IN ORDER TO MAKE A MACHINE THAT CONVERTS WRITTEN TEXT INTO SPOKEN LANGUAGE (READING MACHINE), IT WOULD BE NECESSARY TO MAP SENTENCES OF WRITTEN ENGLISH ONTO CORRESPONDING SENTENCES OF SPOKEN ENGLISH, IN A CONVERSION WHICH THE AUTHOR CALLS "SYNTHETIC ELOCUTION." IN THIS TYPE OF CONVERSION, THE ASSIGNMENT OF PROSODIC FEATURES TO SENTENCES IN CONNECTED TEXTS REQUIRES THE APPLICATION OF RULES WHICH ARE OPERABLE ACROSS SENTENCE BOUNDARIES. THEORETICALLY, A MODEL READING MACHINE MIGHT HAVE A CHARACTER RECOGNIZER AS THE INPUT AND AND A SPEECH SYNTHESIZER AS THE OUTPUT AND WOULD INCORPORATE PROGRAMS FOR SYNTACTIC AND SEMANTIC ANALYSIS, WORD PRONUNCIATION, AND PROSODIC FEATURE ASSIGNMENT. MECHANICAL SPEECH SYNTHESIS BY RULES IS COMPLICATED BY A TWO-PART PROBLEM--(1) DETERMINING THE MINIMAL SET OF PROSODIC FEATURES WHICH HAVE TO BE ASSIGNED, AND (2) DETERMINING THE FORM OF THE RULES WHICH WILL CORRECTLY ASSIGN THEM. THE AUTHOR FOCUSES ON SOME ASPECTS OF THE PROSODIC MODEL AND PROPOSES A MORE THOROUGH MEANS OF DIFFERENTIATING SUCH HOMOPHONOUS EXPRESSIONS AS "LIGHTHOUSE KEEPER" AND "LIGHT HOUSEKEEPER." IT SEEMS THAT ACCENT PLACEMENT IN ENGLISH CANNOT ALWAYS BE DETERMINED WITHIN THE LIMITS OF THE SENTENCE. RATHER IT IS PARTIALLY SUSCEPTIBLE TO PREDICTION BY "CONTEXT SENSITIVE RULES." THIS PAPER WAS READ AT THE 1967 SUMMER MEETING OF THE LINGUISTIC SOCIETY OF AMERICA. (FB)