Translating sentences from 'original' to 'simplified' Spanish

Resumen

Text Simplification (TS) aims to convert complex sentences into their simpler variants, which are more accessible to wider audiences. Several recent studies addressed this problem as a monolingual machine translation (MT) problem (translating from 'original' to 'simplified' language instead of translating from one language into another) using the standard phrase-based statistical machine translation (PB-SMT) model. We investigate whether the same approach would be equally successful regardless of the type of simplification we wish to learn (given that different target audiences require different levels of simplification). Our preliminary results indicate that the standard PB-SMT model might not be able to learn the strong simplifications which are needed for certain users, e.g. people with Down syndrome. However, the phrase-tables obtained during the translation process seem to be able to capture some adequate lexical simplifications.