AutoTutor Lite is capable of assessing student responses in real time by using a lightweight language analyzer based on a Domain Specific Semantic Processing Portal.
Generally speaking, the natural language processing component of
AutoTutor Lite is based on semantic space theory, specifically Latent Semantic Analysis (LSA). If
you are unfamiliar with LSA, consider the following quote from J.R
Firth (1957) "You shall know a word by the company it keeps." LSA is ran
on a large amount of text documents (corpus), where each term in the
corpus is represented numerically. The "distance" between two terms in
the semantic space can be used to judge the accuracy of student/user
input during assessment in AutoTutor Lite. The "semantic answer"
provided by the author is what the student/user responses are compared
against. A central goal for AutoTutor Lite
authors should be to prompt their students/users to provide more
elaborative answers. Self-explanations and deep-level reasoning
questions have been shown to promote learning (Chi, De Leeuw, Chiu,
& LaVancher, 1994; Craig, Sullins, Witherspoon, & Gholson,
2006). Equipped with this knowledge, AutoTutor Lite authors should
carefully configure the semantic engine to best provide feedback to
students/users that prompts them to produce more elaborate
self-explanations. Semantic Space Selection First,
an AutoTutor Lite author should select an appropriate corpus for their
Semantic Space. The "Free Association (by human) corpus is based on the
free association norms database developed at the University of South
Florida (Nelson, McEvoy, & Schreiber, 2004). The TASA corpora were
developed by the Touchstone Applied Science Associates and are broken
down at the grade level. These corpora consist of textbooks commonly
used for specific grade levels. Both the TASA corpora and the Free
Association (by human) corpus are useful for more general domains. Using
these corpora provides a good starting point for AutoTutor Lite
authors. Domain Selection A
unique feature of AutoTutor Lite is the ability to select a specific
domain for the semantic space. These domains are generated by a Domain
Specific Semantic Processing Portal, which essentially places a weight
bias for specific terms found more frequently within the selected domain
(e.g, health, science & mathematics). Weight Criteria This
is the minimum weight an associated term in the "Semantic answer" or
"expectation" must have to be considered or included in the space. Minimum Rank (weight*Strength) The
minimum rank is the product of the weight criterion and strength
criterion, and is essentially the value used to determine how many terms
in the semantic space should be considered. The higher the value for
minimum rank, the smaller percentage of the semantic space is used. Minimum Association Strength The
minimum association strength determines the smallest association to be
considered between the "semantic answer" or expectation and associated
words (i.e., nearest neighbors, synonyms, or related words). A high
value here will only consider the most closely related terms to the
target answer. Minimum Item Weight The
minimum item weight value is used as the cutoff point for the weights
of student/user input. This can be used to eliminate the consideration
of high frequency terms like "I" and "and". For
a case study that explores the usage of these semantic space
configurations please see "The development and analysis of tutorial
dialogues in AutoTutor Lite" (Wolfe, Widmer, Reyna, Hu, Cedillos,
Fisher, Brust-Renck, Williams, Vannucchi, & Weil, 2013). References Chi,
M. T., De Leeuw, N., Chiu, M. H., & LaVancher, C. (1994). Eliciting
self-explanations improves understanding. Cognitive science, 18(3),
439-477.Craig, S. D., Sullins, J., Witherspoon, A., & Gholson, B. (2006). The deep-level-reasoning-question effect: The role of dialogue and deep-level-reasoning questions during vicarious learning. Cognition and Instruction, 24(4), 565-591. Nelson, D. L., McEvoy, C. L., & Schreiber, T. A. (2004). The University of South Florida free association, rhyme, and word fragment norms. Behavior Research Methods, Instruments, & Computers, 36, 402-407. |
Help Pages (EN) >