United States Government. Approved for public release; distribution unlimited.

When Newell introduced the concept of the knowledge level as a useful level of description for computer systems, he focused on the representation of knowledge. This paper applies the knowledge level notion to the problem of knowledge acquisition. Two interesting issues arise. First, some existing machine learning programs appear to be completely static when viewed at the knowledge level. These programs improve their performance without changing their "knowledge." Second, the behavior of some other machine learning programs cannot be predicted or described at the knowledge level. These programs take unjustified inductive leaps. The first programs are called symbol level learning (SLL) programs; the second, non-deductive knowledge level learning (NKLL) programs. The paper analyzes both of these classes of learning programs and speculates on the possibility of developing coherent theories of each. A theory of symbol level learning is sketched, and some reasons are presented for believing...

Plan recognition techniques frequently make rigid assumptions about the student's plans, and invest substantial effort to infer unobservable properties of the student. The pedagogical benefits of plan recognition analysis are not always obvious. We claim that these difficulties can be overcome if greater attention is paid to the situational context of the student's activity and the pedagogical tasks which plan recognition is intended to support. This paper describes an approach to plan recognition called situated plan attribution that takes these factors into account. It devotes varying amounts of effort to the interpretation process, focusing the greatest effort on interpreting impasse points, i.e., points where the student encounters some difficulty completing the task. This approach has been implemented and evaluated in the context of the REACT tutor, a trainer for operators of deep space communications stations.

Plan recognition techniques frequently make rigid assumptions about the student's plans, and invest substantial effort to infer unobservable properties of the student. The pedagogical benefits of plan recognition analysis are not always obvious. We claim that these difficulties can be overcome if greater attention is paid to the situational context of the student's activity and the pedagogical tasks which plan recognition is intended to support. This paper describes an approach to plan recognition called situated plan attribution that takes these factors into account. It devotes varying amounts of effort to the interpretation process, focusing the greatest effort on interpreting impasse points, i.e., points where the student encounters some difficulty completing the task. This approach has been implemented and evaluated in the context of the REACT tutor, a trainer for Operators of deep space communications stations.

: The application of machine learning (ML) to solve practical problems is complex. Only recently, due to the increased promise of ML in solving real problems and the experienced difficulty of their use, has this issue started to attract attention. This difficulty arises from the complexity of learning problems and the large variety of available techniques. In order to understand this complexity and begin to overcome it, it is important to construct a characterization of learning situations. Building on previous work that dealt with the practical use of ML, a set of dimensions is developed, contrasted with another recent proposal, and illustrated with a project on the development of a decision-support system for marine propeller design. The general research opportunities that emerge from the development of the dimensions are discussed. Leading toward working systems, a simple model is presented for setting priorities in research and in selecting learning tasks within large projects. Cen...

A new student modeling system called Assert is described. Assert is a general purpose algorithm which uses domain independent techniques to perform student modeling and to automatically construct libraries of common bugs. For its modeling component, Assert uses a machine learning technique which is an extension of the Either theory refinement algorithm. A common library of bugs is constructed by extracting commonalities across multiple student models. Initial experimental data suggests that Assert is a more effective modeling system than other techniques previously explored, and that the automatic bug library construction significantly enhances subsequent modeling efforts. 1 Introduction Almost since the advent of the computer age, researchers have recognized the computer's enormous potential as an educational aid. Early efforts to use computers as educational tools resulted in a paradigm now generally known as computer aided instruction (CAI). Such programs are used to automate the ...

v Chapter 1 Introduction 1 1.1 Part I---Mental arithmetic : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : 1 1.2 Part II---Multicolumn multiplication : : : : : : : : : : : : : : : : : : : : : : : : 2 1.3 Structure of arithmetic skills : : : : : : : : : : : : : : : : : : : : : : : : : : : : 3 1.4 Aims : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : 3 Part I Mental arithmetic Chapter 2 Memory for arithmetic facts 6 2.1 Phenomena : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : 6 2.1.1 The production task : : : : : : : : : : : : : : : : : : : : : : : : : : : : : 7 2.1.2 Neuropsychological constraints : : : : : : : : : : : : : : : : : : : : : : 9 2.1.3 Rule based processing : : : : : : : : : : : : : : : : : : : : : : : : : : : : 11 2.1.4 Summary : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : 12 2.2 Previous models : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : : 13 2.2.1 D...

Regularitics in the world give rise to regularities in the way in which we deal with the world. That is to say, we fall into routines. I have been studying the phenomena of routinization, the process by which institutionalized patterns of interaction with the world arise and evolve in everyday life. Underlying th;.s evolution is a dialectical process of internaiization: First you build a n;odel of some previously unarticulated emergent aspect' of :m existlug routine. Armed with an incrementally more global view of the interaction, you can often formulate an incrementally better informed plan of attack. A routine is not a plan in the sense of the classical planning literature, except in the theoretical limit of this process. I am implementing this theory using running arguments, a technique for writing rule-based programs for intelligent agents. Because a running argmnent is compiled into TMS networks it proceeds, incremental changes in the world require only incremental recomputa- tion of the reasoning about what actions to take next. The system supports a style of programming, dialectical argumentation, that has many important properties that recommend it as a substrate for large AI systems. One of these might be called additivity: an agent can modify its reasoning in a cls of situation by adducing arguments as to why its previous arguments were incorrect in those cases. Because no side-effects are ever required, reflexive systems based on dialectical argumentation ought to be les's fragile than intuition and experience suggest. I outline the remaining implementation problems.

A review of previous work in the field of learner modelling reveals an emphasis on the surface descriptions of errors of understanding with only partial consideration of the underlying misconceptions that might explain the cause of the errors. Within a general framework for learner modelling we propose a technique, involving the use of a structured interface, for the effective acquisition of a richer and deeper model of learners' errors. The technique is considered from the perspective of a concrete domain: second language learning. Key-words: Learner Modelling, Knowledge Acquisition, Intelligent Tutoring Systems, Second Language Learning. 1. Introduction Problems associated with the acquisition of knowledge about "what" and "how" a learner understands are a stumbling block to the provision of more flexible instruction which is better adapted to the learner. This paper presents the results of initial experiments on a technique to support the acquisition of a richer knowledge about le...

. This paper presents a case-based tutor, CECELIA 1.1, that is based on techniques from CELIA, a computer model of case-based apprenticeship learning [Redmond 1992]. The teaching techniques include: interactive, step by step presentation of case solution steps, student predictions of an expert's actions, presentation of the expert's steps, student explanations of the expert's actions, and presentation of the expert's explanation. In addition, CECELIA takes advantage of a technique from VanLehn's [1987] SIERRA -- presenting examples in an order so that solutions only differ by one branch, or disjunct, from previously presented examples. CECELIA relies on its teaching strategy encouraging greater processing of the examples by the student, rather than on embedding great amounts of intelligence in the tutor. CECELIA is implemented using Hypercard on an Apple Macintosh, and has been pilot tested with real students. The tests suggest that the approach can be helpful, but also suggest that el...