智能导学系统AutoTutor:理论、技术、应用和预期影响

AutoTutor:Theories%2C Technologies%2C Applications and Potential Impacts

AutoTutor 是一种基于自然语言的智能导学系统，在模拟人类教师教学基础上采用自然语言与学 生对话。其使用预期-误解定制式对话，对学生的答案给予及时反馈，并根据学生对问题理解的程度随时调整对 话内容， 有效地引导学生构建理想答案。本文首先概述了AutoTutor 的开发动因及自然语言导学的优势，然后介 绍了其背后的理论和技术，回顾了已有应用研究，并展望了未来的预期影响。AutoTutor 的开发以认知科学和学 习科学理论为基础， 充分利用最新的文本分析技术和对话技巧， 使对话过程自然有效，其后期发展还纳入了学 习情绪相关理论。AutoTutor 对自然语言对话的分析主要采用正则表达式、潜在语义分析和言语行为分类器，并 使用xAPI 标准记录学习活动数据，以方便与其他系统进行信息交流。目前，AutoTutor 已被应用于多个学习领 域，学习效果可达0. 8 个标准差;在旁观者图灵测试中，被试不能区分对话脚本是由系统生成还是由真实教师生 成的;已实现脚本的在线协同编辑。这种基于自然语言的定制式对话可与已有学习平台整合，实现优势互补。

AutoTutor is a natural language tutoring system that simulates a human tutor by holding a conversation with learners in natural language. Most tutors in school systems are not highly trained in tutoring techniques and have only modest expertise on tutoring topics， but they are surprisingly effective in producing learning gains in students. Researchers have dissected the discourse and pedagogical strategies these unskilled tutors exhibit by analyzing approximately 100 hours of naturalistic tutoring sessions. These mechanisms and other ignored learning principles are implemented in AutoTutor. AutoTutor presents questions and problems from a curriculum script， attempting to comprehend learner contributions that are entered by keyboard， formulating dialog moves that are sensitive to the learner's contributions (such as short feedback， pumps， prompts， elaborations， corrections， and hints)， and delivering the dialog moves with a talking head. This conversational structure has been termed expectation - and misconception -tailored ( EMT) dialogue. Tutors give feedbacks to students according to how well their contributions match the expectations or misconceptions. In this paper， the rationale for developing AutoTutor was outlined first and the advantages of natural language tutoring were presented. Next， we reviewed theories and technologies behind the system， together with the applications and potential impacts that have evolved from AutoTutor. Theories on early versions of AutoTutor absorbed cognitive learning principles and practices of teaching strategies. Systems that evolved from AutoTutor added additional theories of emotional learning that have been evaluated with respect to learning and motivation. The technologies that support natural-language tutoring includes latent semantic analysis， part-of-speech classifiers， speech act classifiers， Coh-Metrix and others. The structure of semantic messages is based on FIPA (Foundation for Intelligent Physical Agents) and the Advanced Distributed Learning xAPI (Experience API) specifications. AutoTutor is designed to assist college students in learning the fundamentals of hardware， operating systems， and the Internet in an introductory computer literacy course， and until now it has produced learning gains across multiple domains (e. g. ， computer literacy， physics， critical thinking). On a ‘bystander Turing test'， AutoTutor was indistinguishable from a human tutor when individual conversational turns were evaluated by third-person bystanders who examined transcripts of humantutor interactions. AutoTutor can integrate and largely enhance existing content with its interactive and individualized tutoring conversation.