Published online by Cambridge University Press: 26 August 2021
A diagnostic component for natural language utterances has been devised which allows one to integrate (possibly inconsistent) evidence from a wide range of knowledge sources into a unique decision procedure. It is based on a procedure for the disambiguation of dependency structures using graded constraints. Different description levels like syntax, semantics, and others are treated by separate structural representations, which are disambiguated simultaneously. Due to the complete symmetry of the system architecture the approach allows for the successful handling of a wide variety or student errors.
This article is an extended version of a paper presented at the NLP+IA 98 conference in Moncton, Canada.