SIMPLEX-PB: A Lexical Simplification Database and Benchmark for Portuguese.

Lexical Simplification has the function of changing words or expressions for synonyms that can be understood by a larger number of people. It is very common to have in mind a target audience which will benefit from the task, such as children, low-literacy audiences, and others. In recent years there has been great activity in this field of research, especially for English, but also for other languages such as Japanese and multilingual and cross-lingual scenarios. Few works have children as target audience. Currently, in Brazil, the Programa Nacional do Livro Didatico (PNLD) is an initiative with a broad impact on education, as it aims to choose, acquire, and distribute free textbooks to students in public elementary schools. In this scenario, adapting the level of complexity of a text to the reading ability of a student is a determinant of his/her improvement and whether he/she reaches the level of reading comprehension expected for that school year. On the other hand, there have not been publicly available resources on lexical simplification for Portuguese as yet. Therefore, the development of this material is urgent and welcome. This work compiled the SIMPLEX-PB, the first available corpus of lexical simplification for Brazilian Portuguese. We also make available a benchmark for evaluating the most well-known methods of LS in our dataset.