Robert L. Mercer
We take an information-theoretic approach to the problem of translating passages from French to English. We imagine a French sentence to be a garbled version of an intended English sentence. We seek the most probable English sentence given the French sentence and take this most probable sentence to be the proper translation. If it is not a good translation, then we fault our statistical model of the process by which English is garbled to become French, and we seek improvements.