Spatial and Temporal Information Retrieval in Textual Corpora - Geographical Information Retrieval in Textual Corpora - page 36

11March1957; July1957; 11March; March;springof1787; summerof1799;
winter;2002;18 th Century;18h00;endof1892;from1781;since1787;1997-
1998; end of August 1801; from 1950 to 1960; early 1950s; toward the end of
spring 1800; April and May 1798; first days of autumn; toward the end of 1794
or the beginning of 1795; 12, 15 and 20 July 1916; 1950,52,57; from 29 to 30
September; before the Christian era; since the revolution; during 18 years.
As a first step, we have decided not to include the expressions referring to hours
of the day (less convinient level of detail, in our contexts of use), to durations
(anaphoric references difficult to analyze) and to historical events (events needing
specific thesauri). In the remaining subset, some expressions explicitly refer to a
calendar entry: these are the dates (“11 March 1957”; “spring of 1901”; “18 th
Century”). We will call these equally absolute entities or absolute features. Others
correspond to an adaptation of one or more absolute temporal entities (“end of
1892”; “toward the end of spring 1800”). We will call these relative temporal entities
or relative temporal features (RTFs). The finest temporal range is the day. The
temporal entities can be complete or incomplete. Incomplete entities (“March”; “11
March”; “winter”) cannot be located on an absolute time scale since they are not
associated with a year in an explicit way in the corresponding noun phrase.
2.4.2. Core spatial model and core temporal model
Following the target/site hypothesis described in the state of the art [BOR 98], we
considerthatanimportantpartofspatialortemporalinformationexpressedinatextis
composed of at least one NE and a variable number of spatial or temporal indicators,
specifying its range. The developmentof the model has also been guided by the study
of corpora. Following this study, we have proposed a unified model to describe the
spatial features (SFs) and temporal features (TFs). This model is referred to as core
since it will be used for all the steps of information indexing and retrieval. Figure 2.5
gives an overview of the model.
SFs as well as TFs can be absolute or relative. A relative feature is described
recursively from one or more features. Recursion confers to this model the power of
expression necessary for the representation of complex entities. Each entity is
associated with one or more representations.
An absolute spatial feature (ASF) corresponds to a geolocalizable spatial NE
(“the city of Pau”, “Laruns”) for which there are numerous classifications [HIL 90].
If the administrative entities (cities, roads, for example) are precisely
georeferenceable, the entities corresponding to the physical environment (mountains,
lakes, rivers, valleys, for example) have sometimes ill-defined limits or are without
consensus (e.g. fiat and bona fide borders [SMI 01]). An RSF corresponds to a site
NE accompanied by spatial indications allowing us to identify the target zone: spatial