Abstract: The annotation of dialogues in term of Dialogue Acts (DA) is an important task in the development of dialogue systems. Recently, the N-gram Transducers (NGT) technique showed a better performance than other techniques in the annotation of unsegmented dialogue transcriptions. However, this technique has not been employed in an incremental fashion, which is closer to the annotation framework. In this work, we checked the performance of NGT in this incremental framework and the influence of the size of the partitions in the effort of annotating the whole SwitchBoard corpus.
Index Terms: Dialogue annotation, dialogue systems, statistical models.