This track has specified only the necessary attributes for
each action.
Utterances such as "we want a xxxx in yyyy ..." show that at the point
of hearing "we want" we are unaware which action speaker wants the actor
to perform (i.e. it could be a move or a paint).
The disambiguation of intended action occurs later in the utterance.
This means that the actual annotation of the intented action must
occur when there is no doubt of what the speaker meant.
Nodes |
Transcript
Container
|
|
Object Layer
Container
|
To incrementally annotate an object in this domain, we make use
of an anchor at the beggining of the refering expression. This
anchor will be the first word uttered at the begging of the
object (i.e. "a", "the", etc.).
|
Location Layer
Container
|
Location layer has the following attributes: Vertical Direction, Horizontal Direction,
Topological Properties and Relation Modifiers. This layer describes the spatial
relation between two objects; an object and its ground. It is used to describe
location semantic roles for Move and Rotate actions.
|
Atomic Layer
Container
|
This layer has a list of all possible role values not included in
location and object layer. Thus we have: Distance, Heading
and Endpointorientation. The following already have entries at the
object layer: Color, Decoration type, Decoration location, Size.
|
Id-Role
Container
|
This layer identifies the roles that go with and link to
objects or actions. If there is a role relation between to objects one of them is the anchor and the other is the role value.
To know which one is which, one can say "the role of [something] is [that]". Then [something] is the anchor,
and [that] is the role value.
Id-role has the following three attributes:
Anchor (link to object or action whose role we are attaching);
Role Relation (obj, src, destination ... all semantic roles);
Role Value (appropiate value);
|
Domain Actions
Container
|
|
Speech Act
Container
|
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".
|