speakers position themselves relative to each other in discourse is first developed into a reliable coding scheme agreement between human annotators . We also provide a computational model for automatically annotating text using this coding scheme using supervised learning enhanced by constraints implemented with Integer Linear Programming. We show that this constrained model s analyses of speaker authority correlates very strongly with expert human judgments r2 coefficient of . 1 Introduction In this work we seek to formalize the ways speakers position themselves in discourse. We do this in a way that maintains a notion of discourse structure and which can be aggregated to evaluate a speaker s overall stance in a dialogue. We define the body of work in positioning to include any attempt to formalize the processes by which speakers attempt to influence or give evidence of their relations to each other. Constructs such as Initiative and Control Whittaker and Stenton 1988 which attempt to operationalize the authority over a discourse s structure fall under the umbrella of positioning. As we construe positioning it also includes work on detecting certainty and confusion in speech Liscombe et al. 2005 which models a speaker s understanding of the information in their statements. Work in dialogue act tagging is also relevant as it seeks to describe the ac-1018 tions and moves with which speakers display these types of positioning Stolcke et al. 2000 . To complement these bodies of work we choose to focus on the question of how .