Modeling Language Learners ’ Knowledge State: What Are Language Students ’ Free Written Productions Telling Us?

Language learner models usually provide intelligent tutoring systems with information about the learner’s knowledge state, i.e. the individual’s weaknesses and strengths in the target language. The information is normally collected via answers to pre-defined written production types. However, storing information from essay-type questions is more challenging, as this requires the use of instruments that can differentiate errors from mistakes. This paper investigates whether observing incorrect, as well as correct forms in the learners’ input provide us with a better insight into the learners ’ competence. It explains how an analysis of an error-tagged and part-of-speech encoded learner’s input help compute the ratio of incorrect to correct forms. The results of a preliminary analysis focusing on morpho-syntactic errors show that the exclusive use of this ratio may, in some cases, be inadequate to discriminate errors from mistakes and, therefore, to represent the learner’s knowledge in terms of competence