CHAPTER 3. METHODOLOGY
3.3. Towards a Distant Reading
3.3.2. Visual Models: Graphs, Maps, Trees
Graphical conventions help us frame a system of meaning, through indicators for grou-ping contiguous objects, or line differentiators, color codes, thematic sections to distinguish non-contiguous elements. Because of the variety of applications found in information design, data vi-sualization comes with a long history of graphical user interfaces. As part of a Digital Humanities project, it is important to reach visual representation and analysis with a broad understanding of the tools that such a tradition made available over time for designing graphical representations.
An understanding of visual languages, graphical forms and visual epistemology at large is fun-damental. Similarly, it is necessary to provide the background motivation that drove the choice of a visual model as opposed to others. Franco Moretti, when outlining his theoretical and
methodo-logical framework for distant reading, notably focused on three main visual models: i) graphs, which he uses to assess change in the genre of historical novels; ii) maps, for depicting the geo-graphical dimension of fictional narratives; iii) trees, to generate a classification of detective sto-ries. Graphs, maps and trees were adopted by Moretti to explore the corpus, illustrate its characte-ristics and explain results. Nevertheless, each one of these models for knowledge design presents dominant uses, with a consolidated tradition. Here, I will discuss the visual models proposed by Moretti in relation to both their history and their possible applications in the context of the pre-sent research. For the review of their previous applications, I will notably refer to the studies of the information designer Isabel Meirelles and of the Digital Humanist scholar Johanna Drucker.
While Meirelles approached the study of visual models as structures (Meirelles 2013), Drucker rather examined them as forms in the cultural oriented sense of the term, as “visual forms of knowledge production.” (Drucker 2014) In both cases, instead of presenting a unidirec-tional chronology of graphic design, they opt for an overview on the main principles and theories behind the visual information structures and forms. Graphs for instance are traditionally used to display relational structures (Meirelles 2013: 48) and are more often referred to as networks.
Drawing upon graph theory, a network is defined, both visually and metaphorically, as a set of points, symbolizing actors (e.g. individuals, groups, institutions, texts, etc.), and a set of lines, symbolizing the relations between these actors (Beauguitte 2016: 2-3). Laurent Beauguitte gives a relatively precise description of what constitutes this branch, by defining network analysis as the body of methods, notions and concepts used for studying a given relational phenomenon iden-tified as network. As he also points out, “analyzing a network does not necessarily involve using
network analysis methods, and conversely network analysis methods can be used to study literary works, ecological systems, and so on.” (ivi: 1-2, my translation)
Graphs and networks have already been proposed as theoretical and methodological tools in human sciences. In particular, such a theoretical and visual framework has been widely applied to social sciences, from sociometry (Moreno 1951) to analysis of social networks (Wasserman and Faust 1994), but also to the study of cultural dynamics, with historical (Schich et al. 2014) or industrial perspectives (Yucesoy et al. 2018). The question of a complex structure, being it an ecosystem or a network, and of complexity in general emerges as a central problem in contempo-rary academic research, in several areas of knowledge. Because of the complexity of real-world systems and patterns of connection, as well as the mathematical theory that explains them, visual models for graph and network analysis are inherently dependent on empirical data and they are quantitative in nature (Meirelles 2013: 48). Among other computational tools, programs like Ge-phi help introducing Digital Humanities scholars to network analysis. And yet, the effort for vi-sualization will be useless without a valid dataset able to account for at least one of the functions studied in network analysis (e.g. path length, network centrality, hubs, level of connectivity).
While I did work on network/graph visualizations over the course of my doctorate, they will not be included in this dissertation, given their experimental and introductory nature which does not allow for a reliable analysis of the scenario they generate. Such a parallel, ongoing re-search project is still available in the publication “An introduction to network visualization for television studies: models and practical applications” (Taurino 2019). In the lack of insight on data from major media conglomerates and streaming giants like Netflix and Amazon Video, for the research presented here I therefore relied on graphic visualizations of networks provided by
external, reliable sources like Vox and Variety which do not take the classic shape of a graph (see chapter 5). Instead, they can be described more as a maps of networks.
This brings us to the second abstract model outlined by Moretti for his work on literary studies, that of maps. Maps suggest a geographic dimension, which is used to display processes of “space-making” (Drucker 2014: 76), geo-localization and distribution and the definition of a
“spatial structure” (Meirelles 2013: 115) Geographic mapping has many subfields. One can either work with cartographic coordinates based on real-world geography (MacEachren 1995) or simply plot (fictional) geographic spaces (Boni 2016b). Maps can be used forhighlighting density or ex-pansion, for illustrating travel routes, relationships between places, geo-localized activity. Meirel-les notably discusses more specifically thematic maps, which came from a long history of origi-nally displaying data in mathematics, natural and physical sciences (from navigation purposes to cosmology), to then land in social, economic, political science starting from the mid-1800s, as a result of the increasing use of data for determining population planning and growth (Meirelles 2013: 117). Post-colonial human societies eventually grew into a geographic complexity, made of processes of globalization and increasing interconnectedness. Visual schemes for mapping beca-me even more important with the advent of the Internet and social platforms, which allows for collection of a high amount of data about geolocalized socio-cultural practices.
The spatial turn (Warf and Arias 2008; Bodenhamer et al. 2010) in the Humanities further supported mapping as an essential practice for situating academic research. As of today, desi-gning visual models for maps involve “three basic areas: projection, scale, and symbolization.” (Meirelles 2013: 118) These three areas are fundamental for observing spatial attributes not only in terms of physical component, but also in terms of a cultural and media
eco-logy, which is the approach taken here. Over the course of my doctorate, I notably underwent a mapping project as part of a broader project carried out by the research group Labo Télé (Univer-sity of Montreal) and supervised by professor Marta Boni. The outcomes of such a collaboration are published in the paper “Maps, Distant Reading and the Internet Movie Database: New Ap-proaches for the Analysis of Large-Scale Datasets in Television Studies” (Taurino and Boni 2018). However, for the purposes of this dissertation, the work on maps I pursued remains, at the current stage, too broad. I therefore opted for collecting maps visualization of the United States reported on external sources from different television eras (see Chapter 4 and 5). The purpose was to reconstruct a topology of infrastructures, by comparing different maps of the U.S. tele-communication network and showing how the constructed cultural scape of the television indu-stry still contains a strong geophysical component in the hierarchies of powers.
The concept of hierarchy, which I already discussed in Chapter I relation to Levine’s neo-formalist approach to political forms, is particularly useful to introduce the third model proposed by Moretti, as he adopts trees for classifying literary objects and doing text analysis. Meirelles defines trees precisely as hierarchical structures that can be represented visually as either stacked or nested schemes (Meirelles 2013: 18). “In a nutshell, hierarchical systems are ordered sets whe-re elements and/or subsets awhe-re organized in a given whe-relationship to one another, both among them-selves and within the whole. Relationships vary according to the field domain and type of system, but, in general, we can describe them by the properties of elements and the laws that govern them (e.g., how they are shared and/or related).” (ivi: 17) With regards to stacked schemes, the geome-tries adopted most often for information design of hierarchical structures are displayed as a group of interconnected, directional lines - i.e. vertical/horizontal/central, superior/inferior,
center/peri-phery - (ivi: 18) that resemble trees structures, such as the name itself suggest. Or as Drucker ph-rased it “The tree’s root and branch structure echo morphologies from natural and cultural worlds pressed into the service of a graphical one.” (Drucker 2014: 32)
Drucker additionally suggests that a tree-structured visualization should be supported by the dataset accordingly to its genealogical nature. With this she does not imply that all tree repre-sentations are genealogies, but that such visual models do account for common concepts such as continuities and derivation (Drucker 2014: 87) Hence why, if I would want to approach my cor-pus via a stacked scheme tree, it would result in the impossibility of creating any visualization at all, given that the television anthology form evolved in the United States in a rather discontinuous way, due to industrial and technological disruptions. Even a nested scheme does not seem appro-priate. “Elements in nested schemes are positioned within containers assembled according to their interdependency and subordination.” (Meirelles 2013: 18) A nested graphic form of my corpus translates like this:
Figure 7. Nested graphic visualization of the dataset extracted from the Wikidata Query Service, created using the treemap visual model available on the Wikidata Query Service Platform.
It is evident that such a visual model appears rather obscure and quite difficult to read.
Since the aim of a methodology based on data visualization is to facilitate the cognitive load in-stead, I decided to pursue a different path, and evaluate another model of visualization based on time, able to account for the evolutionary - but non-derivative - path of the anthology form in te-levision. The next paragraph will discuss the representation I adopted for describing the time-component of my corpus, thus serving as a preliminary introduction to the additional qualitative analysis (media industry analysis, platform analysis) presented in chapter 4 and 5.