| The ICICLE project is an effort to create an adaptive tutoring system for deaf users of American Sign Language who are attempting to improve their proficiency in written English. Its primary objective is to syntactically analyze samples of user-produced English text and construct appropriate tutorial instruction for the errors found in that text. One of the challenges faced by the analytic component of the system is the need to discriminate between multiple competing syntactic analyses of ambiguous sentences. This problem is particularly relevant to this application because the different choices may represent different sources of a perceived error, and the system needs to be as accurate as possible when identifying user errors and delivering instruction.; This dissertation describes the development of a user modeling component to address the intelligent parse disambiguation problem in ICICLE. The user model hybridizes two approaches to user modeling, the individualized overlay approach and the population-based stereotype approach. It tracks user language usage in order to best represent current language proficiency, and supplements direct observations with general knowledge about stereotypical second language acquisition sequences.; Through this work, we have enabled the ICICLE system to adapt its parse selection process to the individual, selecting those syntactic interpretations which are the closest match to the user's expected performance levels. The user model adapts to the individual over time, following the user dynamically across the learning journey. As part of this work, we have also established a methodology for empirically deriving stereotypic information about a user population. |
