- 學位論文
利用言語行為理論設計問題本位學習中討論輔助精靈
Designing a Problem-Based Learning Conversation Agent based on Speech Act Theory
摘要
討論是學生互相交換知識、分享心得的重要學習過程。本論文最終目的在設計一個輔助討論的智慧型代理人（conversation agent），試著將此輔助精靈機制應用於問題本位學習（Problem-Based Learning, PBL）教學過程的討論步驟中，並且期望在討論過程中達到引導學生討論並監控討論過程的目的。 本研究首先將討論資料分成兩大部分：發言話語相關與發言話語無關部分。發言話語相關主要運用的是語用學（Pragmatic）與語意學（Semantic）兩種理論做基礎來分析討論話語內容；發言話語無關的部份則是分析發言者與發言時間這兩種類型的討論狀態。 本研究根據文獻與研究，以言語行為理論（Speech Act Theory）中的不同語用行為四種分類：詢問型（request）、回覆型（reply）、評價型（evaluation）、招呼型（chat），相對應地建立語用關鍵字彙（pragmatic keywords）來切割語用段落類型，另外配合使用知識地圖（Knowledge Map）之概念階層方式來儲存語意關鍵字彙（semantic keywords），進一步設計演算法來分析討論，產生語用語意分析報告（pragmatic-semantic analysis report），提供給教師作為教學參考依據。 經由以上的語用語意分析為基礎，建構討論精靈則包含有三個主要流程步驟：感知討論狀態、規則庫條件萃取、產生輔助語句。首先感知討論的不同種狀況，接下來將討論的狀況當作條件，輸入規則庫（Rulebase），經由規則的萃取，輸出輔助精靈的話語對策，最後根據這樣的對策從話語的資料庫中取出相對應的輔助精靈話語，輸出至聊天室中，輔助學生的討論。 透過討論輔助精靈的參與，在小組討論時，教師本身可以負責較為需要照護的組別，而將某些組別交由討論輔助精靈監控整個討論過程。精靈一方面偵測發言話語相關的狀態：計算討論的概念數量、概念類型、語用數量三種不同方向；另一方面，也可以偵測與發言話語無關的狀態：分析討論參與人數、討論經過時間以及討論頻率等，最後再根據這些討論狀態發出相對應語句輔助小組的討論。
並列摘要
Discussion is a good way for students exchanging their idea and finding out the solution of a problem. In the chat-room of an e-learning platform, a teacher could not participate in all discussion of different groups at the same time. An intelligent agent which has ability to analyse student’s conversation could help teachers solve this problem. This research focuses on how to generate a conversation agent with the abilities of participating in students’ discussion and navigating the direction of the discussion. The method of analyzing conversation data is divided into two types: sentence content-dependent analysis and sentence content-independent analysis. The sentence content-dependent analysis of a conversation is related to the theories of Pragmatic and Semantic. The sentence content-independent part is related to conversation participants and speaking timing. Based on these analyses, this research constructs a conversation agent for navigation in students’ discussion. The process of conversation agent for navigating conversation includes the following steps: sensing conversation, extracting rules, and generating guidance sentence. In the end, the conversation agent could not only participate in a chat-room on web, but also generate a Pragmatic-Semantic Analysis Report for teachers. When the agent involves in the discussion of chat-room, it would speak sentences with different sentence strength based on the sentence content-dependent and sentence content-independent analyses. After the conversation, the agent would generate the Pragmatic-Semantic Analysis Report according to the speech act of the conversation and the keywords discussion in this conversation. This report would help teachers know the discussion style and the most discussion concepts of each discussion group.