A Simple and Generic Belief Tracking Mechanism for the Dialog State Tracking Challenge: On the believability of observed information

摘要: This paper presents a generic dialogue state tracker that maintains beliefs over user goals based on few simple domainindependent rules, using basic probability operations. The rules apply to observed system actions and partially observable acts, without any knowledge obtained from external resources (i.e. requiring training data). core insight is maximise the amount of information directly gainable an errorprone itself, so as better lowerbound one’s expectations performance more advanced statistical techniques for task. proposed method evaluated in Dialog State Tracking Challenge, where it achieves comparable hypothesis accuracy machine learning systems. Consequently, with respect different scenarios belief tracking problem, potential superiority weakness approaches general are investigated.