Detecting and Diagnosing Grammatical Errors for Beginning Learners of German: From Learner Corpus Annotation to Constraint Satisfaction Problems.
Adriane Boyd · OhioLink ETD Center (Ohio Library and Information Network) · 2012
This thesis presents a corpus of beginning learner German with a reliable error annotation scheme and an approach for detecting and diagnosing grammatical errors in learner language.A constraint-based dependency parser provides the foundation for a flexible and modular analysis of German by representing parsing as a constraint satisfaction problem.The grammar checker, Fledgling, detects and diagnoses errors using constraint relaxation with a general-purpose conflict detection algorithm.Fledgling is developed and evaluated using authentic learner productions from the learner corpus.It judges grammaticality correctly for 80% of sentences and is 82-91% accurate in determining whether a sentence contains selection, agreement, or word order errors.