How to represent a word and predict it, too: Improving tied architectures for language modelling | Publicación