Supplement to IMIX dialogue act annotation manual

Dealing with multi-sentence utterances

In some cases, the sentences can meaningfully be seen as a single unit standing for a single dialogue act class. In that case, there is no problem.

In other cases though, each sentence may meaningfully be annotated with a different class. However, we may only annotate one class. This is a deliberate limitation of the annotation scheme. We show how to deal with these cases in the following section.

In case each sentence has a different class, we annotate the most important dialogue act class only. We use a priority scheme to determine which is the most important. The following priority order shall be used, highest first:

• followup question. If one or more of the sentences are followup questions, the act should be annotated with one of the followup question classes.
• reformulation.
• verify-question.
• negative-question.
• negative-statement.
• acknowledge.

Finally, we have to define how to deal with the different followup question classes, in case multiple sentences are followup questions:

• If one of these sentences is of class missing-referent, then the class is missing-referent.

• Otherwise, all sentences that are followup questions should be dealt with as if it were a single followup request. That is, the sequence of questions should be understandable and answerable as a single followup request. This is the starting point from which you can determine how it should be rewritten:
  • If the followup request becomes self-contained by applying one of the four rewrite methods to only one of the sentences, then it shall be annotated as the class corresponding to the rewrite method.

  • Otherwise, it is of class referencing-other. This includes the case in which multiple sentences will need to be rewritten, either using the same rewriting method or different ones.

  Example:

  U1:Welke complicaties kunnen optreden bij mazelen?
  S1:Soms kunnen echter complicaties optreden (hersenontsteking, longontsteking), reden waarom kinderen tegen mazelen worden ingeënt.
  U2:Hoe vaak komen die complicaties voor? En hoevaak ga je eraan dood?
  (rewrites to: Hoe vaak komen complicaties voor bij mazelen? En hoevaak ga je eraan dood?)

  So, this should be annotated as other-pp.