Making Sense Of Distributed Representations With Activation Spectroscopy