Solving Multiple Choices for English Exams by Web Search

In this paper the author introduces a novel method that automatically solves multiple choices for English exams. The algorithm is based on the idea that if a language usage instance is correct, it will appear on the world wide web frequently. Therefore, the Web hit counts of such a language usage instance, which is composed of the stem of the multiple choice together with one of the chosen options, is used as the main evidence for selecting the best answer. Moreover, the author suggests a cascade model that includes a number of different decision functions that not only consider the Web hit count of each instance, but also the domains it appears, the windows size for searching and whether to use exact or inexact match in searching, etc. Evaluations are done on a set of multiple choice questions from College Entrance Examinations for high school students. Evaluation results indicate that this approach can also be applied to improve English writing or other related applications.