Towards Adaptive Learning Environments

Existing intelligent learning environments for programming represent a step towards comprehensive adaptive learning environments that support all activities in learning programming. In most of these systems, however, only the tutoring component is adaptive. The user interface usually looks the same for the novice and for the advanced learner, while the student's knowledge of the subject matter strongly changes from the beginning to the end of a course. We argue that a next step towards adaptive learning environments is to make all its components adaptive . In this paper, we discuss some problems of creating adaptive environment components for intelligent learning environments and present our current work into this direction . Introduction Towards Adaptive Learning Environments Peter Brusilovsky, Marcus Specht, and Gerhard Weber University of Trier E-Mail : Iplb I specht I weber) Ocogpsy.uni-trier.de Intelligent learning environments are a relatively new kind of intelligent educational systems . In addition to more traditional system-driven intelligent tutoring components that support students' learning, an intelligent learning environment (ILE) includes one or more studentdriven components that support students' doing. These components form the environment. A special case of an ILE with a complex and powerful environment component is an ILE for programming (here referred as Intelligent Programming Environment, IPE) . An IPE contains both an intelligent coach or tutor to support student learning and a programming environment to support student programming activity . A number of IPEs have been described : APT (Corbett & Anderson, 1992), SYPROS (Herzog, 1992), DISCOVER (Ramadhan & du Boulay, 1993), ABSYNT (M6bus, Thole, & Schr6der, 1993), GEL (Reiser, Kimberg, Lovett, & Ranney, 1992), ELM-PE (Weber & Mollenberg, 1995), and ITEM/11? (Brusilovsky, 1992) . Existing IPEs can be considered as the second step (after such classic ITSs as PROUST (Johnson, 1986) and Lisp Tutor (Anderson & Reiser, 1985)) towards "teachers' dream"comprehensive adaptive learning environments to support all activities in learning programming. Most ofthese IPEs, however, are only partly adaptive . In most cases, only the tutoring component uses the student model for the purpose of adaptation . The user interface usually looks the same for the novice and for the advanced learner, while the student's knowledge of the subject matter strongly changes while learning . We argue that the next step towards an adaptive learning environment is to make all the components of an IPE adaptive . In this paper, we discuss some problems of creating an adaptive interface for IPEs . We present our current work on the next version of ELM-PE where we expect to implement the idea ofcomprehensively adaptive IPE . In the following section, we briefly review the current stage of the ELM-PE project which forms the background of our current work. We describe the existing components that are most interesting in the scope of this paper. We then discuss the application of adaptive interface technology along with using the student model of the ILE as the user model for creating an adaptive interface for environment components . A particularly powerful student model is ELM-the episodic learner model used in ELM-PE . The possibilities of making an adaptive interface based on ELM are demonstrated at hand of ALFRED, the structure editor of ELM-PE . We consider some interface features of ALFRED and discuss how these features can be adapted using different types of knowledge about the student represented in ELM. PROGRAM WINDOW FUNCTION PANEL F (cow ((EMOP LIBTE) MIL) ((EQUAL (FIRST LIBTE) W10UIID(T) (oEEP-REMOVE FAOUMEMT (REST LIBTE))) ((RTOM (FIRST LIVE)) Coo li (FIRST LIBTE) (DEEP-REIVUE RROU)EMT (REST LISTS)))) 323 COMMENT LINE POP-UP-MENUS BUFFER WINDOW Components of ELM-PE Figure 1 : The syntax-driven structure editor ALFRED The knowledge-based programming environment ELM-PE is designed to support novices learning the programming language LISP . It has several features which are especially useful in learning to solve problems in a new, complex domain. Some of these features will be sketched briefly in this section . ALFRED-a syntax-driven structure editor. In order to reduce syntax errors, coding function definitions is supported by a syntax-driven LISP-editor. In the program window of the structure editor, program code can be produced by filling in slots of LISP expressions (Figure 1) . These schemata can be accessed from buttons in the function panel or from typing in the first part of a function call . Additionally, expressions can be typed in directly via keyboard or can be pasted from the buffer window . Example-basedprogramming. Examples of LISP-code can be displayed in a separate example window . Anexample can be selected both by the user and by the system's analogy component from examples presented in the learning materials of the LISP-course as well as from the set of function definitions the student has already coded. Visualization. One central principle in learning programming is to visualize the flow of data during evaluation of programs (e .g ., Eisenstadt, Price, & Domingue, 1993) . This is supported by a stepper showing all evaluation steps indicating corresponding expressions of self-defined function definitions-including definitions of sub-functions-in the program window . The stepper is supposed to help students envision dynamic program properties and to encourage active and self-directed learning . Additionally, those parts of the program code which are responsible for errors or are explained by the knowledge-based diagnosis can be highlighted in the program window . Automatic cognitive diagnosis . The knowledge-based component of the programming environment consists of a cognitive diagnosis of the program code produced by a student . This diagnosis is based on ELM, a case-based learning model (Weber, 1994b) . The diagnosis results in an explanation of how the program code could have been produced by the learner. This explanation is the basis for offering hints to the learner concerning which plans must be followed to solve subgoals during problem solving and what part of the code is incorrect .