In this article, we present a framework for obtaining rephrasings for short text spans. Good candidates include paraphrases, but also more generally phrases that could help a writer revise a text with some shifts in meaning. The presented framework uses as its knowledge source bilingual aligned phrases learnt from parallel corpora. We present several models for selecting rephrasings, and we evaluate the selection power of candidate rephrasings on grammaticality, meaning preservation and authoring value. The approach is then discussed and future work is described.