Robust Plan Recognition in Dialogue

该项目的目标是开发一种更稳健的计划识别过程模型，而不受以前系统的限制性假设。为了有效地进行合作对话，自然语言系统必须能够推断用户的计划识别做出了几个限制性假设，以阻止他们处理自然发生的对话。这些假设包括：1)用户不能打算让系统识别他/她的目标，除非它是在给定情况下可以解释当前对话的唯一目标，以及2)由系统推断的模型总是与用户正在构建的任务相关计划完全一致。这个项目将研究一种更强大的计划识别模型，而不需要这些不切实际的假设，因此它将能够处理不太理想的对话。我们的方法将是研究如何使用证据推理技术来促进在存在多种选择的情况下理性和自然地识别用户意图，以及如何将处理不良输入的技术应用于识别用户计划的系统模型中的错误并从错误中恢复的问题。

The objective of this project is to develop a more robust process model of plan recognition without the restrictive assumptions of previous systems. In order to effectively engage in a cooperative dialogue, a natural language system must be able to infer a user's plan recognition have made several restrictive assumptions that prevent them from handling naturally occurring dialogues. These assumptions include: 1) a user cannot intend that the system recognize his/her goal unless it is the only goal that can account for the current dialogue in the given situation, and 2) the model inferred by the system is always in total agreement with the task related plan under construction by the user. This project will investigate a more robust model of plan recognition without these unrealistic assumptions and which therefore will be able to handle dialogues that are less than ideal. Our approach will be to study hjow evidential reasoning techniques can be used to facilitate rational and natural recognition of user intention in the presence of multiple alternatives and how techniques for handling ill formed input can be applied to the problem of recognizing and recovering from errors in the system's model of the user's plan.