[philipkglass]

I think that we also needed a much larger training corpus than was available in the 1980s. Back then the largest textual data sets were orders of magnitude smaller.