An absolute prosodic modelisation for an English text-to-speech synthesis system implies the study of three main levels: pragmatics, semantics and syntax. Such levels can provide accentuation potentials. Indeed, they show some possibilities but they cannot predict how a given sentence will be uttered. The purpose of this article is to demonstrate the main characteristics that are necessary to the prosodic marking of a text. Furthermore, we will show how this marking can improve the text-to-speech synthesis Kali, developed at the university of Caen.
Firstly, we will delimit the four prosodic levels we used in our study. Secondly, we will show how to locate these levels with a few textual clues including syntactic, semantic and pragmatic information. Thirdly, we will present the use of the linguistic development software Unitex and how it is complementary with Kali.