Hacker News Reader

And bases the argument on a single feed forward network, nothing about RNNs which most good NN NLP results are based on.

Google's SyntaxNet (as of Andor et al 2016) achieves state-of-the-art accuracy for dependency parsing without RNNs.