This is really insightful, there's a pretty large library of algorithms and approaches that have been sitting on the shelf for decades (or used in very few plac

This is really insightful, there's a pretty large library of algorithms and approaches that have been sitting on the shelf for decades (or used in very few places) that become far more interesting with modern hardware. The recent resurgence with Bloom filters for example, a technology form the 1970s, used in a few database products, is now everywhere. But there's also the curious problem of expensive and difficult to produce technology that doesn't have a market large enough to sustain it. Example, the crop of now dead or dying enterprise entity extraction companies. It turns out it's expensive to build those systems, but the market isn't that large, which forces the price up to try and survive, which then further reduces the market and so on into a death spiral. Nowadays people just try to get by with OpenNLP...which was easily bested by tools like ThingFinder and ClearForest a decade earlier.