Database Reference
In-Depth Information
The use of TimeML is motivated by the fact that it is a textual meta-language - thus easy to read and
to index by common Information Retrieval techniques - and can be employed in a discovery service
context in order to represent the metadata contents for a successive search and discovery. It is enough
flexible to allow the annotation (description) of the kind of event and observations, and their temporal
information, possibly imprecise and approximate.
In the following, let us first describe the representation within fuzzy set theory and possibility theory
of time expressions, then, introduce TimeML and, specifically, the adopted tags, and finally, the proposed
partial matching mechanism.
Definition of Imperfect Time Indications
The representation of temporal information requires the choice of a numeric scale to describe the order
of events and phenomena; this scale is defined by an origin (time t=0) and a time unit, so that events
occurred before t=0 have negative time values and those occurred after have positive values.
The choice of the time unit depends on the cycles of an observed periodic phenomenon, such as the
movement of the sun.
Nevertheless, almost every time scale shares the basic unit
second
, and other units with lower
granularity such as
minute
,
hour
and
day
. Besides these units of time, there are other natural language
time indications such as
week
,
month
,
season
,
year
,
century,
that are artificial notions, defined within
a calendar such as the Julian and Gregorian ones. Other notions of time, related to human experience
such as
now
,
soon
,
recent
,
often
,
end of the year
, are approximate time indications.
Several time models have been proposed (Tsotras and Kumar, 1996), their common limitation is that
they deal with crisp time indications.
A flexible framework has been defined within fuzzy set and possibility theory (Zadeh, 1978) to ex-
press approximate hierarchical time indications close to natural language at distinct level of granularity
(De Caluwe et al., 1999; De Caluwe et al., 2000). It takes inspiration from the time hierarchy proposed
in (Maiocchi et al., 1992) and by the time granularity defined as a mapping from the set of positive
integers to the absolute time proposed in (Wang et al., 1997): it builds up a multi-granular hierarchical
time structure in which also approximate and imprecise time granules can be defined. s
The proposal in this chapter is based on this approach. This way it is possible to express temporal
indications in several time units with distinct granularities, the less refined ones obtained by grouping
units of the higher level granularity.
A
basic domain
G
0
, consists of granules of time points below which the elements are not discernable.
For example, if G
0
is
hour
we cannot discern the
minutes
within an hour. Notice that, a granule of a domain
G' (e.g. G'=
week
) that is not the basic domain G
0
(e.g. G
0
=
hour
) can be defined by recursively grouping
granules (G=
day
and G
0
) of its parent domains in the hierarchy. For example, G'=7*G=7*24
*
G
0
=168
hours
.
The adopted temporal specifications are listed here following:
•
A
time point
indication is defined as a pair [t, G] in which t is an ordinal indicating the position
with respect to the time origin on a domain G of granules. An example is: [t=2,
day
] that indicates
the second day from the time origin. Also approximate time points can be specified and repre-
sented by fuzzy sets on the basic temporal domain. A fuzzy time point is [t={0.8/3, 1./4, 0.7/5},
day
] that means around the fourth day after the time origin.Specifically, a fuzzy or imprecise time
point defined on a basic temporal domain assigns a membership degree in [0,1] to each element
