Ontology Learning Process as a Bottom-up Strategy for Building

Domain-specific Ontology from Legal Texts

Mirna El Ghosh

, Hala Naja

, Habib Abdulrab

and Mohamad Khalil

INSA, Rouen, France

Lebanese University, Faculty of Sciences, Tripoli, Lebanon

Lebanese University, Faculty of Engineering, Tripoli, Lebanon

Keywords: Ontology Learning, Semi-automatic Extraction, Natural Language Processing, Legal Ontologies, Domain-

specific Ontologies.

Abstract: The objective of this paper is to present the role of Ontology Learning Process in supporting an ontology

engineer for creating and maintaining ontologies from textual resources. The knowledge structures that

interest us are legal domain-specific ontologies. We will use these ontologies to build legal domain ontology

for a Lebanese legal knowledge based system. The domain application of this work is the Lebanese criminal

system. Ontologies can be learnt from various sources, such as databases, structured and unstructured

documents. Here, the focus is on the acquisition of ontologies from unstructured text, provided as input. In

this work, the Ontology Learning Process represents a knowledge extraction phase using Natural Language

Processing techniques. The resulted ontology is considered as inexpressive ontology. There is a need to

reengineer it in order to build a complete, correct and more expressive domain-specific ontology.

1 INTRODUCTION

It is commonly known that the knowledge of the

legal domain is expressed and conveyed in texts

using domain-specific terminology. However, this

terminology does not provide a well-defined

structure to be used by machines for reasoning tasks.

Meanwhile, the extracting and mining of this

terminology will lead to a certain domain

representation model such as ontology (Mädche,

2000). Ontology is defined as a conceptualization of

a domain into a human understandable, machine-

readable format consisting of entities, attributes,

relationships and axioms (Guarino, 1995). This

definition imposes that the concepts and relations

among them have to be explicitly represented and

expressed using formal language such as Web

Ontology Language (OWL). This formal structure

representation leads to specify axioms for reasoning,

in order to define constraints in ontologies (Wong,

2009). Building and maintaining ontologies

manually remains a resource-intensive, time

consuming and costly task. This is due to the

difficulty in capturing knowledge, also known as the

“knowledge acquisition bottleneck”. Even with

some reuse of Core or Upper ontologies. Therefore,

there is a need to automatic or semi-automatic

techniques that support the building process. These

techniques have become to be known as Ontology

Learning (OL) (Cimiano, 2004). OL has the

potential to reduce the cost of creating and

maintaining ontologies using semi-automatic

methods and tools. Actually, we motivate to develop

legal domain ontology for the Lebanese criminal

domain. In a previous work (El Ghosh, 2016), a

middle-out approach is proposed for building this

ontology for a legal knowledge based system that

performs reasoning and information retrieval tasks

(Figure 1). Accordingly, we proposed to modularize

the legal domain ontology into four modules or

ontologies: upper, core, domain and domain-

specific. The upper module represents the most

general concepts and relations that cover all the

domains (such as Agent, Act and Action). The core

module provides a definition of structural

knowledge in the legal domain. For instance,

concepts, such as Legal_Source, Legal_Act and

Legal_Document, are common for all the legal fields

(criminal, civil, etc.). The concepts of the domain

module, in turn, such as Offence, Infraction and

Offender, describe the conceptualization of the

criminal domain. Finally, in the domain-specific

El Ghosh M., Naja H., Abdulrab H. and Khalil M.

Ontology Learning Process as a Bottom-up Strategy for Building Domain-speciﬁc Ontology from Legal Texts.

DOI: 10.5220/0006188004730480

In Proceedings of the 9th International Conference on Agents and Artiﬁcial Intelligence (ICAART 2017), pages 473-480

ISBN: 978-989-758-220-2

473

module, we learn the knowledge of the Lebanese

criminal system from textual resources such as the

criminal code. Furthermore, an alignment process

will be applied to complete the global ontology by

linking the concepts of the different modules. In

order to develop the different modules, two different

strategies are applied (top-down and bottom-up).

The top-down represents the conceptual modeling

process based on reusing foundational and core

ontologies (El Ghosh, 2016). Meanwhile, the

ontology learning process from textual resources is

depicted by the bottom-up strategy that aims to

develop the domain-specific ontology module.

Domain-specific ontologies specify formally

concepts and relations of a specific subject domain

(Hatala, 2012). They cannot be reused unlike other

kinds of ontologies (upper and core).

Figure 1: Middle-out approach for building modularized

ontology.

What is important that these ontologies are

useful in systems involved with artificial reasoning

and information retrieval. In this context, the OL

process from unstructured legal documents could be

useful for building the criminal domain-specific

ontology. Meanwhile, the main obstacle that exists

is to reduce the efforts required for creating the

ontology by defining a convenient semi-automatic

development process and ontology learning tool. In

order to achieve the goal, we started by discussing

the ontology learning from unstructured texts in

section II. In section III, we overviewed existent

ontology learning methods and tools. The

experimental work is presented in section 4. The

section 5 discusses similar works. We finished by

section 6 for the discussion and section 7 for the

conclusion.

2 ONTOLOGY LEARNING

FROM UNSTRUCTURED TEXT

The term Ontology Learning (OL) was introduced in

(Mädche, 2005) and is considered as an important

task in Artificial Intelligence, Semantic Web and

Knowledge Management. It is the dynamic process

of building ontologies. OL is a data model that

represents a set of concepts and relations within a

domain (Yang, 2008). More specifically, OL is

considered as a subtask of Information Extraction

(IE), which is a type of Information Retrieval (IR)

(Rogger, 2010). The main purpose of OL process is

to apply methods from various fields such as

linguistic analysis, machine learning, knowledge

acquisition, statistics and information retrieval in

order to extract knowledge and support the

construction of ontologies. This dynamic process,

depicted in the Figure 2, takes as input implicit and

unstructured knowledge and produces as output

explicit structured knowledge (Cimiano 2005).

Generally, OL is a semi-automatic process where the

ontology engineer and the domain expert can be

involved to achieve better results (Rogger, 2010).

Thus, the techniques used in the ontology

development process will be under their supervision.

Their expertise and background knowledge helps in

verifying the obtained information and decide the

valuable information.

Figure 2: The dynamic process of ontology learning,

(Buitelaar, 2005).

2.1 Input

As aforementioned, ontologies can be learnt, by

applying the OL process, from various sources of

data types: structured (such as databases), semi-

structured (e.g. XML) and unstructured textual

documents. The domain application of this work is

the Lebanese criminal code which is an unstructured

text resource. This type of resources is the most

available format as input for ontology learning

processes. They reflect mostly the domain

ICAART 2017 - 9th International Conference on Agents and Artiﬁcial Intelligence

474

knowledge for which the user is building the

ontology. In addition, they describe the terminology,

concepts and conceptual structures of the given

domain. However, some authors, such as (Rogger,

2010), consider that processing unstructured data is

the most complicated problem because most of the

knowledge is implicit and allows conceptualizing it

by different people in different manner. Specifically,

in the legal domain, the implicit knowledge of the

natural language is one of the main obstacles to

progress in the field of artificial intelligence and law

(McCarty, 2007).

2.2 Output

Ontology learning from text is the process of

deriving concepts, relations and axioms from textual

resources to build ontologies. The main output of the

OL process is a structured content represented in an

explicit formal way. For (Cimiano, 2004), the tasks

in ontology learning from text are organized in a set

of layers (Figure 3). These tasks aim at returning six

main outputs: terms, synonyms, concepts, taxonomic

relations, non-taxonomic relations and axioms.

These outputs represent the main elements of

ontology.

Figure 3: Ontology learning from text, layer cake

(Buitelaar, 2005).

Terms are the most basic building blocks in

ontology learning (Wong, 2009). Concepts can be

abstract or concrete, real or fictitious. Concept

hierarchies or taxonomies are crucial for any

knowledge based system (Cimiano, 2005). Non-

taxonomic or non-hierarchical relations represent the

interactions between concepts (e.g. meronymy,

thematic roles, attributes, possession and causality)

(Wong, 2009). Finally, the axioms are defined as

propositions or sentences that are always taken as

true. Axioms act as a starting point for deducing

other truth, verifying correctness of existing

ontological elements and defining constraints (Wong,

2009).

3 ONTOLOGY LEARNING

METHODS AND TOOLS

There are many works in the literature that deal with

ontology learning from textual resources. The focus

of this paper is to discuss and evaluate existent

methods and tools to develop (semi-)automatically

text-based domain ontologies. Furthermore, we will

define a (semi-) automatic approach for building our

legal domain-specific ontology.

3.1 Methods

In order to obtain high-quality ontologies, the

development process has to be driven by a

methodology (Hatala, 2012). In this section, we

discuss briefly the most known ontology learning

methodologies from textual resources. In the work

of (Sabou, 2005), the ontology learning process is

based on three major tasks: term extraction,

conceptualization and enrichment. For (Mädche,

2005), the OL process is composed of four different

phases: extract concepts, prune, refine and Import or

reuse. In other studies, such as (Mazari, 2012) and

(Ge, 2012), the ontology learning tasks are resumed

in three: documents preprocessing, concepts

extraction and relations discovery. Actually, these

tasks discover only taxonomic relations (parent-

child, hyponymy (is-a) and meronymy (part-of)).

However, some authors such as (Novelli, 2012),

(Balakrishna, 2010) and (Serra, 2012) propose

methods to solve the problem of learning non-

taxonomic relations of ontologies from text. In the

legal domain, most of the methodologies focus on

concepts extraction as a main step of the ontology

development process (Lenci, 2009). The approach of

(Walter, 2006) is based on the exploitation of the

frequency of definitions in legal texts.

3.2 Tools

In the literature, a long list of ontology learning tools

has been proposed. The existent tools differ

according to input data types, output formats and

mainly the methods and algorithms used in order to

extract the ontological structures. The main goal of

using ontology learning tools is to reduce the time

and cost of ontology development process. In this

section, we discuss mainly the existent ontology

learning tools from unstructured textual resources.

Terminae is a method and tool that generates

standard OWL ontologies (Biebow, 1999). Terminae

integrates linguistic and knowledge engineering

Ontology Learning Process as a Bottom-up Strategy for Building Domain-speciﬁc Ontology from Legal Texts

475

tools to guide the knowledge acquisition from texts

and to build terminological and ontological models.

Text2Onto, successor of Text-to-Onto (Mädche,

2001), is a data-driven, ontology learning tool that

supports automatic development of ontologies from

textual documents (Cimiano, 2005). Text2Onto is

built upon the GATE

framework. Accordingly,

Text2Onto implements linguistic processing and

machine learning statistical techniques to extract

domain concepts and relations. This tool features

also algorithms for generating concepts, taxonomic

and non-taxonomic relations. OntoGen is a semi-

automatic and data-driven ontology editor that helps

the users to build ontologies by suggesting concepts

and relations. This system integrates machine

learning and text mining algorithms. OntoGen offers

two main features: concept suggestion and naming

and ontology and concept visualization. T2K

extracts domain–specific information from texts

using natural language processing techniques in

three main phases: preprocess text and extract terms,

form concepts using POS patterns and relations or

knowledge organization (Dell’Orletta, 2014).

CRCTOL is Concept-Relation-Concept tuple-based

ontology learning system from domain-specific text

documents. The tool adapts a full text parsing

technique and incorporates both statistical and

lexico-syntactic methods (Jiang, 2005). We

conclude that most of these tools rely on linguistic

and statistic methods to learn ontologies. The focus

is on extracting concepts and taxonomies. Thus, we

need to learn more semantic relations and axioms.

Table 1: Summary of ontology learning tools.

Tool

Elements extracted

Techniques

Terminae

(2005)

Terms, synonyms,

concepts, taxonomies,

non-taxonomic

relations

Linguistic and

knowledge

engineering

Text2Onto

(2005)

Terms, synonyms

concepts, taxonomies,

non-taxonomic

relations, instances

linguistic processing

statistical text

analysis

machine learning

association rules

OntoGen

(2006)

Terms, concepts,

taxonomies

Machine learning

text mining

T2K

(2008)

Terms, concepts

,taxonomies

statistical text

analysis and

machine learning

CRCTOL

(2010)

Concepts,

taxonomies,

non-taxonomic

relations

Statistical lexico-

syntactic association

rules

https://gate.ac.uk/

4 OUR WORK

Even after a comprehensive literature review, we

found a difficulty to define a complete approach or

tool that can totally extract domain-specific

ontologies from textual resources. This is due to

two reasons. First, we could not find a complete

(semi-)automatic tool or approach that carries the

ontology development process. Second, there is no

guarantee that the (semi-)automatically generated

ontology is correct and precise enough to

characterize the domain in question (Rudolph, 2007).

Since the focus of the current research is mainly on

extracting the elements of a criminal domain-

specific ontology from textual resources, using an

existent semi-automatic ontology learning can help

to extract an OWL ontology including the basic

elements (concepts, taxonomies, relations and

disjointness axioms). Meanwhile, and based on what

is found in the literature, incomplete and not

satisfactory results are expected. For this reason, the

intervention of ontology engineer and legal expert

during the ontology learning process is required in

order to supervise the work and to verify the

obtained information. Furthermore, a reengineering

methodology is needed in order to enhance the

results by transforming the resulted ontology into a

new more correct, complete and expressive ontology.

The general idea of the reengineering approach is

depicted in figure 4. In the current work, mainly the

ontology learning process, from texts, is discussed.

The reengineering phase will be the study of further

works.

Figure 4: Reengineering phase for updating domain-

specific ontology.

In this section, we introduce the main

components of the ontology learning process used in

the preparation and execution of the criminal

domain-specific ontology.

ICAART 2017 - 9th International Conference on Agents and Artiﬁcial Intelligence

476

4.1 Material Selection

Actually, the domain-specific ontology that we aim

to build, using (semi-)automatic ontology learning

tool guided by an approach, represents the domain-

specific module in the modularized legal domain

ontology. The context of interest is the Lebanese

criminal system. The domain related material is the

Lebanese penal code that consists of legal natural

language texts. The Lebanese penal code contains

the general penal laws of Lebanon. First enacted in

1943 and it remains in effect today. It is translated to

French and English versions. Concerning the

structure of the code, it is divided into two main

books composed of 770 articles (Figure 5).

Figure 5: Excerpt of the Lebanese criminal code.

4.2 Tool Selection

After exploring the literature and collecting the

state-of-the-art for the most frequently used

ontology learning tools, we met some access

difficulties in our experimentations. In fact, three of

the tools were publicly available on the internet to

download and install: Terminae, OntoGen and

Text2Onto. In this section, we discuss briefly the

usability of each tool. Concerning the input type, all

the tools accept simple text files (.txt), Text2Onto

and Terminae accept also PDF files (.pdf). For

OntoGen, there are additional input file types that

need to be pre-processed, such as Named Line-

Document and Bag of Words. Terminae and

OntoGen need preprocessing efforts. Starting with

Terminae where the linguistic tool extract terms

automatically from the corpus based on their

occurrences. Meanwhile, the rest of the steps are

processed manually which is too resource

demanding and too time consuming. For this reason,

this tool is discarded. Furthermore, we face some

difficulties while using OntoGen. We could not

control the system that generates sequences of terms

that are not well related. In addition to this, the

suggestion of concepts is limited to single-word

terms, proposed only from the input documents (no

external resources), and the relations extraction is

limited as well to taxonomic. Meanwhile, OntoGen

provide a visualization and exploration of concepts

only and not of the whole ontology. OntoGen is

discarded too. We finished our experiments by

Text2Onto. According to (Gherasim, 2013),

Text2Onto is an ontology learning tool that covers

the entire process of extracting OWL ontologies.

Furthermore, it provides a long list of proposed

concepts and relationships along with their weights

in a tabular form. Meanwhile, Text2Onto does not

have any mechanism to filter the concepts irrelevant

to goal (Hatala, 2012). The user input is limited to

removing concepts and relationships extracted from

the supplied course. In Text2Onto, the visualization

of the structure of the resulted ontology is missing.

Regarding the external resources, Text2Onto uses

WordNet to improve and enrich the algorithms of

pattern-based relation extraction. However, some

authors found that WordNet lacks the richness of

named relations (Fouad, 2015). For this reason, they

decided to use online ontologies as an alternative to

WordNet. Regarding the limitations of Text2Onto,

this tool still answers the main requirements of our

work: automatic extraction, usability, scalability,

and reusability. Based on this selection, we proposed

to apply a reengineering phase that consists of

evaluating the ontology extracted using Text2Onto,

correcting the detected errors, refine the ontology

model and finish by enrich the semantic relations

and axioms. We will study deeply this point in

further works.

Table 2: List of experimented tools.

Tool

Terminae

OntoGen

Text2Onto

User Input

Add, remove,

modify

Add, remove,

modify

Remove

Visuali-

zation

Not available

Concepts

Not available

External

Resources

Not available

WordNet

4.3 Ontology Extraction Process

In this section, we present the main phases of the

criminal domain-specific ontology extraction

process using Text2Onto. Actually, the process is

composed of two main phases: linguistic

preprocessing and extraction of modeling primitives.

In the following, we discuss briefly each phase and

the algorithms used to achieve the resulted domain-

specific ontology.

Ontology Learning Process as a Bottom-up Strategy for Building Domain-speciﬁc Ontology from Legal Texts

477

4.3.1 Preprocessing

The purpose of the preprocessing phase (Figure 6) is

to prepare the corpus and remove the ambiguity by

filtering out worthless symbols and words, in order

to extract meaningful textual content from the input

documents. In Text2Onto, there is a combination of

machine learning approaches with basic linguistic

processing such as tokenization or lemmatizing and

shallow parsing (Cimiano, 2005). In addition to this,

Text2Onto benefits from GATE by the integration of

JAPE that provides finite state transduction over

annotations based on regular expressions (Mädche,

2001).

Figure 6: Preprocessing phase in Text2onto.

4.3.2 Extraction of Modelling Primitives

In this section, we describe briefly the extraction

phase of the ontology modelling primitives. For this

purpose, Text2Onto implements series of algorithms.

Five main modeling primitives are considered in this

tool: concepts, instances, taxonomies, general

relations and disjoint axioms. In this section, the

extraction process of each primitive is discussed

briefly. For extracting concepts, three algorithms are

implemented. Based on experiments,

TFIDFConceptExtraction algorithm is selected. 486

single and multi-word concepts are extracted such as:

Probation, Criminal, Crime, Term Penalty and

Violence. Concerning the taxonomies (subclass-of

relations), Text2Onto provides three algorithms to

classify concepts based on Vertical Relations,

WordNet, and Patterns. For better results, the three

algorithms are combined.

Table 3: Excerpt of the hierarchies extracted using

Text2Onto.

Domain

Range

Divorcee

Wife

Offender

Person

Death penalty

Penalty

Regarding the Instances, Text2Onto identifies

proper nouns as instances. Technically, it filters the

terms tagged as Instance from the GATE result.

Long list of instances are extracted such as Lebanon,

April and Friday. Text2Onto relies on

SubcatRelationExtraction algorithm to extract

general relations. This algorithm uses syntactic

pattern matching technique to extract general

relations.

Table 4: Excerpt of general hierarchies extracted using

Text2Onto.

Label

Domain

Range

involve

Residence

Placement

require

Activity

License

exceed

Offence

Bound

For the disjointness axioms, they are extracted in

Text2Onto based on lexico-syntactic patterns.

Table 5: Excerpt of disjoint axioms extracted using

Text2Onto.

Domain

Range

Confidence

Measure

Penalty

0.013

Felony

Disposal

0.013

Person

Association

0.06

4.3.3 Ontology Visualization

After applying the algorithms of Text2Onto, the

results are exported, as output, in OWL format.

Subsequently, we have looked for an ontology

visualization tool to visualize the resulted ontology.

Different tools are tested such as OWLViz

, a plug-

in for Protégé, and COE cmap tool

, and OWLGrEd

The resulted ontology is visualized correctly in

OWLGrEd (Figure 7).

Figure 7: Ontology visualization using OWLGrEd.

http://protegewiki.stanford.edu/wiki/OWLViz

http://coe.ihmc.us/

http://owlgred.lumii.lv/

ICAART 2017 - 9th International Conference on Agents and Artiﬁcial Intelligence

478

5 RELATED WORK

There are many works in the literature that proposed

the (semi-)automatic building of ontologies from

textual resources using ontology learning methods

and tools. The most related works are (Francesconi,

2010) in the legal domain and (Ortiz, 2007) in the

political domain. In the work of (Francesconi, 2010),

the authors have used two different tools for term

extraction: GATE for English texts and T2K for

Italian. The rest of the phases, such as evaluation of

terms and link them to concepts, extraction of

lexical relations were processed under the

supervision of ontology engineers and domain

experts. For the work of (Ortiz, 2007), the authors

applied Text2Ontofor creating domain ontology

from texts. They concentrated mainly in their study

on concepts extraction. In addition to this, the

authors proposed a reengineering methodology

based mainly on reusing online ontologies. What

differs our work first is the domain application,

which is the Lebanese criminal code. The context of

the code is composed of legal norms written in legal

language. Secondly, we used Text2Onto to extract

all the essential elements of a domain-specific

ontology. Finally, we expect to build an expressive

domain-specific ontology for reasoning system,

which is difficult using only an ontology learning

tool, for this reason we have proposed a

reengineering approach, based not only on online

ontologies, to correct the errors and to enrich the

extracted ontology with relations and axioms in

order to make it more expressive.

6 DISCUSSION

The aim of this paper if to extract domain-specific

ontology elements from texts using ontology

learning tool. Text2Onto is selected for this purpose.

The tool applies an automatic extraction process

based on list of algorithms and NLP techniques

using GATE applications. In addition to this, the

results can be exported as OWL ontology ready to

edit and update in ontology editor frameworks such

as Protégé. After applying list of algorithms to

extract the elements of the domain-specific ontology,

we obtained some results to discuss. Starting with

concepts, the tool extracted 486, single and multi-

word, concepts. The domain expert filtered the list

and removed the errors. We can resume the

identified errors in some examples. Some verbs like

stay, incur and abort were identified as concepts by

Text2Onto. Some domain-specific concepts were

identified as instances such as Confiscation,

Detainee and Terrorist. For the instances, the

extraction is limited because of the corpus quality.

Actually, the experiment is based on criminal code

written in legal language, which is authoritative and

contains legal speech acts accompanied by rituals of

various types. Text2Onto identified only 20 semantic

relations and 86 disjoint axioms. A reengineering

phase is needed to enrich the extracted ontology.

From this perspective, the reengineering

methodology is proposed to correct, enrich and

refine the resulted ontology and to build correct,

complete and more expressive domain-specific

ontology.

7 CONCLUSIONS

In this paper, we have briefly described the field of

ontology learning from textual resources as a

bottom-up approach for building a domain-specific

ontology for the criminal law. The mechanism of

ontology learning process from unstructured text

was identified. Furthermore, we have presented an

overview of the existent ontology learning methods

and tools. We also discussed our work followed by a

summarizing comparison of the ontology learning

tools used in our experiments. Based on the

experiments, Text2Onto is selected as a tool for the

ontology learning process. In fact, this tool answers

the main requirements of the study. Using

Text2Onto, the main elements of the domain-specific

ontology are extracted (concepts, taxonomies,

relations and axioms). The results were essentials,

but inexpressive. A reengineering process is needed

to build a more expressive ontology.

ACKNOWLEDGEMENTS

This project has been funded with support from the

European Union with the European Regional

Development Fund (ERDF), the National Support

from the National Council for Scientific Research in

Lebanon (CNRS) and Lebanese University

REFERENCES

Mädche, A. and Staab, S., 2000. “Mining ontologies from

text,” Proceeding of EKAW '00, pp. 189-202.

Ontology Learning Process as a Bottom-up Strategy for Building Domain-speciﬁc Ontology from Legal Texts

479

Guarino, N. and Giaretta, P., 1995. “Ontologies and

Knowledge Bases: Towards a Terminological

Clarification,” Proceeding of Toward Very Large

Knowledge Bases: Knowledge Building and

Knowledge Sharing, pp.25-32.

Wong, W., Y., 2009. “Learning Lightweight Ontologies

from Text across Different Domains using the Web as

Background Knowledge,” PhD thesis, University of

Western Australia, School of Computer Science and

Software Engineering.

Cimiano, P. Mädche, A, Staab, S., and Volker, J., 2004.

Ontology Learning. Handbook on Ontologies,

Springer.

Hatala, M., Gaevi, D., Siadaty, M., Jovanovic, J. and

Torniai, C., 2010. “Ontology Extraction Tools: An

Empirical Study with Educators”, IEEE Transactions

on Learning Technologies, vol. 5, no. 3, pp. 275-289.

El Ghosh, M., Naja, H., Abdulrab, H. and Khalil, M.,

2016. “Towards a Middle-out Approach for Building

Legal Domain Reference Ontology,” International

Journal of Knowledge Engineering, vol. 2, no. 3, pp.

109-114, September.

Gaevi, D., Jovanovic, J. and Devedzi, V., 2007.

“Ontology-based annotation of learning object

content,” Interactive Learning Environments, vol. 15,

pp. 1-26.

Serra, I., Girardi, R. and Novais, P., 2010. “Reviewing the

Problem of Learning Non-Taxonomic Relationships of

Ontologies from Text,” International Journal of

Semantic Computing, vol. 6, December.

Mädche, A. Staab, S., 2005. Ontology learning for the

semantic web, IEEE Intelligent Systems, vol 16, pp.

72-79.

Yang, H. and Callan, J., 2008. “Human-Guided Ontology

Learning,” Proceeding of HCIR, pp. 26-29.

Rogger, M. and Thaler, S., 2010, “Ontology Learning,”

Seminar paper, Applied Ontology Engineering,

Leopold–Franzens–University Innsbruck.

Cimiano, P., Hotho, A. and Staab, S., 2005. “Learning

concept hierarchies from text corpora using formal

concept analysis” Journal of Artificial Intelligence

research, vol. 24, pp. 305-339.

Buitelaar, P., Cimiano, P. and Magnini, P., 2005.

“Ontology Learning from Text: Methods, Evaluation

and Applications”, Vol. 123, IOS Press, July.

Sabou, M, 2005. “Visual Support for Ontology Learning:

an Experience Report,” Proceeding of IV05, London.

Mazari, C., Aliane, H. and Alimazighi, Z., 2012.

“Automatic Construction of Ontology from Arabic

Texts,” Proceeding of ICWIT, pp.193-202.

Ge, J., Li, Z. and Li, T., 2012. “A Novel Chinese Domain

Ontology Construction Method for Petroleum

Exploration Information,” Journal of Computers, Vol

7, No 6, pp. 1445-1452.

Novelli, A. and Oliveira, J., 2012. “Simple method for

ontology automatic extraction from documents,”

International Journal of Advanced Computer Science

and Applications, v. 3, p.44-51.

Balakrishna, M., Moldovan, D., Tatu, M. and Olteanu, M.,

2010.“Semi-Automatic Domain Ontology Creation

from Text Resources,” Proceeding of LREC’10.

Lenci, A., Montemagni, S., Pirrelli, V. and Venturi, G.,

2009. Ontology learning from Italian legal texts, in

Law, Ontologies and the Semantic Web – Channelling

the Legal Information Flood, Frontiers in Artificial

Intelligence and Applications, Springer, Volume 188,

pages 75–94.

Walter, S. and Pinkal, M., 2006. ”Automatic extraction of

definitions from German court decisions,” Proceeding

of Workshop on Information Extraction Beyond The

Document, pp. 20–28, Sydney, Australia.

Biebow, B. and Szulman, S., 1999. “TERMINAE: A

linguistics-based tool for the building of a domain

ontology,” in Proc. EKAW ’99 - Proceeding of the

11th European Workshop on Knowledge Acquisition,

Modeling, and Management, Berlin, Germany, pp. 49–

66.

Dell’Orletta, F., Venturi, G., Cimiano, A. and

Montemagni, S., 2014. “T2K²: a System for

Automatically Extracting and Organizing Knowledge

from Texts,” proceeding of LREC, pp. 26-31, Iceland.

Jiang, X. and Tan, A. H., 2005. “ Mining Ontological

Knowledge from Domain-Specific Text Documents,”

proceeding of IEEE International Conference on Data

Mining, USA.

Cimiano, P. and Volker, J., 2005. “Text2Onto,” in Natural

Language Processing and Information Systems, ed:

Springer, pp. 227-238.

Mädche, A. and Volz, R., 2001. “The Text-To-Onto

ontology extraction and maintenance system,”

proceeding of 1st International Conference on Data

Mining.

Rudolph, S., Volker, J. and Hitzler, P., 2007. “Supporting

lexical ontology learning by relational exploration,”

Proceeding of ICCS, pages 488–491.

Fouad, Z., Grigoris, A., Mathieu, A., Giorgos, F.,

Haridimos, K. and Enrico, M., 2015. “Ontology

evolution: a process-centric survey,” The Knowledge

Engineering Review, pp. 45–75.

Gherasim, T., Harzallah, M., Berio, G. and Kuntz, P.,

2013. “Methods and Tools for Automatic Construction

of Ontologies from Textual Resources: A Framework

for Comparison and Its Application,” Advances in

Knowledge Discovery and Management. Springer, pp.

177–201.

McCarty, L., T., 2007. “Deep semantic interpretations of

legal texts,” proceeding of ICAIL, pp. 217-224.

Ortiz, A., 2007. “Polionto: Ontology reuse with automatic

text extraction from political documents,” proceedings

of the 6th doctoral symposium in informatics

engineering.

Francesconi, E. ,Montemagni, S., Peters, W., Tiscornia,

D., 2010. “Integrating a bottom–up and top–down

methodology for building semantic resources for the

multilingual legal domain,” Semantic Processing of

Legal Texts: where the Language of Law Meets the

Law of Language, Springer-Verlag, Berlin,

Heidelberg.

ICAART 2017 - 9th International Conference on Agents and Artiﬁcial Intelligence

480