An exam robot for sentence completion in high school English tests

Addressed in this paper is the problem of sentence completion in Chinese national college or high school entrance English examinations in which the most appropriate word or phrase from a given set shoud be chosen to complete a sentence. Although a variety of methods have been developed to solve this problem in the literature, these approaches mainly focused on language modeling (LM) and latent semantic analysis (LSA) to the best of our knowledge. An exam robot prototype was built by extending the language modeling and latent semantic analysis methods to verb tense analysis and long distance phrase extraction. Specifically speaking, the syntactic, lexical and semantic features are extracted separately using by means of LM and LSA as well as verb tense analysis and phrase extraction two methods developed by the authors. These features are then fed into a learning to rank model to build the exam robot. The proposed approach outperforms LM and LSA models by 4.0 percentage points, achieving 78% accuracy on the question sets for senior entrance exams and 76% accuracy on the question sets for college entrance exams.