TOWARDS A GENERAL TEMPORAL ONTOLOGY FOR
KNOWLEDGE INTEGRATION
1
Yi Qiang,
2
Femke Reitsma and
1
Nico Van de Weghe
1
CartoGIS Cluster, Department of Geography, Ghent University, Ghent, Belgium
2
Department of Geography, University of Canterbury, New Zealand
Keywords: Temporal ontology, Fundamental ontology, GTO, OWL.
Abstract: Practically, temporal information is related to every aspect of our world. A temporal ontology may
effectively negotiate the meanings between different time concepts. Though some temporal ontologies have
been developed, their uses are still narrow and cannot apply into a broader range of knowledge domains.
Our work aims to develop a general ontology of time which can negotiate the heterogeneities in different
time conceptualizations. It is not only a framework for annotating everyday temporal terms on the Web but
also lays a foundation for knowledge infrastructures with more domain-specific time concepts.
1 INTRODUCTION
Practically, temporal information can be found in
every aspect in our daily life. However,
heterogeneities in time conceptualizations cause
ambiguities when people are exchanging time-
related information. For example, if you are
searching through the Web for holiday promotions
this summer, errors may occur when searching from
the Northern Hemisphere for countries in the
Southern Hemisphere. Also, temporal information in
ancient documents is recorded using different
calendars and year-marking systems, which may
cause misunderstandings when they are integrated
together. In recent years, the development of
Semantic Web and ontologies has greatly improved
information sharing and interoperation in many
fields such as Web Services interoperation (Traverso
and Pistore 2004), knowledge management (Takeda
2004; Brodaric et al. 2008) and information retrieval
(Jones et al. 2001). Many ontologies have been
developed and proved their advantages in facilitating
the communication between various information
domains (Hiramatsu and Reitsma 2004; Bard et al.
2005; Raimond et al. 2007; Brodaric et al. 2008).
Also, some attention has been paid to developing
temporal specifications or ontologies for temporal
information on the Web, such as KSL-time (Zhou
and Fikes 2000), OWL-Time (Hobbs and Pan 2006),
TimeML specification (Pustejovsky et al. 2003) and
temporal parts of fundamental ontologies (Navigli et
al. 2003; Herre et al. 2006). The most complete
work of temporal ontology is OWL-Time developed
by Hobbs and Pan (2006), which represents the
commonly-used temporal concepts as well as
temporal aggregates composed of simple time
entities. OWL-Time restricted to temporal concepts
that are frequently used in Web content and Web
Services, but is insufficient in representing time
concepts in some particular domains such as
archaeology, geology and music.
The goal of our work is building a General Temporal
Ontology (GTO) in order to overcome this problem.
The idea of GTO is similar to that of most
fundamental ontologies (e.g. DOLCE
1
, BFO
2
, GFO
3
,
SUMO
4
) which attempt to describe very general
concepts that are the same across all domains. These
fundamental ontologies are designed for integrating
heterogeneous knowledge coming from different
sources, most of which already involve very basic
temporal portions. Similarly, GTO is also built at the
most general level of abstraction, but particularly for
time conceptualizations (Figure 1). In other words,
GTO can be understood as a temporal portion of a
fundamental ontology. With GTO, heterogeneous
temporal semantics can be negotiated. Extensions or
sub-ontologies can be developed from it in order to
1
http://www.loa-cnr.it/DOLCE.html
2
http://www.ifomis.org/bfo
3
http://www.onto-med.de/ontologies/gfo/
4
http://www.ontologyportal.org/
275
Qiang Y., Reitsma F. and Weghe N. (2009).
TOWARDS A GENERAL TEMPORAL ONTOLOGY FOR KNOWLEDGE INTEGRATION.
In Proceedings of the International Conference on Knowledge Engineering and Ontology Development, pages 275-280
DOI: 10.5220/0002307402750280
Copyright
c
SciTePress
annotate domain-specific time concepts. Since the
goal of our work is implementing GTO with the
most prevalent ontology language (i.e. OWL), the
expressiveness of the ontology is restricted to
Description Logic.
Figure 1: GTO plays similar role as fundamental
ontologies.
The paper is structured as follows: the paper starts
with introducing the basic theories and general
taxonomy of GTO. And then some issues in GTO
have been discussed specifically. Finally,
conclusions are drawn and future work is pointed
out.
2 BASICS OF GTO
2.1 Time Theory Adopted by GTO
Two of the most fundamental questions in building a
temporal ontology are choices of time models and
time primitives. Considering the purpose of GTO is
annotating temporal information rather than
answering complex temporal queries, GTO puts
more emphasis on the representation of temporal
semantics than maintaining reasoning
inconsistencies. GTO is mainly based on the linear
model of time. Cyclic time concepts (such as month,
season, day etc.) are viewed as recurring concepts on
the infinite time line. For example, in the sentence
‘flowers bloom in spring’, ‘spring’ is treated as a
period that regularly reoccurs every year, which is a
kind of non-convex time region in GTO.
In terms of time primitives, GTO adopts both time
instant and time interval. The relation and distinction
between time interval and instance are always
controversial. In one view, a time instant is
considered as an infinitesimal point which is only
used in dividing two time intervals. In the other view,
whether a time region is viewed as an instant or an
interval is a granularity decision that may vary
according to the context of use. A time instant is
undividable and occupies the minimum time unit
under a certain granularity, while a time interval is a
dividable segment of time line and contains more
than one instant. GTO adopts the latter view because
people prefer to use time instants to describe those
instantaneous events such as shooting a gun, turning
off a light. An interval starts at an instant and ends at
an instant, which are called beginning point and end
point respectively. In other words, an interval is
defined by two instants. This time theory may cause
inconsistencies in temporal reasoning such as
Divided Instant Problem (DIP) but is more
expressive in representing temporal semantics in the
natural language.
2.2 Taxonomy of GTO
In many fundamental ontologies (e.g. DOLCE and
BFO), time entities are viewed as regions in time
space, which is the root of temporal concepts.
Convex region and non-convex region are the most
general classes of time regions. Convex regions are
connected and have no gap in it. Non-convex
regions are not connected regions with gaps in it,
which can be further categorized into regular non-
convex regions and irregular non-convex regions.
Non-convex regions are useful in representing time
concepts like ‘the opening hour of the clinic is 9am
to 6pm, from Monday to Friday’. All cyclic time
concepts can be represented by regular non-convex
regions, for example, every spring, every Monday.
Irregular non-convex regions are used to describe
irregularly scattered time regions. Each temporal
region may be described by one or more temporal
descriptions. The general taxonomy of GTO is
illustrated in the form of UML diagram (Figure 2).
With UML diagram, not only the hierarchy of the
ontology is shown, properties, objects of properties
and cardinalities are also shown. Take Instant as an
example, the upper part of the box contains the title
of the class (i.e. Instant), the lower part of the box
contains its properties (DescribedBy) and the object
class (TemporalDescription). The number in []
denotes the cardinality of properties. For example,
[1] denotes that the property has one objects and
[1..] denotes that the property has at least one objects.
For saving space, the diagram only displays some
important properties of classes.
3 ISSUES IN REPRESENTING
TEMPORAL SEMANTICS
GTO aims to providing a general and widely-
applicable framework for representing temporal
KEOD 2009 - International Conference on Knowledge Engineering and Ontology Development
276
Figure 2: General taxonomy of the temporal ontology.
semantics. We have attempted to make GTO
adoptable to temporal concepts in a broad range of
domains. We adopted merits from relevant work (e.g.
KIF-Time, OWL-Time and fundamental ontologies),
but also posed our solutions on some issues (e.g.
time description, non-convex regions, vague time
regions and links between time and other
individuals). This section discusses about how GTO
solves these specific issues.
3.1 Time Description
Position and duration are the two main properties of
time regions. These two properties can be described
differently. For instance, ‘the first Monday in 2009’
and ‘5
th
January 2009’ actually point to the same day;
‘7 days’ and ‘1 week’ are intervals of the same
length. Similar to OWL-Time, we defined a class of
temporal descriptions to describe time position and
duration (Equation 1). In this way, time regions may
have diverse position or duration descriptions. We
only create the description class for crisp convex
region (i.e. instant and interval) because all other
time regions can be described by instant and interval
descriptions in some ways. Instants have position
descriptions but no duration description (Equation 2).
Intervals have duration descriptions but no position
description (Equation 3). The positions of intervals
are derived from position descriptions of their
beginning and end instants. Class
TemporalDescription uses integer data properties to
describe time regions. For example, in Figure 2 the
ISPositionDescription (i.e. international standard
time position description) uses properties such as
Year, Month, Date and so on to describe time
positions. For instance, 1
st
Jan 2009 can be
represented as [Year (2009), Month (1), Date (1)].
However, if you look at some ancient text in China,
time positions and durations are described
differently from that in western world. GTO is also
open for adding classes for such special time
descriptions. Additionally, each position description
has a property HasGranularity to denote the
granularity of the instant.
(,) (, )Describes D T DescribedBy T D
(1)
() () ( )[ (, )
()]
i Instant i p DescribedBy i p
PositionDescription p
≡∃
(2)
() () ( )[ (, )
()]
I Interval I d DescribedBy I d
DurationDescription d
≡∃
(3)
Most time regions can be located in the absolute
time line of the real world, in the most common
knowledge, started from the Big Bang and flowing
to the infinite future. But there are some exceptions.
For example, when we say there is a drum beat at
the 13
th
second in a music track, it is impossible to
locate the drum beat in the time line of the world.
Therefore, we defined a class of time lines where
position descriptions can be located (Equation 4).
Instances of Class TimeLine could be the time line in
a music track, a workflow of automatic control or
the 90 minutes of a football game. Then we are able
to express the temporal semantics in the sentence
like ‘the rocket discard its fuel container at the 15
th
minute after fire’ or ‘the manager usually substitutes
the No.10 player at the 75 minute in a football game’.
Here actions take place in the time line of the rocket
launching process or the football game.
The taxonomy of temporal descriptions is illustrated
in Figure 3. ISDurationDescription and
ISPositionDescription stand for international
standard time duration description and international
standard time position description respectively.
TOWARDS A GENERAL TEMPORAL ONTOLOGY FOR KNOWLEDGE INTEGRATION
277
More parallel subclasses can be developed at the
same level in order to describe time durations and
positions in diverse time systems.
()[ ()
( )[ ( , ) ( )]]
d PositionDescription d
s
InTimeLine d s TimeLine s
∀→
∃∧
(4)
Figure 3: The sub-ontology of time descriptions.
3.2 Temporal Relations
In GTO, temporal relations are represented as
properties between classes. Because we take account
of the time granularity, the temporal relations
between time regions under different granularity are
getting complicated. For example, two instants can
only be equal when they not only have the same
position but also have the same granularity
(Equation 5). More complicated situations may arise
for other relations. Considering the purpose of GTO
is not performing complex reasoning about temporal
relations, we only represent temporal relations as
properties between time regions but have not given
complete definitions. This task will be left in the
future work. The thirteen relations between interval
and interval are based on Allen’s model (Allen
1983). Additionally, three relations (before, equal
and after) are defined between instants and instants,
eight relations (before, starts, during, finishes, after
and the inverse ones) are defined between instant
and interval.
1212
11 1
22 2
12 11
22 1 2
() () (, )
(, ) ( )
(, ) ( )
() (,)
(, ) ( )
Instant i Instant i Equal i i
HasDescription i d PositionDescription d
HasDescription i d PositionDescription d
d d HasGranularity i g
HasGranularity i g g g
∧∧
∧∧
∧∧
=∧
∧=
(5)
3.3 Non-Convex Time Regions
Non-convex time regions have gaps in them. In
other words, non-convex time regions are composed
of many convex time regions. Hobbs (2006) also
proposed representations for this kind of temporal
regions (called temporal aggregate in his work).
However, the representation in GTO is simpler but
more expressive. Non-convex time regions are
categorised into regular non-convex regions and
irregular non-convex regions. Zhou and Fikes (2000)
also used this distinction but gave no concrete
representation for them. Irregular non-convex
regions are composed of irregularly scattered convex
regions. To the contrary, regular non-convex
intervals are composed of regularly scattered regions,
for example, ‘every Monday in May of 2009’ and
‘every Christmas’. In GTO, a regular non-convex
region consists of a regular region and a context
region. The regular region is the regularly recurring
region, while the context region is the range in
which the regular region is recurring. Taking ‘every
Monday during May 2009’ as an example, ‘every
Monday’ is the regular region recurring in the
context region ‘May 2009’. The frequency of the
regular region is set by a float data property (i.e.
HasFreqency). This property is used to describe the
semantics like ‘every other Monday in May of 2009’
where the frequency of the regular region is 0.5.
There may be more than one regular region in a
context region, for example, ‘every Monday,
Wednesday and Friday in May 2009’. Additionally,
context regions can be convex intervals or non-
convex intervals. Non-convex context regions are
used to represent nested regular non-convex
intervals such as ‘every Monday in May’. Here, May
(actually means May every year) is the context
which itself is a regular non-convex interval. Non-
convex regions can be described on the basis of
convex regions. Thus we did not create a description
class especially for non-convex regions. Following
is pseudocode of “every Monday in May”.
EveryMondayinMay
Type: RegularNonconvexRegion
HasRegularRegion: Monday
HasContextRegion: EveryMay
HasFrequency: “1”
EveryMay
Type: RegularNonconvexRegion
HasRegularRegion: May
HasContextRegion: Null
HasFrequency: “1”
May
Type: Interval
HasDescription: 1Month
KEOD 2009 - International Conference on Knowledge Engineering and Ontology Development
278
HasEnd: EndofMay
HasBeginning: BeginningofMay
3.4 Vague Time Intervals
Many domains (geology, history and geography) are
faced with the problem of having vague temporal
information. In these cases, instants have no precise
position and intervals have no precise beginning and
end. This may also refer to the granularity issue. A
crisp time interval may become vague when the
working context shifts from a coarser granularity to
a finer granularity. In GTO, the class of vague
convex regions is used for representing the vague
temporal information (Figure 2). Rough set (Pawlak
1982) and fuzzy set (Zadeh 1965; Pawlak 1982) are
currently the most frequently used theories in
dealing with vague temporal information. The main
difference between them is that fuzzy set has
gradually-changing confidence (between 0 and 1)
according to a function while rough set only has
triplex value (0, 1 or uncertain). In GTO rough set
regions have properties such as upper approximation,
lower approximation whilst fuzzy set regions have
properties such as core, support and kernel (Figure
2).
3.5 Linking Time and Individuals
Because temporal information only makes sense
when it is associated with atemporal individuals (e.g.
process, event or object), it is important to formalise
the links between time and individuals. Currently,
most fundamental ontologies accept the distinction
between endurant individuals and perdurant
individuals, which are called differently in
fundamental ontologies (Table 1). The difference
between endurants and perdurants derives from their
relations to time (Bittner et al. 2004). Endurants are
wholly present at any time they are present, for
example, a book, a lake. Perdurants are wholly
present at any time they are present but extend in
time by accumulating different time parts (Navigli et
al. 2003), for example, a war, a storm. All
individuals are located in time regions that are
similar to spatial locations in the physical space. In
most fundamental ontologies, there is a basic link
between time and individuals (Table 1). For example,
GFO and SUMO only defines the most general link
between time and individual. DOLCE views time as
a subtype of quality like colour, size or weight. This
representation is unintuitive and also problematic
because other qualities also (e.g. colour, size) exist in
time. In our view, both endurants and perdurants are
located in time regions. More specifically, endurants
are wholly present during intervals or present at
instants, whilst perdurants persist during intervals
(e.g. state, process) or happen at instants (e.g. event,
changes). All other specific links between time and
individuals can be developed from them.
Table 1: Distinction of Individuals in Fundamental
Ontologies.
Fundamental
Ontology
Perduring
Individual
Enduring
Individual
Links between Time and
Individuals
DOLCE Perdurant Endurant has-quality (individuals,
temporal-quality)
q-location (temporal-quality,
temporal region)
GFO Process Persential project-to (entities, temporal
region)
BFO Occurrent Continuant N/A
SUMO Process Object when (Individuals, Time)
4 CONCLUSIONS AND FUTURE
WORK
This paper sketched GTO, which is a framework of
an upper ontology for temporal concepts. We
integrated merits from existing temporal ontologies
but also proposed our view on some specific issues
(general taxonomy, time description and non-convex
region, granularity and vague time intervals).
Compared with existing temporal ontologies, GTO
aims to provid a more complete framework of time
abstraction that can be applied into in a broad range
of domains. It not only can annotate everyday
temporal terms on the Web, but can also be further
extended for temporal concepts in particular domains
such as history, geography and archeology. Thus,
GTO may be useful in a knowledge infrastructure
which stores temporal information in different time
systems, for example, cooperating with the SKI
ontology (Brodaric et al. 2008). GTO emphasizes on
the representation of more complete temporal
semantics but ignores some reasoning problems such
as granularity and topological relations.
In the next step, more work is needed for improving
the GTO ontology, including defining temporal
relations, representing more complex non-convex
regions and coupling GTO with fundamental
ontologies. Additionally, some use cases will be
developed to assess the utility of GTO in negotiating
different temporal semantics. Its applications in
knowledge management will be further studied,
which may lay a foundation for a temporally robust
knowledge infrastructure.
TOWARDS A GENERAL TEMPORAL ONTOLOGY FOR KNOWLEDGE INTEGRATION
279
ACKNOWLEDGEMENTS
The research work of Yi Qiang is funded by the
Research Foundation – Flanders.
REFERENCES
Allen, J. F. 1983. Maintaining Knowledge about Temporal
Intervals. Communications of the ACM 26(11): 832-
843.
Bard, J., S. Y. Rhee and M. Ashburner 2005. An ontology
for cell types. Genome Biology 6(2).
Bittner, T., M. Donnelly and B. Smith 2004. Endurants
and perdurants in directly depicting ontologies. AI
Communications 17(4): 247-258.
Brodaric, B., F. Reitsma and Y. Qiang 2008. SKIing with
DOLCE: toward an e-Science Knowledge
Infrastructure. In proceedings of 5th international
conference on formal ontology in information systems,
Saarbrücken, Germany.
Herre, H., B. Heller, P. Burek, R. Hoehndorf, F. Loebe
and H. Michalek 2006. General Formal Ontology
(GFO): A Foundational Ontology Integrating Objects
and Processes. Part I: Basic Principles (Version 1.0),
University of Leipzig.
Hiramatsu, K. and F. Reitsma 2004. GeoReferencing the
Semantic Web: ontology based markup of
geographically referenced information.
EuroSDR/EuroGeographics workshop on Ontologies
and Schema Translation Services. Paris.
Hobbs, J. R. and F. Pan. 2006. Time Ontology in OWL.
W3C Working Draft 27 September 2006, from
http://www.w3.org/TR/2006/ WD-owl-time-
20060927/.
Jones, C. B., H. Alani and D. Tudhope 2001. Geographical
Information Retrieval with Ontologies of Place. In
proceedings of the International Conference on
Spatial Information Theory: Foundations of
Geographic Information Science.
Navigli, R., P. Velardi and A. Gangemi 2003. Sweetening
Ontologies with DOLCE. Intelligent Systems, IEEE
18(1): 22-31.
Pawlak, Z. 1982. Rough Sets. International Journal of
Information and Computer Science 11(5): 341-356.
Pustejovsky, J., J. Castano, R. Ingria, R. Sauri, R.
Gaizauskas, A. Setzer, G. Katz and D. Radev. 2003.
TimeML: Robust specification of event and Temporal
Expressions in Text In proceedings of the IWCS-5
Fifth International Workshop on Computational
Semantics, Stanford, CA.
Raimond, Y., S. Abdallah, M. Sandler and F. Giasson
2007. The Music Ontology. In proceedings of the First
Conference on Social Semantic Web.
Takeda, H. 2004. Semantic web: a road to the knowledge
infrastructure on the internet. New Generation
Computing 22(4): 395-413.
Traverso, P. and M. Pistore 2004. Automated Composition
of Semantic Web Services into Executable Processes.
In proceedings of the 3rd International Semantic Web
Conference – ISWC 2004, Hiroshima, Japan.
Zadeh, L. A. 1965. Fuzzy Sets. Information and Control 8:
338-353.
Zhou, Q. and R. Fikes 2000. A Reusable Time Ontology.
In proceedings of the AAAI workshop on Ontologies
for the Semantic Web
KEOD 2009 - International Conference on Knowledge Engineering and Ontology Development
280