Word2net: Deep Representations of Language

This work proposes word2net, a method that replaces their linear parametrization with neural networks and learns interpretable semantic representations and, compared to vector-based methods, better incorporates syntactic information.