On the Training Data Requirements for an Automatic Dialogue Annotation Technique

Carlos D. Martínez-Hinarejos
2007 SIGDIAL Conferences  
When constructing a task-oriented dialogue system, it is usual to perform an acquisition of dialogues for the system's task. This acquisition can be used to define the behaviour of the dialogue system, and it can be rulebased or corpus-based. In the corpus-based case, the models that define the behaviour are automatically inferred from annotated dialogues. The annotation process is timeconsuming and error-prone, and the use of assistant tools for the annotation can reduce the effort in this
more » ... ess. In this work, the data requirements of a previously presented annotation tool are presented, and the results show that the technique obtains its maximum performance even with a relative small amount of annotated dialogues.
dblp:conf/sigdial/Martinez-Hinarejos07 fatcat:y2zfp7q6nrgsbplemda4dierwe