Prosody modelling in concept-to-speech generation: methodological issues

We explore three issues for the development of concept-to-speech (CTS) syst ems. We identify information available in a language-generation system that has the potential to impact prosody; investigate the role played by differ ent corpora in CTS prosody modelling; and explore different methodologies f or learning hom linguistic features impact prosody. Our major focus is on t he comparison of two machine learning methodologies: generalized rule induc tion and memory-based learning. We describe this work in the context of mul timedia abstract generation of intensive care (MAGIC) data, a system that p roduces multimedia briefings of the status of patients who have just underg one a bypass operation.