A major motivation is to achieve in man-machine interactions the efficiency of speech communication among humans. Continuous speech is more difficult to understand than are isolated words. Commercially available speech recognition systems of the latter type are highly successful despite their limited capability. To recognize continuous speech, more information is needed than is contained in acoustic waves alone. The linguistic and contextual knowledge that must be supplied or programmed into a computer to accomplish speech interpretation is the subject of several research activities which are described. Speech synthesis systems face similar problems but are further advanced.