An Incremental Method for the Lexical Annotation of Domain Ontologies
Sonia Bergamaschi, Laura Po, Maurizio Vincini
DII - Universit`a di Modena e Reggio Emilia, via Vignolese 905, Modena, Italy
Paolo Bouquet, Daniel Giacomuzzi
DIT - Universit`a di Trento, Via Sommarive, 14, Trento, Italy
Francesco Guerra
DEA - Universit`a di Modena e Reggio Emilia, v.le Berengario 51, Modena, Italy
Semantic integration, automatic lexical annotation, Semantic Web, Ontology matching.
Abstract: In this paper, we present MELIS (Meaning Elicitation and Lexical Integration System), a method and a
software tool for enabling an incremental process of automatic annotation of local schemas (e.g. relational
database schemas, directory trees) with lexical information. The distinguishing and original feature of MELIS
is its incrementality: the higher the number of schemas which are processed, the more background/domain
knowledge is cumulated in the system (a portion of domain ontology is learned at every step), the better the
performance of the systems on annotating new schemas.
MELIS has been tested as component of MOMIS-Ontology Builder, a framework able to create a domain on-
tology representing a set of selected data sources, described with a standard W3C language wherein concepts
and attributes are annotated according to the lexical reference database.
We describe the MELIS component within the MOMIS-Ontology Builder framework and provide some ex-
perimental results of MELIS as a standalone tool and as a component integrated in MOMIS.
The growth of information available on the Internet
has required the development of new methods and
tools to automatically recognize, process and man-
age information available in web sites or web-based
applications. One of the most promising ideas of
the Semantic Web is that the use of standard formats
and shared vocabularies and ontologies will provide a
well-defined basis for automated data integration and
reuse. However, practical experience in developing
semantic-enabled web applications and information
systems shows that this idea is not so easy to imple-
ment. In particular, we stress two critical issues: on
the one hand, building an ontology for a domain is a
very time consuming task, which requires skills and
competencies which are not always available in orga-
nizations; and, on the other hand, there seems to be an
irreducible level of semantic heterogeneity, which has
to do with the fact that different people/organizations
tend to use “local” schemas for structuring their data,
and ontologies if available at all are often designed
to fit locally available data rather than aiming at be-
ing general specifications of domain knowledge. The
consequence is a situation where data are organized
to comply to some local schema (e.g. a relational
schema, or a directory tree) and no explicit (formal)
ontology is available; or – if an ontology is available
– it is tailored on local data/schemas and therefore of
little use for data integration.
The two issues above led the Semantic Web
and Database communities to address two very
hard problems: ontology learning (inducing on-
tologies from data/schemas) and ontology match-
ing/integration (bridging different ontologies). For
our argument, we only need to observe that several
methods and tools developed to address the two prob-
lems rely in different ways on the use of lexi-
cal information. The reason is simple: beyond the
syntactic and semantic heterogeneity of schemas and
ontologies, it is a fact that their elements and prop-
erties are named using natural language expressions,
and that this is done precisely because they bring in
useful (but often implicit) information on the intended
Bergamaschi S., Po L., Vincini M., Bouquet P., Giacomuzzi D. and Guerra F. (2007).
MELIS - An Incremental Method for the Lexical Annotation of Domain Ontologies.
In Proceedings of the Third International Conference on Web Information Systems and Technologies - Web Interfaces and Applications, pages 240-247
DOI: 10.5220/0001280902400247
meaning and use of the schema/ontology under con-
struction. Therefore, it should not come as a surprise
that a large number of tools for ontology learning and
schema/ontology matching include some lexical re-
sources (mainly WordNet
) as a component, and use
it in some intermediate step to annotate schema el-
ements and ontology classes/properties with lexical
knowledge. To sum up, lexical annotation seems to
be a critical task to develop smart methods for ontol-
ogy learning and matching.
In this context, we developed MELIS (Meaning
Elicitation and Lexical Integration System), a method
and a software tool for the annotation of data sources.
The distinguishing feature and the novelty of MELIS
is its incremental annotation method: the more
sources (including a number of different schemas) are
processed, the more background/domain knowledge
is cumulated in the system, the better the performance
of the systems on new sources. MELIS supports three
important tasks: (1) the source annotation process,
i.e. the operation of associating an element of a lexi-
cal reference database (WordNet in our implementa-
tion, but the method is independent from this choice)
to all source elements, (2) the customization of the
lexical reference with the introduction of new lexi-
cal knowledge (glossa, lemma and lexical relation-
ships), and (3) the extraction of lexical/semantic re-
lationships across elements of different data sources.
Works related to the issues discussed in this paper
are in the area of languages and tools for annotations
((Bechhofer et al., 2002), (Staab et al., 2001) and
(Handschuh et al., 2003) where an approach similar
to our is adopted), techniques for extending WordNet
((Gangemi et al., 2003), (Montoyo et al., 2001) and
(Pazienza and Stellato, 2006) where a system coupled
with Prot`eg`e
for enriching and annotating sources
is proposed), and systems for ontology management
(see the the Ontoweb
and the Knowledgeweb Net-
work of Excellence
technical reports for complete
In most real world applications, ontology elements
are labeled by natural language expressions. In our
opinion, the crucial reason for this aspect of ontol-
for more in-
formation on WordNet.
3, in particular deliverable 1.4
ogy engineering is the following: while conceptual
annotations provide a specification of how some ter-
minology is used to describe some domain (the stan-
dard role of OWL ontologies), natural language la-
bels (lexical annotations) provide a natural and rich
connection between formal objects (e.g. OWL classes
and properties) and their intended meaning. The in-
tuition is that grasping the intended interpretation of
an ontology requires not only an understanding of the
formal properties of the conceptual schema, but also
knowledge about the meaning of labels used for the
ontology elements. In other words, an OWL ontology
can be viewed as a collection of formal constraints be-
tween terms, whose intended meaning also depends
on lexical knowledge.
In most cases, lexical knowledge is used for an-
notating schema/ontology labels with lexical infor-
mation, typically WordNet senses. However, lexical
annotation is a difficult task, and making it accurate
may require a heavy user involvement. Typical prob-
lems are: coverage (a complete lexical database in-
cluding all possible terms does not exist); polysemy
(in natural language, many terms have multiple mean-
ings); compound terms (schemas and ontologies are
often labeled with compound nominal expressions,
like “full professor”, “table leg”, “football team”, and
the choice of the right lexical meaning often depends
on determining the relationship between terms); in-
tegration (a standard model/language for describing
lexical databases does not exist).
That is why several tools which were developed
for annotating sources only provide a GUI for sup-
porting the user in the manual execution of the task.
However, this manual work can be highly time con-
suming, and very tedious for humans.
MELIS tries to make annotation as automatic
as possible by providing a candidate lexical anno-
tation of the sources as the combination of lexical
knowledge (from WordNet) and domain knowledge
(if available). In addition, MELIS uses the WNEdi-
tor (Benassi et al., 2004) to support customized ex-
tensions of WordNet with missing words and senses.
In the following we describe the MELIS method,
its heuristic rules and the main features of WNEditor.
2.1 The MELIS Method
The way MELIS works is depicted in Figure 1. We
start from a collection of data sources which cover re-
lated domains, e.g. hotels and restaurants. In general
we do not assume that a domain ontology is initially
available, though this may be the case. The process is
a cycle which goes as follows:
1. a schema, which can be already partially anno-
MELIS - An Incremental Method for the Lexical Annotation of Domain Ontologies
Figure 1: Functional representation of MELIS.
tated with lexical information, is given as input to
MELIS, together with a (possibly empty) domain
ontology (called reference ontology in the figure).
Lexical information is extracted from WordNet
which may be extended with words/senses which
are not available by interacting with WNEditor;
2. the automatic lexical annotation process starts; its
output is a partial annotation of schema elements,
together with a list of discovered relationships
across different elements. This annotation, whose
main rules are described below, is obtained by us-
ing two main knowledge sources: WordNet (for
lexical senses and relationships across them), and
the reference ontology, if not empty (it provides
non-lexical domain dependent relationships
across senses, e.g. between “hotel” and “price”).
Pre-existing lexical annotations are not modified,
as they may come either from manual annotation
or from a previous annotation round;
3. the resulting annotated schema is passed to a user,
who may validate/complete the annotation pro-
duced by MELIS;
4. the relationships discovered across terms of the
schema are added to the reference ontology
(which means that an extended – and lexically an-
notated version of the domain ontology is pro-
duced, even if initially it was empty);
5. the process restarts with the following schema, if
any; otherwise it stops.
The process is incremental, as at any round the
lexical database and the reference ontology may be
extended and refined. As we said, the process might
even start with an empty reference ontology, and
the ontology is then constructed incrementally from
2.2 The Rules for Generating New
A crucial part of the process has to do with the rules
which are used to produce the MELIS lexical an-
notations. The core rules are derived from CTX-
MATCH2.0, and are described in previous publica-
tions. However, to improve the precision and recall of
MELIS, we added a few specialized heuristic rules.
In what follows, we use the following notational
Letters: capital letters (A, B, C, ...) stand for
class labels, low case letters (a, b, c, ...) stand
for datatype property labels, letters followed by
“#n” (where n is a natural number)refer to the n-th
sense of the label for which the letter stands (e.g.
b#2 is the second sense of the word occurring in
the label “b”).
Arrows: red arrows denote a subclass relation,
black arrows denote datatype properties, blue ar-
rows denote object properties.
Ontologies: O is used for the ontology to be anno-
tated, DO
for the i-th domain ontology available
for the current elicitation process.
The annotation process takes as input a schema
O and works in two main steps: first, for every label
in O, the method extracts from WordNet all possible
senses for the words composing the label; then, it fil-
ters out unlikely senses through some heuristic rules.
The remaining senses are added as lexical annotation.
Ideally, the system produces just one annotation, but
in general it may be impossible to select a single an-
notation. Below is a general description of the heuris-
tic rules used by MELIS.
Rule 1 Couple class–property: if in O we nd a class
labeled A with a datatype property b, and in some
we find a class annotated as A#i and a datatype
property annotated as b#j, then we conclude that the
annotations A#i and b#j are acceptable candidate an-
notations for A and b in O.
Rule 2 Inheritance parent–child: if in O we find a
class labeled
with a datatype property
, and in
some DO
we find a class annotated as
, with a
datatype property annotated as
and a subclass
In the paper we consider a subclass property both an
object oriented definition and a WordNet hyponym relation.
In WordNet we say that a noun X is a hyponym of a noun
Y if X is less general than Y (X is a specialization of Y);
conversely, we say that X is a hypernym of Y if X is more
general than Y. Other relationships across nouns are: X is
a holonym of Y (Y is a part of X), and X is a meronym of
Y (X is a part of Y). Different relationships are used for
WEBIST 2007 - International Conference on Web Information Systems and Technologies
, then we conclude that the annotations
are acceptable candidate annotations for
in O.
Rule 3 Inheritance child–parent: if in O we find a
class labeled A with a datatype property b, and in
some DO
we find a class annotated as A#i , with a
subclass B#j, and the latter has associated a datatype
property annotated as b#k and i, then we conclude
that the annotations A#i and b#k are acceptable can-
didate annotations for A and b in O.
Rule 4 Inheritance in sibling classes: if in O we find
a class labeled A with a datatype property b, and in
some DO
we find a class annotated as C#k with two
subclasses annotated as A#i and B#j, and there is a
datatype property annotated b#h associated to B#j,
then we conclude that the annotations A#i and b#k
are acceptable candidate annotations for A and b in
Rule 5 Propagation through object properties: if in
O we find a pair of classes labeled A and B, connected
through any object property, and in some DO
we find
a pair of classes annotated as A#i and C#k, and C#k
has a subclass B#j, then we conclude that the anno-
tations A#i and B#j are acceptable candidate annota-
tions for A and B in O.
Rule 6 Inheritance of parent–child relationship: if
in O we find a pair of classes labeled
subclass of
), and in some DO
we find a subclass
hierarchy in which two classes are annotated as
(with none, one or more intermediate classes
in between), then we conclude that the annotations
are acceptable candidate annotations
in O.
When all heuristic rules are applied, then we dis-
charge any candidate pair of annotations which is not
supported by any of the rules above.
2.3 The WNEditor
WNEditor aids the designer to extend WordNet to
coverdomain specific words which are not in the orig-
inal WordNet database.
Since WordNet is distributed as-it-is and external
applications are not allowedto directly modify its data
files, WNEditor addresses two important issues: (i)
providing a physical structure where WordNet and all
its possible extensions are stored and efficiently re-
trieved; (ii) developing a general technique which can
support users in consistently extending WordNet. The
other grammatical types, e.g. for verbs and adjectives. See
for more details.
first issue is technically solved by storing the orig-
inal WordNet (and all its possible extensions) in a
relational database. The second issue is addressed
by giving ontology designers the possibility to in-
sert new synsets (a synset is a new “concept”, which
can be expressed by several words); insert new lem-
mas (based on an approximate string matching algo-
rithm to perform the similarity search on the whole
synset network); inserting new relationships between
synsets (given a source synset, the designer is assisted
in searching for the most appropriate target synset,
see (Benassi et al., 2004) for details).
WordNet extensions may then be exported and
reused in other applications (though they are always
presented separately from the original WordNet, and
the source of the new annotations is always made ex-
We tested the MELIS approach by coupling it with
. MOMIS is a framework that starts from a
collection of data sources and provides a collection of
tools for:
1. semi-automatically building a customized ontol-
ogy which represents the information sources;
2. annotating each source according to the resulting
3. mapping the created ontology and the original
sources into a lexical database (WordNet) to sup-
port interoperability with other applications.
MELIS has been experimented in MOMIS to
show that it can improve the MOMIS methodology
in two main directions: by supporting the semi-
automatic annotation of the original data sources (cur-
rently the process is manually executed), and by
providing methods for extracting rich relationships
across terms by exploiting lexical and domain knowl-
MOMIS provides a double level of annotation for
data sources and the resulting ontology: for each
source, conceptual annotations map the original struc-
ture into a formalized ontology and lexical annota-
tions assign a reference to a WordNet element for
each source term. Moreover, the ontology structure
is formalized by means of a standard model and each
concept is annotated according to a lexical reference.
MELIS inside MOMIS allows a greater automation
See for references
about the MOMIS project.
MELIS - An Incremental Method for the Lexical Annotation of Domain Ontologies
Figure 2: Functional representation of MOMIS and MELIS.
in the process of source annotation, and provides a
way for discovering relationships among sources ele-
Figure 2 shows how the MELIS component is in-
tegrated into the MOMIS architecture. The process
of creating the ontology and defining the mappings
is organized in five step (each task number is corre-
spondingly represented in figure 2) : (1) local source
schema extraction, (2) lexical knowledge extraction
performed with MELIS, (3) common thesaurus gen-
eration, (4) GVV generation, and (5) GVV and local
sources annotation. The following sections describe
the details of these steps.
Local source schema extraction. To enable
MOMIS to manage web pages and data sources,
we need specialized software (wrappers) for the
construction of a semantically rich representations of
the information sources by means of a common data
Lexical knowledge extraction. The extraction of
lexical knowledge from data sources is typically
based on an annotation process aiming at associating
to each source element an effective WordNet mean-
MELIS supports the user in this task by provid-
ing an effective tool for decreasing the boring manual
annotation activity.
Common thesaurus generation. The common the-
saurus is a set of relationships describing inter- and
intra-schema knowledge about the source schemas.
The common thesaurus is constructed through a
process that incrementally adds four types of relation-
ships: schema-derived relationships, lexicon derived
relationships, designer-supplied relationships and in-
ferred relationships.
Schema-derived relationships. The system au-
tomatically extracts these relationships by analyz-
ing each schema separately and applying a heuris-
tic defined for the specific kind of source man-
Lexicon-derived relationships. These relation-
ships, generated by MELIS, represent complex
relationships between meanings of terms anno-
tated with lexical senses. These relationships
may be inferred not only from lexical knowl-
edge (e.g. by querying WordNet for relationships
across senses), but also from background knowl-
edge (e.g. domain ontologies) which are avail-
able at the time of the annotation. As we will say
later (section 2), at any step MELIS can (re)use
any piece of ontology generated by the current ex-
traction process as a source of domain knowledge
to incrementally refine the extraction of new rela-
Designer-supplied relationships. To capture
specific domain knowledge, designers can supply
new relationships directly.
Inferred relationships. MOMIS exploits de-
scription logic techniques from ODB-Tools (Ben-
eventano et al., 1997) to infer new relationships.
GVV generation. The Global Virtual View (GVV)
consists of a set of classes (called Global Classes),
plus mappings to connect the global attributes of each
global class and the local sources attributes. Such a
view conceptualizes the underlying domain; you can
think of it as an ontology describing the sources in-
WEBIST 2007 - International Conference on Web Information Systems and Technologies
GVV and local sources annotation. MOMIS au-
tomatically proposes a name and meanings for each
global class of a GVV (Beneventano et al., 2003)
Names and meanings have to be confirmed by the on-
tology designer. Local sources are conceptually an-
notated according to the created GVV.
We tested MOMIS integrated with MELIS by build-
ing an ontology of a set of data-intensive websites
containing data related to the touristic domain (see
figure 3), which have been wrapped and data from
them have been structured and stored into four rela-
tional databases off-line available. The main classes
of these sources are: hotel (of the “venere” database),
restaurant (“touring” database), camping (guidaC
database) and bedandbreakfast (BB database).
As discussed before, the incremental annotation
process starts with the annotation of parts of the data
sources, i.e. for each source element the ontology
designer selects one or more corresponding WordNet
synsets. For example, WordNet tells us that “hotel”
and “restaurant” are siblings (i.e. they have a common
direct hypernym); that “hotel” , “house”, “restau-
rant” are direct hyponyms of “building”; that “bed
and breakfast” is an hyponym of “building”; and that
the closest hypernym that “campsite” and “building”
share is “physical object”, a top level synset in Word-
Net. As the last relationship does not allow finding
lexical connections between “camping” and the other
classes, we used the WNEditor to add a direct re-
lationships between “campsite” and the hierarchy of
Notice that the annotation process is a critical
process: by annotating the source element “camping”
as the WordNet synset “camping” a mistake would be
generated because it means “the act of encamping”.
The correct synset for camping is “campsite”, i.e. “the
site where people can pitch a tent”. Moreover, in or-
der to test all the implemented heuristics,“hotel” has
been annotated as its hypernym: “building”.
The annotated schema is then given both as in-
put and as the reference ontology to CtxMatch2.0.
The tool starts the meaning elicitation process and
produces a set of inferred lexical annotations of the
schema elements. The resulting annotated schema is
shown to the designer, who may validate and extend
the annotation produced by CtxMatch2.0 and, even-
tually, restart the process using the updated annotated
schema as reference ontology.
Figure 4 illustrates the results of a sample test of
incremental annotation on one of our schemas. It
shows the annotations manually provided by the on-
tology designer,a fraction of the new annotations gen-
erated after a first run of MELIS, and the additional
annotations generated after a second run, when the
outcome of the first run was provided as additional
background knowledge in input; the numbers on the
arrows refer to the heuristic rule which was used to
generate the annotation. Notationally, a square near
a class/attribute means that the element was manually
annotated, a circle means that the element was auto-
matically annotated after the first run, and a rhombus
that it was incrementally annotated after the second
Rule 1: the attribute “identifier” of the class “fa-
cility” in the source “VENERE” is annotated as
“identifier” of the class “facility” in the source
“BB” since both the classes are annotated with the
same synset.
Rule 2: because of the hyponym relationships
generated by the annotations of the classes “ho-
tel”, “campsite”, “bed and breakfast” and “build-
ing”, the attribute “city” of the class “building” in
the source “VENERE” produces the annotation of
the same attribute in the sources “BB”, “touring”,
Rule 3: because of the hypernym relation rela-
tionships generated by the annotations of “build-
ing” and “bed and breakfast”, the attribute “iden-
tifier” of the class “bed
and breakfast” in the
source “BB” generates the annotation of the same
attribute in the source “VENERE”. By execut-
ing a second run of the MELIS process, the at-
tribute “identifier” on the class “building” gener-
ates the annotation of the same attribute on the
classes “campsite” and ”restaurant” of the sources
“guidaC” and touring (application of the heuris-
tic rule 2).
Rule 4: because of the new relationship intro-
duced in WordNet, “campsite” is a sister term of
“restaurant”. Consequently, the attribute “local-
ity” is annotated in the same way in the sources
“guidaC” and “touring”.
Rule 5: in the source “VENERE” the class “map”
has a foreign key: the attribute url” that refer-
ences to the class hotel”. Because of this rela-
tionship joins with hierarchical relationships “ho-
tel”, “campsite”, “bed and breakfast” and “build-
ing”, the annotation of attribute “url” of the class
“map”, applying Rule 3 and Rule 5, generates the
MELIS - An Incremental Method for the Lexical Annotation of Domain Ontologies
Figure 3: Sources used for evaluating MELIS.
same annotation for “url” in the classes “camp-
site”, “bed
and breakfast” and “restaurant” of the
other sources.
Notice that heuristic 6 and 7 are not exploited in
this example. Such rules may be exploited in nested
structures as hierarchies, and they may not be applied
in flat structures as relational databases.
The results are highly dependent on the annotation
manually provided by the user as MELIS input . For
this reason, it is not meaningful to give any evalua-
tion in terms of number of new annotations discov-
ered. Concerning the evaluation of the new annota-
tions generated, our experience highlights that all of
the new annotated elements have a correct meaning
w.r.t. WordNet.
In this paper we presented MELIS, a method and
tool for incrementally annotating data sources accord-
ing to a lexical database (WordNet in our approach).
MELIS exploits the annotation of a subset of source
elements to infer annotations for the remaining source
elements, this way improving the activity of manual
MELIS is based on the integration and the exten-
sion of the lexical annotation module of the MOMIS-
Ontology Builder (Benassi et al., 2004) and some
components from CTXMATCH2.0, a tool for elicit-
ing meaning and matching pairs of nodes in hetero-
geneous schemas, using an explicit and formal rep-
resentation of their meaning (Bouquet et al., 2005;
Bouquet et al., 2006). CTXMATCH2.0 was extended
with respect to (Bouquet et al., 2005; Bouquet et al.,
2006) with a set of heuristic rules to generate new an-
notations on the basis of the knowledge provided by
a given set of annotations; WNEditor was modified in
order to jointly work with CTXMATCH2.0, by pro-
viding a customized lexical database.
We experimented MELIS in conjunction with the
MOMIS system in order to improve the MOMIS
methodology for semi–automatically creating a do-
main ontology from a set of data sources. The first re-
sults show that MELIS and MOMIS working in con-
junction are an effective tool for creating a domain
ontology. The testing was performed within the WIS-
DOM project
, for creating an ontology from several
data-intensive websites about hotels and restaurants.
As we noticed in the introduction, MELIS can be
used to provide valuable input not only for ontology
learning, but also for ontology matching tools. Here
we want to notice that these tools can greatly ben-
efit from the integration of MELIS, as MELIS pro-
vides a highly accurate collection of lexical annota-
tions which can be exploited in the matching phase.
Future work on MELIS will be addressed on im-
proving the annotation technique in order to deal with
compound terms (like “full professor”, “table leg”,
“football team”). Compound terms do not appear
in any lexical database, unless they form a stable
compound (e.g.“station wagon”). Their annotation is
therefore more difficult, as the choice of the right lex-
ical meaning often depends on determining the rela-
tionship between terms.
Moreover, we will introduce in MELIS more ac-
curate stemming techniques in order to improve the
matching among input terms and the words of the lex-
ical reference database. Finally, we are developing a
methodology for building and sharing among a com-
munity new lexical database entries, e.g. by establish-
ing how and when a new noun/meaning can be “pro-
moted” to be part of the common lexical reference.
WEBIST 2007 - International Conference on Web Information Systems and Technologies
Figure 4: Annotations generated with MELIS.
Bechhofer, S., Carr, L., Goble, C. A., Kampa, S., and
Miles-Board, T. (2002). The semantics of seman-
tic annotation. In Meersman, R. and Tari, Z., ed-
itors, CoopIS/DOA/ODBASE, volume 2519 of Lec-
ture Notes in Computer Science, pages 1152–1167.
Benassi, R., Bergamaschi, S., Fergnani, A., and Miselli, D.
(2004). Extending a lexicon ontology for intelligent
information integration. In de M´antaras, R. L. and
Saitta, L., editors, ECAI, pages 278–282. IOS Press.
Beneventano, D., Bergamaschi, S., Guerra, F., and Vincini,
M. (2003). Synthesizing an integrated ontology. IEEE
Internet Computing, pages 42–51.
Beneventano, D., Bergamaschi, S., Sartori, C., and
Vincini, M. (1997). ODB-QOPTIMIZER: A tool
for semantic query optimization in oodb. In
Int. Conference on Data Engineering - ICDE97.
Bouquet, P., Serafini, L., and Zanobini, S. (2005). Peer-to-
peer semantic coordination. Journal of Web Seman-
tics, 2(1).
Bouquet, P., Serafini, L., Zanobini, S., and Sceffer, S.
(2006). Bootstrapping semantics on the web: meaning
elicitation from schemas. In WWW2006 Conference
Proceedings. W3C.
Gangemi, A., Navigli, R., and Velardi, P. (2003). The
ontowordnet project: Extension and axiomatization
of conceptual relations in wordnet. In Meers-
man, R., Tari, Z., and Schmidt, D. C., edi-
tors, CoopIS/DOA/ODBASE, volume 2888 of Lecture
Notes in Computer Science, pages 820–838. Springer.
Handschuh, S., Staab, S., and Volz, R. (2003). On deep
annotation. In WWW, pages 431–438.
Montoyo, A., Palomar, M., and Rigau, G. (2001). Word-
net enrichment with classification systems. In In
Proc. of WordNet and Other Lexical Resources: Ap-
plications, Extensions and Customisations Workshop,
NAACL-01, pages 101–106, Carnegie Mellon Univ.,
Pittsburgh, USA.
Pazienza, M. T. and Stellato, A. (2006). An open and scal-
able framework for enriching ontologies with natural
language content. In Ali, M. and Dapoigny, R., edi-
tors, IEA/AIE, volume 4031 of Lecture Notes in Com-
puter Science, pages 990–999. Springer.
Staab, S., Maedche, A., and Handschuh, S. (2001). An
annotation framework for the semantic web. In S.
Ishizaki (ed.), Proc. of The First International Work-
shop on MultiMedia Annotation. January. 30 - 31.
Tokyo, Japan.
MELIS - An Incremental Method for the Lexical Annotation of Domain Ontologies