The future of natural language processing lies in finding a more compressed representation of word distribution and training algorithms to understand the infinite number of plausible continuations of multiple frames in a high dimensional continuous space.