John Searle and other influential theorists have argued that machine translation (MT) and other natural language processing (NLP) programs can never appreciate meaning in the deepest sense – in other words, that they can never truly exhibit semantics. It is true that MT and many other NLP systems have made steady and impressive progress while use of explicit semantic processing has undergone a rise and fall. It is also true that consensus among researchers on the meaning of meaning has remained elusive. In this chapter, however, we observe renewed interest in semantic representation and processing. Moreover, we foresee gradual adoption of semantic approaches grounded upon audio, visual, or other sensor-based input. We distinguish such perceptually-grounded semantic approaches from most current methods, which have tended to remain perception-free. With respect to philosophical implications, we suggest that perceptually-grounded approaches to automatic natural language processing can display intentionality, and thus foster a truly meaningful semantics. As background for these predictions and suggestions, we survey the role of semantics in machine translation to date in terms of three paradigms: rule-based, statistical, and neural MT. A section on each paradigm discusses its treatment of semantics: rule-based methods have generally emphasized symbolic semantics; statistical methods have generally avoided semantic treatment or employed vector-based semantics; and neural methods have handled meaning as implicit within networks.