Demand-Driven NLP for ICALL

When the user input is first received by the erver, th system performs a string match comparing the user's answer to pre-stored answers. Each pre-stored answer is accompanied by a previously computed analysis for student model updates. Storing multiple answers allows for freer student input and, at an introductory level, involves little extra work. At an advanced level where exercises become even less constrained, alternatives to pre-storing correct answers can be considered (see Gerbault, 1999). If the student input matches any of the stored answers, then the user's student model is updated and the Language Server returns the session back to the Java applet. A previous study (Heift & Nicholson, 2000a) has shown that approximately 40% of all answers submitted are initially correct. In these instances, the complete error checking mechanism is unnecessary if it can be quickly determined that the sentence is correct. If the string match fails, the sentence is passed to the Punctuation Check.