Geoscience Reference
In-Depth Information Relations in the discourse
As before, the relations between spatial, temporal and/or thematic references
annotated in a text have to be pointed out. The objective is to review a set of
documents pronounced relevant with respect to one or more geographic criteria
mentioned in the query.
First, we focus on the relations between the temporal component and the spatial
component of the information contained within a document. An annotation model of
the beginning and end of scope of a spatial or temporal frame in the text has to be
developed as well as a corresponding mechanism of analysis and annotation of such
textualsections.Weproposelinkingthesemanticstothestreamoftext[BAT 11]inthe
goal of attaching a piece of thematic information to a spatial area and/or to a calendar
The propositions considered would be based on the work relative to the theory of
Charolles [CHA 97, CHA 05b] picked up in recent contributions [BIL 06a, BAT 11].
An approach targeting the identification of discourse frames must therefore be
developed; considering the temporal or spatial adverb phrases in front of nouns as the
starting point of, respectively, a temporal or spatial frame. The scope of such frames
can be extended beyond the sentence introduced by these adverb phrases to the
beginning of the next frame, for example. The construction of a frame consists of
identifying and tagging the temporal (respectively spatial) introducers followed by
evaluating its scope and defining a temporal (respectively spatial) universe for every
delimited textual section. This process necessitates the completion of the information
available in the spatial and temporal indexes with the study of the structure of the text
as well as the identification of the different linguistic tags of transition and
connection between the parts of the text.
For a retrieved document, the analysis of its spatial and temporal frames as well
as the spatial, temporal and thematic features having served for the selection of the
document has to allow us to increase or decrease the corresponding relevance score.
The simplest case concerns a document whose relevant SFs are located in a single
spatial frame s and the TFs in a single temporal frame t. In the most frequent cases of
our corpus, a temporal frame encompasses one or more spatial frames. Some cases
can be described following the relations presented in Figures 2.1 (Allen relations
[ALL 84]) and 2.2 (RCC-8 [RAN 92] model). These range from the non-connection
of a frame s and a frame t to being equal through the partial overlapping of these
frames, from the viewpoint of their scope within the text. We thus propose a
score-reinforcement coefficient for a document, which is negative if the frames s and
t are disjoint, weak if the overlap is partial and significant if there is complete
Search WWH ::

Custom Search