tailieunhanh - Báo cáo khoa học: "Demonstration of a POMDP Voice Dialer"

This is a demonstration of a voice dialer, implemented as a partially observable Markov decision process (POMDP). A realtime graphical display shows the POMDP’s probability distribution over different possible dialog states, and shows how system output is generated and selected. The system demonstrated here includes several recent advances, including an action selection mechanism which unifies a hand-crafted controller and reinforcement learning. The voice dialer itself is in use today in AT&T Labs and receives daily calls. . | Demonstration of a POMDP Voice Dialer Jason Williams AT T Labs - Research Shannon Laboratory 180 Park Ave. Florham Park NJ 07932 UsA j dw@ Abstract This is a demonstration of a voice dialer implemented as a partially observable Markov decision process POMDP . A realtime graphical display shows the POMDP s probability distribution over different possible dialog states and shows how system output is generated and selected. The system demonstrated here includes several recent advances including an action selection mechanism which unifies a hand-crafted controller and reinforcement learning. The voice dialer itself is in use today in AT T Labs and receives daily calls. 1 Introduction Partially observable Markov decision processes POMDPs provide a principled formalism for planning under uncertainty and past work has argued that POMDPs are an attractive framework for building spoken dialog systems Williams and Young 2007a . POMDPs differ from conventional dialog systems in two respects. First rather than maintaining a single hypotheses for the dialog state POMDPs maintain a probability distribution called a belief state over many possible dialog states. A distribution over a multiple dialog state hypotheses adds inherent robustness because even if an error is introduced into one dialog hypothesis it can later be discarded in favor of other uncontaminated dialog hypotheses. Second POMDPs choose actions using an optimization process in which a developer specifies high-level goals and the optimization works out the detailed dialog plan. Because of these innovations POMDP-based dialog systems have in research settings shown more resilience to speech recognition errors yielding shorter dialogs with higher task completion rates Williams and Young 2007a Williams and Young 2007b . Because POMDPs differ significantly from conventional techniques their operation can be difficult to conceptualize. This demonstration provides an accessible illustration of the .