Finding the Blank with Sequence Labeling for English Learning

Abstract

Previous approaches to generate fill in the blanks (FITB) questions for English learning mostly are limited to corpus-based approaches that generate Cloze collocates or most frequent co-occurring words. In this study, we propose a Natural Language Processing based approach to generate Fill In The Blank English learning exercises. First, we classify the nature of the problem, and second, we look into the generation of learning exercises. We limited the scope of this research to verb conjugation exercises but it can be extended to other types of learning exercises as well. We looked into the generation of English learning exercises as two types of problems, a sequence labeling problem where we label each token from a sentence whether it could be a potential blank or not and as a sequence to sequence generation problem where we measure its effectiveness in generating such English learning exercises. Generation of FITB for English grammar through these approaches can be useful in the field of Education and can act as a baseline for future work on this problem.