People who don’t know history are doomed to say ignorant things about it. Here’s a neural network model of sentence comprehension, once widely cited, from 30 years ago. People have been working on this for a long time. https://www.tandfonline.com/doi/abs/10.1080/01690968908406371 …https://twitter.com/LanguageMIT/status/1056865214568763392 …
-
Pretty ridiculous to suggest that Ted Gibson "doesn't know history." And how does 1989 paper invalidate the claim that a vast majority of the surge of interest (+compute) in modern DL has focused on easier problems? (object/speech recognition, surface-level/"cheatable" NLP tasks)
and yes most work has focused on easier problems but often with the premise and promise that that work would solve harder problems. read the 2017 alphago nature paper for an prime example of that sort of rhetoric and promise.