Analysing literature through the lens of information theory and network science
0 likes318 views
This document introduces a new tool for interactively analyzing literature that constructs temporally ordered networks of character occurrences. It uses natural language processing to automatically detect characters and relationships in texts. Researchers used the tool to produce annotated graphs of novels that identified important narrative episodes and sequences. Their annotations provided insights into how digital tools can cultivate new collaborative interpretive practices for literary analysis.
Analysing literature through the lens of information theory and network science
- 1. Twenty Thousand Leagues Above the Book: An Interactive Visual Analytics Approach to Literature n words n words k slices of n words (e.g. n=1,000) ... 1 2 3 k Abel Magwitch Joe Gargery, Mrs Joe Gargery, Pip Joe Gargery, Mrs Joe Gargery, Pip + Lexicon of characters (can also be a list of other entities such as places or phrases) - network visualisations - statistics and other measures Directed network Dynamic network visualization Source texts from Project Gutenberg Matched information • Abel Magwitch • Joe Gargary • Mrs Joe Gargary • Pip Matched information • Abel Magwitch automatic/semi-automatic methods to detect match entities in text, topic detection, sentiment analysis, part-of-speech tagging) configure analyze In recent years data-driven analysis has emerged as a growing methodology within literary studies. These distant reading practices harness available technology to open new avenues for how we understand literary texts. Whereas traditional literary scholarship is generally grounded in the interpretation of the specific language of a text or body of texts, macroanalytic approaches present new ways of seeing texts, both individually and in the aggregate. Here we introduce a novel tool for collaborative interaction with literature that transcends the boundaries of traditional close and data-driven distant reading. Our approach constructs temporally ordered networks of information occurrence, which we configured to match characters as the unit of information. This creates a unique view of the narrative structures within novels and opens a variety of possibilities for visual as well as information theoretic analysis. Figure 2: Overview of the data processing pipeline of our tool prototype. The user can choose between a supervised approach (lexicon of information to match needed) or an unsupervised approach (using state-of-the-art natural language processing). Markus Luczak-Roesch (@mluczak), Adam Grener, Emma Fenton As part of our initial user studies humanities scholars collaboratively produced a set of annotated graphs based on the visualisations that were provided to them (see Figures 3 and 4 for two examples). These annotated graphs show to great detail what kind of insights humanities scholars are looking for.Annotating groups of nodes to mark up larger episodes orsequences in the novel was a common task theyperformed. The interactions between the derived networks and the original texts flowed both ways; the scholars began with moments of known narrativesignificance and moved to the network to understand that significance, but they also identified elements of the network that presented as significant to identify areas to begin textual analysis. These observations on self-designed annotations are evidence for an emergent methodological practice, as the humanities scholars were given a novel tool and then developed their own analytical process around it. Our users studies also demonstrate the potential of our tool to cultivate collaborative interpretive practices for readers. Project page: https://vuw-sim-stia.github.io/computational-literary-science/- Source code repository: https://github.com/vuw-sim-stia/lit-cascades Figure 3: Manual annotation of the Bleak House network that identifies moments of convergence for important characters. Figure 4: Manual annotations of the David Copperfield network that identify moments of narrative significance. Figure 1: Deployment of our prototype on a 49'' multi-touch screen. Individuals and groups can convene in front of this setup and use the tool while working with one of the analysed novels. DOI of the paper describing this work: https://doi.org/10.1145/3148330.3154507 Demo: https://goo.gl/yFoZ6U