The Effect of Semantic Difference on Non-expert Judgments of Simplified Sentences

Abstract

Advances in text simplification depend on reliable judgments of sentence difficulty. The ability of untrained native English speakers to judge sentence difficulty in the presence of variation in semantic similarity is examined using cloze tests and a forced-choice comparison task. Judgments from participants in web-based experiments demonstrate ability to assess sentence difficulty of professionally leveled sentence pairs with 84% accuracy. The comparison task results suggest that participants' ability to judge comparative sentence difficulty is inversely related to semantic similarity; that is, contrary to our intuition, speakers appear more accurate at judging sentence difficulty for sentences that are dissimilar than for those that are similar.