Adaptive Assessment and Evaluation:

We have noted that adaptive instruction has has at least two quite different meanings -- 1) adapting instructional or training services to meet special needs of physically or otherwise challenged individuals and 2) adapting services based on an individual's given progress and needs with regard to guidance, presentation, or evaluation of progress. Adaptive assessment actually has two alternative common meanings as well. Most frequently the term used is actually "adaptive testing," which refers to a rather sophisticated statistical strategy by which testing levels-of-difficulty or mastery are constantly adjusted to find the point which best defines an individual's current level of skill or knowledge in relation to pre-defined norms. This strategy's primary objective is that of certification or evaluation for predictive/placement purposes, not instruction.

Alternatively, one may adapt the content of a test to best facilitate the progress of an individual's learning of a given body of knowledge. That is, if an instructor can but assess what a student already knows well, what the student misunderstands, and what the student knows nothing of, then instruction may be individually tailored to meet precisely that individual's need for content and further assessment of progress. Thus, in tutorial interviews, an instructor will frequently change a line of questioning in order to "track" his developing evaluation of what a student understands about a given knowledge domain. By finding strengths and weaknesses, an individually targeted, or customized, tutoring plan may be established. This meaning of the term adaptive assessment thus focuses more on individualizing the ongoing educational processes, rather than focusing on establishing where a student resides among some comparative or normative group of peers.

This latter meaning, where adaptive testing defines a strategy for educational purposes, describes the process incorporated into MediaMatrix's design for delivering adaptive tutoring. Importantly, MediaMatrix incorporates an automatic knowledge generation engine which is capable of 1) tracking all interactions of an individual student with all types of MediaMatrix objects; and 2) developing a mirror image of an individual student's semantic network as a representation of his/her knowledge constructions. On the basis of this mirroring system, questions may be dynamically selected and presented to the student based upon both content criteria as well as the student's current skill-level in learning. This selection on the basis of content offers just-in-time queries aimed at, in respective order of importance for tutoring purposes, 1) supplying information on unassessed areas of current relevance for study; 2) remedial presentations to eliminate prior errors or misdirected learning; and 3) continuing confirmation of previously assessed knowledge in order to increase confidence levels in the system's judgements that the student has correctly learned desired associations.

In all cases, question items are dynamically selected on the basis of an individual student's knowledge, progress, and learning skill level. Further, the fact that an expert's semantic networks have been incorporated into the system when items were originally constructed allows the system to constantly compare a student's emerging semantic network with the expert's complete network. This opens yet another avenue for adaptive guidance and presentation, in that items of known high correlations are rarely asked again when a student is working at the highest skill levels and is being assessed by random probes. By emphasizing questions of prior difficulty or items which do not match the expert's knowledge network, a student either corrects former problems or the system prompts the student with opportunities to navigate to those problem areas for further study. This is the ultimate integration of adaptive guidance and adaptive tutoring with adaptive assessment and evaluation. Together, they define MediaMatrix's meaning of adaptive instruction.

Next Topic... Automatic Knowledge Generation