A Comprehensive Disfluency Model for Multi-Party Interaction

We present a disuenc y model derived from analysing transcriptions of the AMI meeting corpus. Our model goes beyond previous work in that it discriminates several classes that are elsewhere regarded the same. Furthermore, we provide a formal account for naturally occurring phenomena that are rarely modeled in other schemes. Our annotations show signicant occurrences of these classes. An evaluation of the annotations from four different annotators reveals a high agreement,