Punker: predicting punctuation in Dutch Speech Recognizer output

Jo Meyhi

The presentation deals with a set of experiments in the field of automatic punctuation insertion into Speech Recognizer output. The system inserts punctuattion into Dutch speech and makes use of a memory based learning algorithm. The Punker system is based on a combination of features: word identity, POS-tags, basic chunk information (restricted to noun phrase and verb phrase), a set of cue words and some basic prosodic features