Verfahren und Anordnung zur Spracherkennung bei Wortkomposita enthaltenden Sprachen

摘要

A speech recognition process and device are disclosed for languages that contain composite words. Only the components of composite words are stored in a linguistic model and only said components are managed in the vocabulary. Different processing paths are established to recognise possible composite words, both for the corresponding composite word components and for possible individual words. Specific linguistic model statistics are calculated based on said processing paths. The linguistic model statistics are based on a probability decomposition, during which the context and the components of a composite word are separately taken into account. Use is made of the fact known in linguistics according to which grammatically determinant parts of a composite word are generally found at the end of the composite word and indicate the gender, case and number of the composite word. The invention is particularly suitable for real time speech recognition during discrete and continuous dictation.