PAEK, T. AND HORVITZ, E. 2000. Conversation as action under uncertainty. Proceedings of UAI-2000, 455-464.  Home/Search   Document Not in Database   Summary   Related Articles   Check

....Networks Several systems have used Bayes Nets to integrate uncertain inference into plan recognition. In Albrecht, Zukerman and Nicholson [AZN98] dynamic belief networks were trained to predict a user s goal based on observed actions in a multi user dungeon video game. Horvitz and Paek [HP99, PH00, HP00, PHR00] use dynamic Bayesian Networks to recognize user intentions in several dialogue domains. Charniak and Goldman [Gol90, CG91, CG93] built an entire natural language understanding system, including plan recognition, in a unified dynamic belief network. Plan hypotheses were generated by ....

Tim Paek and Eric Horvitz. Conversation as action under uncertainty. In Proceedings of the 16th Conference on Uncertainty in Artificial Intelligence (UAI-2000), Stanford, CA, June 2000.

.... other facial cues relevant for interaction [22] 24] Although our current architecture does not include representations of social cues, Clark s work highlights the need for eventually incorporating such cues into a model of language grounding (for a computational approach along these lines, see [25]) To understand language, it is often necessary for the listener to assume the speaker s point of view. Points of view can involve seeing something from a different physical vantage point, as well as less literal perspective shifts in order to see things from different social or cultural ....

T. Paek and E. Horvitz, "Conversation as action under uncertainty," in Proceedings of the 16th Conference on Uncertainty in Artificial Intelligence, 2000.

....existence of a plan corpus. If a plan corpus exists or can be created for the new domain (see section below) all we have to do is use it to train models for the new domain. On the other hand, most goal recognizers (e.g. Vilain, 1990; Carberry, 1990a; Kautz, 1991; Charniak and Goldman, 1993; Paek and Horvitz, 2000] require a complete, hand crafted plan library in order to perform recognition, which can require a significant amount of knowledge engineering for each domain. Granted, these systems are performing plan recognition and not just goal recognition, which makes the comparison unfair. However, ....

....goals, even if one is more likely than the other. There are several lines of research which incorporate probabilistic reasoning into plan and goal recognition. Carberry, 1990a] and [Bauer, 1994] use Dempster Shafer theory and [Charniak and Goldman, 1993] Pynadath and Wellman, 1995] and [Paek and Horvitz, 2000] use Belief Networks to represent the likelihood of possible plans and goals to be attributed to the user. All of these methods, however, require a complete plan library as well as the assignment of probability distributions over the library. Appelt and Pollack, 1991] and [Goldman et al. 1999] ....

Tim Paek and Eric Horvitz. Conversation as action under uncertainty. In Proceedings of the 16th Conference on Uncertainty in Artificial Intelligence (UAI-2000), Stanford, CA, June 2000.

....Networks Several systems have used Bayes Nets to integrate uncertain inference into plan recognition. In Albrecht, Zukerman and Nicholson [AZN98] dynamic belief networks were trained to predict a user s goal based on observed actions in a multi user dungeon video game. Horvitz and Paek [HP99, PH00, HP00, PHR00] use dynamic Bayesian Networks to recognize user intentions in several dialogue domains. Charniak and Goldman [Gol90, CG91, CG93] built an entire natural language understanding system, including plan recognition, in a uni ed dynamic belief network. Plan hypotheses were generated by ....

Tim Paek and Eric Horvitz. Conversation as action under uncertainty. In Proceedings of the 16th Conference on Uncertainty in Arti cial Intelligence (UAI-2000), Stanford, CA, June 2000.

....part of the system s overall processing. Still, the overall impact of many such fine grained decisions can be important. A number of publications are available that include descriptions of techniques such as the ones discussed here in the context of descriptions of entire systems (see, e.g. [ 13]; 11] 16] In the present article, to focus attention on questions concerning the methods themselves, we start by describing a simple example system, a type of decision that this system has to make regularly, and an experiment in which we gathered data relevant to this type of decision. ....

E. Horvitz and T. Paek. Conversation as action under uncertainty. In C. Boutilier and M. Goldszmidt, editors, Uncertainty in Artificial Intelligence: Proceedings of the Sixteenth Conference. Morgan Kaufmann, San Francisco, 2000.

....where veri cations are necessary but not executed, and over veri cation represents the opposite situations. The last major question concerning dialog strategies is how to fuse multiple information sources to make an optimal decision. An obvious approach, Bayesian networks, was reported in [12, 13, 14]. The architecture Quartet consists of two main modules: the maintenance module for handling uncertainties of channel and signals levels, and the intention module for handling the uncertainties of the user intentions. The conversation control thus maintains both the intention status (the simulated ....

T. Paek and E. Horvitz. Conversation as Action Under Uncertainty. In Proceedings of the 16th Conference on Uncertainty in Articial Intelligence (UAI-2000), Stanford, CA, June 2000.

....the potential for such interactions depends upon the embodiment of the agent and the types of possible communicative action. Adaptation An MP SR module might need to adapt to changing conditions. An agent s ability to adapt its global communication strategy is a factor in the agent s usefulness [21], but adaptation with respect to the characteristics of the surface realizations that the agent s MP SR module derives are also important [6] Certain surface realizations are more beneficial than others (where benefit can be expressed by measures such as likelihood of being interpreted correctly, ....

T. Paek and E. Horvitz. Conversation as action under uncertainty. In Proceedings of the 16th Conference on Uncertainty in Artificial Intelligence (UAI-2000), pages 455--464, Stanford, CA, June 2000.

....part of the system s overall processing. Still, the overall impact of many such fine grained decisions can be important. A number of publications are available that include descriptions of techniques such as the ones discussed here in the context of descriptions of entire systems (see, e.g. [13]; 11] 16] In the present article, to focus attention on questions concerning the methods themselves, we start by describing a simple example system, a type of decision that this system has to make regularly, and an experiment in which we gathered data relevant to this type of decision. ....

E. Horvitz and T. Paek. Conversation as action under uncertainty. In C. Boutilier and M. Goldszmidt, editors, Uncertainty in Artificial Intelligence: Proceedings of the Sixteenth Conference. Morgan Kaufmann, San Francisco, 2000.

No context found.

T. Paek and E. Horvitz. Conversation as action under uncertainty. In Proceedings of the Sixteenth Conference on Uncertainty in Artificial Intelligence, pages 445--

No context found.

T. Paek and E. Horvitz. Conversation as action under uncertainty. In Proceedings of the Sixteenth Conference on Uncertainty in Artificial Intelligence, pp. 445-464. AUAI, Morgan Kaufmann, August 2000.

No context found.

T. Paek and E. Horvitz. Conversation as action under uncertainty. In Proceedings of the Sixteenth Conference on Uncertainty in Artificial Intelligence, pp. 445-464. AUAI, Morgan Kaufmann, August 2000.

....to learn key probabilistic dependencies from data and the extension of the system to consider additional sensory information about a user s context and attention. Some of this work is being pursued in the related Quartet project, focused on developing a multilevel conversational architecture [11]. Fig. 5. Troubleshooting of the overall conversation following a sequence of poor recognitions and detection of background noise. 7 Summary We presented methods for extending traditional speech recognition systems with representations and inference machinery that consider the intentions ....

T. Paek and E. Horvitz. Conversation as action under uncertainty. In Proceedings of the Sixteenth Conference on Uncertainty in Artificial Intelligence, pages 445--

No context found.

PAEK, T. AND HORVITZ, E. 2000. Conversation as action under uncertainty. Proceedings of UAI-2000, 455-464.

No context found.

Paek, T., and Horvitz, E. 2000. Conversation as Action Under Uncertainty. In Proceedings of the 16th Conference on Uncertainty in Artificial Intelligence (UAI-2000), 455-- 464.

No context found.

Tim Paek and Eric Horvitz. Conversation as action under uncertainty. In Proceedings of the Sixteenth Conference on Uncertainty in Artificial Intelligence, 2000.

No context found.

TimPaek and Eric Horvitz. 2000. Conversation as action under uncertainty. In Proceedings of the 16th Conference on Uncertainty in Artificial Intelligence (UAI2000) , pages 455--464, Stanford, CA, June.

Online articles have much greater impact   More about CiteSeer.IST   Add search form to your site   Submit documents   Feedback  

CiteSeer.IST - Copyright Penn State and NEC