15th International Congress of Phonetic Sciences (ICPhS-15) Barcelona, Spain August 3-9, 2003

Improving Text-Based Prediction of German Phrase Level Accent by Incorporating Concepts of Novelty and Contrast

Uwe D. Reichel, Irene Jacobi

University of Munich, Germany

In this study semantic and discourse level motives for accentuation are addressed by providing new automatic methods to extract contrast constructions and to determine novelty of information without requirement of manually prepared world knowledge. Compared with a baseline model employing information content a C4.5 decision tree incorporating also contrast and novelty outperforms the former by about 4% in classification accuracy.

Full Paper

Bibliographic reference.  Reichel, Uwe D. / Jacobi, Irene (2003): "Improving text-based prediction of German phrase level accent by incorporating concepts of novelty and contrast", In ICPhS-15, 2449-2452.