Although research on artificial natural language processing has made great progress, so far it is avoiding the most salient feature of language: Meaning of words.
For sure, among some circles a call for taking the semantics seriously would stir (superficial) jeers. However, at the end of the day, the meaning of words is the most obvious thing to tackle if you want to take natural language seriously.
Phenomenologically speaking, meanings of words are instances of intentional qualia. As is true for qualia in general, statistical approach cannot uncover the fundamental aspects of the meaning of words.
Statistical learning has made great progress in recent years. Feats like GPT-2 are impressive. However, these statistical methods cannot tackle the semantic sides of language no matter how impressive their results might appear.
Semantics is a part of the mind-brain problem. Technical separation of language processing from consciousness studies is good for a while, but it cannot be the ultimate route for understanding why and how we speak.
Related video:
Meaning of words and intentionality