Ask HN: Any breakthrough in NLP in 2018?

As an interested (but amateur) outsider, I think I've seen many exciting developments in NLP in the mid 2010's, such as Word2vec (2013), GRU (2014), SyntaxNet (2016), and its second version, DRAGNN (2017). It almost looked like new ideas were introduced every quarter.

However, I don't remember seeing anything similarly "breakthrough" in the past year. It looks like NLP has moved to the "boring but mature" stage (for now), with bigger datasets and incremental improvements over already known methods.

So, what do you guys think? What would be the most surprising/innovative new ideas on NLP that showed up in 2018 (or late 2017)?