Toward More Precision in Correction of Grammatical Errors

We describe a system for detecting and correcting instances of a small class of frequently occurring grammatical error types in a corpus of essays which have been manually annotated for these errors. Our system employs a precise broad-coverage grammar of English which has been augmented with a set of mal-rules and malentries to explicitly license certain types of erroneous expressions. The derivation tree produced by a parser using this grammar identifies the location and type of an error in an ill-formed sentence, enabling a postprocessing script to use the tree and the inventory of error types to delete and/or insert tokens in order to produce a corrected version of the original sentence.