A grammatical analyzer has been implemented to the specific end of providing the syntactic information required, in phonetic transcription and prosody generation, by a high-quality text-to-speech system performing real-time speech synthesis from unrestricted Italian text, relying on a small dictionary. After detecting the correspondencies between syntax and phonology, contextual rules have been stated, supported with statistical analysis of written texts, predicting the syntactic role of content words on the basis of adjacent function words. Two sample applications of the grammar are reported: the solution of stress ambiguities caused by non-homophone homographs and the grouping of words into "phonological words".