對於祈使機械控制命令的理性預期模型設計

Abstract

自然語言處理的相關研究可以追朔到1950年代，為了讓一般使用者能夠用自然語言與電腦系統溝通，許多研究學者投注於基於自然語言之對話系統的研究。現有的研究依其設計，主要可以分為三種基本的對話方法- 基於有限狀態（Finite State Machine）、基於框架（Frame），以及基於資訊狀態更新（Information State Update）等對話方法。在這篇論文中，我們定義了祈使控制機械命令，用來組成一個任務，一個任務包含了受控元件、行為、受控元件的描述。我們提出一個理性期望模型，結合有限狀態機與分散式代理人等設計概念，分析我們所定義的任務。其中，有限狀態機是用來處理任務中的受控元件、行為、描述子等關係，簡化後續對話方法中各分散式代理人（包括元件代理人、行為代理人、回應代理人、命令代理人等）所負責的處理程序。最後，我們並以各種不同任務分析，結果可以明顯看出本文所提之方法可以幫助對話管理者減少對話的次數。

Since 1950s, natural language processing and dialog approaches, concerning the interactions between human users and computers, have been researched and evolved towards a more friendly user interface. By their designs, the existing dialog approaches can be classified as Finite State-based, Frame-based and Information State Update (ISU)-based. In this thesis, we define imperative machine control commands to construct a task with operated device ,behavior and description of the device. we propose a rational expectation model, integrated with a finite state machine and distributed agents, to analyze the task we defined. A finite state machine is designed to construct relationships of a operated device , behavior and description describing the device. Such a design simplifies the process of distributed agents, including the component agent, behavior agent, response agent and command agent, in a dialogue system. In the end of the thesis, we analyze REM in terms of different case studies of tasks. From the case studies, we show that the REM can decrease the conversation of dialogue manager.