=Paper=
{{Paper
|id=Vol-1918/declerck
|storemode=property
|title=Software Projects for Developing Digital Humanities Resources
|pdfUrl=https://ceur-ws.org/Vol-1918/declerck.pdf
|volume=Vol-1918
|authors=Thierry Declerck
|dblpUrl=https://dblp.org/rec/conf/gldv/Declerck17
}}
==Software Projects for Developing Digital Humanities Resources==
https://ceur-ws.org/Vol-1918/declerck.pdf
Software Projects for developing Digital Humanities Resources
Thierry Declerck
DFKI GmbH, Language Technology Lab
Stuhlsatzenhausweg, 3
D-66123 Saarbrücken
declerck@dfki.de
Abstract programme2 . We think that a main aspect of this
success story lies in the fact that the students had
In this short paper we report on experiences to work together, building teams for working on
gained from bachelor and master theses, modules and meeting for integrating the work done
and from a series of software projects con- so far.
ducted in cooperation with the Department In all the 4 different software projects conducted
of Computational Linguistics of the Saar- until now, we could observe that the folktale topic
land University. Those bachelor/master the- was a driver calling for participation of a larger
ses and software projects were dealing with group of students (they can choose between differ-
the application of Natural Language Pro- ent software projects). We describe in the following
cessing and Semantic Web technologies to sections the types of approaches we followed and
the representation and analysis of folktales. the results that the students generated and made
Data, codes and results of the software available on various repository management ser-
projects have been made available in vari- vices, like GitLab, GitHub or Bitbucket. The idea
ous repository management services, like of having software projects as a platform followed
GitLab, GitHub or Bitbucket. We think that the work done by two students in their master and
it will be important to discuss the design bachelor theses, which were written in the context
of such openly accessible repositories in of their Research Assistant appointments within a
order to ensure their re-usability and fur- larger national project3 . We describe briefly the
ther extensions across various educational results of all those endeavours in the following sec-
institutions. tions.
1 Introduction
2 Annotations
In the past 3-4 years we proposed in cooperation
with the Computational Linguistics (CL) depart- In the context of cooperation between the past D-
ment of the Saarland University a series of bache- SPIN4 and AMICUS projects5 a master thesis was
lor/master theses and software projects, which were written by the student Antonia Scheidel on the
dealing with various aspects related to the wider
or http://www.abitur-und-studium.de/
field of folktales and therefore introducing Digital Bilder/Jana-Ott-Christian-Eisenreich-
Humanities (DH) topics to students trained primar- und-Christian-Willms-Studenten-von-
Thierry-Declerck-haben-ein-Programm-
ily to learn and apply computational methods of entwickelt-das-Maerchen-vorlesen-
language technologies. kann.aspx
2 See
Our diagnosis was that the approach building http://kulturellebildung.de/fa/
user/Fachbereiche/Literatur_Sprache/
on software projects for introducing CL students, Aktuelles/141121_PRESSE_Erzaehlen.pdf
and some few students from other departments, to 3 We do think that involvement of students as Research
Digital Humanities topics has been very successful. Assistant in projects is an important aspect to be considered.
4 D-SPIN was a predecessor of CLARIN-D. See
It is also the case that some of the projects we
https://weblicht.sfs.uni-tuebingen.de/
conducted have gained the interest of a broader englisch/index.shtml
public, including press coverage1 and a broadcast 5 AMICUS: Automated Motif Discovery in Cultural Her-
itage and Scientific Communication Texts, was a Dutch project
1 http://derstandard.at/2000004368363/ dealing partly with the annotation of folktales with recurrent
Wenn-der-Computer-zum-Maerchenonkel-wird Motifs. See https://ilk.uvt.nl/amicus/
23
annotation of fairy tales with Propp’s functions6 . written a program that analyzes the text according
Vladimir Propp “was a Soviet folklorist and scholar to linguistic criteria, with the aim of recognizing
who analysed the basic plot components of Rus- the (main) characters in it, and storing those in a
sian folk tales to identify their simplest irreducible database. This database is of the “Ontology” type,
narrative elements.”7 . Those basic plot elements on the base of which logical operations can be per-
are called by Propp “functions” and he identified formed. The background is a formal description of
31 such functions, like “Interdiction”, “Delivery” what can be found in these fairy tales, including an
or “Rescue”, etc. Propp also introduced circa 150 ontology about family relations. Thus, the system
sub-functions that are specialisations of the 31 top- can recognize that in the text “the daughter” is the
level functions. Complementary to the functions, same person as the “sister” when this is suggested
Propp identified 7 broad characters, like “the vil- by the context. This way, recognized characters in
lain”, “the donor” or the “hero”. The “morphology fairy tales are semantically annotated with more
of the tale” described by Vladimir Propp was based general categories, like “Woman”. And we then
on a subset of the so-called Afanasyev collection know in which contexts (or situations) a specific
of Russian Folktales8 . family member (for example the “daughter”) is in-
Antonia Scheidel developed a new annotation volved (see (Declerck et al., 2012) and (Koleva
scheme according to which fairy tales can be et al., 2012) for more details on the results of her
queried for texts, temporal structures, characters, work.).
dialogues, and Propp’s functions9 . The annota- Once we had those resources, i.e an annotation
tion scheme has been named APftML, standing for framework for folktales, based in a first instance of
“Augmented Propp fairy tale Mark-up Language”. the mark-up of Proppian functions, and an ontology
Antonia Scheidels’ work is documented in (De- framework in which characters playing a role in
clerck and Scheidel, 2010) and (Declerck et al., folktales are stored as instances of domain-specific
2011). Annotated fairytale textual data is important classes, the idea was to extend those to a larger
in that automated systems have a data set against framework supporting DH application scenarios.
which they can map their results (see, for exam-
ple (Scheidel and Declerck, 2010), describing an 4 Approaches to Story Segmentation
information extraction application in the folktale
In a first software project which was building on the
domain)10 . If fairy tales are manually annotated
top of the two resources mentioned in the previous
with the annotation scheme, the results of the auto-
sections, a division of work could be established
matic processing can be compared with the human
between the four members of the project team. One
annotation.
task consisted in offering a meaningful segmenta-
tion of the tales. The approach for this consisted in
3 Syntactic Analysis and a first Ontology
automatically segmenting the tales along the lines
Based on the annotation framework mentioned in of the dialogue structure. This had one motivation:
the previous section, Nikolina Koleva has worked to offer a base for the integration of a text-to-speech
for her bachelor thesis on an automated system system supporting the “read aloud” of a tale, in
for processing fairy tale texts. She considered for which voices are associated to each contributors to
her work two tales, “The Magic Swan Geese”, an the dialogues (and for sure one voice for the narra-
English version of the Russian fairy tale “Gusi- tor). This application is described in more details
lebedi”, and “Väterchen Frost”, a German version in the next section.
of the Russian fairy tale “Djed Moros”. She has The students worked in this project mainly on
the English version of the “Froschkönig” tale (The
6 See (Propp, 1968)
7 https://en.wikipedia.org/wiki/ Frog Prince)11 . Following those new steps, the
Vladimir_Propp initial annotation format has been augmented with
8 See https://en.wikipedia.org/wiki/ detailed dialogue descriptions. And the ontology
Alexander_Afanasyev has also been extended, including now a descrip-
9 The annotation scheme can be downloaded
at http://www.coli.uni-saarland.de/ tion of dialogues (questions, answers, monologues
˜ascheidel/APftML.xsd etc.), including the encodings of the participants
10 Examples of such annotated data can be down-
loaded at http://www.coli.uni-saarland.de/ 11 See https://en.wikipedia.org/wiki/The_
˜ascheidel/APftML.xml Frog_Prince
24
and the dialogue turns. In the two most recent 6 Iterative Ontology Developments
and currently still running software projects the
students are implementing a strategy on addition- We described in sections 4 and 5 how the original
ally segmenting a tale by the locations in which ontology has been enriched with additional features.
events are occurring. There is an interesting corre- In a second software project, work was dedicated
lation between the segmentation by dialogues and in the ontologisation of classical knowledge – in-
the one by locations, as in this kind of narratives dexation and classification – resources in the field
the participants to a dialogue are often sharing a of folklore. We were considering in this software
location. project two such resources: The “Motif-index of
folk-literature” (Thompson, 1955 1958) and the
5 Emotions Detection and “Types of International Folktales” (Uther, 2004).
Text-to-Speech Modules The first resource, which we abbreviate as TMI, is
available as an on-line resource17 . A folktale motif
One student had the task to implement a program
can be defined as a “repeated story element, e.g., a
able to detect emotions. For this the original anno-
character, an object, an action, or an event that can
tation scheme has been extended, supporting the
be found in several stories”18 . In TMI all motifs
mark-up of 6 basic emotions (fear, grief, joy, etc.),
are organized in a tree structure, so that each motif
which are also encoded in the ontology. The auto-
has a more abstract class that describes a span of
matic processing of the text (based in this case on
subordinated motifs. One motif entry consists of a
the NLTK package12 ) was then marking the emo-
motif-id, motif name, motif description (optional),
tion detected in one sentence, on the base of a emo-
and references to literature where it occurs.
tion lexicon build from annotated examples that
The second resource builds on former work by
served a as seed that was completed by consulting
Antti Aarne (Aarne, 1961) and Stith Thompson.
the WordNet13 module implemented in NLTK14 .
This classification system was extended by Hans-
A major extension of the past work in this soft-
Jörg Uther (see (Uther, 2004)), and in the following
ware project was that synthetic voices also play a
we are using the acronym ATU for referring to this
role. Once a character has been recognized, for
resource. A folktale type can be described as a
example the princess (in the fairy tale “Frog King”)
main story line that can be found in several cultures.
additional features are coded (for example age, gen-
The parts of this story line can refer to specific story
der, emotion, etc.). Then a previously defined syn-
elements also known as motifs. A folktale type is
thetic voice is automatically added to the charac-
therefore a bigger unit than a motif.
ter. And when the text is processed by the system,
the story can be “told” by the voices. If there is Our approach consisted in extracting from those
no detected character in a dialogue situation, it is knowledge resources, which are stored in different
assumed that the narrator is the speaker and the formats, classification relevant information and to
reader the receiver. A demo can be heard in the cor- re-organize them in two interrelated ontologies, us-
responding Bitbucket repository15 . In this software ing for this the W3C standards OWL19 , RDF(s)20
project we made use of the “Mary” Text-To-Speech and RDF21 .
System16 . The overall results of the projects are The integrated ontology resulting from the
described also in (Eisenreich et al., 2014). software project, also after curation done in
12 NLTK stands for “Natural Language Toolkit” and is writ-
the context of an internship at DFKI, contains
ten in Python, including a lot of corpus processing and statisti- 46,950 motifs for the TMI domain and 2802
cal libraries. See http://www.nltk.org/ elements for the ATU domain, most of them
13 See https://wordnet.princeton.edu/ for
interrelated by corresponding properties. Re-
more details.
14 See http://www.nltk.org/howto/wordnet. sults of this software project are available in a
html for more details.
17 https://sites.ualberta.ca/
15 The data, algorithms and results of the ˜urban/
projects are stored in https://bitbucket. Projects/English/Motif_Index.htm.
18 https://en.wikipedia.org/wiki/Motif_
org/ceisen/apftml2repo. A demo of
the TTS application is available at: https: (folkloristics)
19 See http://www.w3.org/TR/owl-
//bitbucket.org/ceisen/apftml2repo/src/
cbf4d71de7f96146d17c4c84572ceb9a99cd300f/ semantics/.
example%20output/audio_output.mp3?at= 20 See http://www.w3.org/TR/rdf-schema/
master&fileviewer=file-view-default formoredetails.
16 See http://mary.dfki.de/ for more details. 21 See https://www.w3.org/RDF/ for more details.
25
GitLab repository: https://gitlab.com/ tation of characters in folktales. In Kalliopi Zer-
folktaleclassification/. vanou and Antal van den Bosch and, editors, Pro-
ceedings of the 6th Workshop on Language Tech-
An application of this new integrated ontology
nology for Cultural Heritage, Social Sciences, and
for the classification of characters in folktales has Humanities (LaTeCH 2012), pages 30–35, 209 N.
been presented in (Declerck et al., 2016) and more Eighth Street Stroudsburg, PA 18360 USA, 4. Asso-
recent developments related to this integrated on- ciation for Computational Linguistics (ACL), ACL.
tology are described in (Declerck et al., 2017). Thierry Declerck, Tyler Klement, and Antónia Kostová.
2016. Towards a wordnet based classification of ac-
7 Conclusion tors in folktales. In Verginica Barbu Mititelu, Co-
rina Forascu, Christiane Fellbaum, and Piek Vossen,
We did report on specific teaching activities in the editors, Proceedings of the Eighth Global WordNet
field of the representation and processing of folk- Conference. Global WordNet Association, GWA, 1.
tales by students (mainly) in the field of computa-
Thierry Declerck, Antónia Kostová, and Lisa Schäfer.
tional linguistics. The specificity of the experiences 2017. Towards a linked data access to folktales
we are reporting is that those activities took place in classified by thompsons motifs and aarne-thompson-
the context of software projects or internships, thus uthers types. In Proceedings of Digital Humanities
with a focus on practical implementation and devel- 2017. ADHO, 8.
opment works. We noticed that this kind of team Christian Eisenreich, Jana Ott, Tonio Sdorf, Christian
work, or also compact work done in the context Willms, and Thierry Declerck. 2014. From tale to
of an internship, is delivering a very large amount speech: Ontology-based emotion and dialogue anno-
tation of fairy tales with a tts output. In Proceedings
of resources that are potentially very relevant for
of ISWC 2014. Springer.
being reused in other type of teaching activities.
Maybe also a coordinated action between univer- Nikolina Koleva, Thierry Declerck, and Hans-Ulrich
sities and other educational institutions toward the Krieger. 2012. An ontology-based iterative text pro-
cessing strategy for detecting and recognizing char-
organization of such software projects could be an acters in folktales. In Jan Christoph Meister, edi-
idea to discuss and implement. Last but not least, tor, Digital Humanities 2012 Conference Abstracts,
many of the results presented in this short paper pages 467–470, Hamburg, 7. University of Hamburg,
have been submitted to and accepted at relevant Hamburg University Press.
workshops and conferences, bringing the students Vladimir Propp. 1968. Morphology of the folktale.
thus also closer to this type of academic achieve- Trans., Laurence Scott. 2nd ed., University of Texas
ments. Press.
Antonia Scheidel and Thierry Declerck. 2010. Apftml
- augmented proppian fairy tale markup language. In
References Sándor Darányi and Piroska Lendvai, editors, First
Antti Aarne. 1961. The Types of the Folktale: A Clas- International AMICUS Workshop on Automated Mo-
sification and Bibliography. The Finnish Academy tif Discovery in Cultural Heritage and Scientific
of Science and Letters. Translated and Enlarged by Communication Texts. Szeged University.
S. Thompson. Second Revision (FFC 184). Stith Thompson. 1955−1958. Motif-index of folk-
Thierry Declerck and Antonia Scheidel. 2010. An in- literature: A classification of narrative elements in
formation extraction approach to the semantic anno- folktales, ballads, myths, fables, medieval romances,
tation of folktales. In Sándor Darányi and Piroska exempla, fabliaux, jest-books, and local legends. Re-
Lendvai, editors, First International AMICUS Work- vised and enlarged edition, Indiana University Press.
shop on Automated Motif Discovery in Cultural Her- Hans-Jörg Uther. 2004. The Types of International
itage and Scientific Communication Texts. Univer- Folktales: A Classification and Bibliography. Based
sity of Szeged, Hungary. on the system of Antti Aarne and Stith Thompson.
Suomalainen Tiedeakatemia.
Thierry Declerck, Antonia Scheidel, and Piroska Lend-
vai. 2011. Proppian content descriptors in an in-
tegrated annotation schema for fairy tales. In Lan-
guage Technology for Cultural Heritage. Selected
Papers from the LaTeCH Workshop Series, Theory
and Applications of Natural Language Processing,
pages 155–169. Springer.
Thierry Declerck, Nikolina Koleva, and Hans-Ulrich
Krieger. 2012. Ontology-based incremental anno-
26