| ▲ | schoen an hour ago | |
I have a friend who is a great expert on rule-based machine translation and natural language understanding in the classical sense (intellectually describing, in detail, how individual natural languages work and are structured), and none of her techniques have achieved state-of-the-art results for some years. The famous bitter lesson. It sounds like you were already doing statistical NLP (the thing that has been clobbering everything rule-based in performance) but have seen that field itself start to require much greater resources than what you were working with. | ||