Scale and digital tools

In my last few posts, I’ve shared brief introductions to Voyant,, and Palladio based on my own exploration of these tools. All of these tools make it relatively easy to a certain kind of DH research – Voyant does text analysis, makes mapping easy, and Palladio creates network graphs. But how do these different approaches interrelate? What can one tool tell us to ask deeper questions of another tool?

When I explored Voyant,, and Palladio, I used a dataset shared by Stephen Robertson which represented the WPA slave narratives. Thinking through what each of these different tools highlighted in the same data helped me to understand the differences involved.

Fundamentally, Voyant,, and Palladio made me think about scale differently. Part of this was due to the data I used. The texts I plugged into Voyant included WPA interviews from many states while I used a more limited set for and Palladio, which focused solely on Alabama. This led me to examine the differences between interviews in different states using Voyant, while I used and Palladio to examine data within one state (see below).

The more I thought about these differences of scale, I realized the difference wasn’t just the data I start with but also how the tools needed different kinds of inputs and created different kinds of outputs. You could, after all, map all of the WPA interview locations or include them all in a single network, but how would you break up the interview texts other than by state? You could break it up by interview or interviewee, but this is something you have to do before you upload your text to Voyant. (Still, the possibility is exciting!)

When mapping with or graphing with Palladio, you have the opposite approach to scale. They both map and graph individual points, or incidents of data, but allow you to generalize it through different kinds of grouping and counting. So the scale at which you work with data in these different tools differs based on what you give it and the ability to move between scales also differs. Though I only explored it briefly, Palladio even includes the functionality to tie your individual data pieces to outside resources such as webpages and pictures. This allows you to realize individual data points, whether they be peoples, places, or texts, very personally.

While Palladio and make it easier to work up from individual data points, they have virtually no sense of anything other than metadata. understands where a thing should be on a map and can sort data or visualize it according to certain traits (say the gender or nationality of a subject), and Palladio can create a graph connecting people based on the same traits. However, neither can really examine what a person said or wrote. Voyant can do that because it works at the scale of texts and their words. So when I noticed that a significant number of interviewers only interviewed “house slaves” in a Palladio graph (below), I realized that I could compare those interviews with others using Voyant‘s text analysis abilities to get a sense of how much interviewers influenced the interviews. To do that, we could “chunk” the interview texts not by state but by interviewer and then by the form of enslaved labor. By comparing the results from the interviewer “chunking” and the form of labor “chunking,” we could get a sense of how each factor influenced the interview.

Graph of interviewers to type of enslaved labor performed by interviewees created in Palladio

Fundamentally, Voyant works at the level of texts, counting words, not individual people and places, while Palladio and see the world as an aggregate of individual people and places but not what people said or wrote. These are all potentially useful perspectives depending on your research question, but moving between these perspectives can provide an even deeper analysis and understanding. Historical figures after all wrote texts at the same time they were embedded in geographical spaces and were a part of networks. And using tools such as Voyant,, and Palladio can help us recreate their lives in different but ultimately interrelated ways.