Auto Tutor is an intelligent tutoring system that holds conversations with learners in natural language dialog. Auto Tutor uses Latent Semantic Analysis (LSA) to evaluate the conceptual quality of student contributions so that it adapts to student discourse. The correctness of a student contribution is decided by a threshold cosine value that must be reached between student answer and one or more expectations in a good answer template. We have identified two parameters that robustly influence the threshold values in assessments that are validated by subject matter experts: the length of student answer and the length of the expectation. Auto Tutor therefore needs to be dynamically sensitive to the length of the student answer and the length of the expectation.