Is deeper always better? Replacing linear mappings with deep learning networks in the Discriminative Lexicon Model