A Platform for English Assessment Testing Using Corpus Material

Over the years learner corpora have proven itself as a reliable source of data for foreign language learning. Corpus-based item generation has recently become trend in design of language assessment tools. In this work an approach to test item generation utilizing learner corpus resources is presented on the basis of a specially developed web-based testing platform LangExBank. LangExBank utilizes data from REALEC corpus which contains English texts written by Russian university students. The corpus contains error annotation, which includes error tagging and correction. The main research task of this work is the generation of feasible multiple choice questions from error-annotated sentences. Two technics of option generation (rule-based and neural network-based) are implemented for different types of errors and are evaluated using expert assessment.