KDI: Learning Through Writing Using Adaptive Tutoring Systems: Modeling Knowledge Representations from Open-Ended Questions

该项目结合了心理学、教育学、人工智能和计算语言学在学习和知识表示方面的相关最新研究成果，并将它们应用于智能教学系统(ITSS)的开发。学习理论可以通过研究不同类型的知识表征和程序表征的演变来扩展。因此，这项研究将考察学习，即学生知识表征的变化，因为他们整合了多个信息源。尽管ITSS被开发用于检查需要有限可能回答集的明确定义的问题的学习，但ITSS在评估要求学生提供自然语言回答的开放式问题时遇到了问题。然而，要求学生提供作文问题的书面答案可以提供更丰富的学生知识，并提高学生的写作能力。因此，这项研究将证明，ITSS可以扩展到简单的回答之外，并且这些ITSS可以更容易地适应为学生提供个性化的教学。为了开发这些ITSS，本研究使用计算语言学技术潜在语义分析(LSA)从辅导系统中以作文为基础的答案中模拟学生的知识表征。随着学生的知识表征的变化，他们的作文反应也应该改变，包括更多的知识和质量的提高。因此，这样的ITS可以为学生(在领域知识和写作技能方面)的提高提供必要的反馈。此外，智能交通系统可以初步评估学生的能力，并为学生提供额外的帮助，随着学生获得更多的知识(如支架式)，帮助逐渐消失。因此，通过将LSA纳入ITS，个性化教学可以调整提供的反馈水平和呈现的内容。因此，本研究的目的是通过将LSA(一种通过作文表征学生知识的方法)整合到智能教学系统中来研究复杂技能的习得。由于本研究关注的是来自知识丰富领域的复杂技能，因此本项目的结果扩展了心理学中的学习理论，提供了关于学生在学习过程中如何将多种知识来源整合到知识表征中的信息。这项研究对教育心理学也有理论意义，包括对两个内容领域的脚手架进行系统的检查。此外，由于LSA是一种自动派生知识表示的方法，因此本研究对开发任何内容知识领域的培训系统具有广泛的意义。

This project combines relevant, recent findings from psychology, education, artificial intelligence, and computational linguistics in learning and knowledge representation and applies them to the development of intelligent tutoring systems (ITSs). Learning theory can be extended through the study of the evolution of different types of knowledge representations together with procedural representations. Thus, this research will examine learning, the change in students' knowledge representations, as they integrate multiple sources of information.Although ITSs have been developed to examine learning for well-defined problems requiring a limited set of possi-ble responses, ITSs have had problems evaluating open-ended questions that require the student to provide natural language responses. However, asking students to provide written answers to essay questions can provide a much richer representation of student knowledge and also improving students' writing abilities. Thus, this research will demonstrate that ITSs can be developed which extend beyond simple responses and that these ITSs can adapt more readily to provide individualized instruction to students. In order to develop these ITSs, this research uses the computational linguistic technique Latent Semantic Analysis (LSA) to model student knowledge representations from essay-based answers within tutoring systems.As students' knowledge representations change, so too should their essay responses, including more knowledge and improving in quality. Therefore such an ITS can provide the necessary feedback for student improvement (both in domain knowledge and in writing skill). In addition, the ITS can initially assess the students' ability and provide extra help to the students with help gradually fading as the student gains more knowledge (e.g., scaffold-ing). Thus, by incorporating LSA into an ITS, individualized instruction is possible for adjusting both the level of feedback provided and the content presented. The goal of this research is thus to study the acquisition of complex skills through incorporating LSA, a method for representing students' knowledge representations through their essays, into intelligent tutoring systems.Because this research focuses on complex skills from knowledge-rich domains, results from this project extend theories of learning in psychology by providing information about how multiple sources of knowledge are integrated into students' knowledge representations as they are learning . The research also has theoretical implications for educational psychology including the systematic examination of scaffolding for two content areas. In addition, because LSA is an automatic method for deriving knowledge representations, this research has broad implications for developing systems for training in any area of content knowledge.