Error Recognition and Parsing of Syntactically Mildly Ill-formed Natural Language
Abstract
This paper reports on the error-recognition module of an interactive CALL (Computer assisted language learning) system. To reach the so-called `communcative compentence' a language learner needs a stimulating environment which allows him/her to use the foreign language creatively with adequate feedback to learn from production errors. To support both aspects the system presented here allows free formed input and is able to give adequate feedback for mildly ill-formed sentences. The grammar is a traditional LFG-style grammar which is used by a standard Earley-parser. Error recognition is based on a modified unification algorithm which is able to retract attributes if the values contradict and to add a new attribute for feedback purposes containig the type of error. It is argued, that LFG is especially suited for such a system, because it is computationally feasible and because it uses concepts and terminology which can easily be transformed into learner-oriented explanations.