Converting Mikrokosmos frames into Description Logics
P.J. Beltr·an-Ferruz and P.A. Gonz·alez-Calero and P.Gerv·as
GAIA - Applied Arti cial Intelligence Group
Dep. Sistemas Inform·aticos y Programaci·on
Universidad Complutense de Madrid
C/ Juan del Rosal, 8. 28040 Madrid (Spain)
pablo@fdi.ucm.es, fpedro,pgervasg@sip.ucm.es
Abstract
Mikrokosmos contains an ontology plus a number
of lexicons in different languages that were origi-
nally developed for machine translation. The under-
lying representation formalism for these resources
is an ad-hoc frame-based language which makes it
dif cult to inter-operate Mikrokosmos with state-of-
the-art knowledge-based systems.
In this paper we propose a translation from the
frame-based representation of Mikrokosmos into
Description logics. This translation allows us to
automatically transform Mikrokosmos sources into
OWL and thus provide a powerful ontology in the
formalism of the semantic web. Furthermore, the
reasoning mechanisms of Description Logics may
also support knowledge acquisition and mainte-
nance as well as its application in natural language
processing systems.
1 Introduction
The Mikrokosmos project was originally an interlin-
gual system for Knowledge-Based Machine Trans-
lation (KBMT) (Nirenburg, 1987) developed in the
Computing Research Laboratory from New Mexico
State University. Although KBMT was conceived
for translation of domain speci c texts, no further
restrictions are imposed in the contents of the text.
Therefore the creators of Mikrokosmos built a rich
ontology that contains a lot of general concepts,
more than 4.700 concepts that are connected with
an average of other 14 concepts using attributes and
relations (de Quesada, 2001).
KBMT is an expensive approach that requires
a big effort on knowledge acquisition, and it has
been considered impractical by some authors. For
that reason, the creators of Mikrokosmos were es-
pecially concerned about developing real-size sys-
tems that would demonstrate the feasibility of their
approach. Generating contents for the ontology was
their  rst concern, while the use of a rigorous for-
malism for knowledge representation was not con-
sidered a priority (Moreno-Ortiz et al., 2002).
The work presented here is an effort to port
Mikrokosmos into Description Logics (DL) in or-
der to incorporate this resource into the systems we
are developing. Our work on natural language gen-
eration integrates ontologies and case-based reason-
ing (CBR) (Diaz-Agudo et al., 2002), an approach
that heavily relies on classi cation-based reasoning
from DL.
Representing Mikrokosmos in DL should bring
several bene ts. Since DL is the underlying knowl-
edge representation approach in the Semantic Web,
a big number of supporting tools are being de-
veloped for acquiring and maintaining ontologies
represented in some version of DL, such as OWL
(Bechhofer et al., 2004). Giving a well-founded
formal representation to Mikrokosmos should im-
prove its quality by uncovering inconsistencies. Fi-
nally, porting Mikrokosmos to a formalism as pop-
ular as OWL will de nitively increase its potential
user community.
There are other efforts that convert ontologies
from different representations to Description Log-
ics languages. OntoMap is a web-site that pro-
vides access to upper-level ontologies and hand-
crafted mappings between them (Kiryakov et al.,
2001). The initial set of ontologies contains Cyc,
EuroWordnet and Mikrokosmos ontology, but it
only deals with the top-level hierarchy. In case of
Mikrokosmos it only contains 13 concepts. Their
main effort involves the mapping between differ-
ent ontologies. They also provide the resources in
DAML+OIL. The project’s goal is to facilitate easy
access, understanding, and reuse of general-purpose
ontologies and upper-level models.
The rest of this paper runs as follows. Next
section describes the frame-based language used in
Mikrokosmos ontology, and Section 3 describes the
DL which is the target of the translation. Section 4
is dedicated to the mapping process and Section 5
evaluates this process. Section 6 points out future
work, and  nally Section 7 concludes the paper.
Concept Slot Facet Filler(s)
REPLACEMENT-FOR DEFINITION VALUE  when x is a replacement for y 
IS-A VALUE PHYSICAL-OBJECT-RELATION, EVENT-RELATION
INVERSE VALUE REPLACED-BY
DOMAIN SEM EVENT, OBJECT
RANGE SEM EVENT, OBJECT
Table 1: Example frame: REPLACEMENT-FOR
Figure 1: Mikrokosmos top hierarchy
2 Mikrokosmos ontology
In Mikrokosmos, ontology lists the de nitions of
concepts that are understood as corresponding to
classes of thing and events in the world. Concepts
are primitive symbols of a world model which in-
cludes objects, events and properties organized in
a complex hierarchy of language-independent con-
cepts (See top hierarchy of Mikrokosmos in Figure
1). The concepts are constructed following super or-
dinates, or hyponymy relations (IS-A links). In ad-
dition to its organization into a taxonomy via IS-A
links, the ontology contain numerous other links be-
tween concepts, such as links using properties (Lon-
ergan, 2001). For example DECEMBER has a rela-
tion with WINTER using the property PART-OF-
OBJECT.
Each concept that makes up the ontology is lan-
guage independent and is represented using frames.
For example we can see the frame for concept
REPLACEMENT-FOR in Table 1.
The format of Mikrokosmos Ontology is de-
scribed in detail in (Nirenburg and Raskin, 2004).
It formally introduces the syntax and the semantics
of the ontology using a BNF grammar. We are most
interested in how we can access to this information.
Ontology is saved in a text  le using Spencer no-
tation that is based on XML. There is another nota-
tion called Beale notation that is based on Lisp, but
we will focus on Spencer notation.
In the XML based format we have the whole on-
tology represented in a list of RECORD entries.
De nition of one CONCEPT requires one or more
of these RECORD entries. Each entry contains four
 elds, that are: CONCEPT, SLOT, FACET, and
FILLER.
The CONCEPT  eld can be  lled by any Name
of a concept of the ontology.
The second  eld in each entry is SLOT. This
 eld can be  lled with PROPERTY or any of its
subclasses using IS-A links. There are two kind
of slot  llers. One type are descendants of AT-
TRIBUTE or RELATION, that represent links be-
tween concepts in the hierarchy. The other type are
descendants of ONTOLOGY-SLOT. We will call
them special slots, and all of them have the sense of
determining the structure of the ontology. Possible
descendants of ONTOLOGY-SLOT are: DEFINI-
TION, DOMAIN, INSTANCES, INVERSE, IS-A,
RANGE, SUBCLASSES and some others that are
less important; later in this section we will explain
them in detail.
The third  eld is FACET, and it describes some
 ner distinctions between the possible  llers of the
slot. Possibles FACET  llers are: VALUE, SEM,
DEFAULT, INV, NOT, DEFAULT, DEFAULT-
MEASURE and RELAXABLE-TO.
The last  eld is FILLER, and its value depends
on the other  elds, but generally it contains either a
Name of a concept of the ontology or an instance.
Initially we can think that there are no restrictions
in these representations, but there are some spe-
cial slots that limit expressiveness. All CONCEPT
frames have non-special and special slots. Special
slots for all kinds of concepts are:
 DEFINITION: De nition in English of the
concept.
 IS-A: It is used for asserting parents in the hi-
erarchy.
 SUBCLASSES: It is used for listing concept
children.
 INSTANCES, SPANISH1, ENGLISH1: They
are only used in the leaves of OBJECT and
EVENT, and contains words of the dictionary.
Special slots which can only be present in
all PROPERTY and only in PROPERTY concept
frames are:
 DOMAIN: It has  llers usually  lled with
EVENTs1 and/or OBJECTs and it determines
whether a CONCEPT can have it as a SLOT.
 RANGE: It is used in RELATIONs and AT-
TRIBUTEs. In RELATIONs the RANGE slot
has only the SEM facet. The  llers of the SEM
facet are the names of concepts that are in the
range of this RELATION. In ATTRIBUTEs
the RANGE slot has only a VALUE facet. The
VALUE facet is  lled by all the possible literal
or numerical values permissible for that AT-
TRIBUTE. The  ller can also be a numerical
range speci ed using appropriate mathemati-
cal comparison operators (such as >, <, ...).
 INVERSE: It is de ned only for RELATIONs.
It is mandatory for all RELATION frames. The
INVERSE slot has only the Value facet which
is  lled by the name of the RELATION which
is the Inverse of the given RELATION.
 MEASURED-IN: It is de ned only for the de-
scendants of the SCALAR-ATTRIBUTE con-
cept frame. The MEASURED-IN slot is used
to add a measuring unit for the number or
scalar range that  lls facets of the RANGE slot
in SCALAR-ATTRIBUTE concept frames.
The facet  llers of the MEASURED-IN
slot are the daughters of the MEASURING-
UNIT concept. The MEASURED-IN slot
is used only in those SCALAR-ATTRIBUTE
frames where MEASURING-UNIT has physi-
cal sense (e.g. for SIZE, AGE, etc.).
3 Description logics language:SHIQ
DL are a family of logical formalisms that origi-
nated in the  eld of arti cial intelligence as a tool
for representation of conceptual knowledge. Since
then, DLs have been successfully used in a wide
range of application areas such as knowledge repre-
sentation, reasoning about class-based formalisms
(e.g. conceptual database models and UML dia-
grams), and ontology engineering in the context of
the semantic web. The basic syntactic entities of DL
are concepts, which are constructed from concept
1In this paper when we say a concept name in plural we
are refering to this concept and his children, using links IS-A
de ned in the ontology.
names (unary predicates) and role names (binary
predicates) using the set of concept and role con-
structors provided by a particular DL (Lutz, 2003).
Our interest in Mikrokosmos ontology is to map
its contents to a DL language. We have chosen
ALCQHIR+ also known as SHIQ (Horrocks et
al., 2000).
SHIQ is the basic logic ALC augmented with
qualifying number restrictions, role restrictions,
role hierarchies, inverse roles, and transitive roles.
ALC comprises concepts  denoting sets as
well as roles  denoting binary relations. Unlike
roles, concepts can be compound. Compound con-
cepts are constructed by the following operators: in-
tersectionu, uniont, complementation: taking
concepts as arguments , and the value restrictions
8, and 9  taking a role and a concept as their ar-
guments. Formally,ALC is given by the following
formation rules, where c denotes a concept symbol
and r a role symbol (Schild, 1991):
C;D !cj>jCuDj:Cj8R:C
R !r
DL SHIQ is implemented in the RACER sys-
tem (Haarslev and Moller, 2003). This makes it a
desirable target representation for our ontology. For
describing our ontology in SHIQwe will use the
notation explained in Table 2, that contains denota-
tional semantics for our language translation.
4 Mikrokosmos mapping toSHIQ
Once we have identi ed DL language we want
to use  SHIQ and we have described the
Mikrokosmos ontology, we can proceed to map it.
The  rst step is to determine whether a concept
is a class or a slot. Although in the Mikrokosmos
ontology everything is a concept we need to dis-
tinguish between Mikrokosmos concepts that cor-
respond to unary predicates  which map to DL
classes and Mikrokosmos concepts that corre-
spond to binary predicates  which map to DL rela-
tions. EVENT, OBJECT and all of their subclasses
will be unary predicates so they will be classes.
Meanwhile PROPERTY and all its hierarchy ex-
cept ONTOLOGY-SLOTs (see Figure 1) will be bi-
nary predicates so they will be slots. There are
a few exceptions: concept ALL is top in DL and
ONTOLOGY-SLOT and all of their subclasses are
not mapped to DL language because they have the
2 (C) is the interpretation of a concept. Interpretation of a
concept is the set of all individuals in the domain that satis es
description of the concept.
class-def (primitivejde ned) CN CN(vj:=)>
subclass-of C1:::Cn u 2(C1)u:::u (Cn)
slot-constraint1 u (slot-constraint1)
... ...
slot-constraintm u (slot-constraintm)
topjthingjbottom Ct:CjCt:CjCu:C
(C1 and ::: and Cn) ( (C1)u:::u (Cn))
(C1 or ::: or Cn) ( (C1)t:::t (Cn))
(not C) (: (C))
(one-of i1 ::: in) (Pi1 t:::tPin)
slot-constraint SN >
has-value C1:::Cn u9SN: (C1)u:::u9SN: (Cn)
value-type C1 :::Cn u8SN: (C1)u:::u8SN: (Cn)
max-cardinality n C u nSN: (C)
min-cardinality n C u nSN: (C)
cardinality n C u nSN: (C)u nSN: (C)
has- ller d u9SN: (d)
slot-def SN
subslot-of SN1:::SNn (SNvSN1):::(SNvSNn)
domain C1:::Cn 9SN:>v (C1)u:::u (Cn)
range C1:::Cn >v8SN: (C1)u:::u (Cn)
inverse RN (SN vRN)(RN vSN)
properties transitive SN2S+
properties symmetric (SNvSN )(SN vSN)
properties functional >v 1SN
disjoint C1 C2:::Cn ( (C1)v: (C2))
covered C by C1 :::Cn  (C)v (C1)t:::t (Cn)
disjoint-covered C by C1:::Cn ( (C1)v: (C2))
( (C)v (C1)t:::t (Cn))
equivalent CC1 :::Cn ( (C) =  (C1)):::( (Cn 1) =  (Cn))
instance-of iC1:::Cn Piv (C1)u:::u (Cn)
related SNij Piv9SN:Pj
Table 2: Denotational semantics for language de nition
sense of structuring the ontology. ONTOLOGY-
SLOT and all of their subclasses encode the struc-
ture of the Mikrokosmos ontology. They are not
mapped as DL classes or slots. Instead they are in-
corporated into the DL de nition of the Mikrokos-
mos concepts that they refer to.
Mikrokosmos has some information that can not
be mapped to a DL language. We will address this
problem in two ways. First we will make some an-
notations to class and slots that are not supported
by DL language, but which could be provided by
RDFS based languages. Second, extra information
about slots that is not supported by DL language
will be stored in special concepts created from the
corresponding slots.
4.1 Building DL classes
Now we will discuss how we extract information
stored in the XML based  le to build classes in DL
language.
The information that has to be extracted is:
class-def (primitivejde ned) CN
subclass-of C1:::Cn
slot-constraint1
...
slot-constraintm
Having identi ed the set of DL classes we need
to identify their superclasses and slot-constraints.
Information about superclasses is encoded in XML
records of the form shown in Figure 2. Additional
sources of information about superclasses  such as
RECORDs where CN appears as FILLER and SUB-
CLASSES appears as SLOT actually encode re-
dundant information and are therefore discarded.
<RECORD>
<CONCEPT>CN</CONCEPT>
<SLOT>IS-A</SLOT>
<FACET>VALUE</FACET>
<FILLER>Ci </FILLER>
</RECORD>
Figure 2: XML encoding of superclass information
Information about slot-constraints is encoded in
records having PROPERTYs as a slot. But there are
also some ONTOLOGY-SLOT used in class de ni-
tion and we will assign them a representation.
We collect information about slot-constraints
from XML records of the form shown in Figure 3:
<RECORD>
<CONCEPT>CN</CONCEPT>
<SLOT>SN</SLOT>
<FACET>FACET </FACET>
<FILLER>C</FILLER>
</RECORD>
Figure 3: XML encoding for slot-constraints
We obtain different information depending on the
value of FACET
 If FACET = DEFAULT-MEASURE
CN slot-constraint SN value-type C is added to
the corresponding class de nition.
 If FACET = DEFAULT. This information is
stored as an annotation
 If FACET = INV. This information comes from
another slot, that it is inverse to SN. There is no
need to handle this information here because
DL has automatic handling for such type of in-
formation.
 If FACET = NOT. This entry appears when we
restrict inheritance of one SLOT in the hier-
archy. Information contained in Mikrokosmos
about these is af rmative information and neg-
ative information, DL only uses af rmative in-
formation to handle it, so we do nothing with
this information.
 If FACET = RELAXABLE-TO. This informa-
tion is stored as an annotation
 If FACET = SEM
CN slot-constraint SN value-type C is added.
 If FACET = VALUE
CN slot-constraint SN has-value C is added.
Additional information encoded in terms of
records with ONTOLOGY-SLOTS  as slots ,
must be handled and incorporated into the corre-
sponding class de nitions.
The ONTOLOGY-SLOTs to be identi ed are
DEFINITION, SPANISH1 and ENGLISH1.
 If SLOT = DEFINITION. We will make an an-
notation in class de nition.
 If SLOT = SPANISH1 or ENGLISH1. We
create two SLOTs called SPANISH1 and EN-
GLISH1, so we can assert:
slot-constraint ENGLISH1 has- ller d. 3
4.2 Building DL relations
Information required to build DL relations is en-
coded in XML records with ONTOLOGY-SLOTS
in their SLOT  eld of the form shown in Figure 4
<RECORD>
<CONCEPT>SN</CONCEPT>
<SLOT>SLOT</SLOT>
<FACET>FACET</FACET>
<FILLER>X</FILLER>
</RECORD>
Figure 4: XML encoding of slot information
Possible relevant  llers of the ONTOLOGY-
SLOTS are:
 DEFINITION, IS-A and SUBCLASSES: This
information is handled for DL relations in the
same way as for DL classes.
 INVERSE: It can be used with SEM and
VALUE FACET and represents inverse slots.
slot-def SN inverses X is added.
 DOMAIN: As before when there is a restric-
tion in inheritance Mikrokosmos asserts af r-
mative and negative information so there is
a FACET NOT that is rejected, and has no
translation to DL language. There are more
possibilities for  lling the FACET: VALUE,
DEFAULT, RELAXABLE-TO and SEM, we
make no distinction among them:
slot-def SN domain disjoint X1:::Xn is
added.
 RANGE: FACET NOT is treated as above.
When we have other FACETs there are two
possible kinds of FILLERs: CONCEPTS or
numeric ranges. For CONCEPTS
slot-def SN range disjoint X1:::Xn is added.
For numeric range we create a subclass of
Numeric-Range (See Figure 5 and example in
Figure 6).
 MEASURED-IN: This information is consid-
ered the same as RANGE. It can only have
SEM or DEFAULT FACETs.
slot-def SN range X is added.
3These slots encode cross indexing with lexical informa-
tion. Another possible mapping would have been to add them
as instances, but this would result in loss of this cross indexing
information.
class-def primitive Numeric-Range
slot-constraint Left-Range-Margin
max-cardinality 1 int
slot-constraint Right-Range-Margin
max-cardinality 1 int
slot-def Numeric-Left-Margin
range int
slot-def Numeric-Right-Margin
range int
class-def de ned Numeric-Right-Range
subclass-of Numeric-Range
slot-constraint Right-Range-Margin
min-cardinality 1 int
class-def de ned Numeric-Left-Range
subclass-of Numeric-Range
slot-constraint Left-Range-Margin
min-cardinality 1 int
class-def de ned Numeric-Closed-Range
subclass-of Numeric-Right-Range
subclass-of Numeric-Left-Range
Figure 5: Range de nitions
<RECORD>
<concept>VISCOSITY</concept>
<slot>RANGE</slot>
<facet>SEM</facet>
< ller>(<;>; 0 1)</ ller>
<uid>256</uid>
</RECORD>
class-def VISCOSITY
subclass-of Numeric-Range
slot-constraint Left-Range-Margin
has- ller 0
slot-constraint Right-Range-Margin
has- ller 1
Figure 6: Example of range restriction
4.3 Building Mikrokosmos PROPERTYs as
DL classes
As we have seen in previous subsection, not all in-
formation about PROPERTYs can be mapped easily
to slots. Because of that we have decided to include
an extra hierarchy of concepts created from PROP-
ERTYs.
For each slot we create a class that inherits
from CLASS-SLOT called CLASS-<PROPERTY-
NAME>. These classes contain all information
about the PROPERTYs that we could not represent
in a DL relation.
For each SLOT applied to a CONCEPT we will
create a class that inherits from CLASS-SLOT-
CONCEPT called CLASS-<PROPERTY-NAME>-
<CONCEPT-NAME>. These classes have slot-
constraints in order to de ne information not cap-
tured in the respective concept.
With this structure of classes we do not lose any
information about slots and slot-constraints but al-
most all information stored in that way is not useful
for reasoning in current tools like RACER (Haarslev
and Moller, 2001).
5 Evaluation of the translation process
DL provide the way to carry out complex inference
and reasoning tasks. In order to achieve this goal
our DL language is less expressive than Mikrokos-
mos. Among all restrictions in the expressiveness of
DL languages we will mention two. DL languages
are not able to reason with default values for the re-
strictions. And they do not manage  nite domains
such as enumerates or sets.
These differences in expressivity between
Mikrokosmos and our DL language has as a result
some interesting points in the translation process.
There were two possible solutions to this problem.
First one was to discard all information that has not
a direct mapping to our DL language. And second
one  which we have chosen is to make some
arti ces in order to preserve all information, but
obviously we cannot reason with this information.
There are two places where we have made this
kind of arti ces:
 Default values: Mikrokosmos is able of man-
aging default values for restrictions while DL
is not. So we have decided to store it as an
annotation.
 Numeric restrictions: For example in
Mikrokosmos we can restrict the age of a
person to be plus than 0 and minus that 120,
but our DL language is not capable. Because
of that we have created the complex structure
of Numeric-Range concepts.
So we can say that we have no loss of information
in the translation process. But we were incapable to
use all information contained in Mikrokosmos for
reasoning and inference tasks.
6 Applications of Mikrokosmos and future
work
One of the distinguishing features of the origi-
nal Mikrokosmos resources for machine transla-
tion was the explicit isolation between the pseudo-
taxonomical structure used to represent the concepts
on one hand, and the particular lexical information
ALL
EVENT
MENTAL-EVENT
PASIVE-COGNITIVE-EVENT
REMEMBERKNOW
EMOTIONAL-EVENT
ACTIVE-COGNITIVE-EVENT
CONSIDER
SOCIAL-EVENT
CHANGE-EVENT
DIVIDE
ABSTRACT-SOCIAL-ABILITY
ELIMINATE
PHYSICAL-EVENT
APPLY-FORCE
CHANGE-LOCATION
CUTJUMP
CHANGE-POSITION
CLIMB
MOUNT
SALTAR CORTAR
PENSARSABER
WORRY
Figure 7: Mikrokosmos ontology with some instances
that was associated with those concepts for realiza-
tion in a particular language. This peculiarity al-
lowed relatively easy bidirectional translation be-
tween different languages via an intermediate con-
ceptual representation.
Subsequent uses and/or transformations of these
resources must take into account this peculiarity. In
our case, the work carried out so far in transport-
ing the Mikrokosmos ontology to OWL has been
restricted to the part of the ontology concerned
with the conceptual representation. Although this
transformation already opens up avenues of re-
search for knowledge representation for problem
solving (D· az-Agudo and Gonz·alez-Calero, 2002),
the number of useful applications of the results of
this process in the  eld of natural language process-
ing will increase greatly once the corresponding lex-
icons  there are currently versions in Spanish and
English are also transformed into OWL.
For instance, use of this resource provides the
means for intelligently substituting a given word for
a different one - as required for example in our po-
etry generation system (Diaz-Agudo et al., 2002)
during the adaptation of the structure of poem from
the case base to obtain a verse approximation of a
user query. Assuming that a structure such as:
Sabed que en mi perfecta edad y armado
con mis ojos abiertos me he rendido
al ni no que sab·eis ciego y desnudo.
needs to be adapted, and the adaptation requires the
substitution of the verb  sabed for a related one out
of a list of candidates - possibly obtained from the
given user query - such as  pensad ,  cortad and
 saltad . By conuslting the structure of the ontology
 see extract in Figure 7 for illustration the sys-
tem may correctly select  pensad as a preferable
candidate in view of its proximity in the ontology to
the original word.
Our future lines of research in this  eld will focus
in a deeper study of which concepts are primitive
and which ones are de ned. Now we have decided
that all concepts having any restriction are de ned
concepts. This decission was taken in orden to rea-
son with the ontology but it is necessary to examine
it in detail.
7 Conclusions
Mikrokosmos ontology is a rich and extensive
knowledge resource that was developed with a pro-
prietary formalism, with a weak theoretical founda-
tion. We have analysed the contents of the ontology
which have lead us to propose a possible translation
into description logics.
All this effort of understanding Mikrokosmos
ontology and mapping it to a description logics
language has resulted in a concrete implementa-
tion. We have chosen OWL  an RDFS based
language , in its version idOWL DL. This version
implements reasoning using JENA (McBride, 2001)
and the DIG interface (Bechhofer et al., 2003).
There are two inference engines that implement the
DIG interface: RACER and FaCT4. As part of this
implementation we have developed an import plu-
gin for Prot·eg·e 2.0 (See Figure 7).
With this work we can pro t from all the knowl-
edge stored in the Mikrokosmos ontology for other
tasks related to Arti cial Intelligence. These tasks
are Natural Language Processing and Knowledge-
Intensive Case-Based Reasoning. We still have to
translate the Mikrokosmos lexicon in order to fully
exploit the original resource.
Acknowledgements
The  rst author is supported by a FPI Predoctoral
Grant form Universidad Complutense, Madrid. The
4http://dl-web.man.ac.uk/dig/
Figure 8: Screen capture of Prot·eg·e 2.0 with
Mikrokosmos ontology.
work was partially funded by the Spanish Commit-
tee of Science & Technology (TIC2002-01961).

References
Sean Bechhofer, Ralf Moller, and Peter Crowther.
2003. The DIG description logic interface. In
Description Logics 2003, CEUR Workshop Pro-
ceedings.
Sean Bechhofer, Frank van Harmelen, Jim Hendler,
Ian Horrocks, Deborah L. McGuinness, Pe-
ter F. Patel-Schneider, and Andrea Stein. 2004.
Owl web ontology language reference, February.
W3C http://www.w3.org/TR/2004/REC-owl-ref-
20040210/.
Mercedes Garc· a de Quesada. 2001. Estructura
de nicional terminogr ca en el subdominio de
la oncologa cl· nica. In Estudios de Linguistica
Espa nola, volume 14.
Bel·en D· az-Agudo and Pedro Antonio Gonz·alez-
Calero. 2002. CBROnto: a task/method ontol-
ogy for CBR. In S. Haller and G. Simmons, edi-
tors, Procs. of the 15th International FLAIRS’02
Conference (Special Track on CBR), pages 101 
106. AAAI Press.
B. Diaz-Agudo, P. Gerv·as, and P A. Gonzalez-
Calero. 2002. Poetry generation in COLIBRI. In
S. Craw and A. Preece, editors, ECCBR 2002,
Advances in Case Based Reasoning, pages 73 
102. Springer. Lecture Notes in Arti cial Intelli-
gence.
Volker Haarslev and Ralf Moller. 2001. Descrip-
tion of the RACER system and its applica-
tions. In Proceedubgs International Workshop on
Description Logics (DL-2001), pages 132 142,
Stanford, USA.
Volker Haarslev and Ralf Moller, 2003. RACER
User s Guide and Reference Manual Version
1.7.7. Concordia University and Univ. of Appl.
Sciences in Wedel, November. http://www.sts.tu-
harburg.de/ r.f.moeller/racer/racer-manual-1-7-
7.pdf.
I. Horrocks, U. Sattler, and S. Tobies. 2000. Rea-
soning with individuals for the description logic
SHIQ. In David MacAllester, editor, Proceed-
ings of the 17th International Conference on Au-
tomated Deduction (CADE-17), number 1831,
pages 482 496, Germany. Springer Verlag.
Atanas K. Kiryakov, Marin Dimitrov, and
Kiril Iv. Simov. 2001. Ontomap - the
guide to the upper-level. In Proceedings of
SWWS’01, The  rst Semantic Web Work-
ing Symposium, Stanford University, Cal-
ifornia, USA, July 30 - August 1, 2001.
http://www.ontotext.com/publications/swws01.pdf.
E. Lonergan. 2001. Lexical knowledge engineer-
ing: Mikrokosmos revisited. In PACLING2001
- Paci c Association for Computational Linguis-
tics 2001, Kitakyushu, Japan.
C. Lutz. 2003. Description logics with concrete
domains a survey. In Advances in Modal Log-
ics Volume 4. King’s College Publications.
Brian McBride. 2001. Jena: Implementing the rdf
model and syntax speci cation. In Proceedings
of the Second International Workshop on the Se-
mantic Web - SemWeb’2001. Hongkong, China,
May 1, 2001. http://SunSITE.Informatik.RWTH-
Aachen.de/Publications/CEUR-WS/Vol-
40/mcbride.pdf.
Antonio Moreno-Ortiz, Victor Raskin, and Sergei
Nirenburg. 2002. New developments in onto-
logical semantics. In Proceedings of LREC-02,
Spain, June.
Sergei Nirenburg and Victor Raskin, 2004. On-
tological Semantics, chapter 7, pages 159 207.
The MIT Press, September.
S. Nirenburg. 1987. Knowledge-based machine
translation, the cmu approach. In Machine Trans-
lation: theoretical and methodological issues,
Studies in Natural Language Processing, pages
68 69, Cambridge. Cambridge University Press.
Klaus Schild. 1991. A correspondence theory
for terminological logics: preliminary report.
In Proceedings of IJCAI-91, 12th International
Joint Conference on Arti cial Intelligence, pages
466 471, Sidney, AU.
