Automatic recognition of configurations in plays – Training a Machine Learning Algorithm

Have a look at the Introduction to this series of blogposts on Automatic recognition of configurations in plays

I Train an algorithm to identify action information in stage directions (enter-, exit-, dead- and aside)

In a first step stage directions of 15 dramatic texts from Textgrid (www.textgridrep.de) were classified by hand. (https://github.com/pouyana/teireader/tree/ner/drama/hand_classified). This was done in large part by undergraduates participating in the course “Quantitative Dramenanalyse” taught in winter term 2013/2014 together with Dr. Christof Schöch in the Digital Humanities Bachelor-Program at the University of Würzburg. I am very grateful for their work. They were enriched with one of the following type-Attributes in the tag <stage>: enter, exit, aside, dead. This resulted in a total number of 1986 (https://github.com/pouyana/teireader/tree/ner/drama/corpora) stages. Secondly the stages were extracted from the documents and linguistically classified with the help of the Natural Language Toolkit (http://www.nltk.org/). NLTK consits of several text-processing libraries for python scripts to classify and tokenize utterances as well as adding part of speech tags. and analyzing them semantically.

Other 1504
Enter 176
Exit 230
Aside 68
Dead 8

Table 1: numbers of the different stage types in the manually classified corpus

In a second step Random Forest was trained with all these types of stages in / with http://scikit-learn.org with all types of satges (exit, enter, aside, dead). This was done with NLTK-Trainer (http://nltk-trainer.readthedocs.org/en/latest/; http://www.nltk.org/howto/data.html; https://github.com/japerk/nltk-trainer)The training corpus consisted of the following plays:

Lessing, Gotthold Ephraim Emilia Galotti 1772
Hafner, Philipp Mägera, die förchterliche Hexe 1761
Gryphius, Andreas Horribilicribrifax 1663
Lessing, Gotthold Ephraim Der Freigeist 1749
Lessing, Gotthold Ephraim Der Misogyn 1748
Lessing, Gotthold Ephraim Die Juden 1749
Lessing, Gotthold Ephraim Die alte Jungfer 1748/49
Lessing, Gotthold Ephraim Miss Sara Sampson 1755
Lessing, Gotthold Ephraim Nathan der Weise 1779
Lessing, Gotthold Ephraim Der junge Gelehrte 1747
Möser, Justus Die Tugend auf der Schaubühne oder Harlekins Heirath 1798
Schiller, Friedrich Die Räuber 1781
Schiller, Friedrich Don Carlos 1787
Schiller, Friedrich Kabale und Liebe 1784
Weise, Christian Masaniello 1682

Table 2: Dramatic texts with hand-classified stages

Then a first automatic classification of new material was done for the following plays, complementing the corpus to 29 dramatic texts, representing a variety of dramatic  subgenres in 17th and 18th century :

Bostel, Lukas von Der hochmüthige, gestürzte und wieder erhabene Crösus 1684
Goethe, Johann Wolfgang Stella 1775
Gryphius, Andreas Absurda Comica oder Herr Peter Squentz 1658
Gryphius, Andreas Verlibtes Gespenst, geliebte Dornrose 1660
Kurz, Joseph Felix von Prinzessin Pumphia 1767
Lessing, Gotthold Ephraim Der Schatz 1748
Lessing, Gotthold Ephraim Damon 1748
Lessing, Gotthold Ephraim Minna von Barnhelm 1767
Lessing, Gotthold Ephraim Philotas 1774
Schiller, Friedrich Don Carlos 1787
Stranitzky, Joseph Anton Türckisch bestraffter Hochmuth 1725
Weise, Christian Bäurischer Machiavellus 1681
Weise, Christian Der niederländische Bauer 1700
Weise, Christian König Wentzel 1700

Table 3: Dramatic texts in the test corpus for automatic classification

The automatically classified action information of the stages (see Table 4: before) then was revised by hand (see Table 4: after):

Before After Count
enter other 76
aside enter 1
aside dead 1
aside other 4
other dead 2
enter exit 3
exit other 12
other aside 21
other enter 274
enter aside 4
other exit 80
exit dead 2

Table 4: Correction of the automatically classified action information

To improve the performance of that algorithm for an intended use in a bigger context we did some further training with the help of cross-validation. We used 75% of the corpus with the extracted stages for training and validated the results on the remaining 25%. Having classified the whole corpus beforehand it was thus possible to see if the algorithm gave the same results as classified. This procedure is done four times, always with another quarter as a testing corpus.

Here are the results (for an explanation of ‚precision‘, ‚recall‘ and ,f-measure’ see: https://en.wikipedia.org/wiki/Precision_and_recall)

aside precision 0.878788
aside recall 0.878788
aside f-measure 0.878788
enter precision 0.936842
enter recall 0.631206
enter f-measure 0.754237
exit precision 0.820000
exit recall 0.694915
exit f-measure 0.752294
other precision 0.866667
other recall 0.972222
other f-measure 0.916415

Table 5: Results of cross-validation

Cross-validation lead to excellent results for those stages, containing an “aside”. For “enter” and “exit” theres were high precsion values (meaning that those entities which were found trully prooved to be “enter” or “exit”). But there were also quite low values for recall, showing there were not few cases which were not found but classified as “other”.

To improve the results we thought of doing named entity recognition for two purposes:

  • To eliminate such cases were stages only consist of names (as they are “enter”)
  • To train random forest again with stages containig the information “Person Name” instead of various names.

These effort will be the topic of my next blogpost.

 

 

 

 

German Comedies of the Old Empire

History of German Comedy is usually based on about 100-150 canonical printed plays of the north and middle part of (today’s) Germany. Looking at the extension of performances of German Comedy as well as on the circulation of printed and handwritten plays reveals another territorial structure in question: The Old Empire (Das Heilige Römische Reich deutscher Nation).

Recent publications on German literature during this last period of the Old Empire show the importance of the this political, juridical, social and communicative structure for the production, distribution, reception of literature and other cultural products (see for the impact of the Old Empire on the thinking and writing of authors like Wieland, Herder, Schiller, Goethe, Jean Paul, Eichendorff and Kleist: Markus Hien: Altes Reich und Neue Dichtung. Literarisch-politisches Reichsdenken zwischen 1740 und 1830. Berlin u.a. 2015; See for a history of German Enlightment with repeated references on structures of the Old Empire: Steffen Martus. Aufklärung. Das deutsche 18. Jahrhundert – ein Epochenbild. Berlin 2015).

Here are some of the challenges for a History of German Comedy taking the production of the area of the Old Empire into account:

  • A vast amount of plays. The 35 volumes of Reinhart Meyers bibliography on drama in 18th century may give a first impression what the production in the Old Empire was like (see Reinhart Meyer, Bibliographia dramatica et dramaticorum. Kommentierte Bibliographie der im ehemaligen deutschen Reichsgebiet gedruckten und gespielten Dramen des 18. Jahrhunderts nebst deren Bearbeitungen und Übersetzungen und ihrer Rezeption bis in die Gegenwart. 1. Abteilung: Werkausgaben, Sammlungen, Reihen. 3 Bde. Tübingen 1986; 2. Abteilung: Einzeltitel. Tübingen 1993, ff.)
  • Importance of libretti and discussions on Opera and “Singspiel” for the development of theatrical expression (See for example Bernhard Jahn: Die Sinne und die Oper. Sinnlichkeit und das Problem ihrer Versprachlichung im Musiktheater des nord- und mitteldeutschen Raumes (1680-1740). Tübingen 2005; Jörg Krämer: Deutschsprachiges Musiktheater im späten 18. Jahrhundert: Typologie, Dramaturgie und Anthropologie einer populären Gattung. 2 Bde. Tübingen: Niemeyer 1998)
  • No clear demarcation between genres: difficulties to identify plays with good ending and comical structures. Question if such a concept of ‘comedy’ is a mere systematical construct.
  • Interrelation of different types of theatre with own traditions like Opera, puppet play, strolling players, Catholic and Protestant school plays
  • Impact of Italian, Latin and English Influences (additional to the French influences better known for that period), Importance of translations and adaptations
  • Discontinuous, non-teleological and non centralistic developments often using a single  work in differing version with differing intentions
  • Need for a structure of presentation / narration taking into account local developments as well as processes of cultural influcence and exchange.

Automatic recognition of configurations in plays – Introduction

Sure, there are some quantitative, computer-based methods applicable to dramatic texts which do not require any data mining, as for example stylometry. Others, like topic modelling, require a basic linguistic tagging which can be done with existing tools like tree tagger. But today I want to present my small experiments to extract the scenic presence of the characters of a play which I believe is a crucial structural aspect of dramatic texts. This is fundamental for network analysis but also for matrix information. [1]

The idea of charting the scenic presence in a configuration matrix goes back to Solomon Marcus who proposed to make a matrix for every single play which [2] holds one row for every character and one column for every scene. If a character appears on stage, it is registered in the corresponding cell. Based on these matrices Marcus computes for instance, the co-presence of characters, their lack of encounters or their scenic presence over long periods as well as the overall configuration density of a play.[3] The latter is the result of dividing the number of cells holding a 1 by the total number of cells (i.e., the number of characters multiplied by the number of scenes). In other words the configuration density indicates how many of the potential character entrances have actually been realized. It can therefore be designated as a measure of a play’s ‘population density’.[4]

This a simple task if the texts are marked up like the ones from the Folger Shakespare library. http://www.folgerdigitaltexts.org In this data characters in stage directions are tagged with a genuine who-tag, scenes are marked as <scene>, so that it is easy to extract information about who is present on stage in every scene. As far as I can see the markup of character-names in the stage directions is done by hand. For the case of German drama in the 17th and 18th century the markup of the Texts in Textgrid https://textgridrep.de  or the Gutenberg Collection do not not provide this information but only tags for scenes, acts, stage directions and speakers. In some cases there are no scene divisions at all.[5] In others there are, but often characters are entering and leaving the stage during a scene so that it is not possible to assume the scenic presence of all characters just for the <scene>-elements.

A research agenda in this area may look as follows.

  1. Classify stages by hand and train an algorithm to identify action information in stage directions.
  2. Train an algorithm to identify names of characters
  • Bring the information of I and II together
  1. Write a script which reads the results of III and checks if there are contradictions with speaker-information. For example: Show cases where a character identified as dead by a stage directions appears as a speaker afterwards. These cases might be wrong or might denotate a ghost’s appearance.

During the next weeks I will present the results of a small project with Pouyan Azari, student of computer science and Johnathan Gaede, student of digital humanities, working as graduate student researchers, funded by the Bavarian Academy of Sciences and Humanities.[6]


 

[1] Peer Trilcke: Social Network Analysis (SNA) als Methode einer textempirischen Literaturwissenschaft. In: Philip Ajouri, Katja Mellmann u. Christoph Rauen (Hg.): Empirie in der Literaturwissenschaft, Münster 2013, S. 201-247; Frank Fischer, Dario Kampkaspar, Mario Göbel, Peer Trilcke: Digitale Literaturwissenschaftliche Netzwerkanalyse trilcke.de/digitale-lina/; Wilhelm, T., Burghardt, M., & Wolff, C. (2013). “To See or Not to See” – An Interactive Tool for the Visualization and Analysis of Shakespeare Plays. In R. Franken-Wendelstorf, E. Lindinger, & J. Sieck (Eds.), Kultur und Informatik: Visual Worlds & Interactive Spaces (pp. 175–185). Glückstadt: Verlag Werner Hülsbusch. Retrieved from http://epub.uni-regensburg.de/28417/; Katrin Dennerlein “Measuring the average population densities of plays. A case study of Andreas Gryphius, Christian Weise and Gotthold Ephraim Lessing”. In: Semicerchio. Rivista di poesia comparata LIII (2015), S. 80-88.

[2]   Solomon Marcus, Mathematische Poetik (Frankfurt am M.: Athenäum 1973 [first published Bucuresti: Editura Academiei 1970]), p. 290.

[3]   See Marcus, Mathematische Poetik, 292–301.

[4]    Katrin Dennerlein “Measuring the average population densities of plays. A case study of Andreas Gryphius, Christian Weise and Gotthold Ephraim Lessing” Semicerchio. Rivista di poesia comparata LIII (2015).

[5]   In cases where just the tags of the scenes are missing, it is possible to identify scenes automatically via XPath because Acts don’t have a text child.

[6]     http://www.badw.de (08.12.2015).

A digital corpus of German comedy between 1650 and 1815 – Part 1

Which plays between 1650 and 1815 are digitized and freely available? The biggest corpus of German Texts is the Textgrid Corpus. Texts are normalized and therefore usable for stylometry as well as topic modelling and are encoded following the guidelines of TEI P5 thus providing the markup for analyzing certain structural elements. For the above mentioned period Textgrid contains about 200 plays (including adaptations but excluding translations).

The corpus shows a a wide range of genre labels in titles and subtitles.

Allegorisches Drama (1), Bardiet (1), Bürgerliches Trauerspiel (1), Drama (14), Dramatische Skizze (1), Dramatisches Gedicht (6), Dramenfragment (1), Duodrama (1), Familiengemälde (2), Farce (2), Festspiel (2), Knittelversspiel (1), komische Oper (5), Komödie (7), Gelegenheitsstück (1), Lustspiel (45), Maerchen (2), Melodrama (1), mit Musik vermischtes Drama (1), Monodrama (1), musicalisches Dramate (1), Nachspiel (2), Oper (2), Operette (1), Parodie (2), Ritterschauspiel (1), Satire (1), Schäferspiel (3), Schattenspiel (1), Schauspiel (18), wunderliches Schauspiel (1), Schauspiel mit Gesang (1), scherzhaftes Zwischenspiel (1), Scherzspiel (1), Schimpfspiel (1), Singspiel (4), Sittengemälde (1), Tragikomödie (1), Tragödie (4), Trauerspiel (47), Volksmärchen (1), Vorspiel (3)

What catches the eye is that those labels obey very different rules of classification.
They refer to

  • the content (“Bardiet”, “Bürgerliches Trauerspiel”, “Familiengemälde”, “Ritterschauspiel”, “Märchen”, “Schäferspiel”),
  • to the media used (“Mit Musik vermischtes Drama”, “musicalisches Dramate”, “Schauspiel mit Gesang”, “Oper”, “Operette”, “Schattenspiel”),
  • to their length (“Dramatische Skizze” or “Dramenfragment”)
  • to the number of characters (“Monodrama”, “Duodrama”)
  • to their relationship to another play (“Parodie”)
  • to the form (“Knittelverspiel”)
  • to their relationship to society (“Satire”)
  • to emtions they intend to produce (“Komödie”, “Lustspiel”, “Farce”, “scherzhaftes Zwischenspiel”, “Scherzspiel”, “Schimpfspiel”, “Trauerspiel”, “Melodrama”, “Tragikokmödie”, “Tragödie”).

The remaining labels are quite unspecific, but are used for about 35 plays (mostly “Drama” and “Schauspiel”).

The labels point to genres in that sense that they designate historical terms and series of texts. In Textgrid  there are a lot of genres with only one specimen. I think it is not problematic to summarise the following genres under the genre-term ‚comedy‘ because they all have comical elements and a happy end:

Farce (2), Knittelversspiel (1), komische Oper (5), Komödie (7), Lustspiel (45), Parodie (2), Satire (1), scherzhaftes Zwischenspiel (1), Scherzspiel (1), Schimpfspiel (1)

In addition I decided to add the following plays to this category:

  • Der niederländische Bauer (1700) of Christian Weise, which is designated as „wunderliches Schauspiel“ in the original title but is mentioned in all Histories of German Comedy I know, has a good ending and treats the comical motive of the king for one day.
  • Three plays which are likely to be comedies judging from title and characterlist: Gottlieb von Stephanie one-act play Der Schauspieldirektor (1786), named „Gelegenheitsstück“ and the two aftermaths Der Witzling (1745) fronm Luise Adelgunde Victorie Gottsched and Die Tugend auf der Schaubühne oder Harlekins Heirat (1763) von Friederike Caroline Neuber.

This results in a total amount of 7 comedies in this corpus for the period from 1650-1815.  In grouping these plays together I moved from a concept of genre as a historical label and a series of texts to a concept of genre as a classification term in literary studies. To tell those two usecases of ‘genre’ apart I will use ‘genre(1)’ and ‘genre(2)’  [1]

[1] Harald Fricke uses the term ‚genre‘ to designate „historisch begrenzte[] literarische[] Institutionen“ und ‚Gattung‘ to designate „rein systematische[] literaturwissenschaftliche[] Ordnungsbegriff[e]“. (Harald Fricke: Norm und Abweichung. Eine Philosophie der Literatur, München 1981, S. 132). As there is only one term in Englisch, I decided to use genre(1) and genre(2).

Computing the comedies I will need other genres(2) to test if my findings on comedy are really specific for the genre(2) and not for the period as a whole. Taking the following genres(1) together under the label of ‘tragedy’ may provide a good corpus for comparative questions: Bardiet (1), Bürgerliches Trauerspiel (1), Tragödie (4), Trauerspiel (47). This results in a number of 53 tragedies. By tragedy I mean that there is a protagonist which becomes guilty without being guilty and the end of the play with the dead of the protagonist with none of the spectators wishing for it.

A “Bardiet” is a very special German genre(1) as it was termed by Friedrich Gottlieb Klopstock for a play with patriotic content in high style (‘ein vaterländisches Gedicht’). The only existing representative of this genre(1) is his own play Hermanns Schlacht published from 1769 .

The remaining genre labels in Textgrid are:

Drama (14), Dramatische Skizze (1), Dramatisches Gedicht (6), Familiengemälde (2), Festspiel (2), Maerchen (2), Melodrama (1), mit Musik vermischtes Drama (1), Monodrama (1), musicalisches Dramate (1), Oper (2), Operette (1), Ritterschauspiel (1), Schäferspiel (3), Schattenspiel (1), Schauspiel (18), Schauspiel mit Gesang (1), Singspiel (4), Sittengemälde (1), Tragikomödie (1), Volksmärchen (1), Vorspiel (3)

Do those plays do have enough in common except of a good ending (what is the case for all the plays in my corpus in this third group)? There are some which seem to be more simliar to tragedy although there is no tragic ending. This ist especially the case for operas where we had the convention in 18th century to end even tragic plots with a good ending (see for example: Florian Mehltretter: Die unmögliche Tragödie. Karnevalisierung und Gattungsmischung im venezianischen Opernlibretto des siebzehnten Jahrhunderts. Frankfurt u.a. 1994; Ursula Kramer (Hg.): Lieto fine? Musik-theatralische Schlussgestaltung um 1800. Tübingen 2009). Here the protagonist begins in a state of good fortune, falls into bad fortune, and is restored back to good fortune by the plays conclusion. An example ist Christoph Martin Wielands Alceste (1773).There are other plays showing a good ending and having comical passages. Should they be added to the comedies although they are not mentioned in histories of comedy? Examples are: Lucas von Bostels Der hochmuetige, gestuerzte und wieder erhabene Croesus (1711) or Die Zaubrflöte (1791) by Johann Emanuel Schikaneder.

It stands to reason just to test if the plays in the third group are more likely to behave like comedies or like tragedies or like outliers when computing, for example,  word frequency, grammatical structures, topics, density or complexity measures.

Remains the task to read the plays to see if they have a good ending and if they contain comedy…

Minutes of the Workshop Computer-based analysis of drama and its uses for literary criticism and historiography (March 2015, Munich)

Over the last years, some developments paved the way for the computer-based analysis of dramatic texts. On the one hand, more and more texts are available electronically as, for example, in the collection Théâtre classique for French drama, the complete works of Shakespeare and a collection of German drama and libretti. On the other hand, we can now see the emergence of new methods and tools that allow us to gain and compute information like, for example, word frequency, speech length, configuration structures or topics automatically, also from large amounts of texts.

Only rarely, results of computer-based text analysis find their way into literary history (except for some studies on authorship attribution as, for example, in the Molière/Corneille and the Shakespeare/Marlowe debate). The books of, for instance, Franco Moretti, Matthew Jockers, Matt Erlin/Lynne Tatlock and Stephen Ramsay may serve as evidence that this is now changing but they do not deal with drama. As questions of literary criticism and history are at the heart of literary studies and dramatic texts have genuine textual structures, the workshop aimed at evaluating the possibilities of computer-based drama analysis for theses domains. What can be the use of the mostly quantitative data for questions such as style, popularity, canonicity, genre, periods of literary history, and individual authorial periods of production? What kind of new questions, which new micro- or macronarratives are arising from these approaches? Where are the differences to prior non-computational approaches to quantitative aspects of drama (see e.g. the works of F.v.Cube/W. Reichert, Solomon Marcus, Manfred Pfister)?

These were the underlying questions of the workshop on Computer-based analysis of drama and its uses for literary criticism and historiography, which took place on 12-13 March 2015 in Munich at the Bavarian Academy of Sciences and Humanities. It was a first inquiry in this emerging field with scholars from the disciplines of literary criticism, computer science and linguistics involved. I am gratful for the funding on behalf of the Young Scholars Program of the Bavarian Academy of Sciences and Humanities for this workhop.

In the beginning of the workshop Martin Mueller the TEI-expert and Co-Author of The Chicago Homer and WordHoard emphasized the necessity of corpus curation. In his paper Shakespeare His Contemporaries: A corpus-wide approach he showed that linguistic as well as semantic annotation are the prerequisites for later analysis of a corpus. On a corpus of 550 English plays written between 1576 and 1642 he showed not only the usefulness of linguistic annotation (like Tokenization, Lemmatization, POS-tagging, Standardization of Spelling, Division into N-grams) but also presented the idea of building a meta-castlist that associates each of the ~13,500 speaking roles with a corpus-wide ID and properties of sex, age, kinship, and social status. In developing the labels for social status together with undergraduates, he aims at understanding Shakespeares plays from a todays point of view. He also plans to compute who speaks to whom in verse or prose and at what length. For further reading see his Blog-entry.

Gerhard Heyer and Thomas Efer provided interesting insights in their Toolbox for Drama Mining (Information Theory and Structural Analysis: A Toolbox for Drama Mining). It allows to compute each of the following features, relate them to one another and visualize the results: block entropy, number of types, number of tokens, number of characters, speech length of characters (median and average), speech length complexity, token complexity and median and maximum length of monologues. Networks of characters are analyzed and visualized as well. Right now the parsing of dramatic texts is adapted for the Folger Shakespeare texts and for drama from the Gutenberg collection. The aim of their project is to gain insights in the linguistic structures and the development of the dramatic works of Shakespeare.
Their findings about differences in complexity of the Hamlet Quartos may be of interest for researchers concerned with questions of dependencies and absolute dating of those early prints. As to the complexity values for several translations of Othello it was astonishing to see that NoFear Shakespeare had a very high complexity measure compared to adaptations which did not try to facilitate the reading.

Toolbox for Dramamining - comparison of Othello-Translations

Toolbox for Dramamining – comparison of Othello-Translations

This stems from the fact that NoFear Shakespeare is much more narrative in its structure and tries to tidy up difficult tropes . This leads to the fact that the words to follow are much less foreseeable than in other translations which results in a higher complexity value. It is clear that this is only one aspect of complexity which has to be combined with other ones to cover ideas of complexity usually refered to in literary criticism. With this in mind it will be most interesting to see how these and other of their findings on plot unfolding and network analysis relate to traditional Shakespeare criticism.

Manuel Burghardt and Thomas Wilhelm (University of Regensburg) presented their Tool for Computing and Visualizing aspects of dramatic texts (Interaktive Analyse und Visualisierung von Dramen). They introduced their project of analyzing Plays of the Folger Shakespeare Library which allows to quickly overview who is present on stage in which scene and who is talking to whom in what length.

Visualising Shakespeare

Visualising Shakespeare

Together we are now heading for extending the tool as to the aim to analyze not only Shakespeares plays but also German plays from TextGrid and computing some more values like speech length, speech number and configuration density to see how they develop over the course of dramatic history of the 17th and 18th century.

In my talk Configuration density as a measurement for differences between comedies and tragedies I tested a measure to tell comedies and tragedies apart which had been computed for the first time by Solomon Marcus. Configuration density is a term for the average population density, i. e. the number of characters on stage in relation to the overall number of characters. As comedies are likely to have more characters on stage in average than tragedies where characters often miss each other tragically I assumed that comedies would have a much higher configuration density than tragedies. As a first step to inquire wether configuration density is a differentia specifica of tragedies and comedies in German plays of the 17th and 18th century I computed the density of dramatic texts from Andreas Gryphius, Christian Weise and Gotthold Ephraim Lessing available in TextGrid. Although there is a trend that the genres differ in the average configuration density there are some outliers. For my small corpus they can be explained by the structure of the plays: Play in play, episodic structure, choires lower the overall configuration density of comedies, for example. During the next months I will try to prove these findings for a set of 200 plays.

Peggy Bockwinkel und Marcus Willand (University of Stuttgart) questioned the assumption that the title-character also is the protagonist of a drama. (Die dramatische Figur im Dramentitel und ihre Funktion für die dramatische Handlung. Eine konsistente Relation?) Therefore they compared the number of speeches and tokens of a character (activity) with the number of references to this very character by others (presence). Thus, they computed an activity marker which allows to contrast activity patterns of prototypical roles (like king, queen, servant et cetera). Another question is wether the title-character is also the central node of the network of a drama. Having the already determined activity of a character in mind, Bockwinkel/Willand pushed the idea of building networks with edges equivalent to the frequency of the dialogue between dramatic agents. By way of example they showed that Emilia Galotti in Lessing’s drama with the same title is not very active compared to the number of speeches of other characters. Rather Marinelli, the intriguer, is the central character of the play regarding to activity and centrality, not Emilia – even though the drama bears her name. However, her indirect presence is very high, which means that she is talked about frequently and is therefore more likely the trigger of what the scholars call the core dramatic issue than the one in charge to solve this issue. On a corpus, preliminary containing mostly bürgerliche Trauerspiele (originating from TextGrid) they are now aiming to show possible patterns in the centrality of the title-character regarding literary periods and genres in further detail.

Peer Trilcke,  Frank Fischer (University of Göttingen) and Dario Kampkaspar (Herzog-August-Bibliothek Wolfenbüttel)  inquired wether the computation of scenic copresenc of characters represents changes in literary history and allows to specifiy the differences between works of the literary mainstream and exceptional ones (Digitale Netzwerkanalyse dramatischer Texte). For this purpose they calculated network information for 150 plays from TextGrid of the time between 1730 and 1799. They showed that there is a gain of the average number of dramatis personae at about 1770 – exactly at the beginning of the Storm and Stress period. At the same  time we can witness a rise of the average degree, the number of direct relationships of a character to other characters. Comedies, tragicomedies and tragedies differ significantly as to thedensity of their networks and the number of dramatis personae. Considering the widley approved exceptionality of the plays of Goethe and Schiller they showed that these plays differ clearly from the mainstream with regard to their number of dramatis personae and their degree. See the slides here.

Jonathan Reeve (Modern Language Association) extracted Dialog from Shakespeare’s kings and compared it statistically with dialog from a number of other character groups, such as Shakespeare’s queens, servants, and fools, in an effort to computationally identify characteristic trends in the language of each group (»Imperial Voices«: Gender and Social Class among Shakespeare’s Characters, a Stylometric Approach). Distinctive word patterns are then identified in each group of characters. He concluded that relationships between these groups may be divined through this analysis, even suggesting implicit Shakespearean class hierarchies and gender relations. As I marvel at the method of this paper I would very much appreciate to test it with material from German drama. Read full paper.

Christof Schoech (University of Würzburg) focused on thematic aspects of genre with the help of topic modelling (Topic Modeling Literary Genre). In analyzing a collection of French Drama of the Classical Age and the Enlightenment he aimed at discovering whether different dramatic genres have distinctive dominant topics and inversely, whether clustering methods based on either most relevant topic words or overall topic scores produce groupings of texts which agree with more traditional genre distinctions.
The three most distinctive topics for comedies, with their top-ranked words, are tp49 (“aimer-parler”), tp56 (“monsieur-homme”) and tp32 (“beau-croire”). The three most distinctive topics for tragedy are tp36 (“coeur-amour-croire”), tp41 (“soin-jour”) and tp00 (“dieu-sang”), followed very closely by tp21 (“coeur-amour-cruel”). This shows that in tragedy, love is no less of an essential issue than in comedy, but in a very different perspective and associated, especially in tp21, with cruelty and misery.
In several cases, interesting patterns of topic importance over text progression can be discovered. This is the case, for example, for tp30 (“hymen-époux-coeur-foi”), a topic related to socially and divinely consecrated marriage and associated with both comedy and tragedy. In tragedy, tp30 is of a much larger concern initially but then continually drops in importance, with marriage becoming either increasingly difficult or less of a concern. In comedy, on the contrary, tp30 is of somewhat lower but constant relevance during the first four fifths of the play, but then increases drastically in the last fifth of the play.
Topic-based principal component analysis shows that the topics structure the text collection along the lines of genre. The first component (horizontal axis) clearly divides the plays into comedies (in the two quadrants to the left) and the tragedies (in the two quadrants to the right). The second component seems to be related to the difference between the two main genres and tragi-comedy, which can only be found in the upper two quadrants of the plot. However, there is substantial overlap between tragicomedy and, especially, tragedy. Based on topics, tragicomedy seems closer to tragedy than to comedy.
The paper has just been accepted for the Göttingen Dialog in Digital Humanities

The workshop showed that there are diverging questions, needs and manners of solutions according to subjects involved. In short it seems to me that

  • linguists think of useful possibilities to enrich their linguistically deep tagged corpora with semantic annotations which allow to pose questions of literary critics.
  • Scholars from computer science either work in collaboration with literary critics and linguists or try to exploit concepts themselves e.g. from information theory for the analysis of dramatic texts.
  • Literary critics either aim to prove assumptions already known or hitherto unquestioned in literary criticism with quantitative data. Or they try to find unknown structures as well as semantic aspects in large corpora accessible with the help of computers for the first time. For them it is most relevant to explalin too which known features of a text aspects of language or computable features of the structure (for example of networks or configurations) correspond or provide deeper insights in. This means that even if one is providing a completely new values it has to be linked to aspects of texts or meaningful structures already known – be it as a synonym, be it as a new aspect.

It seems to me that projects in digitally assisted analysis of drama should always have at least one specialist from each subject (not to forget a person who really knows about statistics).

History of Comedy

This blog is about

a) Literary History of comedy in the Old Empire in 17th and 18th century (German adaptations and German original plays)

b) Digitally assisted analysis of drama and documents of performances with the aim to provide new insights in dramatical texts and the history of drama. This comprises preliminary steps like digitization, caption of metadata and electronic edition as well as the analysis of metadata and, of course, dramatic texts. Methods for close reading are as interesting as insights in patterns of large amounts of texts. All findings shall serve  to enlarge and improve my knowledge about a).