22.6 Interactive communication in adaptive instruction

The response-sensitive strategies in CBI have been mostly applied to simple student-computer interactions such as multiple-choice, true-false, and short-answer types of question-asking and responding processes. However, Al techniques for natural language dialogues have provided an opportunity to apply the response-sensitive strategy in a manner requiring much more in-depth communications between the student and computer (see 19.2.3). For example, many ITSs have a function to understand and generate natural dialogues during the tutoring process. Although the Al method of handling natural languages is still limited and its development is relatively slow, it is certain that future adaptive instructional systems, including ITS, will have a more powerful function for handling response-sensitive strategies.

The development of a powerful response-sensitive instructional system using emerging technology, including Al, requires a communication model that depicts the process of interactions between the student and tutor. As Wenger (1987) defined, the development of an adaptive instructional system is the process of software engineering for constructing a knowledge communication system that causes and/or supports the acquisition of one's knowledge by someone else, via a restricted set of communication operations.

22.6.1 Process of Instructional Communication

To develop a communication model for instruction, the process of instructional communication should first be understood. Seidel and his associates (Seidel, Compton, Kopstein, Rosenblatt & See, 1969) divided instructional communication into teaching and assessment channels existing between the instructor and student. (Fig. 22-2 is adopted from Seidel et al. with modifications.) Through the teaching channel, the instructor presents the student communication materials via the interface medium (e.g., computer display). The communication materials are generated from the selective integration of the instructor's domain knowledge expertise and teaching strategies based on information he or she has about the student. The student reads and interprets the communication materials based on the student's own current knowledge and the perceived instructor's expectation. The student's understanding and learning of the materials is communicated through his or her response or questions. The questions and responses by the student through the interface medium are read and interpreted by the instructor. Seidel et al. (1969; Seidel, 1971) called the communication process from the student to the instructor the assessment channel. Through this process, the instructor updates or modifies his or her information about the student and generates new communication materials based on the most up-to-date information. The student's knowledge successively approximates to the state that the instructor plans to accomplish or expects.

The model of Seidel and his associates (1969) describes the general process of instruction. However, it does not explain how to assess the student's questions or responses and generate specific communication materials. Since specific combinations of questions and responses between the student and instructor occurring in the teaching and assessment process are mostly task-specific, it is difficult to develop a general model for describing and guiding the process.

22.6.2 Diagnostic Questions and Instructional Explanations

Most student-system interactions in adaptive instruction consist of questions that the system asks to diagnose the student's learning needs and explanations that the system provides based on the student's learning needs. Many studies have been conducted to investigate classroom discourse patterns (see Cazden, 1986) and the effect of questioning (Farrar, 1986; Hamaker, 1986; Redfield & Rouseau, 1981). However, few principles or procedures for asking diagnostic questions in CBI or ITS have been developed. Most diagnostic processes in CBI and ITS take place from the analysis of the student's on-task performance. For assessing the student's knowledge state and diagnosing his or her misconceptions, two basic methods have been used in ITS (see also 19.3.2): (a) overlay method for comparing the student's current knowledge structure with the expert's, and (b) buggy method for identifying specific misconceptions from a precompiled list of possible misconceptions. In both methods, the primary source for identifying the student's knowledge structure or misconceptions is the student's on-task performance data.

From the analysis of interactions between graduate students tutoring undergraduates in research methods, Graesser (1993) identified a five-step dialogue pattern to implement in an ITS. They are: (a) Tutor asks question; (b) student answers question; (c) tutor gives short feedback on answer quality; (d) tutor and student collaboratively improve on answer quality; and (e) tutor assesses the student's understanding of the answer. According to Graesser's observation, tutor questions were primarily motivated by curriculum scripts and the process of coaching student's idiosyncratic knowledge deficits. This five-step dialogue pattern suggests only a general nature of tutoring interactions rather than specific procedures for generating interactive questions and answers.

Collins and Stevens (1982, 1983) generated a set of inquiry techniques from analyses of teachers' interactive behaviors in a variety of domain areas. Nine of their most important strategies are: (a) selecting positive and negative examples; (b) varying cases systematically; (c) selecting counter examples; (d) forming hypotheses; (e) testing hypotheses; (f) considering alternative predictions; (g) entrapping students; (h) tracing consequences to a contradiction; and (i) questioning authority. Although these techniques are derived from the observation of classroom teachers' behaviors rather than experienced tutors, they provide valuable implications for producing diagnostic questions.

Brown and Palincsar (1982, 1989) emphasize expert scaffolding (see 7.4.3) and Socratic dialogue techniques in their reciprocal teaching (see also 23.4.1.3.4). While the expert scaffolding provides guidance for the tutor's involvement or provision of aids in the learning process, the Socratic dialogue techniques suggest what kinds of questions should be asked to diagnose the student's learning needs. Five ploys are important to present in the diagnostic questions: (a) Systematic varied cases are asked to help the student focus on relevant facts; (b) counter examples and hypothetical cases are asked to question the legitimacy of the student's conclusions; (c) entrapment strategies are presented in questions to lure the student into making incorrect predictions or premature formulations of general rules based on faulty reasoning; (d) hypothesis identifications are forced by asking the student to specify his or her work hypotheses; and (e) hypothesis evaluations are forced by asking the student's prediction (Brown & Palincsar, 1989).

Leinhardt's (1989) work provides important implications for generating explanations for the student's misconceptions identified from the analysis of on-task performance or response. She identified two primary features in expert teachers' explanations: (a) explicating the goal and objectives of the lessons, and (b) using parallel representations and their linkages. A model of explanation that she developed from the analysis of an expert tutor's explanations in . teaching algebra subtraction problems shows that explanations are generated from various relations (e.g., pre-, co- and postrequisite) between the instructional goal and content elements and the constraints for the use of the learned content.

As the above review suggests, efforts for generating the principles of tutoring strategies (diagnosis and explana tion) have continued from the observation of human tutoring activities (e.g., Berliner, 1991; Borko & Livingston, 1989; Leinhardt, 1989; Putnam, 1987), and from simulation and testing of tutoring processes in ITS environments (Ohlson & Rees, 1991; see Chapter 19.) However, specific principles and practical guidelines for generating questions and explanations in an on-task adaptive system have yet to be developed.

22.6.3 Generation of Tutoring Dialogues

Once the principles and patterns of tutoring interactions are defined, they should be implemented through interactions (particularly, dialogues) between the student and system. However, the generation of specific rules for tutoring dialogues is an extremely difficult task. After having extensively studied human tutorial dialogues, Fox (1993) concluded that tutoring languages and communication are indeterminate, because a given linguistic item (including silence, face and body movement, and voice tones) is in principle open to an indefinite number of interpretations and reinterpretations. She argues that indeterminacy is a fundamental principle of interaction and that tutoring interactions should be nonrule governed. Also, she says that tutoring dialogues should be contexualized, and the contexualization should be tailored to fit exactly the needs of the student at the moment. The difficulty of developing tutoring dialogues in an adaptive system suggests that the development of future adaptive systems should focus on the application of the advantageous features of computer technology for the improvement of the tutoring functions of the adaptive system rather than simulating human tutoring behaviors and activities. As discussed earlier, however, Al methods and techniques have provided a much more powerful tool for developing and implementing flexible interactions required in adaptive instruction than traditional programming methods used in developing ordinary CBI programs. Also, the development of computer technology, including AI, continuously provides opportunities to enrich our environment for instructional research, development, and implementation.