Programmable Corpora: Towards a Combined Words/Networks Analysis of Literary Texts · SUNBELT (Montréal)

### Programmable Corpora:
Towards a Combined Words/Networks Analysis of Literary Texts

[Frank Fischer](https://www.hse.ru/en/org/persons/182492735) ¹ ²  ·  Eugenia Ustinova ¹

¹ Higher School of Economics, Moscow 
² DARIAH-EU

This presentation:

**bit.ly/2WPybvF**

[SUNBELT 2019](https://www.fourwav.es/view/717/info/)  ·  Montréal  🇨🇦  ·  20 June 2019

### TOC

1. Network Research on Drama
2. Programmable Corpora: Networks for Literary Studies
3. Catching Protagonists: Quantitative Dominance Relations

---

## Chapter 1.

## Network Research on Drama

## Literary Network Data as Example Files (1993)

Donald Knuth: The Stanford Graph Base: A Platform for Combinatorial Computing. ACM Press, 1993. Networks extracted from fiction („Anna Karenina“, „Les Misérables“, etc.) are based on co-occurrences.

### One of the First Literary Network Analyses (1998)

![Simple Storys, Network Graph, with Pajek (1998)](images/simple-stories-network.png)

Thomas Schweizer, Michael Schnegg: *Die soziale Struktur der „Simple Storys“. Eine Netzwerkanalyse.* 1998. ([PDF](https://www.ethnologie.uni-hamburg.de/pdfs-de/michael-schnegg/simple-stories-publikation-michael-schnegg.pdf)) Network extracted from [Ingo Schulze’s](https://en.wikipedia.org/wiki/Ingo_Schulze) novel „Simple Storys“, done by ethnologists. 38 nodes (characters). Edges: positive, negative or exchange relationships. Visualisation with [Pajek](http://vlado.fmf.uni-lj.si/pub/networks/pajek/).

### „Les Misérables“ as Example File in Gephi (2008)

![Gephi Screenshot, CC](images/gephi-les-miserables.png)

(Screenshot taken from [workshop material](https://jasonheppler.org/courses/csu-workshop/gephi.html) by Jason Heppler, 2016. Licensed under CC BY-NC-SA 4.0.)

### Franco Moretti’s Analysis of „Hamlet“ (2011)

![Network of Hamlet selon Moretti](images/moretti-hamlet.gif)

(Source: [newleftreview.org](https://newleftreview.org/II/68/franco-moretti-network-theory-plot-analysis).)

## Distant-Reading Showcase (2016)

![465 drama networks at a glance](images/distant-reading-showcase-poster.jpg)

*Distant-Reading Showcase* (released at DHd2016, Leipzig). Download via Figshare. DOI: [10.6084/m9.figshare.3101203.v2](https://dx.doi.org/10.6084/m9.figshare.3101203.v2).

### Extraction of Literary Network Data: A „Digital Spectator“

- following an idea of Solomon Marcus, „Poetica matematică“, 1970
- operationalisation of ‘interaction’: „Two characters interact with one another if they perform a speech act within the same segment of a drama (usually a ‘scene’).“

---

## Chapter 2.

## Programmable Corpora: Networks for Literary Studies

## **DraCor**

- DraCor: Drama Corpora Platform (https://dracor.org/)
- DraCor is a showcase for the concept of „Programmable Corpora“
- Programmable Corpora: collection of full-text literary corpora with an API and Linked Open Data

## Frontend dracor.org

![DraCor-Frontpage)](images/dracor-frontpage.png)

https://dracor.org/ (public beta!)

## Connected Corpora

- German, Russian, Spanish, Swedish, Italian, Shakespeare, Ancient Greek, Roman, …
- full texts encoded in TEI (XML standard format in the Humanities, ~ 550 elements for the encoding of documents)
- example: [Shakespeare’s „Hamlet“](https://dracor.org/shake/hamlet)

## DraCor Technology Stack

![DraCor Technology Stack](images/dracor-drawio.svg)

All repos are open source: https://github.com/dracor-org

## DraCor API

- provides metadata and network data for all plays in CSV or GEXF format
- character constellations per segment (*dynamic graphs!*)
- spoken text (all characters, only female/male, or per character)
- live documentation via Swagger: https://dracor.org/documentation/api/

### Pushkin’s „Boris Godunov“ (1/2)

![betweenness](images/boris-godunov-gephi.png)

Correlations: Label size = betweenness centrality; heat of nodes = word-based measures. (Data source: GEXF file from https://dracor.org/rus/pushkin-boris-godunov.)

### Pushkin’s „Boris Godunov“ (2/2)

![Boris Godunov (dynamic graph)](images/pushkin-boris-godunov.gif)

**Dynamic graph**, generated with [**ndtv**](https://cran.r-project.org/web/packages/ndtv/index.html) package. Data coms directly from the **DraCor API**. Script by Ivan Pozdniakov ([source code on RPubs.com](https://rpubs.com/Pozdniakov/godunov)).

## DraCor Shiny App

![Shiny App](images/shiny-kaethchen.png)

https://shiny.dracor.org/ (by Ivan Pozdniakov, based entirely on DraCor API).

## Small-World Phenomenon in Russian Drama

![Small-World Phenomenon in Russian Drama)](images/sw_size_v2.svg)

Testing all plays against the criteria in Watts/Strogatz 1998. Repo: https://github.com/pixelmagenta/rusdracor-small-worlds.

## Gamification: „Brecht Beats Shakespeare!“

![card game](images/card-game-dh2018.jpg)

„Brecht Beats Shakespeare!“ (released at DH2018, México). Download via Figshare. DOI: [10.6084/m9.figshare.5926363.v1](https://doi.org/10.6084/m9.figshare.6667424.v1).

---

## Chapter 3.

## Catching Protagonists: Quantitative Dominance Relations

## What Are „Quantitative Dominance Relations“?

- term coined by Manfred Pfister (1977)
- modelled by us as multi-dimensional vector of network- and word-based features

## Calculation of Rankings for 8 Measures (For Every Character in Every Play)

**5 network-based**:
- Degree
- Weighted degree
- Closeness centrality
- Betweenness centrality
- Eigenvector centrality

**3 word-based**:
- Number of words
- Number of speech acts
- Number of appearances

## Example: Character Rankings for „Hamlet“ (Excerpt)

|Character |Words|Speech Acts|Appea- rances|Between- ness|Close- ness|Weighted Degree|Degree|Eigenvector|TEXT (total)|NETWORK (total)|
|----------|-----|-----------------|------------------|------------------|----------------|--------------------|------|-----------|------------------|---------------------|
|Horatio |4 |4 |4 |1 |1 |4 |1 |3 |4 |1 |
|Gertrude |7 |6 |3 |3 |2 |3 |2 |1 |5 |2 |
|Claudius |2 |2 |2 |4 |3 |2 |3 |2 |2 |3 |
|Hamlet |**1**|1 |1 |2 |4 |1 |**4** |4 |1 |4 |
|Laertes |5 |5 |7 |5 |5 |6 |5 |6 |6 |5 |
|Polonius |3 |3 |5 |6 |6 |5 |6 |5 |3 |6 |
|Ophelia |6 |7 |8 |7 |7 |9 |7 |7 |7 |7 |
|… |… |… |… |… |… |… |… |… |… |… |

Top-7 characters of the play regarding their degree values (out of 38 characters in total). Hamlet ranks 1st for the number of words, but only 4th for degree (too many monologues? 🤔).

## Correlation of Word- & Network-Based Rankings

![Spearman’s rank correlation coefficient](images/cor_coeff.svg)

Distribution of **Spearman’s rank correlation coefficient** for word- vs. network-based measures in Russian Drama Corpus.

## Ranking Results

- for > 50% of plays, a definite *quantitative* main character is identified, ranking first in at least 7 of the 8 measures (in Russian and German Drama Corpus)
- yet literary texts can have more than one definite protagonist

## Major and Minor Characters

- division into **four groups** (following „Reallexikon der deutschen Literaturwissenschaft“, 2003):
 - I. main characters
 - II. secondary characters
 - III. marginal characters
 - IV. functional characters
- using ```cut``` function (in R)

## Groups of Importance

![Groups of importance](images/percentages_five.svg)

Shares for each group of characters in the entire Russian Drama Corpus.

## The „Chekhov Effect“

![Chekhov effect](images/first_group_degree.svg)

Percentages of first-group characters divided by **degree** in the entire corpus **by decade**.

---

Thanks.

https://dracor.org/

#ProgrammableCorpora