“To make its 2007 machine translation system practical, Google developed an in-memory, compressed data structure for a 5-gram language model trained on a 2 trillion word corpus.”