Proceedings of the 3rd Workshop on Scalable Natural Language Understanding, pages 33–40,
New York City, June 2006. c©2006 Association for Computational Linguistics
Scaling Natural Language Understanding via User-driven Ontology
Learning
Berenike Loos
European Media Laboratory, GmbH
Schloss-Wolfsbrunnenweg 33, 69118 Heidelberg, Germany
firstname.lastname@eml-d.villa-bosch.de
Abstract
Non-statistical natural language under-
standing components need world knowl-
edge of the domain for which they are ap-
plied in a machine-readable form. This
knowledge can be represented by manu-
ally created ontologies. However, as soon
as new concepts, instances or relations
are involved in the domain, the manually
created ontology lacks necessary informa-
tion, i.e. it becomes obsolete and/or in-
complete. This means its “world model”
will be insufficient to understand the user.
The scalability of a natural language un-
derstanding system, therefore, essentially
depends on its capability to be up to
date. The approach presented herein ap-
plies the information provided by the user
in a dialog system to acquire the knowl-
edge needed to understand him or her ad-
equately. Furthermore, it takes the posi-
tion that the type of incremental ontology
learning as proposed herein constitutes a
viable approach to enhance the scalability
of natural language systems.
1 Introduction
To let a computer system understand natural lan-
guage one needs knowledge about objects and their
relations in the real world. As the manual modeling
and maintenance of such knowledge structures, i.e.
ontologies, are not only time and cost consuming,
but also lead to not-scalable systems, there exists a
demand to build and populate them automatically or
at least semi automatically. This is possible by an-
alyzing unstructured, semi-structured or fully struc-
tured data by various linguistic as well as statistical
means and by converting the results into an ontolog-
ical form.
In an open-domain scalable natural language un-
derstanding (NLU) system the automatic learning
of ontological concepts and corresponding relations
between them is essential, as a complete modeling
of the world is neither practicable nor feasible, as the
real world and its objects, models and processes are
constantly changing along with their denotations.
This paper assumes that a viable approach to this
challenging problem is to learn ontological concepts
and relations relevant to a certain user in a given con-
text by the dialog system at the time of the user’s
inquiry. My central hypothesis is that the infor-
mation about terms that lack any mapping to the
employed knowledge representation of the language
understanding component can only be found in top-
ical corpora such as the Web. With the help of this
information one can find the right node in the on-
tology to append the concept corresponding to the
unknown term in case it is a noun or to insert it as an
instance in case it is a proper noun or another named
entity.
The goal of the ontology learning component is to
extend the knowledge base of the NLU system and
therefore it will gradually adapt to the user’s needs.
An example from the area of spoken dialog sys-
tems would be that of a user walking through the
city of Heidelberg and asking: “How do I get to
33
the Auerstein”. This would lead to the detection
of Auerstein as being neither recognizable by the
speech recognizer nor mappable to the knowledge
representation of the system. Therefore, the cor-
responding hypernym of Auerstein has to be found
on the internet by recourse to additional information
about the context of the user. In this case, the ad-
ditional information consists of the location of the
user, namely Heidelberg. Once found, the hyper-
nym is mapped to a corresponding concept, which
already exists in the ontology. If there is no such
corresponding concept, the concept for the hyper-
nym thereof has to be determined. The formerly un-
known term is mapped to a concept and is integrated
into the system’s ontology as a child of the concept
for the found hypernym. In case the unknown term
is a proper noun, it is integrated as an instance of the
concept for the hypernym. So far, the research un-
dertaken is related to nouns and proper nouns, also
more generally referred to as terms in this paper.
In the following section, I will describe related
work undertaken to solve the task of ontology learn-
ing, followed by some remarks of the distinction
between ontology learning and natural language in
Section 3. Thereafter, I will sketch out the minimal
stages involved in the type of ontology learning pro-
posed herein in Section 4.
2 Related Work
The capability to acquire knowledge exactly at the
time it is needed can be regarded as an important
stepping stone towards scalable natural language un-
derstanding systems. The necessity of scalability
in NLU became more and more obvious in open-
domain dialog systems, as the knowledge base in-
tegrated into those can never be complete. Before
the emergence of open-domain systems, more or
less complete ontologies were modeled manually for
the domain needed in the NLU system and were
therefore not scalable to additional domains, un-
less modeled in advance in a manual fashion or by
means of off-line ontology learning. Nonetheless,
numerous off-line ontology learning frameworks ex-
ist, which alleviate the work of an ontology engineer
to construct knowledge manually (Maedche, 2002),
(Schutz and Buitelaar, 2005), (Cimiano et al., 2005).
Most of these frameworks apply hybrid methods to
optimize their learning results.
For example, the ontology population method On-
toLearn (Navigli et al., 2004) is based on text min-
ing and other machine learning techniques and starts
with a generic ontology like WordNet and docu-
ments in a given domain. The result is a domain
extended and trimmed version of the initial ontol-
ogy. For this, the system applies three phases to
learn concepts:
• First, a terminology extraction method, using
shallow techniques that range from stochas-
tic methods to more sophisticated syntactic ap-
proaches, is applied, which extracts a list of do-
main terms (mostly nouns and proper nouns)
from a set of documents representative for a
given domain.
• Second, a semantic interpretation takes place
which makes use of a compositional interpreta-
tion and structural semantic interconnections.
• After these two phases the extending and trim-
ming of the initial ontology takes place. With
the help of the semantic interpretation of the
terms they can be organized in sub-trees and
appended under the appropriate node of the ini-
tial ontology applying linguistic rules.
The text understanding system SYNDICATE
(SYNthesis of DIstributed Knowledge Acquired
from Texts) uses an integrated ontology learning
module (Hahn and Marko, 2002). In this approach
new concepts are learned with the help of text un-
derstanding, which applies two different sources of
evidence, namely, the prior knowledge of the topic
domain of the texts and grammatical constructions
in which unknown lexical items occur in the texts.
In an incremental process a given ontology is up-
dated as new concepts are acquired from real-world
texts. The acquisition process is centered on the lin-
guistic and conceptual “quality” of various forms of
evidence underlying the generation and refinement
of concept hypotheses. On the basis of the quality of
evidence, concept hypotheses are ranked according
to credibility and the most credible ones are selected
for assimilation into the domain knowledge base.
The project Disciple (Stanescu et al., 2003) builds
agents which can be initially trained by a sub-
ject matter expert and a knowledge engineer, in
34
a way similar to how an expert would teach an
apprentice. A Disciple agent applies two differ-
ent methods for ontology learning, i.e. exception-
based and example-based ontology learning. The
exception-based learning approach consists of four
main phases:
• First, a candidate discovery takes place, in
which the agent analyzes a rule together with
its examples, exceptions and the ontology and
finds the most plausible types of extensions of
the latter that may reduce or eliminate the rule’s
exceptions.
• In the second phase the expert interacts with the
agent to select one of the proposed candidates.
• Afterwards the agent elicits the ontology exten-
sion knowledge from the expert and finally a
rule refinement takes place, in which the agent
updates the rule and eliminates its exceptions
based on the performed ontology extension.
• When the subject matter expert has to specify a
fact involving a new instance or new feature in
the agent teaching process, the example-based
learning method is invoked. In this process
the agent tries to find example sentences of
the words next to a new term through various
heuristics. For instance, he finds out that X is
member of Y, and consequently can ask the ex-
pert. If he affirms, the new term can be memo-
rized.
All of the approaches described above exhibit the-
oretical as well as practical (in the light of the task
undertaken herein) shortcomings. The theoretical
problems that have not been resolved in a satisfac-
tory manner by the works described above (as well
as numerous others) are:
• a clear separation of the linguistic and ontolog-
ical subtasks involved in the overall ontology
learning endeavor
• systematic ways and methods for evaluating the
individual learning results
• rigorously defined baselines against which to
evaluate the ensuing learning approaches.
In the following I will describe how these is-
sues can be addressed within the user-driven
ontology learning framework proposed herein.
3 Natural Language versus Ontology
Learning
Before describing the actual ontology learning
process it is important to make a clear distinction
between the two fields involved: This is on the one
hand natural language and on the other hand ontol-
ogy learning.
The corpora to extract knowledge from should
come from the internet as this source provides the
most up-to-date information. The natural language
texts are rich in terms, which can be used as labels
of concepts in the ontology and rich in semantic re-
lations, which can be used as ontological relations
(aka properties).
The connection between the two areas which are
working on similar topics but are using different ter-
minology needs a distinction between the extraction
of semantic information from natural language and
the final process of integrating this knowledge into
an ontology.
Figure 1: Natural Language versus Ontology Learn-
ing
Figure 1 shows the process of ontology learning
from natural language text. On the left side relevant
natural language terms are extracted. During a trans-
formation process they are converted into labels of
concepts and relations of an ontology. Proper nouns
are transfered into instance labels in the ontology1.
1In our understanding the term ontology denotes both the
instance model as well as the ground ontology.
35
4 Scaling NLU via User-driven Ontology
Learning
A user-driven ontology learning framework should
be able to acquire knowledge at the run time of the
NLU system. Therefore, terms which are not under-
stood by the system have to be identified. In dialog
systems this is true for all terms uttered or written by
a user, which are not presently contained in the lex-
icon or can be derived by means of derivational or
flexional morphology. In the following I will refer
to these terms as unknown terms2.
When a user of an open-domain spoken dialog
system makes an utterance, it happens regularly, that
the term is not represented in the system’s lexicon.
Since it is assumed, in this work, that the meaning
of terms is represented by means of a formal on-
tology, a user-driven ontology learning framework
is needed to determine the corresponding concepts
for these terms, e.g., via a search on topical corpora.
For instance, a term such as Auerstein could be em-
ployed to query a search engine. By applying natural
language patterns, as proposed by Hearst (1992) and
statistical methods, as proposed by Faulhaber et al.
(2006) possible hypernyms or sets of hypernym can-
didates of the term can be extracted. For these a cor-
responding concept (or set of possible concepts) in
the ontology employed by the dialog system need to
be found. Last but not least the unknown term has to
be inserted into the ontology as either an instance or
a subclass of that concept. This process is described
in greater detail in Section 5.4).
It is important to point out that terms often have
more than one meaning, which can only be deter-
mined by recourse to the context in which it is ut-
tered/found (Widdows, 2003), (Porzel et al., 2006).
Therefore, information about this context needs to
be added in order to make searching for the right
hypernym feasible3 as shown in Section 5.3. For ex-
ample, the term Lotus can refer to a flower, a specific
type of car or among copious other real world enti-
ties to a restaurant in Heidelberg. Therefore, a scal-
able ontology learning framework in a dialog system
requires at least the following ingredients:
2This closely corresponds to what is termed out-of-
vocabulary (OOV) words in the automatic speech recognition
community.
3Of course, even in the same context a term can have more
than one meaning as discussed in Section 5.7.
• A formal explicit model of a shared conceptual-
ization of a specific domain of interest (Gruber,
1993), i.e. an ontology;
• processing methods which indicate the un-
known terms;
• a corpus, as the starting point to retrieve hyper-
nyms;
• methods for mapping hypernyms to concepts in
the ontology;
• an evaluation framework;
Figure 2 shows the steps involved in on-demand
ontology learning from the text to the knowledge
side.
Figure 2: From text to knowledge
5 On-demand learning
From the cognitive point of view learning makes
only sense when it happens on-demand. On-demand
means, that it occurs on purpose and that activity
is involved rather than passivity. As pointed out by
Spitzer (2002) for human beings activity is neces-
sary for learning. We cannot learn by drumming
data into our brain through listening cassettes when
sleeping or by similar fruitless techniques. The rea-
son for this is, that we need active ways of struc-
turing the data input into our brain. Furthermore, we
try only to learn what we need to learn and are there-
fore quite economic with the “storage space” in our
brain.
36
It makes not only for humans sense to simply
learn whatever they need and what is useful for
them. Therefore, I propose that ontology learning,
as any other learning, is only useful and, in the end,
possible if it is situated and motivated by the given
context and the user needs. This can entail learning
missing concepts relevant to a domain or to learn
new concepts and instances which become neces-
sary due to changes in a domain.
However, the fundamental ontological commit-
ments should be adhered to. So, for example, the
decision between a revisionary and a descriptive on-
tology should be kept in the hand of the knowledge
engineer, as well as the choice between a multiplica-
tive and a reductionist modeling4. As soon as the
basic structure is given new knowledge can be in-
tegrated into this structure. Thus, for a reduction-
ist ontology a concept such as Hotel should be ap-
pended only once, e.g. to an ontological concept as
PhysicalObject rather than NonPhysicalObject.
In the following I will describe the various steps
and components involved in on-demand ontology
learning.
5.1 Unknown terms in dialog systems
In case the dialog system works with spoken lan-
guage one can use the out-of-vocabulary (OOV)
classification of the speech recognizer about all
terms not found in the lexicon (Klakow et al.,
2004). A solution for a phoneme-based recog-
nition is the establishment of corresponding best-
rated grapheme-chain hypotheses (Gallwitz, 2002).
Those can be used for a search on the internet. In
case the dialog system only works with written lan-
guage it is easier to identify terms, which cannot be
mapped to ontological concepts, at least if they are
spelled correctly. To evaluate the framework itself
adequately it is useful to apply only correctly writ-
ten terms for a search.
Later on in both cases - i.e. in spoken and written
dialog systems - a ranking algorithm of the best, say
three, hypotheses should be selected to find the most
adequate term. Here methods like the one of Google
“Did you mean...” for spelling errors could be used.
4More information on these and other ontological choices
can be found summarized in (Cimiano et al., 2004)
5.2 Language Understanding
All correctly recognized terms of the user utterance
can be mapped to concepts with the help of an analy-
sis component. Frequently, production systems
(Engel, 2002), semantic chunkers (Bryant, 2004)
or simple word-to-concept lexica (Gurevych et al.,
2003) are employed for this task. Such lexica assign
corresponding natural language terms to all concepts
of an ontology. This is especially important for a
later semantic disambiguation of the unknown term
(Loos and Porzel, 2004). In case the information of
the concepts of the other terms of the utterance can
help to evaluate results: When there is more than one
concept proposal for an instance (i.e. on the linguis-
tic side a proper noun like Auerstein) found in the
word-to-concept lexicon, the semantic distance be-
tween each proposed concept and the other concepts
of the user’s question can be calculated5.
5.3 Linguistic and Extra-linguistic Context
Not only linguistic but also extra linguistic context
plays an important role in dialog systems. Thus, to
understand the user in an open-domain dialog sys-
tem it is important to know the extra-linguistic con-
text of the utterances. If there is a context module
or component in the system it can give information
on the discourse domain, time and location of the
user. This information can be used as a support for a
search on the internet. E.g. the location of the user
when searching for, say Auerstein, is advantageous,
as in the context of the city Heidelberg it has a dif-
ferent meaning than in the context of another city
(Bunt, 2000), (Porzel et al., 2006).
Part of the context information can be represented
by the ontology as well as patterns for grouping a
number of objects, processes and parameters for one
distinctive context (Loos and Porzel, 2005).
5.4 Finding the appropriate hypernym on the
internet
For this, the unknown term as well as an appropri-
ate context term (if available) needs to be applied
for searching possible hypernyms on the Web. As
mentioned before an example could be the unknown
term Auerstein and the context term Heidelberg.
5E.g. with the single-source shortest path algorithm of Dijk-
stra (Cormen et al., 2001).
37
For searching the internet different encyclopedias
and search engines can be used and the correspond-
ing results can be compared. After a distinction be-
tween different types of unknown terms, the search
methods are described.
Global versus local unknown terms: In the case
of generally familiar proper nouns like stars, hotel
chains or movies (so to say global unknown terms),
a search on a topical encyclopedia can be quite suc-
cessful. In the case of proper nouns, only common in
a certain country region, such as Auerstein (Restau-
rant), Bierbrezel (Pub) and Lux (Cinema), which are
local unknown terms, a search in an encyclopedia
is generally not fruitful. Therefore, one can search
with the help of a search engine.
As one can not know the kind of unknown terms
beforehand, the encyclopedia search should be ex-
ecuted before the one using the search engine. If
no results are produced, the latter will deliver them
(hopefully). In case results are retrieved by the for-
mer, the latter can still be used to test those.
Encyclopedia Search: The structure of Encyclo-
pedia entries is generally pre-assigned. That means,
a program can know, where to find the most suit-
able information beforehand. In the case of finding
hypernyms the first sentence in the encyclopedia de-
scription is often found to be the most useful. To
give an example from Wikipedia6, here is the first
sentence for the search entry Michael Ballack:
(1) Michael Ballack (born September 26, 1976
in Grlitz, then East Germany) IS A German
football player.
With the help of lexico-syntactic patterns, the hy-
pernym can be extracted. These so-called Hearst
patterns (Hearst, 1992) can be expected to occur fre-
quently in lexicons for describing a term. In example
1 the pattern X is a Y would be matched and the hy-
pernym football player of the term Michael Ballack
could be extracted.
Title Search: To search only in the titles of web
pages might have the advantage, that results can be
6Wikipedia is a free encyclopedia, which is editable on the
internet: http://www.wikipedia.org (last access: 26th January
2006).
generated relatively fast. This is important as real-
time performance is an important usability factor in
dialog systems. When the titles contain the hyper-
nym it still is to be expected that they might not
consist of full sentences, Hearst patterns (Hearst,
1992) are, therefore, unlikely to be found. Alter-
natively, only the nouns in the title could be ex-
tracted and their occurrences counted. The noun
most frequently found in all the titles could then be
regarded as the most semantically connected term.
To aid such frequency-based approaches stemming
and clustering algorithms can be applied to group
similar terms.
Page Search: For a page search Hearst patterns as
in the encyclopedia search can almost certainly be
applied. In contrast to encyclopedia entries the recall
of those patterns is not so high in the texts from the
web pages.
Figure 3: Tasks for the evaluation of ontology learn-
ing
The text surrounding the unknown term is
searched for nouns. Equal to the title search the oc-
currence of nouns can then be counted. With the
help of machine learning algorithms a text mining
can be done to ameliorate the results.
5.5 Mapping text to knowledge by term
narrowing and widening
As soon as an appropriate hypernym is found in a
text the corresponding concept name should be de-
termined. For term narrowing, the term has to be
stemmed to its most general form. For the term
widening, this form is used to find synonyms. Those
38
are, in turn, used for searching ontological concept
names in the ontology integration phase. If the hy-
pernym found is in a language other than the one
used for the ontology, a translation of the terms has
to take place as well.
5.6 Integration into an ontology
After the mapping phase newly learned concepts,
instances or relations can be integrated into any
domain-independent or even foundational ontology.
If no corresponding concept can be found the next
more general concept has to be determined by the
techniques described above.
5.7 Evaluation
An evaluation of such a system can be divided into
two types: one for the performance of the algorithms
before the deployment of the system and one, which
can be performed by a user during the run time of
the system.
Methodological evaluation Before integrating
the framework into a dialog system or any other
NLU system an evaluation of the methods and their
results should take place. Therefore, a representative
baseline has to be established and a gold-standard
(Grefenstette, 1994) created, depending on the task
which is in the target of the evaluation. The ensuing
steps in this type of evaluation are shown in Figure
3 and described here in their order:
1. The extraction of hypernyms of unknown
words from text and the extraction of semantic
relations (other than is-a) between NLU terms.
2. The mapping of a linguistic term to an ontolog-
ical concept.
3. The integration of ontological concepts, in-
stances and relations into the system’s ontol-
ogy.
Depending on the three steps the most adequate
baseline method or algorithm for each of them has
to be identified. In step 1 for the extraction of hyper-
nyms a chance baseline as well as a majority class
baseline will not do the job, because their perfor-
mance would be too poor. Therefore, a well estab-
lished algorithm which, for example applies a set of
standard Hearst patterns (Hearst, 1992) would con-
stitute a potential candidate. For the mapping from
text to knowledge (see step 2) the baseline could be
a established by standard stemming combined with
string similarity metrics. In case of different source
and goal languages an additional machine transla-
tion step would also become necessary. For the base-
line of ontology evaluation a task-based framework
as proposed by (Porzel and Malaka, 2005) could be
employable.
Evaluation by the user As soon as the framework
is integrated into a dialog system the only way to
evaluate it is by enabling the user to browse the on-
tological additions at his or her leisure and to de-
cide whether terms have been understood correctly
or not. In case two or more hypernyms are scored
with the same – or quite similar – weights, this ap-
proach could also be quite helpful. An obvious rea-
son for this circumstance is, that the term in ques-
tion has more than one meaning in the same context.
Here, only a further inquiry to the user can help to
disambiguate the unknown term. In the Auerstein
example a question like “Did you mean the hotel
or the restaurant?” could be posed. Even though
the system would show the user that it did not per-
fectly understand him/her, the user might be more
contributory and less annoyed than with a question
like “What did you mean?”. The former question
could also be posed by a person familiar with the
place, to disambiguate the question of someone in
search for Auerstein and would therefore mirror a
human-human dialogs, which in turn would further-
more lead to more natural human-computer dialogs.
6 Concluding Remarks
In this paper I have shown, that the scalability of
non-statistical natural language understanding sys-
tems essentially depends on its capability to be up to
date when it comes to understand language. Fur-
thermore, I took the position that ontology learn-
ing is viable, when it happens incrementally and in
a context-sensitive fashion. Future work will focus
on implementation and evaluation within a running
multi-modal dialog system. Additionally, a tight in-
tegration with automatic lexicon and grammar learn-
ing is of paramount importance.
39

References
John Bryant. 2004. Scalable construction-based parsing
and semantic analysis. In Proceedings of the 2nd In-
ternational Workshop on Scalable Natural Language
Understanding (ScaNaLU 2004) at HLT-NAACL 2004.
Harry Bunt. 2000. Dialogue pragmatics and con-
text specification. In H.C. Bunt and W.J. Black, ed-
itors, Computational Pragmatics, Abduction, Belief
and Context; Studies in Computational Pragmatics,
pages 81–150. John Benjamins, Amsterdam.
Philipp Cimiano, Andreas Eberhart, Daniel Hitzler, Pas-
cal Oberle, Steffen Staab, and Rudi Studer. 2004. The
SmartWeb foundational ontology. SmartWeb Project
Report.
Philipp Cimiano, G¨unter Ladwig, and Steffen Staab.
2005. Gimme’ the context: Context-driven automatic
semantic annotation with C-PANKOW. In Proceed-
ings of the 14th World Wide Web Conference. ACM
Press.
Thomas H. Cormen, Charles E. Leiserson, Ronald L.
Rivest, and Clifford Stein. 2001. Section 24.3: Dijk-
stra’s algorithm. In Introduction to Algorithms, Sec-
ond Edition, pages 595–601. MIT Press and McGraw-
Hill.
Ralf Engel. 2002. SPIN: Language understanding for
spoken dialogue systems using a production system
approach. In Proceedings of the International Confer-
ence on Speech and Language Processing 2002, Den-
ver, USA.
Arndt Faulhaber, Berenike Loos, Robert Porzel, and
Rainer Malaka. 2006. Open-class named entity clas-
sification in multiple domains. In Proceedings of the
Ontolex Workshop. Genua, Italy.
Florian Gallwitz. 2002. Integrated Stochastic Models for
Spontaneous Speech Recognition. Logos, Berlin.
Gregory Grefenstette. 1994. Explorations in Automatic
Thesaurus Discovery. Kluwer Academic Publishers,
USA.
Thomas Gruber. 1993. A translation approach to
portable ontology specifications. Knowledge Acqui-
sition (5).
Iryna Gurevych, Rainer Malaka, Robert Porzel, and
Hans-Peter Zorn. 2003. Semantic coherence scoring
using an ontology. In Proc. of the HLT/NAACL 2003,
page (in press), Edmonton, CN.
Udo Hahn and Kornl G. Marko. 2002. Ontology and lex-
icon evolution by text understanding. In Proceedings
of the ECAI 2002 Workshop on Machine Learning and
Natural Language Processing for Ontology Engineer-
ing (OLT’2002). Lyon, France.
Marti A. Hearst. 1992. Automatic acquisition of hy-
ponyms from large text corpora. In Proceedings of
COLING 92, Nantes, France.
Dietrich Klakow, Georg Rose, and Xavier Aubert. 2004.
Oov-detection in a large vocabulary system using au-
tomatically defined word-fragments as filler. In Pro-
ceedings of EUROSPEECH’99, Budapest, Hungary.
Berenike Loos and Robert Porzel. 2004. Resolution of
lexical ambiguities in spoken dialogue systems. In
Proceedings of the 5th Workshop on Diascourse and
Dialogue, Cambridge, Massachusetts, USA.
Berenike Loos and Robert Porzel. 2005. Towards
ontology-based pragmatic analysis. In Proceedings of
DIALOR’05, Nancy, France.
Alexander Maedche. 2002. Ontology Learning for the
Semantic Web. Kluwer Academic Publishers, USA.
Roberto Navigli, Paola Velardi, Alessandro Cucchiarelli,
and Francesca Neri. 2004. Extending and enriching
WordNet with OntoLearn. In Proceeedings of Inte-
grated Approach for Web Ontology Learning and En-
gineering. IEEE Computer.
Robert Porzel and Rainer Malaka. 2005. A task-based
framework for ontology learning, population and eval-
uation. Ontology Learning from Text: Methods, Eval-
uation and Applications Frontiers in Artificial Intelli-
gence and Applications Series, 123.
Robert Porzel, Iryna Gurevych, and Rainer Malaka.
2006. In context: Integrating domain- and situation-
specific knowledge. SmartKom Foundations of Mul-
timodal Dialogue Systems, Springer, Cognitive Tech-
nologies.
Alexander Schutz and Paul Buitelaar. 2005. RelExt:
A tool for relation extraction in ontology extension.
In Proceedings of the 4th International Semantic Web
Conference. Galway, Ireland.
Manfred Spitzer. 2002. Lernen. Spektrum Akademis-
cher Verlag.
Bogdan Stanescu, Cristina Boicu, Gabriel Balan, Mar-
cel Barbulescu, Mihai Boicu, and Gheorghe Tecuci.
2003. Ontologies for learning agents: Problems, solu-
tions and directions. In Proceedings of the 2003 IEEE
International Conference on Systems, Man and Cyber-
netics, Volume: 3. Washington D.C.
Dominic Widdows. 2003. A mathematical model for
context and word-meaning. In International and Inter-
disciplinary Conference on Modeling and Using Con-
text. Stanford, California.
