Empirical Identification of Text Simplification Strategies for Reading-Impaired People

摘要

Objective The way in which information is written or presented can exclude many people, especially those who have problems to read, write or understand. However, the process of simplifying texts by hand is extremely time and effort consuming so any attempts to automate part of this process can leverage the access to information. The final objective of this research is the development of an automatic system that eases the process of text adaptation for reading-impaired people. Main Content In this paper we present a preliminary analysis of a parallel corpus containing simplified texts along with their original versions in order to identify what kind of transformations are used when creating the easy-to-read version of a text. The study has been carried out to find out how this process can be automated, at least partially. Results The study has evidenced that a parallel corpus containing easy-to-read versions of texts becomes a very powerful source of knowledge. First, it provides very important information regarding the type of operations carried out during simplifications, and the relative frequencies of each type of operation. Second, it constitutes a valuable resource for identifying empirical instances of the various types of transformations, which can be used as data during development or even training for an automatic or semi-automatic system to simplify texts. Both the frequencies and the types of transformations have been extracted from the study of the corpus. We therefore propose this analysis as a basic resource for the development of an automatic simplification system. Conclusion In this paper we have presented a preliminary analysis of the type of operations involved in text simplification. This first step is previous to any attempt to automatize of the process, as it is required to know what kind of phenomena can be expected. We can conclude that the first step towards a useful formalization of how to simplify texts has been taken.