Next: The representation
Up: Interfaces
Previous: Salsa/Tiger XML
Contents
Index
The underlying model
The syntactic level of TIGER/SALSA XML corresponds to the
syntactic representation of TIGER XML [5]. The
syntactic structure is a tree with both node and edge
labels
. Trees can contain
crossing edges to encode discontinuous constituents.
On the role-semantic level, we model each frame instance as a frame tree of depth one with a root labelled with the frame
name. A frame tree has at least one edge that points to the
frame-evoking element, which is unlabelled in the graphical
representation. All other edges point to frame elements and are
labelled accordingly.
In order to make the annotation more flexible, we keep all frame trees
separate. This means that leaves of frame trees are nodes of the
syntactic structure. In principle, however, frame tree leaves could
also figure as roots of other frame trees, resulting in a nested
semantic structure.
In addition, a model for exhaustive semantic annotation must also be
able to encode the following complications:
- A frame element may consist of more than one constituent. Consequently, our model (see
[2]) allows for frame trees in which multiple edges bear the same label.
- A frame element or target may consist of only part of a word in
the case of (German) compound nouns. For example, the German
compound Mietrechtsdiskussion (tenant law discussion) contains
both the target (diskussion) that introduces a Conversation frame and its Topic role (Mietrecht).
Therefore, our model is able to make reference to sub-word units.
- A frame element may be situated in a different sentence than the
target, as often happens with conversation frames. Hence, frame
trees can refer to entities in adjacent sentences.
- At times, the meaning of a sentence is ambiguous or vague, and
annotators cannot commit to a single tag. For these situations, the
model allows to tag multiple annotation referring to the same entity
as underspecified, both on the level of frames and the level
of frame elements. Consistent with
[4], it is left to the user how to
interpret this representation (e.g. as disjunction or conjunction).
Next: The representation
Up: Interfaces
Previous: Salsa/Tiger XML
Contents
Index
Aljoscha Burchardt
2007-09-04