Reengineering ASFA Thesaurus to OWLPage 1 of 13
Reengineering ASFA Thesaurus to OWL
1. Introduction
The overall objective of the ASFA Partnership1 is to disseminate information on aquatic sciences
and fisheries to the world community, bearing in mind the special interests of developing countries
in this field.
The ASFA Thesaurus2 is an indexing and searching tool. It contains the subject descriptors used to
index the records which are contained in the Aquatic Sciences and Fisheries Abstracts (ASFA)
bibliographic database. ASFA is an abstracting and indexing service covering the world's literature
on the science, technology, management, and conservation of marine, brackish water, and
freshwater resources and environments, including their socio-economic and legal aspects.
ASFA Thesaurus graph currently consists of more than 10,000 nodes. ASFA Thesaurus has a
rather typical structure, made up of descriptors (graph nodes), equivalent terms, and relations
among descriptors (types of edges between nodes: BT, NT, RT, UF) that create a forest structure
(an indirect acyclic graph).
Descriptors are not encoded by means of an ID, but directly via a "preferred" English term, which
makes it originally difficult to achieve multilinguality. An attempt to declare Spanish equivalent
terms has been made (see below), but in the NeOn context a more general scope is required,
which can be reached by mapping ASFA to the RTMS-derived ontologies, which contain
multilingual labels.
When dealing with such resources in an ontology design project, we typically face a reengineering
problem. In the next section, we put the ASFA reengineering problem in the context of NeOn
ontology design methods.
2. The design context
Reengineering data sources for ontology design can take many forms, and there are two main
dimensions, along which these forms can vary. The first syntactic dimension, not addressed in this
section, is the set of reuse and maintainance methods applied to data, for example extracting
ontologies from text or maintaining correspondences between databases and ontologies. The
second, semantic dimension is the set of schema translation methods that are used to interpret the
semantics of the data. A comparison of different approaches to semantic reengineering has been
presented in the NeOn Deliverable D2.2.1 [D221], which also includes examples from ASFA. We
summarize here the basic distinctions.
Thesauri are examples of Knowledge Organizations Systems (KOSes), also called Concept
Schemes in SKOS [MB05]. They usually have a forest-like graph structure, without an explicit
formal semantics. Typical relations that label the edges between the nodes in those graphs include
broader than (BT in ASFA), narrower than (NT), related to (RT), and used for (UF). The nodes
1
http://www.fao.org/fishery/asfa
2
http://www.fao.org/fishery/asfa/8/en
Page 2 of 13NeOn Integrated Project EU-IST-027595
themselves are variously called descriptors (as in ASFA), concepts, etc. SKOS uses concept as a
generic term for nodes.
When reengineering KOSes into ontologies for the semantic web, we need to understand what are
the possible mappings between the (implicit semantics of) their graph structure, and the formal
semantics underlying the graph structure of RDF and OWL models. Deciding on those mappings is
relevant not only to reengineering (T2.2 in NeOn WP2), but also in order to provide multilingual
natural language expressions to ontology elements (T2.4). Forthcoming deliverable D2.3.3 will deal
with these issues.
Declaring the semantics assumed when reengineering KOSes is very relevant to NeOn, since
NeOn use cases, as well as most large organizations all around the world, have been producing
rich KOSes in order to encode distinctions that are currently being made in the form of ontologies
over the Semantic Web. ASFA is one such case.
In previous semantic web work, KOSes have been considered acceptable ontologies. However,
the minimal requirement to use them on the Semantic Web is to reengineer them as OWL
resources, with proper structure and namespace. Therefore, the difficulty in establishing if a KOS is
an ontology or not can be overcome within a reengineering perspective: what counts is not if they
are ontologies or not, but rather how they can be reengineered and exploited by semantic
technologies.
In [D221], we have provided two patterns for KOS reengineering: (1) converting the KOS node and
edge types (i.e. its schema) with the semantics of a description logic TBox, and the KOS nodes
and edges (tokens) with the semantics of a description logic ABox (Fig. 1); versus (2) the KOS
node and edge types with the semantics of OWL metamodel, and the KOS nodes and edges with
the semantics of a description logic TBox (Fig. 2).
Fig. 1: ABox KOS reengineering pattern, e.g. asfa:Descriptor is converted as [
asfa:Descriptor rdf:type owl:Class ], asfa:Dredging is converted as [
asfa:Dredging rdf:type asfa:Descriptor ]
Fig. 2: TBox KOS reengineering pattern, e.g. asfa:Descriptor is converted as [
asfa:Descriptor owl:equivalentTo owl:Class ], asfa:Dredging is converted as [
asfa:Dredging rdf:type owl:Class ]
The ABox pattern leaves the informal semantics of the reengineered resources mostly
“untouched”; on the contrary, the TBox pattern tries to enforce a formal semantics to them, even at
the cost of changing their structure. The following examples illustrate the difference:
(a1) asfa:Descriptor rdf:type owl:Class
Reengineering ASFA Thesaurus to OWLPage 3 of 13
(a2) asfa:Dredging rdf:type asfa:Descriptor
(t1) asfa:Descriptor owl:equivalentTo owl:Class
(t2) asfa:Dredging rdf:type owl:Class
In (a1), the ABox reengineering pattern is applied to ASFA schema: a mapping axiom is added
which states that asfa:Descriptor is an instance of owl:Class. In (a2), a particular
asfa:Descriptor, asfa:Dredging, is said to be an instance of asfa:Descriptor. In
formal semantic terms, ASFA schema has been considered as the “vocabulary” of ASFA, while its
nodes as the “domain” of ASFA.
In (t1), the TBox reengineering pattern is applied to ASFA schema: a mapping axiom is added
which states that asfa:Descriptor is equivalent to owl:Class. In (t2), a particular
asfa:Descriptor, asfa:Dredging, is said to be an instance of owl:Class. In formal
semantic terms, ASFA schema has been aligned to the OWL “metamodel”, and ASFA nodes have
been considered as the “vocabulary” of ASFA. The “domain” of ASFA is considered in this case as
the interpretation of ASFA nodes as formal classes: e.g. Dredging being interpreted as the set of
all dredging activities actually occurring in the fishery world.
It is straightforward to see that the interpretation exemplified by the ABox pattern is actually the
interpretation of the KOS, not of the fishery world: as such, it does not imply any (even plausible)
change in the original semantics, With the TBox pattern, things are more troublesome: how can we
be sure that fishery experts actually intend something that corresponds to a formal (extensional)
interpretation of ASFA nodes as classes of real world activities? With the TBox pattern, we are
changing the semantics of the KOS.
This change can be cost-effective or not. While it is beneficial to have a rigorous extensional
semantics associated with ontologies, pragmatically it can be the opposite: experiences in FOS
[GFK+04] and NeOn, indicate that when a large organization has a lot of conceptual structures
already represented and maintained as KOSes, it is technically and socially difficult to impose a
TBox reengineering pattern. On the contrary, the broad success of SKOS [MB05] indicates the
practical advantage of the ABox reengineering pattern.
SKOS (Simple Knowledge Organization Systems, [MB05]), is the most used metamodel to
reengineering thesauri as OWL or RDF(S) models. SKOS has been thought primarily as an
RDF(S) vocabulary, because it is intended to represent thesaurus nodes as RDF descriptions
(called “concepts”), which are on their turn asserted to be rdf:type rdfs:Class. This ambiguity
cannot be expressed in OWL-DL, thus preventing the full capability of OWL-DL reasoners such as
Fact++ or Pellet. Representing KOSes in SKOS can be considered as one variety of the ABox
KOS reengineering pattern. An example of SKOS conversion for ASFA is the following:
(s1) asfa:Descriptor rdfs:subClassOf skos:Concept
(s2) skos:Concept rdf:type rdfs:Class
(s3) asfa:Dredging rdf:type asfa:Descriptor
It is basically an ABox pattern, but the intermediate class skos:Concept introduces a novelty:
asfa:Descriptor is no more rdf:type owl:Class, as in (a1), but rdfs:subClassOf
skos:Concept,
and
(s2)
skos:Concept
rdf:type
rdfs:Class.
Therefore,
asfa:Descriptor rdf:type rdfs:Class (because of inheritance). rdfs:Class, differently
from owl:Class, is not constrained to one specific logical layer, and can be either a class or a
metaclass.
SKOS reengineering pattern can therefore be used to say consistently the following axioms:
(s4) asfa:Dredging rdf:type asfa:Descriptor
(as in (a2))
(s5) asfa:Dredging rdf:type rdfs:Class
Page 4 of 13NeOn Integrated Project EU-IST-027595
(even more generally than in (t2), since owl:Class rdfs:subClassOf rdfs:Class)
In other words, SKOS reengineering pattern allows in principle to reengineer KOSes with both
ABox and TBox patterns. However, in practice this is not really true, since it generates an OWLFull ontology, which limits the applications of reasoning engines for OWL-DL.
On the reasoning side, we should notice that the ABox pattern is obviously weaker, since the BT
relation alone, even if reengineered as an owl:TransitiveProperty as in
skos:broaderGeneric, does not ensure the powerful inheritance reasoning available with the
TBox pattern. Recent developments of HiDL-Lite [HDL] reasoning are anyway promising enough to
stick in favor to the ABox pattern, without requiring the SKOS “two-layered” pattern.
A further hybrid pattern that makes use of a more complex reengineering procedure is the
following: firstly (ABox reengineering pattern), a KOS is ABox-reengineered by aligning it to a
metamodel called LMM, with explicit mappings to possible formal interpretations; secondly (TBox
interpretation pattern), it can be customly reengineered as a TBox by declaring what mappings to
what formal interpretations are required.
In the next sections, we present previous ASFA reengineering attempts (sect. 3), the reengineering
procedure and its alignment to SKOS, and to LMM (sect. 4), as well as some plans for alignments
to RTMS-derived ontologies and modularization (sect. 5).
3. The ASFA reengineering background
After realizing the need to reengineer ASFA as a Fishery ontology in NeOn, we have evaluated the
state of play of previous reengineering efforts.
3.1 FOS reengineering
In the 2003 FOS project [GFK+04], a TBox reengineering pattern was used, leading to a
DAML+OIL ontology (later migrated to OWL) of 11484 classes, where the BT relation was simply
reengineered as a rdfs:subClassOf relation, for example:
(t3) asfa:Decompression_chambers rdfs:subClassOf asfa:Diving_equipment
sometimes this leads to counterintuitive axioms, e.g.:
(t4) asfa:Blood_vessels rdfs:subClassOf asfa:Circulatory_system
In (t4), the extensional semantics resulting from the reengineering leads to the counterintuitive
interpretation that all blood vessels are circulatory systems, while the intended meaning of ASFA
experts was probably intensional, i.e. that blood vessels are part of the circulatory system. (t4)
axiom also shows an example of ambiguity arising with this kind of reengineering:
asfa:Blood_vessels has to interpreted as the class of individual blood vessels (since plural
names in thesauri do not necessarily refer to collection entities), or as the class of collections of
blood vessels in an organism? The TBox pattern forces to take one of these interpretations, while
the ABox pattern does not.
In the FOS project, after ASFA TBox reingeneering, about 1600 ASFA descriptors that had no BT
relation (“orphans”) have been semi-automatically aligned to external ontologies: other fishery
ontologies when possible (e.g. fos-economy, fos-techniques), otherwise OntoWordNet (with prefix
own in the example) and DOLCE (with prefix dul in the example), and validated with the help of a
fishery expert, Ian Pettman (see sect. 3.2). Some examples include:
(t6) asfa:Agropisciculture rdfs:subClassOf fostechniques:Aquaculture
(t7) asfa:Fish:utilization rdfs:subClassOf foseconomy:PostHarvestUse
(t8) asfa:Entrainment rdfs:subClassOf own:Movement__Motion
Reengineering ASFA Thesaurus to OWLPage 5 of 13
(t9) asfa:Energy_transfer rdfs:subClassOf dul:Process
The problem of orphans is relevant also in ABox-reengineered versions. In the current version,
there are still 1482 orphan descriptors.
3.2 Pettman’s reengineering
In 2008, Ian Pettman from OneFish has produced two new OWL versions of ASFA, by performing
firstly a TBox-reengineering, and afterwords a SKOS alignment.
Pettman’s TBox is basic TBox reengineering, inspired by the FOS one, although contains some
syntactic glitches and namespace issues. The author of the reengineering was also aware of the
“intensional” interpretations existing in ASFA.
Since this version of ASFA is the last and the most correct one from the ASFA partnership, it has
been used to produce the NeOn version (see sect. 4).
3.3 Aquaring’s reengineering
Pettman’s version was created also on demand from the EU Aquaring project. As reported by
Marta Rodriguez, who works in the project, these are some insights in their attempt to reengineer
ASFA in OWL:
– Aquaring uses ASFA ontology (among other ontologies) to semantically annotate resources
– The ASFA ontology used is one programmatically created from a XML file sent by Ian Pettman.
The problem personally found with that ontology was the multilingual support compatibility with
the other ontologies we had
– Aquaring version considers all descriptors as instances of the Descriptor class
– The BT, RT, USE were transformed in properties: hasParent, related_to and synonym
– They found a problem with this approach: BT and RT cannot be always transformed to
hasParent and related_to concept idea
– They extracted 1715 parent terms with no BT, and started aligning them under ASFIS subject
categories classification elements, but they gave up after 139 alignments
3.4 Jacques’ requirement analysis
Within WP7, Yves Jacques has managed to collect a set of requirements for the reengineering of
ASFA. He interviewed the ASFA team and also considered FAO needs for interfacing with the
ASFA thesaurus, and came up with the following requirements. There were a few question marks
where he did not receive input from ASFA, but by and large they were supposed to be correct.
Scope
The thesaurus is currently used in two central ways:
1. To index ASFA records with standard keywords.
2. To search for ASFA records with standard keywords.
Maintenance
-- Editing rights for the thesaurus are held by a group of approx. six people from various
organizations, geographically dispersed
Page 6 of 13NeOn Integrated Project EU-IST-027595
-- ASFA is looking for a new way in which to manage the thesaurus, so that it can be updated more
frequently (an ontological OWL format could be a choice)
-- Digital storage/backup are needed
-- Web-based access is needed
-- Versioning is needed
-- User access rights are needed
-- User roles are needed, e.g. editor, validator, etc.
-- Locking during editing is needed
-- A draft (limited access) version is needed that can when validated replace the current public
version
Output
The thesaurus output is currently in print form and also accessible via two different web interfaces.
These outputs will continue to be needed:
1. The thesaurus publisher receives an XML (RDF?) version of the file for implementation in their
web site:
http://www.csa.com/factsheets/supplements/asfathes.php
2. FAO holds the thesaurus in a relational database
http://www4.fao.org/asfa/asfa.htm
3. Other third parties also provide the thesaurus in various interfaces:
http://www.thesaurusbuilder.com/asfa/index.asp?action=dsplsrch
http://www.pmel.noaa.gov/foci/sebscc/special_issue/csa_thesaurus/ab.html
Modelling
The thesaurus needs to maintain at a minimum its current expressivity as found in Ian Pettman's
OWL version: use, used for, status, audit dates (there are multiple, see OWL file), broader term,
narrower term, related term.
The thesaurus is multi-lingual [so that seems supported in the new version]
The model should be comprehensible to anyone with an indexing background.
Integration
It would be good if thesaurus terms could be mapped to other ontologies, e.g. AGROVOC (general
FAO need)
Non-functional requirements
Performance
Can be opened and edited on a standard spec Wintel computer (Pentium 4, 2.00 ghz, 1gb RAM)
without significant delay. Can open ontology in 0-5 seconds.
Dependencies
Not dependent on outside data sources or models except where these are standard, e.g. SKOS.
Reengineering ASFA Thesaurus to OWLPage 7 of 13
Standards
ASFA is modelled after ISO 2788:
http://www.collectionscanada.gc.ca/iso/tc46sc9/standard/2788e.htm
It would be good that the thesaurus is in or can be converted to a standard model, e.g. SKOS:
http://www.w3.org/2004/02/skos/
It should be expressed in a standard markup language, e.g. OWL, RDF
Modularity
It is desirable that the thesaurus can be easily split into thematic modules, e.g. Biological Sciences
and Living Resources, Ocean Technology, Policy and Non-Living Resources, Aquatic Pollution and
Environmental Quality, Aquaculture, Marine Biotechnology.
3.5 Conclusions on background section
Most lessons learnt and requirements (e.g. efficiency, standards) obtained go to the direction of an
ABox reengineering, with an alignment to SKOS as described in the next section. Mapping and
modularity seem also have a high priority, and examples are provided here for future guidance.
4. Applying the ABox reengineering pattern to ASFA
Two versions of ASFA ABox KOS reengineering pattern have been produced. The first one adopts
the pure ABox pattern, the second aligns the first to SKOS. Further alignment to LMM is provided
by the skos2lmm module (see below).
All versions are based on two namespaces:
http://www.xxx.org/asfam.owl
http://www.xxx.org/asfad.owl
where www.xxx.org is a placeholder for the namespace that will be indicated by ASFA partnership.
asfam.owl is the file with the ASFA schema in OWL; asfad.owl is the file with the complete ASFA
data (descriptors, non-descriptors, and relations between them).
As a temporary solution, the current versions are downloadable from the following CNR URIs. This
one if for the reengineered ASFA schema:
http://www.ontologydesignpatterns.org/ont/fao/asfa/asfam.owl
This is for the reengineered ASFA instances:
http://www.ontologydesignpatterns.org/ont/fao/asfa/asfad.owl
This is for the aligned ASFA schema to SKOS:
http://www.ontologydesignpatterns.org/ont/fao/asfa/asfam2skos.owl
This is for additional alignments of ASFA schema to LMM:
http://www.ontologydesignpatterns.org/ont/fao/asfa/asfam2lmm.owl
The following import axioms hold:
asfad.owl owl:imports asfam.owl
asfam2skos.owl owl:imports asfam.owl
asfam2lmm.owl owl:imports asfam2skos.owl
Page 8 of 13NeOn Integrated Project EU-IST-027595
Changes with versions until now, issues arisen, and things to do are all annotated in the annotation
spaces of the ontologies.
4.1 The basic schema after applying the ABox KOS reengineering pattern
The ASFA schema has been fully reengineered in OWL as follows (Fig. 3):
Fig. 3. The ASFATerms class and subclasses in asfam.owl.
ASFATerms has been represented as:
(1) asfam:ASFATerms rdf:type owl:Class
(2) asfam:ASFATerms rdfs:subClass skos:Concept
ASFA Descriptor has been represented as:
(3) asfam:Descriptor rdf:type owl:Class
(4) asfam:Descriptor rdfs:subClass asfam:ASFATerms
ASFA NonDescriptor similarly as ASFA Descriptor
BT has been represented as:
(5) asfam:BT rdf:type owl:TransitiveProperty
(6) asfam:BT rdfs:subPropertyOf skos:broader
NT has been represented as:
(7) asfam:NT rdf:type owl:TransitiveProperty
(8) asfam:NT rdfs:subPropertyOf skos:narrower
asfam:NT owl:inverseOf asfam:BT
relatedTerm has been represented as:
(9) asfam:relatedTerm rdf:type owl:SymmetricProperty
(10) asfam:relatedTerm rdfs:subPropertyOf skos:related
USE has been represented as:
(11) asfam:use rdf:type owl:ObjectProperty
USED FOR has been represented as:
(12) asfam:usedFor rdf:type owl:ObjectProperty
(13) asfam:usedFor owl:inverseOf asfam:use
Reengineering ASFA Thesaurus to OWLPage 9 of 13
Additional attributes for term management, such as input date, approved date, audit date, etc. have
been represented as instances of owl:DatatypeProperty on the class asfam:ASFATerms.
4.1 The basic schema alignment to SKOS
Alignment to SKOS has been made straightforwardly by declaring rdfs:subClassOf axioms:
(14) asfam:ASFATerms rdfs:subClassOf skos:Concept
(15) asfam:BT rdfs:subPropertyOf skos:broader
(16) asfam:NT rdfs:subPropertyOf skos:narrower
(17) asfam:relatedTerm rdfs:subPropertyOf skos:related
(18) asfam:use rdfs:subPropertyOf skosmapping:exactMatch
(19) asfam:usedFor rdfs:subPropertyOf skosmapping:exactMatch
Fig. 4. The ASFATerms class and subclasses in asfam.owl as aligned to SKOS.
Page 10 of 13NeOn Integrated Project EU-IST-027595
4.2 The SKOS alignment to LMM
SKOS has been independently aligned to LMM (Linguistic MetaModel) [PGG][Gan09], which is
also aligned to many other resources (Wordnet, FrameNet, LMF, DBpedia, etc.), as well as to the
codolight ontology design metamodel, used in order to integrate the different models and tool
descriptions developed in NeOn.
The advantages of aligning SKOS (and then ASFA via inheritance) to LMM include:
(a) the smooth navigation between heterogeneous KOSes, lexica, and relevant knowledge
resources across the Semantic Web
(b) the custom reengineering of fragments (modules) from asfad.owl into regular TBox domain
ontologies (an example is included below in sect. 4.3).
More details on the NeOn reengineering methods built on top of LMM will be presented in
deliverable D2.3.3.
The basic LMM vocabulary has the following namespace:
http://www.ontologydesignpatterns.org/ont/lmm/LMM_L1.owl
The SKOS alignment to LMM contains the following axioms, which hold also for the ASFA classes
and properties that are rdfs:subClassOf or rdfs:subPropertyOf of respective SKOS ones:
(20) skos:Concept rdfs:subClassOf lmm1:Meaning
(21) skos:semanticRelation3 rdfs:subPropertyOf lmm1:relatedMeaning
(22) skos:broader rdfs:subPropertyOf lmm1:specializes
(23) skos:narrower rdfs:subPropertyOf lmm1:isSpecializedBy
(24) skosmapping:exactMatch rdfs:subPropertyOf lmm1:relatedMeaning
3
skos:broader, skos:narrower, and skos: related are rdfs:subpropertyOf skos:semanticRelation
Reengineering ASFA Thesaurus to OWLPage 11 of 13
Fig. 5. The ASFATerms class and subclasses in asfam.owl as aligned to LMM.
4.3 Examples from asfad.owl
We include here some examples from asfad.owl, which contains ASFA terms and their relations;
see also Fig. 6:
(25) asfad:Tidal_barrages rdf:type asfam:Descriptor
(25) asfad:Tidal_barrages asfam:BT asfad:Barrages
(26) asfad:Tidal_barrages asfam:RT asfad:TidalPower
(25) asfad:Farm_ponds rdf:type asfam:NonDescriptor
(27) asfad:Farm_ponds asfam:use asfad:Fish_ponds
(28) asfad_Undercurrents asfam:NT asfad:Water_currents
Page 12 of 13NeOn Integrated Project EU-IST-027595
Fig. 6. An example of a descriptor from asfad.owl with BT, NT, and RT relations.
Reengineering ASFA Thesaurus to OWLPage 13 of 13
5. Ongoing and future work
Future work on ASFA includes reusing the alignment performed that was made in FOS (cf. sect.
3.1), in order to provide lmm1:specializes axioms to the top 1482 descriptors from last version.
Liaisons with the Aquaring project can be envisaged at this stage.
It would be very interesting to experiment with alignment techniques from USFD, INRIA, and OU.
The alignment made in FOS could be in this case considered as a gold standard, in order to test
different automatic alignment techniques.
Once fully aligned, ASFA will require some modularization, e.g. following the RTMS architecture
distinguishing between species, resources, vesseltypes, geartypes, techniques, geographic
entities, etc. In the FOS project, the alignment induced some modularization that can also be
reused.
Another area of improvement is multilinguality: from requirement analysis (sect. 3.4) it seems that
the current version is multilingual, but the Pettman’s version only contains English labels. A
possible multilingual extension can be provided after full alignment to RTMS-derived ontologies.
The integration of a LIR extension (cf. D.2.4.1, T2.4) would then be highly beneficial.
References
[HDL] G. De Giacomo, M. Lenzerini, R. Rosati. Towards Higher-Order DL-Lite. Proc. of DL2008, 2008.
[Gan09] A. Gangemi. What's in a Schema. In C.R. Huang and N. Calzolari and A. Gangemi and A. Lenci and
A. Oltramari and L. Prevot (eds.): Ontologies and the Lexicon, Cambridge University Press, Cambridge, UK,
2009.
[GFK+04] A. Gangemi, F. Fisseha, J. Keizer, I. Pettman, and M. Taconet. A Core Ontology of Fishery and its
use in the Fishery Ontology Ser vice Project. In First International Workshop on Core Ontologies, EKAW
Conference, CEUR-WS, volume 118, 2004.
[MB05] A. Miles and D. Brickley. SKOS Core Vocabulary Specification. Technical report, World Wide Web
Consor tium (W3C), November 2005. http://www.w3.org/TR/2005/WD-swbp-skos-core-spec-20051102/
[D221] M. Sabou et al. D2.2.1Methods for Selection and Integration of Reusable Components from Formal
or Informal User Specifications. NeOn Project Deliverable, available at (2007).
[PGG] D. Picca, A. Gangemi, A. Gliozzo. LMM: an OWL Metamodel to Represent Heterogeneous Lexical
Knowledge. Proc. of the International Conference on Language Resources and Evaluation (LREC)
Marrakech, Morocco, 2008.