Annotation -- Speaker -- Speech Act

primary track
Speaker.Speech Act

These are speech acts which may or may not refer to other actions. If they do, link to the former action. Request speech acts must have a domain action specified on the layer "Domain Actions".

Attributes
  ValueSet "SpeechActs" (9 tokens)    Speech Act   Speech acts are labelled at the point where it is identified. However they span over all the domain specific actions until speech act terminates. This means that SA are not annotated incrementally. We dont mark the start of the speech act but we do mark the end.
  MultiLink    Content   Content of SA request is the domain action entry. Content of SA reject points to actions only. Content of SA self-correct points to id-role only. Content of SA apology does not require a link. Content of SA refine points to id-role which is elaborating the action. Add "in response to" link to the actor action. These actions refer to a another action to which they are linked. These are Accept, Reject, Undo, Refine, Stop, etc. An argument can have a different value from the original speech act.
  ValueSet "OverlapChoices" (2 tokens)    Overlapping   If Overlapping is chosen, this label is on top of a previously introduced label. However, different from simultaneous, it does not share the start and end point of the previous label. The time stamps for this label are indeed the begining and end of this label only, and not the group of overlapping labels. If Simultaneous option is checked, the begining and end of the simultaneous label group are the same for all labels in the group. This is these labels are simultaneous. The first label of the group should be unmarked, and all following in the group should be marked.

 

Attribute Values

Speech Act

Speech acts are labelled at the point where it is identified. However they span over all the domain specific actions until speech act terminates. This means that SA are not annotated incrementally. We dont mark the start of the speech act but we do mark the end.

Request   Speaker can request the actor to perform any of the domain specific actions. Request content links to an entry in the domain action layer.  
Accept   Speaker can accept/confirm an action performed by the actor. Accept should link to an entry in the actor track which must lay in the past (Before the Accept).  
Reject   Speaker may reject a actor action or their own action (as "in forget that"). Content of a Reject action should point to actions either at the speaker or actor track.  
Self-Correct   A Self-Correct action points to the speaker track. The content should point to the id-role which corrects the speaker's requested domain action.  
Refine   A refine action's content is pointing to the id-role entry. The id-role entry in turn points to the previous action that the id-role is elaborating. For instance, we have two utterances: "move the square to the right" and "a little bit more". The first utterance is a move action. The second one has a location id-role and a refine action. The refine action points to the elaborating id-role ("a little bit more"). This action refines arguments of previously given actions. Single utterances such as "a little bit lower" would be labelled as Refine actions. Refine content should point to the action of whoever had the last turn in the dialogue. This means that such action, either speaker or actor layer, is being elaborated. Refine label cover sentences which present ellipsis phenomena. Usually these sentence will further restrict arguments given in previous utterances or fill previously empty arguments. Notice that the new utterance, either a "a little bit lower" or "keep going" are not contradictory with previous actions. If a contradiction were the case, the new utterance would be a Correct or Self-Correct action. This would distinguis Refine from Correct.  
Apology   Speaker apologizes. Does not need a content link.  
Abandon   Utterance is abandoned.  
Monolog   Utterance is not intended for actor interpretation. Example "Let's see".  
Unspecified   Unspecified or uninterpreatable action. Annotator should explain in comments.  

Content

Content of SA request is the domain action entry. Content of SA reject points to actions only. Content of SA self-correct points to id-role only. Content of SA apology does not require a link. Content of SA refine points to id-role which is elaborating the action. Add "in response to" link to the actor action. These actions refer to a another action to which they are linked. These are Accept, Reject, Undo, Refine, Stop, etc. An argument can have a different value from the original speech act.


Overlapping

If Overlapping is chosen, this label is on top of a previously introduced label. However, different from simultaneous, it does not share the start and end point of the previous label. The time stamps for this label are indeed the begining and end of this label only, and not the group of overlapping labels. If Simultaneous option is checked, the begining and end of the simultaneous label group are the same for all labels in the group. This is these labels are simultaneous. The first label of the group should be unmarked, and all following in the group should be marked.

Overlapping   If Overlapping is chosen, this label is on top of a previously introduced labe\ l. However, different from simultaneous, it does not share the start and end point of the previous label. The time stamps for this label are indeed the begining and end of this label only, and not the group of overlapping labels.  
Simultaneous   If Simultaneous option is checked, the begining and end of the simultaneous label group are the same for all labels in the group. This is these labels are simultaneous. The first label of the group should be unmarked, and all following in the group should be marked.  

Annotation -- Speaker -- Speech Act