Open access
Author
Date
2022Type
- Doctoral Thesis
ETH Bibliography
yes
Altmetrics
Abstract
This thesis deals with network science approaches to literary corpus analysis. In an at-scale approach, it uses intermediate resolution levels to combine the advantages of both hermeneutical and statistical tools in a mixed-methods approach to release synergies.
The Deutsche Novellenschatz, a collection of 19th century German language novellas, will be the starting point of the analyses. The texts where edited and published by Paul Heyse and Hermann Kurz between 1871 and 1876 in 24 volumes. The corpus was digitized by Weitin in 2016 and is part of the Deutsche Textarchiv (German text archive). Investigating structural properties of this corpus will lead to analyses that help to understand whether these properties resonate with the intentions of the editors in respect of quality and delimitation for the novella genre. A key tool in the at-scale approach will be network analysis to consider different levels of examination from single texts to the entire corpus.
More concrete, the thesis can be split into three parts. Each part takes a certain viewpoint to disentangle the object of study. First, we try to validate the underlying novella theory by a reading experiment. The editors claim that a novella is characterized by the fact that it can be summarized within a few sentences. The presented study shows that this can only hold under certain circumstances. While it is possible that the claim was made as a general principal, we converted their recommended procedure into a hypothesis. In that way, we tested the summarizability of Hieronymus Lorm’s novella Ein adeliges Fräulein. As a result, we observed that only frequent readers agree on content and ordering of the summaries. In the second research project, we analyze the entire corpus with stylometric distance measures. The goal of this is twofold. On the one hand, it makes it possible to gain new insights into the corpus. On the other hand, we could use the Novellenschatz as a prime example of a closed corpus. As such, it is used to gain a deeper understanding about the implications and effects towards the application of the underlying methods. Based on this, we closer analyzed the resulting subgroups to deliver meaningful results on the subject matter.
The final part of the thesis is dedicated to the creation and examination of character networks. Therefore, we construct these networks in a semi-automated approach from the novellas. These networks are special in manifold ways. First, we do not only consider character names but also enrich our data set by also taking into account synonyms. In this way, we can assure that each character co-occurrence has an underlying contextual interaction. Moreover, it makes it less likely to miss out character interactions. This is important since we do use this information to create two-mode networks. Here, the links are present between characters and paragraphs to generate a dynamic plot of the interactions. These networks are then studied for structural (dis-)similarities. In particular, we test again for the intention of the editors on how the strict form of the novella gives rise to a specific character constellation network. For that reason, we employ a measure that uses the spectra of the adjacency matrix to convert structural similarities in networks into distances. As a result, we show that the distances for the novellas are small in general. However, on a relative scale, we can see differences between the network types. Similar to the second project, we analyze and classify the resulting clusters and subgroups. Show more
Permanent link
https://doi.org/10.3929/ethz-b-000602443Publication status
publishedExternal links
Search print copy at ETH Library
Publisher
ETH ZurichSubject
novella theory; stylometry; Deutscher Novellenschatz; character networks; network analysisOrganisational unit
09610 - Brandes, Ulrik / Brandes, Ulrik
More
Show all metadata
ETH Bibliography
yes
Altmetrics