A Language-independent Metric for Measuring Text Simplification that does not Require a Parallel Corpus

Abstract

Natural language processing encompasses several tasks, one of which is the automatic text simplification. Telling whether one text is simpler than another involves not only knowledge about the language being analyzed, but also a cultural knowledge of the target audience to which the text is being directed. Most of the current metrics used to measure text simplification are based on the use of parallel corpora, prepared by humans, which makes it difficult to apply the metrics in automatic text simplification in real time. In this paper, we present ISiM (Independent Simplification Metric), a metric that dismiss a parallel corpus, is simple, fast, language and human annotation independent, capable of quantifying the simplicity/complexity of a sentence, thus contributing improve automating text simplification. The results of the tests performed indicate that the proposed metric has the potential to be used to evaluate automatic methods of simplification.