A multi-level description of date expressions in German telephone speech

The ability to efficiently and reliably process date expressions is crucial to many speech-based applications. A multi-level description which contains the syntactic deep structure, the orthographic surface form, a citation form representation, and the phonetic transcription is proposed. With this representation, high-level constraints, e.g. on syntactic structure, can be exploited to restrict the search space on the lower levels. This multi-level representation is applied to read and spontaneous date expressions of the German SpeechDat(M) telephone speech corpus, and the results of an analysis of the transcriptions of 3000 recordings are presented