Formal and computational methods for achieving mutual understanding in conversations among humans and computers.
Material type:
TextDescription: 88 pISBN: 0493088059Subject(s): Psychology, Cognitive | Computer Science | Language, Linguistics | Speech Communication | 0633 | 0984 | 0290 | 0459Dissertation note: Thesis (Ph.D.)--Stanford University, 2001. Summary: In conversation, participants establish and maintain their mutual belief that their utterances have been understood well enough for current purposes—a process that has been referred to as <italic>grounding</italic>. In order to make a contribution to conversation, participants typically do more than just produce the right utterance at the right time; they coordinate the presentation and acceptance of their utterances until they have reached a sufficient level of mutual understanding to move on, a level defined by the grounding criterion. Recent interest in employing grounding for use in collaborative dialog systems has highlighted difficulties in rendering hitherto qualitative intuitions about grounding into formal terms. In this dissertation, I propose a formalization of grounding based on decision theory that captures key intuitions about the contribution model while providing an explicit method for determining the grounding criterion. Starting with the Principle of Maximum Expected Utility, I describe how to assess when a contribution has been made by building up the notion of <italic>sufficient</italic> mutual understanding from two perspectives, that of listeners and speakers. Finally, I elucidate how grounding relates to mutual understanding and shared knowledge, or common ground.Summary: Since uncertainties permeate every level of grounding, from attending to what was said and identifying what words were spoken, to understanding the intentions behind the words, I present a domain-independent, multi-modal computational architecture I developed that allows users to collaborate with a spoken dialog system in establishing mutual understanding at multiple, interdependent levels of analysis. The architecture, called <italic>Quartet</italic>, applies the formalization of grounding discussed in the first part of the dissertation. After describing representations, inference procedures, and decision strategies for managing uncertainties within and between the multiple levels, I highlight the advantages of Quartet by describing interactions between a user and two spoken dialog systems that utilize the architecture: <italic>Presenter</italic>, a prototype system for navigating Microsoft PowerPoint presentations, and the <italic>Bayesian Receptionist</italic>, a prototype system for dealing with tasks typically handled by front desk receptionists at the Microsoft corporate campus. Key advantages include: robustness to uncertainties that often cause misunderstandings, hands-free continuous listening of spoken input, and grounding strategies that adapt over time.
Source: Dissertation Abstracts International, Volume: 62-01, Section: B, page: 0573.
Adviser: Herbert H. Clark.
Thesis (Ph.D.)--Stanford University, 2001.
In conversation, participants establish and maintain their mutual belief that their utterances have been understood well enough for current purposes—a process that has been referred to as <italic>grounding</italic>. In order to make a contribution to conversation, participants typically do more than just produce the right utterance at the right time; they coordinate the presentation and acceptance of their utterances until they have reached a sufficient level of mutual understanding to move on, a level defined by the grounding criterion. Recent interest in employing grounding for use in collaborative dialog systems has highlighted difficulties in rendering hitherto qualitative intuitions about grounding into formal terms. In this dissertation, I propose a formalization of grounding based on decision theory that captures key intuitions about the contribution model while providing an explicit method for determining the grounding criterion. Starting with the Principle of Maximum Expected Utility, I describe how to assess when a contribution has been made by building up the notion of <italic>sufficient</italic> mutual understanding from two perspectives, that of listeners and speakers. Finally, I elucidate how grounding relates to mutual understanding and shared knowledge, or common ground.
Since uncertainties permeate every level of grounding, from attending to what was said and identifying what words were spoken, to understanding the intentions behind the words, I present a domain-independent, multi-modal computational architecture I developed that allows users to collaborate with a spoken dialog system in establishing mutual understanding at multiple, interdependent levels of analysis. The architecture, called <italic>Quartet</italic>, applies the formalization of grounding discussed in the first part of the dissertation. After describing representations, inference procedures, and decision strategies for managing uncertainties within and between the multiple levels, I highlight the advantages of Quartet by describing interactions between a user and two spoken dialog systems that utilize the architecture: <italic>Presenter</italic>, a prototype system for navigating Microsoft PowerPoint presentations, and the <italic>Bayesian Receptionist</italic>, a prototype system for dealing with tasks typically handled by front desk receptionists at the Microsoft corporate campus. Key advantages include: robustness to uncertainties that often cause misunderstandings, hands-free continuous listening of spoken input, and grounding strategies that adapt over time.
School code: 0212.
There are no comments on this title.