=Paper=
{{Paper
|id=None
|storemode=property
|title=BauDenkMalNetz - Creating a Semantically Annotated Web Resource of Historical Buildings
|pdfUrl=https://ceur-ws.org/Vol-721/paper-04.pdf
|volume=Vol-721
|dblpUrl=https://dblp.org/rec/conf/esws/Dumitrache011
}}
==BauDenkMalNetz - Creating a Semantically Annotated Web Resource of Historical Buildings==
https://ceur-ws.org/Vol-721/paper-04.pdf
BauDenkMalNetz – Creating a Semantically
Annotated Web Resource of Historical Buildings
Anca Dumitrache and Christoph Lange
Computer Science, Jacobs University Bremen, Germany
{a.dumitrache,ch.lange}@jacobs-university.de
Abstract. BauDenkMalNetz (“listed buildings web”) deals with cre-
ating a semantically annotated website of urban historical landmarks.
The annotations cover the most relevant information about the land-
marks (e.g. the buildings’ architects, architectural style or construction
details), for the purpose of extended accessibility and smart querying.
BauDenkMalNetz is based on a series of touristic books on architectural
landscape. After a thorough analysis on the requirements that our website
should provide, we processed these books using automated tools for text
mining, which led to an ontology that allows for expressing all relevant
architectural and historical information. In preparation of publishing
the books on a website powered by this ontology, we analyze how well
Semantic MediaWiki and the RDF-aware Drupal 7 content management
system satisfy our requirements.
1 Motivation
The architectural landscape of a city is not just made up of well-established
landmarks, but of historical buildings with a rich cultural background that lie
outside the mainstream touristic circuit. People wanting to explore less known
places of a city have little access to information about these hidden architectural
gems and the stories behind them, even though all required data on historical
buildings in Germany has been meticulously collected by the offices for historical
monuments (Denkmalämter). However, this data has generally not been published
in an easily accessible way. Existing databases and form-based search facilities
are often tedious to browse through.1
In Bremen, an effort to collect this information and present it to the general
public was made by the publisher Nils Aschenbeck, who released a series of
city guide books [AW09]. However, for the moment, these books have only been
published in print. By making use of these books, BauDenkMalNetz (German
for “listed buildings web”) proposes a way of discovering Bremen’s architectural
landscape that is suited for the tech-savvy tourist.
1
See, for example, http://194.95.254.61/denkmalpflege/index.htm.
2 Transitioning from Written Text to Digital Media
The purpose of BauDenkMalNetz is to develop a web portal that publishes online
printed text enriched with semantic annotations. Publications usually make use
of a concrete set of concepts, that relate to one particular subject area, and thus
can be reduced to a strict vocabulary. Identifying this vocabulary was a key step
in the process of producing a formal representation of the semantic metadata
that our web portal needs to store. After we have created a conceptual model
of our data, we want to analyze ways of publishing our semantically enriched
text online. Finally, we want to compare and contrast BauDenkMalNetz to other
cultural heritage web applications, and identify possible directions for further
work.
2.1 Building an Ontology
The publications that lie at the basis of our work with BauDenkMalNetz have
been made available to us (but not the general public) in simple HTML files.
There is a file for each individual building, with pictures associated to each file,
and information like the name of the architect being highlighted. Four books
have been published thus far [AW09], with more than one hundred buildings
being described in total.
In order to enable enhanced browsing and querying, the data on Bremen’s
historical buildings needs to be organized, and the proper semantic metadata
needs to be put in place. For this purpose, we have developed the BauDenkMal-
Netz ontology, a formal representation of the metadata vocabulary on historical
buildings and related concepts, together with the relations among them. The
ontology has been formalized and implemented in OWL, and was engineered in
the stages specified by the METHONTOLOGY [FLGPJ97] methodology.
Scenario An example scenario of interacting with a publication backed by the
BauDenkMalNetz ontology involves a tourist, working out an itinerary for visiting
the city of Bremen. For this purpose, she needs to be able to browse through
a particular neighborhood, by filtering the buildings based on their addresses.
Suppose she is interested only in visiting those buildings that were built in the
19th century. Then she finds one particular architect that she is familiar with,
and she wants to add all of his buildings to her itinerary. Finally, during her visit,
she will want to stop at each individual building and read up on its history, like
the years between it was built, and what famous people had been living there.
Requirements Based on this scenario, we have identified a list of requirements
that the BauDenkMalNetz ontology needs to meet in order for the data to be
easily accessible:
– buildings need to be represented as uniquely identified entities, which will be
mapped to individual pages of the website; any knowledge represented using
the BauDenkMalNetz ontology needs to be interconnected, with the building
entity as the central point of the representation;
– information on the physical address and neighborhood needs to be available
for every building;
– the architect and the architectural style of a building have to be highlighted
when that information is available;
– the time and timespan over which a building was built has to be specified for
individual entries.
A more general requirement that the BauDenkMalNetz website needs to
address is browsing from one building to another. This could be supported by
information on the buildings’ physical location (e.g. they are on the same street),
or based on characteristics that they share (e.g. they were built by the same
person).
Text Analysis Starting from these requirements and based on the original
touristic guides, we identified they key concepts of the vocabulary that relates
to historical buildings, by employing n-gram models 2 to find the most likely
occurrences of word groupings. The results of this analysis were used in the
conceptualization phase of the BauDenkMalNetz ontology. The fact that the
accuracy of n-gram models increases with the volume of the processed text was
an advantage that made us consider this approach.
The first step that enabled us to process the text was removing the unnecessary
HTML tags, and stripping it down to a plain-text format. The text is written in
German; we needed to normalize it to plain ASCII characters, as the German-
specific special characters seemed to interfere with the script used to analyze it.
We made use of the LaMaPUn [GJA+09] Perl library for processing the text.
We used a list of the most frequent German stop words in order to filter out the
information that was not meaningful for the domain vocabulary.
We analyzed series of 1 to 4-gram models. The script recognized over 600
possible groupings of words that are likely to occur together. Over 500 of these
groups had a likelihood coefficient larger than 2. This coefficient is computed by
having the number of incidences of the words in the group together divided by
the sum of individual incidences outside of the group.
The text analysis made apparent some clear trends. Most of the likely groups of
words that appeared together referred to one of the following categories: physical
buildings (e.g. Bahnhof (train station) Sankt Magnus, Kirche (church) Sankt
Magni), personal names (e.g. Rudolf Alexander Schroeder), physical addresses
(e.g. Leuchtenburger Strasse (a street), Am Bahnhof Sankt Magnus) and building
features (e.g. Bungalow, Turm (tower)). By identifying these categories, we got a
first impression of what are the key concepts we need to define for our ontology.
2
A probabilistic model that, given the first n − 1 words in a sentence, will predict the
nth word. [MS99]
Conceptualization Based on this analysis, and according to the requirements
identified in the previous section, we conceptualized entities to be represented
in the BauDenkMalNetz ontology3 . Most concepts identified during the n-gram
analysis were transformed into resources, then properties were added to connect
them. The core of the BauDenkMalNetz ontology is the following (concepts
underlined, relations in italics):
– building – a resource identifying a particular building;
– building part – a subconcept of the building entity (e.g. tower, annex);
– building complex – a composite consisting of several building entities;
– building type – different types of constructions (e.g. church, hospital);
– address – the physical location of a building;
– architect – the person or group of people that have designed the building;
– inhabitant – famous person that has lived in that building;
– year – when a building was built; can refer to the year when construction
began, ended, or both.
Fig. 1. A fragment of the BauDenkMalNetz ontology
3
Available at: http://oaff.info/ontology/bdmn#.
Alignment to Other Ontologies The Linked Data community [Hea+] ad-
vocates the reuse of knowledge models and vocabularies, in order to achieve
interoperability across the Web. Indeed, there already exist various ontologies
that model some of the relevant knowledge about historical buildings, out of
which we found the following ones relevant for aligning with the BauDenkMalNetz
ontology:
– The GeoNames [Geo] ontology models geospatial semantic information. In
particular, it assigns to individual locations on the globe a unique URI. For
our purposes, it can be used to uniquely identify each historical building
based on its coordinates. Reusing this ontology brings the added advantage
of explicitly specifying the geolocation of a building, which allows for easier
integration with web mapping services.
– The CIDOC CRM [Cid] ontology represents the detailed scientific docu-
mentation of cultural heritage objects, which include historical monuments.
By aligning our ontology to CIDOC CRM, we can formulate a full description
of the historical information related to a building (e.g. the architectural style
of the monument, the official sources which document the monument etc.).
2.2 Publishing in a Semantic Content Management System
For deploying BauDenkMalNetz, we have so far established requirements and
analyzed how well two semantic content management systems satisfy these
requirements: Semantic MediaWiki (SMW [Sem]) and Drupal 7 [Dru].
Requirements Based on the scenario discussed in the previous section, we
have also analyzed the requirements that our website needs to provide. Digitally
representing publications means that the BauDenkMalNetz web portal needs to
build on the use cases of the written text that lies at its core, and enhance them
with semantic browsing and querying capabilities that will provide for a better
user experience. Therefore, a suitable content management system for deploying
BauDenkMalNetz should offer the following functionality:
1. the possibility of integrating RDF triples, and at least a minimum of ontology
support;
2. support for querying the RDF content of the website (e.g. by using SPARQL);
3. browsing based on the semantic metadata;
4. extensible publishing support for:
(a) users, through enabling PDF and HTML exporting;
(b) machines, by interlinking the publications across the Web, according to
linked data principles;
5. the possibility of importing large amounts of text into the system.
Semantic MediaWiki SMW [Sem] was built as an extension of MediaWiki,
the wiki engine which powers Wikipedia. It provides enhanced features for
browsing and organizing its contents via semantic annotations. We built the first
BauDenkMalNetz prototype using SMW [DLK+10].
Our motivation for using SMW in deploying the initial version of our web
portal was its suitability for rapidly creating a working prototype (cf. [BDH+09]).
SMW allows for easily adding and editing of the necessary data and metadata
available on historical buildings, in keeping with requirements 1 and 3. New
information could be easily incorporated and linked to the already existing data
via SMW’s page creation and editing tools. At the same time, the metadata
vocabulary (i.e. the ontology) could be easily modified, simply by adding in-text
annotations.
Requirement 2 is addressed by a simple query language included in SMW.
The SMW querying functionality does not operate directly on RDF, and instead
uses a syntax that addresses RDF triples based on the names with which they are
declared in the wiki pages. While it provides basic functionality for querying RDF
data, which includes selecting pages in the wiki, together with what properties of
the pages to display, the SMW query language lacks the complexity of SPARQL
(e.g. querying within a particular namespace).
Fig. 2. Screenshot of the SMW prototype
When further assessing requirement 1, we found that the conceptual model of
our metadata was less obvious and never explicitly formalized, as the ontology, to
which the texts adhere, is not necessarily specified explicitly in SMW, but rather
implied from the annotations done directly on the text. In this case, alignment
to other similar ontologies (in keeping with the linked-data philosophy of reuse)
is still possible, yet it is rendered more difficult by the lack of an explicit formal
definition of the ontology.
Requirement 5 was also not addressed by our prototype. SMW provides
some tools suited for database import, however the texts we want to analyze
are stored in simple HTML files. The volume of data that needs to be processed
makes it almost impossible to have the texts annotated manually, like we did for
building the prototype, while also making BauDenkMalNetz rather suited for
the employment of natural language processing techniques in order to get the
needed semantical annotations.
Drupal 7 As our goal is to publish existing content, rather than creating
new content in a collaborative way, we also considered Drupal [Dru], a rather
traditional content management system. Given the BauDenkMalNetz documents
collection and our ontology, we have so far analyzed Drupal’s features w.r.t. the
requirements established above. Deploying BauDenkMalNetz in Drupal remains
to be done in spring 2011.
Requirement 1 is satisfied as the latest version 7 of Drupal provides an
RDF API [CDC+09] that is integrated in the Drupal core. This enabled us to
easily upload our OWL ontology into the website, by using the RDF vocabulary
import feature. The keywords pertaining to each resource were then added to the
taxonomy of our website, and mapped to the corresponding classes and properties
in the ontology. For printed media, where a particular text usually does not
undergo much change after being published, the advantage that Drupal brings
is that, as the structure of the text is already known, its conceptualization can
be set as the core of the website via the RDF API even before the website is
deployed.
Requirement 2 is addressed by the SPARQL module for Drupal, which allows
us to query our external triple store. The task of building meaningful queries is
made even easier by the SPARQL Views [Cla10] module, which supports visual
query building and result display.
Results When comparing SMW to Drupal, we have encountered some drawbacks
of SMW that led us to reconsider our approach. The flexibility and agility of
SMW were not of a particular advantage in our setting. The publication sources
are imported from external sources, and therefore we are not interested in
MediaWiki’s collaboration support. The ontology and its connections to other
ontologies are, for now, created just by us, but they are not evolved or extended
dynamically by a community – therefore we are not interested in giving write
access to the ontology via the content management system. We rather prefer
having a clear conceptual model of the metadata from the beginning. Drupal
supports the initial import of such an ontology before importing the content and
thus is suited for managing annotations to publications that have already existed
before.
Also, we have concluded that using SPARQL to power our query engine
would provide more flexibility for our queries, while also making them portable,
as SPARQL is not platform dependent. While SMW is currently working to
integrate SPARQL4 functionality in its core, for the moment, the support it
provides is limited, whereas Drupal provides SPARQL support through the
modules discussed in the previous section.
Table 1. Comparison of SMW and Drupal based on the requirements list.
Req. SMW Drupal Results
1. inline RDF triples RDF part of the core, Evoc Drupal for bet-
declaration, no explicit module for ontology import ter ontology sup-
ontology support port
2. SMW query language SPARQL, SPARQL Views Drupal for ad-
modules vanced querying
possibilities
3. wiki pages mapped to RDF mapping for content draw
resources and categories types
4a. third-party plugin, not well Printer, e-mail and PDF Drupal
documented versions module in
developmenta
4b. synchronizing with Evoc external vocabulary draw
vocabularies supported by support
SMW through exportb and
importc
5 through page creation, through page creation, but Drupal
with manual semantic with specialized content
annotations types
a
http://drupal.org/project/print
b
http://semantic-mediawiki.org/wiki/Help:RDF_export
c
http://semantic-mediawiki.org/wiki/Help:Import_vocabulary
3 Development and Evaluation Plan
During spring 2011, we continued developing the BauDenkMalNetz website in
Drupal, by uploading the texts of the tourist guides to our website, with the
keywords in the vocabulary highlighted in the resource’s pages. We will make
semantic browsing available, based on these key concepts, achieved through
Drupal’s taxonomy feature. Also, for increased functionality, we will add a
geospatial aspect to the semantic navigation by utilizing the Google Maps
4
http://semantic-mediawiki.org/wiki/SPARQL_and_RDF_stores_for_SMW
API [Goo]. Finally, resources referring to people (e.g. the architect) will be
cross-referenced with Linked Data resources, like DBPedia5 .
For even more advanced querying features, we are considering to make use
of the XSPARQL [AKK+08] query language. XSPARQL combines the XML
query language XQuery with the RDF query language SPARQL, which allows for
generating XML-formatted results for queries over the semantic metadata of our
website and, in future, interlinked websites. By selecting from a list of available
queries, tourists will be able to create personalized guides of historical buildings.
For evaluating the usability of the BauDenkMalNetz website, existing methods
for evaluating (semantic) digital libraries [FTA+07; Kru09] are applicable. A
group of test-users will navigate through the website, providing feedback based on
usability (of the content management system with our extensions) and usefulness
(of the content, in the way our system publishes it). The users will provide
feedback on how easy/difficult it is to find a particular building, by querying the
system based on a criteria of their own choosing (e.g. location, architectural style
etc.), and also about how they managed to find their way from one particular
building to another, based on a common characteristic. They will also be asked to
provide their input on how accurate the query results are in relation to what they
were expecting to find, and also about the informative character of individual
buildings’ pages. Based on this assessment the user-friendliness of the website we
will consider possible improvements. A first release of BauDenkMalNetz, adapted
according to the results of an initial evaluation round, is expected in May.
4 Related Work on Cultural Heritage
There exist a number of projects that process data about cultural heritage using
semantic web technologies. Most approaches encountered gather the information
from a wide array of sources (e.g. historical documents, archaeological excavation
reports etc.), and consequently one of their main issues is developing an ontology
that serves as a common medium for these different types of texts. In contrast,
the BauDenkMalNetz ontology was developed from a singular source – published
texts written in the same style, by the same author, on the topic of cultural
heritage. Therefore, the ontology’s intended use is not to provide a universal
definition of the vocabulary describing historical buildings, but to define the
vocabulary used by this particular series of publications. By studying the related
work on cultural heritage we were able to shed some light on how we could
improve our data model in order to represent a greater pool of sources, therefore
enabling the reusability of our core ontology. For this purpose, the following
applications have been assessed:
MANTIC [MPV10] is a project similar to BauDenkMalNetz, that represents
data on cultural heritage sites of the city of Milan, that was gathered from
historical sources and publications. At its core, it uses the CIDOC CRM ontology
for storing information about the archeology of the city. This information is then
5
http://dbpedia.org
incorporated into the Google Maps API, making for an easy to use application
for browsing Milan’s historical landmarks, that is quite similar in scope to our
work. Unlike BauDenkMalNetz, MANTIC deals with historical sources, which
comprise a great variety of publications, written in different styles and over a
long period of time. MANTIC provides a good example of how CIDOC CRM
can be reused for representing historical landmarks, however, since the sources
MANTIC deals with are so disjointed, identifying a common vocabulary for them
is more difficult, and therefore no special ontology that deals primarily with
historical buildings was devised.
The Fundación Marcelino Botín [Fun] worked on a similar project that
aimed to gather information on eleven cultural heritage sites of Cantabria, a
region of Northern Spain. Like MANTIC, the Cantabria project had to reconcile
information from a heterogeneous set of sources, by adapting the CIDOC CRM
ontology to suit their dataset. However, most of the data populating the ontology
had already been preprocessed (as spreadsheets, web pages etc.), and adding
content to the project website was done in a semi-automated way. Therefore,
unlike BauDenkMalNetz, the Cantabria project is intended as a community
portal, where experienced users can modify or add new data to the website and
to the ontology. Aside from providing another example of how to reuse existing
standards, this project is relevant for us because of the way it makes use of the
various benefits brought by using semantic metadata: a semantic search engine,
an interactive map based on geoposition metadata, and interoperability with
other cultural heritage repositories.
CultureSampo [HMK+09] is an application that publishes cultural heritage
information about Finland. Like BauDenkMalNetz, CultureSampo builds on
existing standards for conceptualizing cultural items, and then extends them
with domain specific information. However, as it covers a larger content (history,
folklore, artifacts etc.), CultureSampo integrates a wide array of domain specific
ontologies, that were developed in a semi-automatic fashion based on existing
thesauri. While the development methodology of CultureSampo is relevant and
can be adapted for BauDenkMalNetz, the scope of the project is too wide to
enable us to reuse their data model.
5 Conclusion and Further Work
After assessing in which ways traditional printed publications on historical land-
marks can be enhanced by transposing them in a digital format and enriched with
semantic annotations, we devised the BauDenkMalNetz ontology, by analyzing
its requirements and processing the texts that were made available to us by using
text mining techniques. In keeping with linked data principles, we aligned our
ontology to other existing representations that relate to our specific domain, like
CIDOC CRM and GeoNames. Once we determined the structure of our metadata,
we compared how different content management systems (SMW and Drupal 7)
satisfy the requirements for deploying the BauDenkMalNetz website. As Drupal
provides a more rigorous way of declaring a conceptual model, which is more
suitable for digital publications, we have chosen it as the medium in which our
web portal will be developed.
Once finished, the BauDenkMalNetz website will provide a comprehensive and
easy-to-use guide to the city of Bremen, and possibly even help boost the touristic
appeal of Bremen. A possible enhancement to the resource will be creating a
mobile version of the website, so that tourists can create virtual itineraries that
they can access on the go. However, the scope of our work is not limited to
Bremen. We believe that both the ontology and the vocabulary will prove general
enough to adapt in order to represent any touristic publication guide on historical
landmarks.
Acknowledgments
The authors would like to thank Deyan Ginev for help with the LaMaPUn library,
Lin Clark for help with assessing Drupal 7, and the anonymous peer reviewers
for their pointers to further related work.
References
[AKK+08] W. Akhtar, J. Kopecký, T. Krennwallner, et al. “XSPARQL: Trav-
eling between the XML and RDF worlds – and avoiding the XSLT
pilgrimage”. In: The Semantic Web: Research and Applications. 5th
European Semantic Web Conference (ESWC) (Tenerife, Spain).
Ed. by S. Bechhofer, M. Hauswirth, J. Hoffmann, et al. LNCS 5021.
Springer Verlag, 2008.
[AW09] N. Aschenbeck and I. Windhoff. Landhäuser und Villen in Bremen.
Bremen: Aschenbeck Verlag, 2009.
[BDH+09] J. Bao, L. Ding, R. Huang, et al. “A Semantic Wiki based Light-
Weight Web Application Model”. In: Proceedings of the 4th Asian
Semantic Web Conference. 2009, pp. 168–183.
[CDC+09] S. Corlosquet, R. Delbru, T. Clark, et al. “Produce and Consume
Linked Data with Drupal!” In: The Semantic Web. 8th International
Semantic Web Conference (ISWC). Ed. by A. Bernstein, D. R.
Karger, T. Heath, et al. LNCS 5823. Springer, Oct. 2009.
[Cid] The CIDOC Conceptual Reference Model. url: http://cidoc.ics.
forth.gr (visited on 2010-03-07).
[Cla10] L. Clark. “SPARQL Views: A Visual SPARQL Query Builder for
Drupal”. In: Poster and Demo Proceedings of the 9th International
Semantic Web Conference (ISWC). 2010. url: http://iswc2010.
semanticweb.org/pdf/518.pdf.
[DLK+10] A. Dumitrache, C. Lange, M. Kohlhase, et al. “Prototyping a
Browser for a Listed Buildings Database with Semantic MediaWiki”.
In: 5th Workshop on Semantic Wikis. Ed. by C. Lange, J. Reu-
telshöfer, S. Schaffert, et al. CEUR Workshop Proceedings 632.
2010. url: http://ceur-ws.org/Vol-632/.
[Dru] Drupal.org – Community plumbing. web page at http://drupal.
org. url: http://drupal.org.
[FLGPJ97] M. Fernández-López, A. Gómez-Pérez, and N. Juristo. “METHON-
TOLOGY: from Ontological Art towards Ontological Engineering”.
In: Proceedings of the Fourteenth National Conference on Artificial
Intelligence AAAI-97. (Stanford, USA). MIT Press, 1997, pp. 33–40.
[FTA+07] N. Fuhr, G. Tsakonas, T. Aalberg, et al. “Evaluation of digital
libraries”. In: International Journal of Digital Libraries 8 (2007),
pp. 21–38.
[Fun] Case Study: An Ontology of Cantabria’s Cultural Heritage. url:
http : / / www . w3 . org / 2001 / sw / sweo / public / UseCases /
FoundationBotin/ (visited on 2011-04-12).
[Geo] GeoNames. url: http : / / www . geonames . org (visited on
2010-04-23).
[GJA+09] D. Ginev, C. Jucovschi, S. Anca, et al. “An Architecture for Lin-
guistic and Semantic Analysis on the arXMLiv Corpus”. In: Ap-
plications of Semantic Technologies (AST) Workshop, Informatik.
2009. url: http://www.kwarc.info/projects/lamapun/pubs/
AST09_LaMaPUn+appendix.pdf.
[Goo] Google Maps. url: http : / / maps . google . com (visited on
2011-01-10).
[Hea+] T. Heath et al. Linked Data – Connect Distributed Data across the
Web. url: http://linkeddata.org (visited on 2010-06-11).
[HMK+09] E. Hyvönen, E. Mäkelä, T. Kauppinen, et al. “CULTURE SAMPO –
A National Publication System of Cultural Heritage on the Semantic
Web 2.0”. In: ESWC. 6th European Semantic Web Conference
(ESWC). Ed. by L. Aroyo, P. Traverso, F. Ciravegna, et al. LNCS
5554. Springer, 2009.
[Kru09] S. R. Kruk. “Semantic Digital Libraries. Improving Usability of
Information Discovery with Semantic and Social Services”. PhD
thesis. National University of Ireland, Galway, 2009.
[MPV10] G. Mantegari, M. Palmonari, and G. Vizzari. “Rapid Prototyping
a Semantic Web Application for Cultural Heritage: The Case of
MANTIC”. In: The Semantic Web: Research and Applications (Part
II). 7th Extended Semantic Web Conference (ESWC). Ed. by L.
Aroyo, G. Antoniou, E. Hyvönen, et al. LNCS 6089. Springer, 2010.
[MS99] C. D. Manning and H. Schütze. “Statistical Inference: n-gram Mod-
els over Sparse Data”. In: Foundations of Statistical Natural Lan-
guage Processing. MIT Press, 1999. Chap. 6.
[Sem] Semantic MediaWiki. url: http : / / semantic - mediawiki . org
(visited on 2010-03-04).