Yeah Norvig seems to be vindicated by deep learning models as a sophisticated markov chain which can give pretty good results purely through probability.

Even norvig acknowledges that there are edges cases which probability will have a hard time on, especially untrained. Chomsky tends to focus on these with his focus on emulation instead of markov simulation.

More context on the debate: http://web.cse.ohio-state.edu/~stiff.4/cse3521/norvig-chomsk...