SARI

An automatic evaluation metric for text simplification systems that compares a system’s output against both the original input sentence and a set of human-written simplification references, rewarding the system for adding appropriate words, keeping important words, and deleting unnecessary words. Introduced by Xu et al. in 2016, SARI specifically addresses the limitations of BLEU (a machine-translation metric) for simplification tasks, where fluency alone is insufficient — a good simplification must also actually make changes. SARI is widely used in the NLP text-simplification literature, though human evaluation with target reader groups remains essential for accessibility applications since automatic metrics cannot directly measure the user experience.