Access by Geographic Content to Textual Corpora: What Orientations? - Geographical Information Retrieval in Textual Corpora - page 9

footprintsortimestamps extractedfromthe documents.Ingeneral, the comparisonis
based on the intersection sizes of these footprints and stamps.
The recognition of “spatial” and “temporal” terms in texts is supported by
techniques of named entity recognition (NER). NER, detailed in [CHI 98], consists
of the retrieval of textual objects called named entities (in other words, proper nouns,
expressions of time and numeric expressions) which can be categorized into classes,
such as people's names, names of organizations or businesses, place names,
quantities, distances, values, dates, acronyms and abbreviations.
Moreover,thestudiesoflinguisticanalysishavefoundapplicationsintheworldof
IR:forinstance, theconcept of target/site describedby Vandeloise[VAN 86] andthat
of concrete entity/reference described by Borillo [BOR 98] shows the particular way
humans describe spatial information when it comes to writing. It is more and more
common to see techniques of linguistic analysis being associated with techniques of
statistical analysis [DEL 04]: for example, the detection of named entities in a text
uses morpho-syntactical processes of linguistic analysis [MIK 99]. Thus, the tools of
NLP support a fine analysis based on the interpretation of the semantics contained in
the textual documents. They contribute to NER and to the extraction of noun phrases
which contain these entities. For example, “the south of Pau” is the phrase evoking
the spatial named entity Pau, and “in the beginning of the 18 th Century” is the phrase
evoking the temporal named entity 18 th Century.
Qualitative spatial reasoning (QSR) and qualitative temporal reasoning (QTR)
complete the study of language by proposing reasoning processes for the acquisition
of additional knowledge. The importance given to the qualitative aspects of spatial
information stems from ancient Greece, as Kowalski et al. [KOW 07] recall. More
recently, the study carried out by Allen [ALL 91] focuses on the temporal reasoning
for a qualitative representation. Propositions for QSR have then adapted this study by
taking into account the specificities and the bigger complexity of spatial information.
Cohn [COH 96] and then later Cohn and Hazarika [COH 01] show the state of the art
of QSR and, in particular, classify the spatial relations (see section 2.3.4). In the
context of GIR, a reinvestment of the study related to QSR and QTR targets the
interpretation of spatial and temporal noun phrases, respectively.
1.4. Toward the construction of a multicriteria IR engine
Accessing the content of textual documents via an IR approach integrating the
spatial, temporal and thematic dimensions (Figure 1.2) is the main challenge of this
study. Its objective is the construction of an IR engine combining these three
dimensions.