Gap-fill Tests for Language Learners: Corpus-Driven Item Generation

A system, TEDDCLOG, which automatically generates draft test items from a corpus, which takes the key (the word which will form the correct answer to the exercise) as input and presents the sentences and distractors to the user for approval, modification or rejection.

Abstract

Gap-fill exercises have an important role\nin language teaching. They allow students\nto demonstrate that they understand\nvocabulary in context, discouraging\nmemorization of translations. It is timeconsuming\nand difficult for item writers to\ncreate good test items, and even then test\nitems are open to Sinclair’s critique of invented\nexamples. We present a system,\nTEDDCLOG, which automatically generates\ndraft test items from a corpus. TEDDCLOG\ntakes the key (the word which will\nform the correct answer to the exercise)\nas input. It finds distractors (the alternative,\nwrong answers for the multiplechoice\nquestion) from a distributional thesaurus,\nand identifies a collocate of the key\nthat does not occur with the distractors.\nNext it finds a simple corpus sentence containing\nthe key and collocate. The system\nthen presents the sentences and distractors\nto the user for approval, modification or rejection.\nThe system is implemented using\nthe API to the Sketch Engine, a leading\ncorpus query system. We compare TEDDCLOG\nwith other gap-fill-generation systems,\nand offer a partial evaluation of the\nresults.