Not exactly sure, but after skimming through Wikipedia, natural language generation sounds pretty interesting. But text-to-speech and speech recognition is probably what I've had in mind.