Journal for Language Technology and Computational Linguistics (JLCL)

Stede, Manfred

English

Local Coherence Analysis in a Multi-Level Approach to Automatic Text Analysis

Gamallo Otero, Pablo

Pereira Lopes, Gabriel

Agustini, Alexandre

Automatic Acquisition of Formal Concepts from Text

Lüngen, Harald

Mehler, Alexander

Storrer, Angelika

iiJLCL - Volume 23(2) - 2008
Following the special editions 22(2) and 23(1) on Foundations of ontologies in text-technolo-
gy, this is the third volume of the LDV-Forum to originate from activities of the research 
unit 437 Text-technological modelling of information, funded by the German Research Foun-
dation (DFG) in its second phase 2005-2009. One of the research goals shared by four out 
of five subprojects within the group was the automated analysis of different types of dis-
course relations and the construction and evaluation of domain ontologies and lexical-
semantic wordnets as knowledge sources in this task.
The subproject HyTex - Text-Grammatical Foundations for the (Semi)-Automated Text-to-
Hypertext Conversion (Principal Investigator: Angelika Storrer) was concerned with the 
identification of thematic development structures in specialised texts for the purpose of 
hypertextualisation. The goal of the subproject Indogram - Induction of document gram-
mars for the Representation of Logical Hypertextual Document Structures (Principal Investi-
gator: Alexander Mehler) was to research methods of learning document and content 
structures from very large corpora of hypertext (web) documents. Finally, the subproject 
SemDok - Generic Document Structures in Linearly Organised Texts (Principal Investigator: 
Henning Lobin) dealt with building a text parser in the framework of Rhetorical Struc-
ture Theory for the complex text type of scientific research articles. As a joint intitiative 
of these three projects, a workshop entitled Ontologies and Semantic Lexica in Automated 
Discourse Analysis was held in conjunction with the Arbeitskreis Korpuslinguistik of the 
GLDV at the GLDV Frühjahrstagung in April 2007 in Tübingen.
The workshop included a most inspiring invited talk by Manfred Stede, entitled “Ge-
wusst wie - Ontologien und Textkohärenz”, which presented an insightful overview of 
the history of automatic text understanding systems from early knowledge-based sy-
stems within the discipline of Artificial Intelligence in the 1970s and 1980s to the rise of 
statistical methods in the 1990s and to more recent hybrid approaches and his own text-
technological, multi-level analysis approach to text processing. A concise version of this 
historical overview is now contained as a part of Stede’s contribution to the present vo-
lume.
Altogether, four of the five contributions to the present volume are paper versions of 
talks held at the workshop, namely the ones by Manfred Stede, Caroline Sporleder, Bä-
renfänger et al. and Diewald et al. The fifth article by Cramer et al. was additionally re-
viewed and included due to its immediate relevance for the field of lexical-semantic res-
sources in automated discourse analysis. Each paper was reviewed for the LDV-Forum 
by two external reviewers; as is customary, reviewers for the first round were chosen by 
the guest editors, and reviewers for the second round were chosen by the regular edi-
tors of the LDV-Forum.
Editorial
Harald Lüngen, Alexander Mehler, Angelika Storrer
Editorial
iii JLCL
Discourse analysis in the title of this volume, and, relatedly, discourse structure and dis-
course relations are used as cover terms for various types of relational structuring of text 
beyond the domain of sentences. Several levels of discourse structure can be identified 
on account of the types of relations used and the linguistic units involved in them. Let 
us briefly review the most important levels of discourse structure and current theoreti-
cal and practical approaches to their (automated) identification in text.
The first one is the level of coreference, or anaphora. Anaphora is a cohesive phenome-
non that occurs intra-sententially as well as inter-sententially. Anaphoric relations are 
usually described to hold between discourse entities that are elements of the semantic 
model of the text world, or alternatively, between the linguistic units that are used to ex-
press them. Current anaphora resolution systems such as described in Vieira (2000) or 
Stuckardt (2005) use large amounts of annotated training corpora. Diewald et al.’s (pp. 
74–92) contribution to this volume describes a novel, web-based multi-user annotation 
tool for semantic relations which can be used to produce corpora for a specialised task 
like anaphora resolution. In Bärenfänger et al. (pp. 49–72), an ontology of anaphoric re-
lation types is introduced, and the interrelationship between anaphora, rhetorical rela-
tions, and thematic development is examined in a corpus study based on linguistic an-
notations on multiple levels.
On another level of discourse structure, which may be called lexical cohesion after Hal-
liday and Hasan (1976), the content words occurring in a text are grouped into lexical 
chains based on lexical-semantic relations holding between them. Lexical chaining is 
a computational linguistic application first introduced by Morris and Hirst (1991). In 
their original algorithm, chains were derived by means of looking up lexical-semantic 
relations between content words in Roget’s Thesaurus. More recent approaches to lexi-
cal chaining use wordnets as a knowledge source (Hirst and St-Onge, 1998), or establish 
semantic relatedness using terminologies or social ontologies such as Wikipedia (Meh-
ler 2009, Waltinger et al., 2008). The third article of this volume (Cramer et al. pp. 34–47) 
describes experiences with the development of the lexical chainer GLexi, which derives 
semantic distances from GermaNet and was tested on the HyTex corpus of German spe-
cialised texts. GLexi is ultimately supposed to function as a component in the generati-
on of topic views as an automated hypertextualisation strategy applied to a given linear 
text. Topic views can be regarded as an approximation to a representation of thematic 
development in a text.
Many researchers use the term discourse structure exclusively as a label for the system 
of coherence relations between text segments of different size, usually with propositio-
nal content. Examples of discourse relations of this type are the causal, the contrastive, 
or the elaboration relation. A number of discourse theories aim at describing the admis-
sible structures of text-type-independent discourse coherence relations, notably SDRT 
(Asher and Lascarides, 2003), RST (Mann and Taboada, 2006), or the ULDM (Polanyi 
et al., 2003). Discourse coherence relations are frequently associated with lexical items 
called discourse connectives, but at the same time, many coherence relation instances in 
ivVolume 24 - 2008
a text lack overt cues. The contribution by Sporleder (p. 20–32) in this volume presents 
an evaluation of machine learning models in which lexical-semantic relations from the 
Princeton WordNet are used to disambiguate discourse coherence relations from SDRT 
that lack overt or unambiguous discourse markers. The contribution by Bärenfänger et 
al. (p. 49–72) explores the question how types of anaphoric relations annotated in a cor-
pus of scientific articles can help identify instances of the RST elaboration relation.
We would like to thank both the review board for the extended abstracts submitted for 
the Tübingen workshop as well the reviewers of the paper versions. Without their sup-
port we would not have accomplished another edition of the LDV-Forum of such high 
calibre: Irene Cramer, Marcus Egg, Christiane Fellbaum, Iryna Gurevych, Anke Holler, 
Kai-Uwe Kühnberger, Peter Kühnlein, Lothar Lemnitzer, Henning Lobin, Vivian Rai-
thel, Georg Rehm, Roman Schneider, Bernhard Schröder, Manfred Stede and Christian 
Wolff. Furthermore, we would like to thank the editors of the LDV-Forum, Alexander 
Mehler and Christian Wolff and their team for their support and advice. We hope that 
the readers of the LDV-Forum will find the included papers as interesting and illumi-
nating as we did.
December 2008
Harald Lüngen
Alexander Mehler
Angelika Storrer
References
Literatur
Asher, N. and Lascarides, A. (2003). Logics of Conversation. Cambridge University Press,
 Cambridge, UK.
Halliday, M. and Hasan, R. (1976). Cohesion in English. English Language Series. Long 
 man, London, 5 edition.
Hirst, G. and St-Onge, D. (1998). Lexical chains as representation of context for the  
 detection and correction malapropisms. In Fellbaum, C., editor, WordNet: An  
 electronic lexical database.
Morris, J. and Hirst, G. (1991). Lexical cohesion computed by thesaural relations as an  
 indicator of the structure of text. Computational Linguistics, 17(1):21–48.
v JLCL
Polanyi, L., van den Berg, M., and Ahn, D. (2003). Discourse structure and sentential  
 information structure. Journal of Logic, Language and Information, 12:337–350.
Stuckardt, R. (2005). A machine learning approach to preference strategies for anaphor
 resolution. In Branco, A., McEnery, T., and Mitkov, R., editors, Anaphora   
 Processing: Linguistic, Cognitive, and Computational Modelling. John Benjamins.
Taboada, Maite and Mann, William C. (2006). Rhetorical Structure Theory: Looking  
 back and moving ahead. Discourse Studies, 8(3):423–459.
Vieira, R. u. M. P. (2000). An empirically-based system for processing definite   
 descriptions. Computational Linguistics, 26(4):525–579.
Waltinger, U., Mehler, A., and Stührenberg, M. (2008). An integrated model of lexical  
 chaining: Applications, resources and its format. In KONVENS 2008 -   
 Ergänzungsband Textressourcen und lexikalisches Wissen, pages 59–70, Berlin.
Alexander Mehler (2009): A Quantitative Graph Model of Social Ontologies by
 Example of Wikipedia. In: Alexander Mehler, Serge Sharoff and Marina 
 Santini (eds.): A Quantitative Graph Model of Social Ontologies by Example of 
 Wikipedia. Genres on the Web: Computational Models and Empirical Studies, 
 Forthcoming.


Editorial

Sporleder, Caroline

Lexical Models to Identify Unmarked Discourse Relations: Does WordNet help?

Diewald, Nils

Stührenberg, Maik

Garbar, Anna

Goecke Serengeti, Daniela

Serengeti - Webbasierte Annotation semantischer Relationen

Chiarcos, Christian

An ontology of linguistic annotations

Mönnich, Uwe

Kühnberger, Kai-Uwe

iiiLDV FORUM – Band 23 (1) – 2008
1 Introduction
The division of scientifi c disciplines into a theory part and a more applied or practical 
part is a rather common conceptualization of classifying academic fi elds and can be 
found in nearly all academic research  traditions ranging from the sciences and enginee-
ring disciplines to the humanities and educational research . Although there seems to be 
a natural order of producing scientifi c results, namely an order of how to develop theo-
ry and practice – fi rst, a theory should be developed, i.e. a conceptualization of a certain 
domain must be provided, and second, this theory can be tested in experiments, imple-
mentations, applications etc. – there are many examples in the history of academic disci-
plines where theory follows practical developments and not vice versa. Text tech nology 
is perhaps such  an example: Markup standards such  as RDF, OWL, or XML, coding initi-
atives like OLAC (Open Language Arch ives Community), and practical applications for 
retrieval purposes (oft en in business–related contexts) seem sometimes to get ahead of 
theoretical ch aracterizations of the underlying standards. For example, a standard like 
OWL Full is at present theoretically not very well understood and it took some time to 
specify the theoretical mach ine models of markup languages (like XML) – actually, at a 
time point aft er the languages themselves have been accepted as de facto standards.
Nevertheless, we decided to follow the natural order of categorizing cutt ing–edge re-
search  in text tech nology into theory and practice for this present double volume “On-
tologies in Text Tech nology” of the LDV–Forum: In the fi rst volume, theory–related pa-
pers are collected, whereas work shedding light on applications is covered in the pre-
sent second volume. Although text tech nology itself and, in particular, its connection to 
coding semantic knowledge in form of ontologies is a rather young discipline and some 
practical developments seem to hurry ahead of their theoretical foundations, we think 
that this order enables the reader to follow a more logical succession of the recent deve-
lopments. This is further supported by the fact that articles contained in the fi rst volu-
me can be considered in many aspects as a basis for results provided in this second vo-
lume. More will be said about these interrelationships between the two types of articles 
in Section 3.
In any case, we as the guest editors are proud to present the second part entitled “Appli-
cations of Ontologies in Text Tech nology” of the double issue of the LDV–Forum to the 
research  community. We hope that the interested reader can profi t from the work coll-
ected here and in the best of all possible worlds can pick  up some ideas for her own re-
search , in order to further promote text tech nology and ontology design.
Editorial
Editorial
Uwe Mönnich, Kai-Uwe Kühnberger
iv
2 The Research Unit 437 Text Technological Information Modeling
During the last six years the development of text tech nology in Germany was strongly 
infl uenced by the research  unit 437 “Text Tech nological Information Modeling” funded 
by the German Research  Foundation (DFG). This research  unit is an interdisciplinary re-
search  endeavor carried by the Universities of Bielefeld, Gießen, Dortmund, Tübingen, 
and Osnabrück . Starting in the year 2001, this group constitutes the largest collaborative 
research  project devoted to text tech nological issues and has provided the basis for text 
tech nological research  in Germany. Currently this research  unit is in its fi nal funding 
year. In order to get a bett er idea of the overall project, a concise overview of the sub-
projects funded during the second phase of this research  unit is given: 
Secondary Structuring of Information and Comparative Analysis of Discourse.• 
Principal Investigator: Dieter Metzing.
Induction of Document Grammars for the Representation of Logical Hypertextual Docu-• 
ment Structures. 
  Principal Investigator: Alexander Mehler.
Text-Grammatical Foundations for the (Semi-)Automated Text-to-Hypertext Conversion.• 
  Principal Investigator: Angelika Storrer.
Generic Document Structures in Linearly Organized Texts: Text Parsing Using Domain • 
Ontologies and Text Structure Ontologies.
  Principal Investigator: Henning Lobin.
Adaptive Ontologies on Extreme Markup Structures.• 
 Principal Investigators: Uwe Mönnich , Kai-Uwe Kühnberger.
Although the research  unit tries to cover all aspects of current text tech nological activi-
ties, it is still possible to identify certain core aspects that play a central role in all sub-
projects. Examples for such  vertical topics of the whole research  unit are ontologies, an-
notations, markup standards, and processing aspects of texts. All these topics play an 
important role in all participating projects. Some aspects of these vertical topics of the 
research  unit are also represented in this double volume of the LDV-Forum. Whereas 
certain sub-projects of the collaborative research  unit mentioned above are represented 
in Volume I focusing on the foundations of theories for developing, ch aracterizing, co-
ding, learning, and adapting ontological back ground knowledge as a crucial ch allenge 
for the semantic annotation of text documents, other sub-projects document aspects of 
their ongoing work in the present Volume II “Applications of Ontologies in Text Tech -
nology”. We think that we can provide in this way not only a representative documen-
tation of text tech nology in general, but also a representative collection illustrating the 
research  unit 437, in particular.
3 The Structure of Volume II
This second volume collects applied work on ontology design and text tech nology. The 
rticles span a fi eld from ontologies in discourse parsing and lexical semantics to anapho-
ra resolution, linguistic annotations, and the automatic acquisition of formal concepts 
v
from textual data. It is important to notice that there are many connections between the 
articles published in the two parts of the double volume. In particular, several foundati-
onal results presented in the fi rst volume provided the basis for applications in the pre-
sent volume. We will try to make some of these obvious connections visible while rough-
ly summarizing important topics of the contributions collected here.
In his article “An Ontology of Linguistic Annotations”, Christian Chiarcos discusses 
necessary design features of ontological resources for annotations mainly intended for 
terminological integration, and ontology–based search  across linguistic resources with 
heterogeneous annotations. By developing a structured ontology involving selfcon-
tained sub-ontologies, which  are linked in a declarative way, he shows how a separation 
between the annotation documentation and its interpretation with respect to the refe-
rence terminology can be ach ieved. The underlying idea is a mapping process of anno-
tations onto ontological representations, such  that the full range of types of information 
in annotations (like syntactic, semantic, phonological etc. information) can be referenced 
by an ontology. A theoretical basis of the ideas spelled out in Chiarcos article can be 
found in the contribution of the fi rst volume “Towards a Logical Description of Trees in 
Annotation Graphs” by Jens Mich aelis and Uwe Mönnich . 
The contribution “OWL Ontologies as a Resource for Discourse Parsing” by Maja Bä-
renfänger, Mirco Hilbert, Henning Lobin, and Harald Lüngen bases discourse annota-
tions on the Rhetorical Structure Theory (Mann and Thompson, 1988) to automatically 
arrange discourse segments and rhetorical relations in a tree structure. The resources to 
extract these tree structures are based on heterogeneous types of information like dis-
course marker lexicons, lexico-semantic ontologies, and annotation layers of input text. 
The article focuses particularly on OWL ontologies and how they can be consulted by 
the discourse parser. An important role plays the usage of an OWL version of Germa-
Net and a taxonomy of rhetorical relations which  was developed by the authors them-
selves. Certain aspects of this paper are practical applications of the theoretical results 
of the contribution “Domain Ontologies and Wordnets in OWL: Modelling Options” by 
Harald Lüngen and Angelika Storrer of the fi rst volume.
The development of automatic extraction procedures for generating ch eap but never-
theless reliable ontologies seems to be one of the most important practical ch allenges 
for text tech nological research  (Perez and Manch o, 2003). In particular, synonymy infor-
mation seems to be a good starting point for such  an endeavor to identify diff erent can-
didates for one and the same concept (word sense). A Kumaran, Ranbeer Makin, Vĳ ay 
Patt isapu, Shaik Sharif, and Lucy Vanderwende examine in their article “Evaluating the 
Quality of Automatically Extracted Synonymy Information” two complementary tech -
niques in order to automatically extract synonymy information from large corpora: First, 
a generic broad–coverage parser for generating bits of semantic information and second, 
their synthesis into sets of synonyms using word-sense disambiguation with latent se-
mantic analysis. The authors evaluate their approach es quantitatively and qualitatively. 
From a general perspective this article is a further step towards the whole cycle of au-
tomatic ontology generation: extracting semantic information, expanding an ontology 
with additional information, and adapting this expanded ontology if necessary. In this 
vi
sense, the present paper complements the article “Automatic Ontology Extension: Re-
solving Inconsistencies” by Ekaterina Ovch innikova and Kai-Uwe Kühnberger in the 
fi rst volume.
A classical ch allenge of natural language processing concerns nominal anaphora reso-
lution, partially because several types of knowledge have to be taken into account as, for 
example, morphosyntactic information and domain knowledge. The paper “Resolving 
Nominal Anaphora Using Hybrid Semantic Knowledge” by Daniela Göck e, Maik Stüh-
renberg, and Tonio Wandmach er proposes a hybrid approach  towards extracting auto-
matically necessary domain knowledge: fi rst, they propose a knowledge-free approach  
of distributional similarity (Paaß et al., 2004) based on latent semantic analysis, in this 
respect comparable to the paper “Evaluating the Quality of Automatically Extracted Sy-
nonymy Information” above, and second, they use Hearst patt erns (Hearst, 1982), i.e. 
predicate-argument relations encoded in the syntactic structure of the text. The integra-
tion of semantic relatedness by combining information about extracted relations and 
cooccurrence information is used to identify the most likely antecedent in anaphora re-
solution tasks. The authors evaluate their approach  on a corpus of German scientifi c and 
newspaper articles.
The fi nal contribution of the present volume “Automatic Acquisition of Formal Con-
cepts from Text” by Pablo Gamallo Otero, Gabriel Pereira Lopes, and Alexandre Agu-
stini uses formal concept analysis (Priss, 2006) in order to implement an unsupervised 
learning procedure for concept acquisition from annotated corpora. The idea is to build 
bidimensional clusters of words and their lexico–semantic contexts. Their procedure re-
sults in a concept latt ice describing a domain–specifi c ontology underlying the training 
corpus. The authors use for their evaluation a large Portuguese corpus where the tokens 
were extracted from a general–purpose journal and an English excerpt of the European 
Parliament Proceedings.
4 Acknowledgments
This volume would not have been possible without the help of many people. In the fi rst 
place, the guest editors want to thank the editors-in-ch ief of the LDV–Forum, Alexan-
der Mehler and Christian Wolff . Their encouragement and support in all phases of the 
emergence of this double volume has been irreplaceable in completing it. Furthermore 
we want to thank the German Research  Foundation for fi nancial support of the research  
unit 437 “Text Tech nological Information Modeling” and particularly the speaker of this 
research  unit, Dieter Metzing.
Last but not least, the editors want to thank the program committ ee for their careful 
evaluations of the submitt ed papers. The quality of this volume is also a direct refl ection 
of the work these reviewers invested. The program committ ee consisted of the following 
research ers (in alphabetical order): Irene Cramer, Thierry Declerck , Stefan Evert, Pascal 
Hitzler, Wolfgang Höppner, Helmar Gust, Marcus Krach t, Edda Leopold, Alessandro 
Mosch itt i, Larry Moss, Rainer Osswald, Olga Pustylnikov, Georg Rehm, Hans-Chri-
stian Sch mitz, Bernhard Sch röder, Uta Seewald-Heeg, Manfred Stede, Markus Stuptner, 
Frank Teuteberg, Yannick  Versley, Johanna Völker, Armin Wegner, and Christian Wolff .
vii
References
Hearst, M. (1982). Automatic acquisition of hyponyms from large text corpora. In 
Proceedings of the Fourteenth International Conference on Computational 
Linguistics.
Mann, W. and Thompson, S. (1988). Rhetorical structure theory: Toward a functional 
theory of text organsation. Text, 8(3):243–281.
Paaß, G., Kindermann, J., and Leopold, E. (2004). Learning prototype ontologies by 
hierarchical latent semantic analysis. In Knowledge Discovery and Ontologies, Pisa, 
Italy.
Perez, G. A. and Mancho, M. D. (2003). A Survey of Ontology Learning Methods and 
Techniques. OntoWeb Delieverable 1.5.
Priss, U. (2006). Formal concept analysis in information science. Information Science 
and Technology, 40:521–543.


Not a member yet

Local Coherence Analysis in a Multi-Level Approach to Automatic Text Analysis

Automatic Acquisition of Formal Concepts from Text

Editorial

Lexical Models to Identify Unmarked Discourse Relations: Does WordNet help?

Serengeti - Webbasierte Annotation semantischer Relationen

An ontology of linguistic annotations

Editorial

Evaluating the Quality of Automatically Extracted Synonymy Information

A hybrid approach to resolve nominal anaphora

Anaphora as an Indicator of Elaboration: A Corpus Study

231

253