The idea came from the fact that I wanted to play with the NLP work and wanted more useful data sets to start from (parsing Shakespeare gets old :-). And the fa

The idea came from the fact that I wanted to play with the NLP work and wanted more useful data sets to start from (parsing Shakespeare gets old :-). And the fact that my fairly simple search engine for PDFs was basically just a classic n-tuple index into the word clouds minus stop words and gave good recall (like the old AltaVista) but not good precision. Another comment in this thread about having a tablet to do that suggests an interesting way to deploy something like this on inexpensive table hardware with a somewhat more bespoke (aka performant) OS and application architecture for appliance type operation. (think running a unikernel rather than a general purpose kernel in order to not spend cycles on things you won't ever need)