This paper gives an overview of a relatively new thematic corpus based on German sacred literature of the Baroque period. At present, the digital collection consists of several texts specific to the memento mori genre. All texts in the Austrian Baroque Corpus (ABaC:us) have been enriched with different layers of structural information and tagged using automated tools adapted to the specific needs of the language of the period. One important achievement of the project is that each occurring historic word form has been electronically mapped to its corresponding lemma in High German and corrected or verified by domain experts. In all phases of the workflow, the interdisciplinary team (literary, linguistic, and text technology specialists) insisted on high quality linguistic and semantic annotation, and worked towards creating a sound basis that would allow for more sophisticated research questions. The current version of the interface can be seen as a case example showing how the ABaC:us team provides improved access to these rare pieces of macabre literature that give fascinating evidence of Baroque culture and attitudes towards Life and Death.
The acronym ABaC:us stands for Austrian Baroque Corpus, a digital thematic collection of German texts published during the Baroque Era, in particular the years from 1650 to 1750. The present corpus was compiled between 2010 and 2015 at the Institute for Corpus Linguistics and Text Technology (ICLTT) and at the newly founded Austrian Centre for Digital Humanities (ACDH) of the Austrian Academy of Sciences, alongside two other associated research projects1.
1 The generated textual data of both the project “Text-Technological Methods for the Analysis of Austrian Baroque Literature“ (March 2012 – September 2014, supported by funds of the Österreichische Nationalbank, Anniversary Fund) and of the project “Mortuary Cult in 17th Century Vienna: Confraternity Studies in the Digital Age” (June 2014 – May 2015, supported by funds of the City of Vienna) as well are part of the existing ABaC:us collection. The first project mentioned started with building the corpus from scratch and aimed at identifying rhetorical patterns and regularities on a lexical and syntactic level. The second project focuses on rare literary and non-literary texts that were produced by a Viennese confraternity. The digital analysis of the printed pamphlets (calendars, statutes and instructions) will lead to evidence-based assumptions on the role of confraternal associations in Counter Reformation Vienna.
Digital textual data in German for the Early Modern period2 remain underdeveloped and underexplored. Therefore it was decided that a historical corpus should contain significant and thematically connected examples of Baroque literature. Accordingly, the ABaC:us collection is based on the prevalence of sacred literature and contains mainly textual sources concerning death and dying. The Baroque transience topos and its memento mori appeal shaped and permeated different literary and non-literary genres. Richly illustrated emblem books3 in prose and verse were a focal point of Baroque culture and were frequently printed. They were meant to remind people of the fragility of their existence and the inevitability of death. By reading these moralizing texts, people were supposed to be directed towards holding themselves in steady expectation of their own demise and admonished to live a life of virtue in order to be prepared for death at all times. 2 Examples for other resources from that time period are some texts of the „Bonner Frühneuhochdeutschkorpus“ http://www.korpora.org/Fnhd/, the “GerManC project: A representative historical corpus of German“ at the University of Manchester http://www.llc.manchester.ac.uk/research/projects/germanc/, and the earlier texts of the huge collection „Deutsches Textarchiv“ at the Berlin Brandenburg Academy of Sciences and Humanities www.deutschestextarchiv.de/.
At present, the thematic ABaC:us collection holds 20 religious writings motivated by the fear of sudden death and the central memento mori theme including sermons, obituaries, devotional books, compilations of prayers, songs and works related to the dance-of-death theme. When building up the corpus from scratch, only original textual data in full length were chosen. As a matter of philological principle, only early and if possible the first known editions (editio princeps) and rare specimens from different monastic and public libraries were selected for the digitalization procedure. 2
The Austrian Baroque Corpus currently contains a total of more than 210,000 running words. The main part, approximately 180,000 tokens (85%), can be attributed to the Augustinian monk Abraham a Sancta Clara (16441709)4, a very popular preacher and widely read author5. Due to his literary talent and his distinctive style, Abraham a Sancta Clara’s books reached a wide audience; they were frequently reprinted and distributed across the Germanspeaking lands so that Abraham a Sancta Clara remained very popular even after his death: “The fact that Pater Abraham’s books sold well, led several publishers to the idea of combining parts of his already published works with texts of other authors, ascribing the literary hybrid to the Augustinian preacher.”6 The 300th anniversary of his death was the motivation to start digitization in part – also including works whose authorship is in doubt – hence some parts of the collection also derive from the preacher’s field of influence, for example from other friars in his religious order, or his publishers and imitators.
Five of his most popular texts dealing with death and dying – MERCKS WIENN (1680), LÖSCH WIENN (1680),
MEUBLIERT- UND GEZIERTE TODTEN- CAPELLE (1710) – have a strong association with the city of Vienna. They constitute the core of ABaC:us and are the first to be made available online.
4 Eybl, Franz M. Abraham a Sancta Clara. In: Killy Literaturlexikon Volume 1. Berlin: de Gruyter, 2008, p. 10-14 and Eybl, Franz M.
Abraham a Sancta Clara. Vom Prediger zum Schriftsteller. Tübingen: Max Niemeyer Verlag 1992. 5 The remaining texts are corresponding to Jacob Balde, Abraham Megerle, Johann Carl Megerle, Johann Valentin Neiner, Franz Peikhart,
Emmerich Pfendtner, and Florentius Schilling.
Traditions – Theology. Ashgate 2009. p. 3.
Text capture using OCR of early modern editions is well-known to be problematic. When possible, existing digitized texts have been used as a basis for ingestion into the corpus (as it was only the case with MERCKS WIENN, freely available on Zeno.org7). For texts that had to be captured using OCR, ABBYY FineReader 7 was the best solution as it is capable of scanning both Roman and Black Letter typefaces (Fraktur).
All original primary sources have been fully digitized, transcribed, and encoded as TEI P5 conformant files8. Considerable attention has been paid to the structural and typographic features of the texts. The manual tagging covered names of historical, mythological and biblical interest as well as place names:
Errors and mistakes in the original texts have been allowed to stand; editorial interventions and suggestions are recorded in the mark-up.
The five core texts ascribed to Abraham a Sancta Clara have been linguistically annotated with Part-of-Speech tags (PoS) and lemmatized. Thus each token was automatically mapped to a word class by TreeTagger9 according to
the guidelines (1999) of the 54-part Stuttgart-Tübingen-TagSet (STTS)10, a standardized tagset for German language resources. In addition, every word was automatically supplied with its lemma.
As the tagger was developed for modern day German language and was used out of domain – namely for a historical variety of Early Modern German, which at that time was not fully standardized – the project group had to cope with many (expected) erroneous mappings and mismatches: Even the slightest deviations in orthographic conventions (e.g. variational suffixes: behutsamb/behutsam (cautious), doubling of consonants: Freundschafft/Freundschaft (friendship), elimination of long vowels Bott/Bote (messenger), etc.) caused wrong annotations and had to be verified and manually corrected.11 Problems arose with so-called multi-word lexemes and with historical conventions for separating words, such as Stephans Domkirchen or Haupt Statt (according to modern German orthographic norms: Stephansdomkirche (St. Stephen’s Cathedral) and Hauptstadt (Capital) or contracted forms (see examples below)). Because of the differences between Modern and Early Modern German the StuttgartTübingen- TagSet had to be adapted and extended with additional categories to deal with contracted forms such as wirstu (old form with clitic instead of wirst du), mans (man es) or machs (mach es). For this purpose multiple tags have been linked by an underscore: tag_tag. Subsequently, lemmatization had to be adjusted as well.
Token wirstu mans machs
PoS
(finite content verb + irreflexive personal pronoun)
(substituting indefinite pronoun + irreflexive personal pronoun)
(imperative content verb + irreflexive personal pronoun) Lemma werden_du become_you man_es one_it machen_es make_it
Historical corpora particularly require a comprehensible lemmatization. Word occurrences such as Fegefeuer, meaning “purgatory”, which appear in several orthographic versions (Feeg=Feuer, Feegfeuer, Feg=Feuer, Fegefeuer, Fegfeuer, Fegfeur, Fegfewer), can only be found easily through their lemma or base form. The manual control and mapping of the data refers to two different common dictionaries: the modern standard dictionary Duden (http://www.duden.de/) and the German dictionary of the Grimm Brothers (http://woerterbuchnetz.de/DWB/). Words labelled FM („fremdsprachliches Material“, elements from a foreign language) had to be lemmatized entirely by domain experts. Most FM words stem from phrases and passages in Latin, for this reason, Stowasser (1998) was used in order to assign the right lemma to each Latin word. Only a minority of words tagged FM were not Latin, but French or Italian.
10 See http://www.sfs.uni-tuebingen.de/resources/stts-1999.pdf 11 Hinrichs and Zastrow have already noticed that – compared to other texts – Abraham a Sancta Clara’s style exhibits by far the highest average sentence length, which might also be a reason why the author’s test data had “the highest number of tagging errors”, see Erhard Hinrichs, Thomas Zastrow. Linguistic Annotations for a Diachronic Corpus of German. In: Linguistic Issues in Language Technology, Volume 7, issue 7 (2012), p. 11.
Words not listed in any of the dictionaries were particularly challenging. The main part of these “out ofvocabulary words” were compounds invented by Abraham a Sancta Clara or his imitators to add a playful element to the texts, for instance vernunftselig (rationally blessed), Tigergemüt (temper of a tiger), or Felsenzucht (breeding of rocks). In other cases of creative language use, we decided to use two lemmata to capture their full meaning in the linguistic annotation. Multiple meanings and ambiguous forms (occurring particularly in puns) were separated by a vertical bar: for instance Kümmernis | Kümmer=Nuß (grievance, containing the German word for “nut”).
All words not found in any of the above-mentioned dictionaries were marked by an asterisk and annotated with a notional modern lemma that comes closest to the particular word form.
In order to identify, process and remove all tagging errors and mismatches, an updated version of the token_editor developed at the ICLTT was used. The tool loads the PoS and lemma data of the corpus and displays them in vertical lists: It allows researchers to read the text, create lists of particular items on the basis of regular expressions and assign new data to these datasets. Although the tagger determines a PoS-tag for each word and the token_editor facilitates the evaluation of the automatic assignment of word labels and allows for the verification of the tagger’s suggestion and if necessary the correction of the results by human annotators in an accelerated way, the semi-automatic linguistic annotation was still a time-consuming process.
After having fully worked with the first three texts, we had a sound basis for the following ones and made first experiments in adapting the specific historical language material for further annotation procedures. Using our manually-corrected annotations as additional training material had a very positive effect for domain adaptation and
improved the performance of the applied tools significantly12. A wordlist generated from the first texts was instrumental in reducing errors in the ongoing process of tagging.
In case of MERCKS WIENN and LÖSCH WIENN improvement was evident:
without the lexicon of with the lexicon of
71,2% 57,5% 82,7% 72,4%
Manually-corrected data13 provided relevant lexical information to positively influence the performance of the tagger and not only improved the PoS results but also the lemmatization (by almost 15%).
Our method of annotating more texts of the same time period and genre was an incremental bottom-up process and resulted in high quality data. Using the described bootstrapping approach together with meticulous revision has made ABaC:us a thoroughly validated and reliable corpus which can be utilized for: • • • • • the annotation of other distinct texts evaluating the quality of automatically generated lexical data from corpora the training of different taggers reusing the corpus for creating lexical data from that time period (ABaC:us could serve as a fundament for generating an expandable computational lexicon) and more sophisticated and complex linguistic research questions (such as identifying stylistic features and rhetorical patterns described in the next paragraph).
The Baroque literature of the corpus is full of stylistic patterns and a start has been made to identify those typical “abrahamic” features in the texts by using the applied PoS annotation as a basis. The figures below document how the search for sequences of PoS tags was conducted in order to get results that can be described as often recurring stylistic patterns such as ostensive comparisons (Figure 8, e.g. wie ein Lambel von den Wölffen / as a lamb among 12 See Resch, Claudia, Declerck, Thierry, Krautgartner, Barbara and Czeitschner, Ulrike. 2014. ABaC:us revisited – Extracting and Linking Lexical Data from a historical Corpus of Sacred Literature. In: Atwell, Eric, Brierley, Claire and Sawalha, Majdi (eds.): Proceedings of the
through reliable data”. 13 Two research assistants have worked independently on the manual correction and a senior researcher supervised their decisions, not only, but especially in cases of doubt.
wolves or [zittern] wie ein Laub von der Espen / tremble like an aspen leaf) and pairs of words consisting of a foreign term, a conjunction, and a noun (Figure 9, e.g. Epilogus vnd Weltschluss / Epilogus and end of the world or Fratrum vnd Lay-Brüder / Fratrum and lay friar).
Although it will not be possible to divide the several works where the authorship of Abraham a Sancta Clara is still in question from those where the author is certain, we hope to uncover the characteristics of Abraham a Sancta Clara’s often imitated literary style by analyzing, measuring and counting these significant features. Stylometric methodologies could become relevant by automatically finding and counting distinctive patterns of “abrahamic style” and can enhance our knowledge about the identified features. The ABaC:us corpus is also a rich source for the study of semantic aspects. “Death” and “dying”, a leitmotif of Baroque texts, can be found in numerous variations: examples for the personification of death are Aschen=Mann (ash man), Dieb der Fröhlichkeit (thief of happiness), Reuter auf dem fahlen Pferd (rider on the pale horse). Together with terms and phrases dealing with the “end of life”, “dying” and “killing” more than 1700 death-related lexical units have been identified in MERCKS WIENN, TODTEN BRUDERSCHAFT and TODTEN- CAPELLE. To organize this rich terminology and to provide enhanced access to it we currently are working on the creation of controlled vocabularies in SKOS (Simple Knowledge Organization System). These taxonomies will also ease the task of semi-automated semantic annotation of additional texts. Not only can the project group use the data for
various research interests, such as taxonomy building and semantic enrichment by LOD-sources14, but other researchers can as well. 5
As texts of this era are usually difficult to access, because of issues of fragility and copies scattered across libraries and institutions, availability is an important issue: The value and uniqueness of the fully annotated corpus precisely consists in its online access and already encoded textual knowledge.
In order to guarantee that the processed materials will be re-usable for different purposes, the project team and the technical task force are working on a web-based interface that can be used for further research. ABaC:us was implemented with the publication framework cr_xq15, a further development of the Scalable Architecture for Digital Editions (SADE) of the Berlin-Brandenburg Academy of Sciences and Humanities. We strived to keep the system as generic and flexible as possible. All functions of the edition are parameterized with simple XPath expressions that are held in a project-specific configuration. This dynamic architecture allows to manage the data in a very flexible way and to publish any kind of XML-structured text as a web application. With this format, we represent the textual structure of the works (e.g. chapters, front and back matter etc.) as well as the physical structure of the documents (pages). METS also provides the framework to document the corpus’ data structure, references metadata records of the single works it is composed of, and acts as a wrapper for the web application’s basic configuration that is necessary for rendering.
The user-friendly application (screenshot below) includes a dual display with digital text and parallel facsimile will enable users to obtain different views of the text, provide different kinds of indices and allow for flexible search strategies on both the linguistic (key words, PoS-labels or lemmata) and semantic level. ABaC:us will be a corpus available for a range of communities: literary and linguistic studies, historical research, theology, but may also meet the interest of a broader public audience.
14 See Czeitschner, Ulrike, Declerck, Thierry, and Resch, Claudia. 2014. Porting Elements of the Austrian Baroque Corpus onto the Linguistic Linked Open Data Format. In: Osenova, Petya, Simov, Kiril, Georgiev, Georgi and Nakov, Preslav (eds.): Proceedings of the Joint
15 cr_xq is built upon proven standards: The edition is described with a METS-Container, a metadata standard developed by the Library of Congress that is widely employed by libraries and other cultural heritage institutions to model compound digital objects.
The annotated ABaC:us data have been selected as an Austrian contribution to CLARIN (Common Language Resources and Technology Infrastructure) and have been released in 2015 as: ABaC:us – Austrian Baroque Corpus, digital edition (2015), edited by Claudia Resch and Ulrike Czeitschner.
In this paper, we have given detailed insights into the corpus building process of the Austrian Baroque Corpus ABaC:us, a collection of 20 texts from different authors of the Baroque era (1650-1750). Although German can be seen as a well-documented linguistic variety in terms of language data and tools, data from historical linguistic stages are still scarce and under-explored. Thus, ABaC:us had to be built from scratch, based on thematically connected examples of sacred Baroque literature with the leitmotif of death and dying. The stages of text selection, digitization and transcription have been accurately described. The focus of the paper, however, rests on the annotation process and particularly on the linguistic information. A large part of the corpus – approximately 180,000 tokens (85%), ascribed to the preacher Abraham a Sancta Clara (1644-1709) – contains Part-of-Speech tags and lemma information; challenges during the annotation process are not concealed and our solutions to various problems are illustrated with examples. A showcase analysis of certain rhetorical and semantic features aims to give an idea of the value that the linguistic annotation adds to the corpus. And finally, the brand-new user interface is
introduced, which enables researchers as well as any interested persons to read, examine and analyze the core texts of ABaC:us. The online edition and its accompanying documentation can be found here: https://acdh.oeaw.ac.at/abacus/ BOOT, Peter, 2009, Mesotext. Digitised Emblems, Modelled Annotations and Humanities Scholarship. Amsterdam: Pallas Publications – Amsterdam University Press CZEITSCHNER, Ulrike, DECLERCK, Thierry, MOERTH, Karlheinz and RESCH, Claudia, 2012, Linguistic and Semantic Annotation in Religious Memento Mori Literature. In: ATWELL, Eric, BRIERLEY, Claire and SAWALHA, Majdi (eds.): Proceedings of the LREC 2012 Workshop: Language Resources and Evaluation for Religious Texts. Paris: ELRA, p. 49-52 CZEITSCHNER, Ulrike, DECLERCK, Thierry, and RESCH, Claudia, 2014, Porting Elements of the Austrian Baroque Corpus onto the Linguistic Linked Open Data Format. In: OSENOVA, Petya, SIMOV, Kiril, GEORGIEV, Georgi and NAKOV, Preslav (eds.), 2013, Proceedings of the Joint Workshop on NLP&LOD and SWAIE: Semantic Web, Linked Open Data and Information Extraction associated with the 9th International Conference on Recent Advances in Natural Language Processing (RANLP 2013). Sofia: p. 12-16 EYBL Franz M, 1992, Abraham a Sancta Clara. Vom Prediger zum Schriftsteller. Tübingen: Max Niemeyer Verlag EYBL, Franz M, 2008, Abraham a Sancta Clara. In: Killy Literaturlexikon Volume 1. Berlin: de Gruyter, p. 10-14 HINRICHS, Erhard, ZASTROW, Thomas, 2012, Linguistic Annotations for a Diachronic Corpus of German. In: Linguistic Issues in Language Technology, Volume 7, issue 7, p. 1-16 RESCH, Claudia, DECLERCK, Thierry, KRAUTGARTNER, Barbara and CZEITSCHNER, Ulrike, 2014, ABaC:us revisited – Extracting and Linking Lexical Data from a Historical Corpus of Sacred Literature. In: ATWELL, Eric, BRIERLEY, Claire and SAWALHA, Majdi (eds.): Proceedings of the 2nd Workshop on Language Resources and Evaluation for Religious Texts / LREC 2014. Reykjavik: p. 36-41 RESCH, Claudia, KRAUTGARTNER Barbara and CZEITSCHNER, Ulrike, (Forthcoming). ABaC:us für LinguistInnen – Morphosyntaktische Annotation im „Austrian Baroque Corpus“. In: RESCH, Claudia and DRESSLER, Wolfgang Ulrich (eds.): Digitale Methoden der Korpusarbeit in Österreich. Tagungsbeiträge der 40. Österreichischen Linguistiktagung. Wien: Verlag der österreichischen Akademie der Wissenschaften ŠAJDA Peter, 2009, Abraham a Sancta Clara: An Aphoristic Encyclopedia of Christian Wisdom. In: Kierkegaard and the Renaissance and Modern Traditions – Theology. Ashgate 2009. p. 1-20 STOWASSER, 1998, Lateinisch-deutsches Schulwörterbuch von STOWASSER Joseph Maria, PETSCHENIG Michael, SKUTSCH Franz. Auf der Grundlage der Bearbeitung 1979 neu bearbeitet und erweitert (Gesamtredaktion: Fritz Lošek). Zug: HPT-Medien AG
The ABaC:us Team consists of several people with different academic backgrounds partly working together since 2010 to build up the Austrian Baroque Corpus: Claudia Resch is a senior researcher and project leader at the Austrian Centre for Digital Humanities of the Austrian Academy of Sciences. Current research focuses on German literature of the early modern period and the application of literary and linguistic computing in a corpus-based approach to textual issues. Key areas covered are historical linguistics, text stylistics, and annotation problems associated with non-standard varieties of Early Modern German.
Ulrike Czeitschner is a senior researcher and project leader at the Austrian Centre for Digital Humanities. With an academic background in cultural anthropology and a particular interest in the impact of digital technologies on all kinds of humanities studies, she has been focusing on structural and semantic annotations of various genres. Eva Wohlfarter is a junior researcher at the Austrian Centre for Digital Humanities and a PhD candidate in linguistics at the University of Vienna. She holds a master’s degree in Applied Linguistics and her main fields of interest are corpus linguistics, historical linguistics, sociolinguistics, and discourse studies.
Barbara Krautgartner is junior researcher at the Austrian Centre for Digital Humanities. She holds a master’s degree in German philology and is currently studying Web- and App-Development in Vienna. Developing scientific web applications and automatic data processing are her special fields of interest.
Since the ABaC:us working group insisted on high-quality linguistic and semantic annotation throughout the project, most phases of the project would not have been feasible without external funding: • • •
Abraham a Sancta Clara and his Danses Macabres (October 2009 – September 2010), research grant by the City of Vienna.
Text-Technological Methods for the Analysis of Austrian Baroque Literature (March 2012 – September 2014), supported by funds of the Österreichische Nationalbank, Anniversary Fund.
Mortuary Cult in 17th Century Vienna: Confraternity Studies in the Digital Age (June 2014 – May 2015), supported by funds of the City of Vienna.