Prosodic analysis: The next generation

Abstract

In many text-to-speech systems, sentence prosody is derived by string-oriented rules, which\nare often rather ad hoc and linguistically unsound. By contrast, a system which employs a syn\ntactic parser allows for highly general and computationaily simple procedures defining the\nmapping between syntax and prosody. The program described in this paper, PROS2, is a first\nattempt towards a "next generation". First, the input sentence is parsed by applying phrase\nstructure rules. Next, the syntactic representation is mapped onto a metrical tree, augmented\nwith "focus" markers. The metrical tree provides an abstract characterization of accent and\nphrasing. While the program, at the time of writing, is still at the prototype stage, it holds\nconsiderable promise for further development and can be expanded in several directions.