An Extensible Crosslinguistic Readability Framework

Automatic assessment of the readability level (i.e., the relative linguistic complexity) of documents in a large number of languages is an important problem that can be applied to many real-world applications, such as retrieving age-appropriate search engine results for kids, constructing automatic tutoring systems, and so on. Unfortunately, existing readability labeling techniques have only been applied to a very small number of languages. In this paper, we present an extensible crosslinguistic readability framework based on the use of parallel corpora to quickly create readability software for thousands of languages, including languages for which no linguists are available to define readability rules or for which documents with readability labels are lacking to train readability models. To demonstrate our idea, we developed a system based on the proposed framework. This paper discusses the theoretical and practical issues involved in designing such a system and presents the results of an experiment conducted with the system.