Size does not matter. Frequency does. A study of features for measuring lexical complexity

Abstract

Lexical simplification aims at substituting complex words by simpler synonyms or semantically close words. A first step to perform such task is to decide which words are complex and need to be replaced. Though this is a very subjective task, and not trivial at all, there is agreement among linguists of what makes a word more difficult to read and understand. Cues like the length of the word or its frequency in the language are accepted as informative to determine the complexity of a word. In this work, we carry out a study of the effectiveness of those cues by using them in a classification task for separating words as simple or complex. Interestingly, our results show that word length is not important, while corpus frequency is enough to correctly classify a large proportion of the test cases (F-measure over 80%).