Can Artificial Neural Networks Do Anything at All?

The very first generation of work was just sort of, can they do almost anything interesting? Can we take some kind of artificial neural network and train it to do any kind of interesting non-trivial behavior in language? The first couple of tasks that people tried them on and where they started to show some signs of life were sentiment analysis. And textual similarity, this is sort of reading two sentences, probably talking about related things and deciding if their meanings were similar.