DAFOE: AN ONTOLOGY BUILDING PLATFORM

From Texts or Thesauri

Sylvie Szulman, Jean Charlet

LIPN, Paris 13 University, Villetaneuse, France

INSERM UMR S 872, Eq. 20, Paris, France

Nathalie Aussenac-Gilles, Adeline Nazarenko

CNRS/IRIT, Universit Toulouse, Toulouse, France

LIPN, Paris 13 University, Villetaneuse, France

Eric Sardet, Valery Teguiak

LISI-ENSMA, CRITT-Informatique, Poitiers, France

Keywords:

Platform, Ontology, NLP, Ontology-based Database.

Abstract:

This paper describes a new platform for building ontologies using many entries (texts, terminologies, thesauri

or database). However, They may be build from scratch and bound with texts afterwards. After a description

of the used data model, a meta-modelisation architecture is presented as well as the database implementation,

which allows to manage large ontologies with large corpora.

1 INTRODUCTION

Although text-based ontology engineering gained

much popularity in the last 10 years, very few ontol-

ogy engineering platforms exploit the full potential of

the connection between texts and ontologies.

We propose DAFOE

, a new platform for build-

ing ontologies using different types of linguistic en-

tries (text corpora, results of natural language pro-

cessing tools, terminologies or thesauri). DAFOE

supports knowledge structuring and conceptual mod-

elling from these linguistic entries as well as ontol-

ogy formalization. DAFOE outputs models with two

main original features: an ontology articulated with a

lexical component and a connection with the text or

linguistic entry that motivated their deﬁnition.

The requirements of the platfom and its develop-

ment focus 1) on integrating various kinds of tools

currently used within a single modelling platform,

2) on guaranteeing persistence and traceability of the

whole ontology building process, and 3) on develop-

ing the platform in an open source paradigm with pos-

sible plugin extensions.

This paper focuses on the terminological and

knowledgerepresentation in DAFOE: after the presen-

http://dafoe4app.fr

tation of various ontology engineering environments

from texts, we describe the data model of the DAFOE

platform and the corresponding meta-modelling ar-

chitecture that allows its database implementation.

2 TEXT-BASED ONTOLOGY

ENGINEERING

2.1 Ontology Engineering

Environments

There is a growing interest for ontologies and related

tools, including Ontology Engineering Environments

(OEEs). Since an early overview of tools support-

ing ontology engineering (Duineveld et al., 2000),

several joint efforts provided extensive state-of-the-

art overviews, like the review of OntoWeb thematic

network (Gomez-Perez et al., 2002), the comparison

of ontology editors (Denny, 2004) or the evaluation

of OEEs at EON 2002 workshop (Sure and Angele,

2002).

In DAFOE perspective, we carried out a survey of

about 15 OEEs. Many of them are pure ontology ed-

itors that support the development of formal ontolo-

372

Szulman S., Charlet J., Aussenac-Gilles N., Nazarenko A., Sardet E. and Teguiak V. (2009).

DAFOE: AN ONTOLOGY BUILDING PLATFORM - From Texts or Thesauri.

In Proceedings of the International Conference on Knowledge Engineering and Ontology Development, pages 372-375

DOI: 10.5220/0002276203720375

 SciTePress

gies (often represented using OWL) but do not as-

sist the tasks of knowledge acquisition or structuring.

Knowledge engineers are supposed to have a ﬁrst of

ontology draft before using such tools.

Among the tools that take text as input knowledge

three major approaches can be identiﬁed.

The ﬁrst one aims at building an ontology auto-

matically out of text analysis (Bozsak et al., 2002).

An alternative approach is based on human-machine

cooperation. Systems like ASIUM (Faure and

Nedellec, 1999), OntoGen

or Cam´el´eon (Aussenac-

Gilles and Jacques, 2008) support parts of an in-

cremental and interactive ontology development. A

third approach, TERMINAE (Aussenac-Gilles et al.,

2008), integrate results from several NLP tools like

Text2Onto

but do not constrain their selection or

combination. The result is more than an ontology, as

long as lexical entries are associated to concepts. It is

called a termino-ontological resource.

2.2 Original Features of DAFOE

The goal of DAFOE is both to extend the variety of

human language technologies that can be used and

to support scalable ontology engineering. It claims

that there are several ways to get an ontology, and

that tools and processes must be selected according to

each ontology case-study. DAFOE will propose tools

similar to those of Text2Onto, but human supervision

will play a major role for selecting tools, validating

their results and conceptualizing. Knowledge con-

ceptualization requires that a human selects and orga-

nizes properly concepts and relations, but this process

can be guided.

To sump-up, this survey of the state-of-the-art em-

phasizes the major motivations for developing the

DAFOE platform:

• we want DAFOE to keep track of the linguistic

items that justify the conceptual modelling; this

would help not only to maintain and update the

ontology, but also to produce richer ontologies as-

sociated with a lexical component. DAFOE will

build termino-ontological resources.

• since we argue that extracting an ontology from

texts is a complex process that requires several

modelling steps and that the models produced

at each stage are worth being stored, we want

DAFOE to support that multi-step modelling pro-

cess and to store the intermediary models as a

modelling track.

http://ontogen.ijs.si/

http://ontoware.org/projects/text2onto/

• we want DAFOE to make it possible to use the

links between those various models to come back

from the ontology to the text that has been used to

build it, to deﬁnitions and comments that justify

the content of the ontology.

• we want DAFOE to support large-scale ontologies

and lexical resources, which is a major argument

towards a database implementation (Sec. ??) and

an important difference with systems like TERMI-

NAE or Prot´eg´e.

3 DATA MODEL

DAFOE data model has to take into account vari-

ous ontology building strategies, whatever informa-

tion source (texts, terminologies, thesauri or human

expertise) is used. Moreover, DAFOE must evolve to

offer new functionalities in satisfaction to new needs,

which means that plugin extensions will be supported

and that the data model allows these extensions.

3.1 Overall Architecture

The data model is based on a valid methodology for

building ontologies from texts, which has inspired

tools such as TERMINAE (Aussenac-Gilles et al.,

2008) or Text2Onto (Cimiano and Volker, 2005). This

methodology takes into account the whole process of

“transforming” textual data into ontologies and split

it into different phases, which correspond to various

input levels if one wants to start with a thesauri rather

than text, for instance. This methodologies relies on

two main ideas: 1/ textual data are an important in-

formation source to build ontologies, especially if the

ontology is to be used to annotate textual documents

but 2/ textual data cannot be mapped directly into an

ontology and the transformation must be mediated.

The data model is therefore structured into four

layers. Each one corresponds to a speciﬁc method-

ological step.

3.2 Corpora Layer

The corpora layer is useful for the knowledge engi-

neer willing to build an ontology from text. He/she

can build a working corpus by selecting different

source documents and browse that corpus, either as

plain documents or as segmented ones. In the data

model the corpus is represented as a sequence of sen-

tences, each one having a unique identiﬁer.

DAFOE: AN ONTOLOGY BUILDING PLATFORM - From Texts or Thesauri

373

3.3 Terminological Layer

The terminological layer gives a view over the do-

main speciﬁc lexicon of the corpus. It gathers the

terms of the domain and their relationships. Termino-

logical knowledge is traditionally produced by NLP

tools such as term extractors applied on the working

corpus. An alternative approach would consist in im-

porting a preexisting terminology of the domain. The

underlying assumption is threefold: text analysis can

extract term candidates that are relevant for a given

domain, those terms are likely to be turned into on-

tology concepts and the distribution of these terms re-

ﬂects their semantics (Harris, 1968).

3.4 Termino-Conceptual Layer

This layer represents a semantic structure of unam-

biguous termino-concepts and termino-conceptual re-

lations. The knowledge engineer may build that layer

by importing a preexisting termino-conceptual re-

source such as a thesaurus or out of the analysis of

the terminological layer. In that case, he/she analyses

the meaning of terms and relations that appear at the

terminological layer with respect to each other and by

looking at their occurrences. He/she clusters terms

and relationships that have the same meaning, dis-

tinguishes the various meanings of ambiguous terms,

compares the contexts in which they are used. He/she

thus deﬁnes non ambiguous termino-concepts (TC)

and termino-conceptual relations (RTC) holding be-

tween CTs.

The termino-conceptual layer is pivotal for trans-

forming linguistic elements into conceptual ones and

tracing the ontology back to the linguistics. This

traceability improves ontology readability and main-

tenance. The terms and terminological relations that

are connected to termino-conceptual elements are

said to be ”conceptualised”.

3.5 Ontology Layer

The ontology data model allows to formalize TCs and

RTCs in a formal language equivalent at OWL-DL.

Concepts are described as classes, individuals as in-

stances of classes, properties between classes as ob-

ject properties and properties between a class and a

value as data properties or attributes. An automatic

process will translate TCs and RTCs into formal con-

cepts in a hierarchy with inherited properties as usual

subsumption in description language. This translation

exploits the structure of the semantic network repre-

sented in the termino-conteptual layer and the differ-

ential criteria associated with TCs and RTCs.

4 META-MODELLING

ARCHITECTURE AND

DATABASE IMPLEMENTATION

DAFOE platform is intended both to support large

volumes of data and to provide a variety of ontol-

ogy engineering methods. As such a diversity can

not be managed in a unique and static model, we

adopted for DAFOE platform an extended Ontology-

Based Database (OntoDB) architecture that supports

model management. In this section, we present the

original OntoDB model, we show how it is extended

to cope with DAFOE speciﬁcities, we illustrate the

resulting architecture with a simple example and we

give some implementation details.

4.1 Ontology-based Database (OntoDB)

The OntoDB model has two characteristics: (1) ontol-

ogy and data are stored in the database and support the

same processes (insert, update,...), (2) every piece of

data is associated with an ontological element that de-

ﬁnes its meaning. The OntoDB architecture is quite

similar to the layers deﬁned in the MOF

architec-

ture, which consists of three levels of modelling: the

M1 level allows to represent the models, the M2 level

represents the meta-models and the M3 level repre-

sents the meta meta model (instances are represented

in M0). An important feature provided by OntoDB

is its capability to support model evolution and data

intensive applications (H. Dehainsala, 2007). This is

one reason why we have proposed an OntoDB-like

architecture for DAFOE.

4.2 DAFOE Architecture: A

Model-based Database (MBDB)

Since corpora and extracted terminologies are gener-

ally large, we opted to store corresponding data in a

database and the choice of a MOF architecture led

us to adopt an OntoDB architecture. The integra-

tion of DAFOE data model into OntoDB is not triv-

ial, however. The main difﬁculty stems from the fact

that an OntoDB architecture relies on a single model,

while the DAFOE approach is based on three different

models: a Terminological Model (TM), a Termino-

Conceptual Model (TCM) and an Ontological Model

(OM). Hence, the idea is to represent each of DAFOE

three models in a separate OntoDB based architec-

ture.

http://www.omg.org/mof/

KEOD 2009 - International Conference on Knowledge Engineering and Ontology Development

374

More generally, this integration requires an effort

of identiﬁcation and distribution of DAFOE data ac-

cording to a MOF architecture.

4.2.1 Data Distribution

The data exploited in DAFOE for building ontologies

from texts is distributed according to the three ab-

straction levels of MOF. The particularity of this ap-

proach is that it provides, for each part (data, model

and meta-schema) of the managed information lay-

ers (terminology, termino-conceptual and ontologi-

cal), a meta-modelling level supporting the evolution

of those part structures. It is represented by the MOF

syntax (Mi/Mi-1), which means that the information

model Mi-1 is represented as instances of the infor-

mation (meta) model Mi.

4.2.2 Model Transformation

Even if the integration of DAFOE into OntoDB solves

the problem of model evolution thanks the various ab-

straction levels deﬁned by MOF, the problem of tran-

sition from one modelling layer to another remains

open. To solve this problem, we propose to use model

transformation mechanisms to enhance the DAFOE

architecture.

5 CONCLUSIONS

DAFOE is a new platform to assist a knowledge engi-

neer throughout the ontology building process. It al-

lows him/her to integrate domain speciﬁc knowledge

sources (text corpora, terminologies, thesauri or hu-

man expertise) and to deﬁne a formal ontology. The

strength of DAFOE approach is i) a precise deﬁnition

of the various steps by which one can design a for-

mal ontology; ii) a data model guaranteeing persis-

tence and traceability of the whole ontologies build-

ing process; iii) the supply of ﬂexible methodological

guidelines that support the knowledge engineer with-

out constraint; iv) an architecture based on the MOF

model and plugins adaptability to ensure extensibil-

ity of the model and processes around a core tool; v)

the speciﬁcation of various modelling strategies based

on different input/output of the platform; vi) the ﬁnal

production of an ontology which is associated to a ter-

minological component.

A prototype of the DAFOE platform is under im-

plementation. We use Model Driven Engineering and

the EXPRESS language.

REFERENCES

Aussenac-Gilles, N., Despres, S., and Szulman, S. (2008).

Bridging the Gap between Text and Knowledge: Se-

lected Contributions to Ontology learning from Text,

chapter The Terminae Method and Platform for On-

tology Engineering from Texts. IOS Press.

Aussenac-Gilles, N. and Jacques, M.-P. (2008). Designing

and evaluating patterns for relation acquisition from

texts with cam´el´eon. Terminology, special issue on

Pattern-based Approaches to Semantic Relation Ex-

traction, 14:45–73.

Bozsak, E., Ehrig, M., Handschuh, S., Hotho, A., Maed-

che, A., Motik, B., Oberle, D., Schmitz, C., Staab, S.,

Stojanovic, L., Stojanovic, N., Studer, R., Stumme,

G., Y. Sure, J. T., Volz, R., and Zacharias, V. (2002).

Kaon - towards a large scale semantic web. In Pro-

ceedings of the Third International Conference on

E-Commerce and Web Technologies (EC-Web 2002),

Aix-en-Provence, France, volume 2455 of LNCS,

pages 304–313. Springer Verlag.

Cimiano, P. and Volker, J. (2005). Text2onto - a framework

for ontology learning and data-driven change discov-

ery. In Montoyo, A., Munoz, R., and Metais, E., ed-

itors, Proceedings of the 10th International Confer-

ence on Applications of Natural Language to Informa-

tion Systems (NLDB), volume 3513 of Lecture Notes

in Computer Science, pages 227–238, Alicante, Spain.

Springer.

Denny, M. (2004). Updated ontology edi-

tor survey results (table). available at

http://www.xml.com/pub/a/2004/07/14/onto.html.

Duineveld, A., Stoter, R., Weiden, M., Kenepa, B., and Ben-

jamins, V. (2000). Wondertools? a comparative study

of ontological engineering tools. International Jour-

nal of Human-Computer Studies, 6(52):1111–1133.

Faure, D. and Nedellec, C. (1999). Knowledge acquisi-

tion of predicate argument structures from technical

texts using machine learning: the system asium. In

Fensel, D. and Studer, R., editors, Proceedings of the

11th International Conference on Knowledge Engi-

neering and Knowledge Management, pages 329 –

334. Springer-Verlag, LNAI.

Gomez-Perez, A., Angele, J., Fernandez-Lopez, M.,

Christophides, V., Stutt, A., and Sure, Y. (2002). A

survey on ontology tools. ontoweb deliverable 1.3.

Technical report, Universidad Politecnica de Madrid.

H. Dehainsala, G. Pierra, L. B. (2007). Ontodb : An

ontology-based database for intensive applications.

pages 497–508.

Harris, Z. (1968). Mathematical Structures of Language.

Interscience Publishers.

Sure, Y. and Angele, J., editors (2002). Proceedings of the

First International Workshop on Evaluation of Ontol-

ogy based Tools (EON 2002), Siguenza, Spain, vol-

ume 62 of CEUR Workshop Proceedings. CEUR-WS

Publication. available at http://CEUR-WS.org/Vol-

62/.

DAFOE: AN ONTOLOGY BUILDING PLATFORM - From Texts or Thesauri

375