A text-driven method for the deletion procedure in cloze passages

This article proposes a text-driven method for the deletion procedure in cloze passages. Contrary to the previous methods, the deletion rate suggested in this study is based on the number of the linguistic and discourse structures of a passage. After determining the proficiency level of the subjects, a passage of appropriate length and difficulty with the deletion rate of 7 was developed. Along with this standard cloze, eight different versions were also constructed. The deletion rate of the eight versions was based on the number of linguistic and discourse clues. Following a counterbalanced design, the cloze test forms were administered to 403 subjects majoring in English at different academic levels. The results sug gested that the text-driven method is superior to the fixed-ratio deletion method on some grounds. It was also revealed that the differences among the subjects' performance on different versions of cloze tests were significant. It is concluded that the random deletion method may lead to a paradox between the difficulty and the reliability of cloze tests.