I attended a two-part workshop titled “Spatial Analysis in Critical Praxis,” hosted by the Digital Ethnic Futures Consortium (DEFCon) — a community of researchers and educators focused on applying DH to the field of ethnic studies. (DEFCon initially organized around the production of The Digital Black Atlantic, the intro to which we read this semester.) This was an intensive, hands-on workshop, limited to a small number of participants (~12). In total, we had 6 hours together, which gave us plenty of time for both theoretical discussion and tool-based practice.

Broadly, the workshop homed in on the idea of using mapping tools (i.e., GIS) to analyze social datasets and ultimately drive societal change. This is where the “critical praxis” part comes in — the teachings were grounded in a theory of praxis (from Paolo Freire’s Pedagogy of the Oppressed) that directly links evaluation or analysis of trends to concrete action. This was framed around the following research questions: 

Accordingly, we weren’t just there to learn how to use GIS tools, but rather to think critically about how we use them, and to what ends. 

Furthermore, we went a step beyond simple cartography to actually conduct spatial analysis, which joins the visual field of mapping with humanities/social sciences-led analysis of the relationships between peoples and social/environmental systems. One way to think about this is with the following frame: “There’s something bad happening. How does that relate to the people I care about?” Spatial analysis draws a line between the phenomenon and the affected population.

In the second half of the workshop, we dove into 2 GIS tools in particular: ArcGIS and Felt. A bit about each:

• ArcGIS is the industry standard, developed by Esri in the ‘90s — it’s incredibly powerful, but also far from user-friendly. It can also be prohibitively expensive (in the thousands of dollars annually), though you may be able to access it through your institution. ArcGIS has an online version of the platform, which is what we used in the workshop. The benefit to ArcGIS Online is that it’s more streamlined, cloud-based, and functions regardless of operating system (ArcGIS Pro, on the other hand, only works on Windows).
• Felt is a relatively new tool that is free (for now) and available only as a browser-based interface. Its interface follows in the footsteps of collaborative tools like Google Docs or Figma, making it easy to work with teammates on a map in real time. Compared to ArcGIS, it’s much easier to learn the basics and get started quickly. However, because of its focus on ease of use, it can be difficult to do more advanced operations. Also, given the product is only a year or two old, it might lack some of the extensive options that other GIS tools offer.

In our workshop, we played around with multiple publicly available datasets, including one from the EPA on sources of pollution, and one from the US census (the American Community Survey). We were able to join these together and ultimately reveal trends in how areas of high exposure to pollutants overlap with various population groups. It was cool to see just how quickly we could get this up and running, again without too much pain on the technical front. The instructors encouraged us to try discovering geographic datasets on our own — it can be as simple as looking up the locality on Google, alongside a couple filtering search terms. For instance: “site:.gov GIS “Long Beach””

Ultimately, as students and researchers of the digital humanities, it may not be necessary to be a pro at ArcGIS or other tools that often have a steep learning curve. However, having proficiency in GIS principles and tools can allow you to take your research to new places and levels.

I recently attended a workshop hosted by The Pudding — a digital publication that puts out lots of great data-driven pieces with a focus on culture and media — on the subject of using ChatGPT to conduct research. I thought I’d share some of their findings and insights, in case it’s helpful for others considering how best to use AI, if at all, in their research.

The Pudding team members walked us through an internal project where they’d designated ChatGPT as their AI “intern.” The goal was not to let ChatGPT do everything, but rather assist and support at each stage of the research process to make the team’s work more efficient.

They defined their research process in five stages: idea, data, storyboard, design, and development. They focused on using ChatGPT for the first three. The team already had a research topic in mind — an analysis of clutch shooting in basketball — but turned to ChatGPT to help flesh it out.

Even at the idea stage, they quickly encountered that ChatGPT has a nasty habit of equivocation. It did not want to commit to making decisions. It also expressed itself verbosely and with a lot of fluff. The researchers were able to minimize some of these tendencies by explicitly telling the system to avoid them — however, this dragged out the process of actually getting any results.

ChatGPT proved useful for generating tons of ideas for research direction, but these were mostly surface-level. The team had to sift through dozens of bad ideas to find one that might have legs. Still, I feel this can be a useful exercise when brainstorming if you feel you’re running up against a wall.

Next came the data work. The Pudding team already had a dataset handy for clutch basketball shots, but they needed to get it into the right shape for analysis. ChatGPT had more success here: It seems to perform well at taking data in one format and converting it to another, such as appending or removing columns in a tabular dataset, modifying text strings, etc. — steps that might otherwise require significant Excel intervention or programming. That being said, it’s crucial to review ChatGPT’s work each time it cleans or processes data. Sometimes it loses track of the most recent dataset and returns to answers from earlier in the chat log. 

Nevertheless, being able to use natural language to engage with code is a huge plus. What used to be locked up in programming languages that you had to learn, you can now do with simple words.

The promise of accelerating data cleaning is a bright spot, especially, in my eyes, for digital humanities researchers. In their article “Against Cleaning” in Debates in the Digital Humanities 2019, Katie Rawson and Trevor Muñoz suggest that the amorphous work of “cleaning” data takes up to 80% of data workers’ time in research. 80%! Even cutting that down by 10 or 20 percentage points would have a massive impact on DH researchers.

Back to The Pudding’s AI intern project: Another area where ChatGPT can deliver impact is exploratory analysis. When first forming or testing hypotheses, the team found that ChatGPT could rapidly summarize data — including visualizing the data with a range of charts — in ways that would otherwise require a lot of programming. 

These two promising use cases (data cleaning and exploratory analysis) point to a broader theme here: that ChatGPT can be very helpful for automating or accelerating rote data work — and very unhelpful when it comes to the more creative work, such as ideation and storyboarding. 

At the end of the project, the team decided against publishing the research, but they plan on releasing the cleaned dataset — a nice service for other researchers hoping to look at the same topic.

Throughout the process, The Pudding researchers were continually confronted with the same question: At what point is it more productive to just do the work, rather than keep prompting ChatGPT and QAing its results? There isn’t an easy answer here — it depends on the nuances of the task at hand.

P.S. Because the subject was using ChatGPT in research, I tried using ChatGPT to help me outline this blog post using the notes I’d taken during the workshop. But the output wasn’t particularly helpful, and I decided it’d be faster to just write the damn thing myself. However, if this were for a longer research article, and if my notes were more extensive/thorough, I could see this being a potential time-saver for scoping out my writing.

I thought it’d be interesting to drop several texts by the same author into Voyant to see if I could learn anything about how that author uses language across their oeuvre. For my sample, I used a few of Fyodor Dostoevsky’s novels — specifically, The Brothers Karamazov, Crime and Punishment, Notes from the Underground, and The Idiot — because I’m somewhat familiar with them and because each text was available via Project Gutenberg. I copy/pasted the plain-text files (excluding the introductory and concluding text added by Project Gutenberg) into the same Voyant window.

The results loaded, and most seemed pretty meaningless. The most common words were, unsurprisingly, “the,” “and,” “to,” “of,” and “I”. I had to dig deeper to start finding some bread crumbs of potential insight. After scrolling through the list of most common terms for a little while, I started to see more substantial terms: “prince,” “old,” “cried,” “heart,” and so on. These all felt very Dostoevskyian. 

I explored a handful of these. “Old,” for instance, is used over 1,000 times across this corpus. From the “contexts” window, I noticed “old” was used in a variety of ways — ”three-year-old,” “old-fashioned,” “an old friend,” and so on. 

Voyant pointed me to a passage — where Alexei (Alyosha) Karamazov visits his father — that is densely populated with the word “old”: 

“Though there was a dining‐room in the house, the table was laid as usual in the drawing‐room, which was the largest room, and furnished with old‐fashioned ostentation. The furniture was white and very old, upholstered in old, red, silky material. In the spaces between the windows there were mirrors in elaborate white and gilt frames, of old‐fashioned carving. On the walls, covered with white paper, which was torn in many places, there hung two large portraits—one of some prince who had been governor of the district thirty years before, and the other of some bishop, also long since dead.”

Here, “old” is repeated so often it practically becomes a parody. But by paying attention to “old,” I also started to see other terms in the passage that strengthen its effect — e.g., “thirty years before” and “long since dead” — which add to the portrait of the father’s home (and the father himself) as stuffy and antiquated. 

Reading this passage, I also became interested in the term “ostentation.” It only appears 3 times in the corpus, all in The Brothers Karamazov. The second and third instances appear close to each other — in the sermons of Father Zosima, who criticizes how people “live only for mutual envy, for luxury and ostentation” and then again for their “gluttony, fornication, ostentation, boasting and envious rivalry of one with the other.” These instances more directly link ostentation with immorality and sin. Now, revisiting the earlier passage, we can see traces of this kind of judgment in the description of Fyodor Pavlovich’s (the father’s) house.

Investigating both of these words (“old” and “ostentation”) drew me into the texts in ways I didn’t expect. Voyant allowed me to shuttle between distant and close perspectives as I engaged with the corpus, seeing linkages between words that appeared many pages or even books apart. 

It also expanded how I thought about Dostoevsky’s writing, helping me see beyond my existing assumptions and notions. To this end, I think one of the key benefits of text mining tools like Voyant is its ability to lead us down paths of inquiry that we might not have considered otherwise. 

In this sense, it reminds me of Richard Jean So and Edwin Roland’s article “Race and Distant Reading.” When So and Roland examine where their model fails to accurately classify an author’s race, they use James Baldwin’s Giovanni’s Room (classified by the model as being written by a white author) as a case study. Among their findings, they realize the word “appalled,” used just once in the novel, has an “outsized influence” on the language model — and this isn’t random. Despite its scarcity, the word “appalled” serves a unique function in Baldwin’s novel, which they would never have noticed if it weren’t for the model (So and Roland 71).

Works cited

So, Richard Jean and Edwin Roland. 2020. “Race and Distant Reading.” PMLA 135.1: 59–73.

Reading “Hybrid Scholarly Publishing Models in a Digital Age” — published by Professors Michael and Karlin, alongside Matthew K. Gold — gave me a newfound appreciation for the ways the Manifold platform has facilitated a more collaborative and student-led style of learning within the classroom this semester.

Coincidentally, this past week, I bought my first-ever e-book — an experience that drew me into a beguiling labyrinth of market forces and motivations. I’ll describe these below, as a way of unpacking some of “the ecological, economic, and logistical traces that books leave as they move through the processes of creation, production, and distribution” (Michael et al. 280). It’s important to note, however, that this e-book was for sale, and that its sales likely pay out in the form of royalties to the author, as opposed to an open access model like much of what appears on Manifold.

A co-worker recently recommended R. F. Kuang’s Babel, or the Necessity of Violence, a work of semi-historical fantasy fiction. I hadn’t yet taken any steps to get my hands on it — I wasn’t sure I wanted to buy it, and I figured there’d be a long line of holds in the public library system — when this co-worker told me about an Amazon Prime Day sale. The book was being sold in e-book form for just $2.99.* Who doesn’t love a good deal? 

I didn’t have a Kindle reader, but I learned there’s a Kindle app for Apple devices, so I downloaded it. I looked up the book, but the app told me the book wasn’t available on my device — even though Amazon’s website said I could use the Kindle app to read it. It was then that another co-worker let me in on a secret: If you buy the e-book on your desktop, it’ll automatically appear on the mobile app. As it turns out, the confusing app experience was intentional: this way, Amazon is able to avoid having to pay app store owners like Apple and Google a cut for e-book purchases, since the transactions are taking place outside the app.

Anyways, I finally had my book. Amazon wasted no time in trying to upsell me: If I paid another $12, I could get the audiobook version, so that I could seamlessly switch between formats without losing my place. This was a strong selling point and speaks to how book publishers and 3rd-party platforms like Amazon can potentially gain from a hybrid / multimodal publishing model by creating sequences of purchases tied to increasing levels of textual value.

As the authors of the “Hybrid Publishing” article describe, “The relative value a text or digital file holds is bound in cultural, legal, and moral contexts” (Michael et al. 288). In this case, Amazon is able to use not just print and digital mediums, but also audio, as individual levers to fuel sales. The price differentials between mediums imply a hierarchy of values: in this case, the e-book is positioned as an entry point, but a diminished one. For the fuller experience, you should really buy the audiobook.

I found this morass of market dynamics fascinating for the ways they simultaneously facilitated and obstructed my journey into the text. Ultimately, I ended up reading (and enjoying) a book I might not have read otherwise, though not without some struggle along the way.

*The book is ~560 pages long, which, at the price of $2.99, equates to just half a cent per page.

Works cited

Michael, Krystyna, Jojo Karlin, and Matthew K. Gold, 2022. “Hybrid Scholarly Publishing Models in a Digital Age.” New Directions in Print Cultures Studies: Archives, Materiality, and Modern American Culture. Bloomsbury Press.

While reading Jennifer Guiliano and Carolyn Heitman’s article, “Difficult Heritage and the Complexities of Indigenous Data,” I was struck by how the open access movement — an ostensibly well-intended shift toward a more democratic approach to using/sharing data — can pose a threat to historically marginalized communities by perpetuating colonial practices of producing and preserving knowledge.

In Catherine D’Ignazio and Lauren Klein’s book Data Feminism, the authors describe a map that I believe exemplifies what Guiliano and Heitman frame as an “Indigenous-centric approach” to (re)constructing cultural memory. The map was researched and designed by Dr. Margaret Pearce, in collaboration with First Nations, Métis, and Inuit communities across Canada, and is titled “Coming Home to Indigenous Place Names in Canada.” The boundaries of the mapped landmass mirror that of the Anglo-Western conception of Canada, but the place names on the map reflect those used historically and contemporaneously by the Indigenous communities that inhabit(ed) those lands.

At multiple points, the map protects or obscures knowledge, as a means of preserving the privilege of knowledge for a select few, rather than as a right for all. The exact locations of place names are not given, so as to limit the ability of outsiders to gain access. (This is achieved through the map’s massive scale of 1:5,000,000, which allows for a certain level of descriptive ambiguity.) Furthermore, in some cases, place names are not translated into English. 

Additionally, as a measure to prevent decontextualization and misuse, the map may not be reproduced at all unless the relevant communities are consulted. As Klein and D’Ignazio describe in Data Feminism, “Each time the map is reproduced…Pearce writes to the communities to whom the names belong, explains the proposed context of the names, and requests permission for the names to be reproduced in that context.” This allows for and maintains the agency and oversight of the communities who have consented to the use of their place names — and disrupts the tendency for open-access platforms to allow “for objects to be divorced from their conditions of production and contexts of interpretation for all forms of reuse” (Guiliano 18).

Link to map here.

I chose to map something not traditionally considered mappable: the creative writing process. 

In 2021, I wrote one haiku (or multiple) every day for 100 days. My goal was simply to write more regularly, and the condensed form of the haiku helped lower the barrier to entry. 

A few weeks into the process, I started logging the daily efforts in Day One, a journaling app that adds metadata — like time, location, and weather conditions — to journal entries. This is what allowed me to eventually map the results. 

I jotted down the haikus as soon as they came to me, which often occurred at my apartment in Brooklyn, as well as on walks with my dog and even trips to visit family in Vermont, Maine, and elsewhere in the US. I chose to use a heatmap approach when visualizing the haikus to show how they tended to cluster around specific locations.   

I used Carto to visualize the distribution of haikus and give viewers a way to read each haiku while being able to see where it was produced. You can check out the result here. 

The Carto platform is closed-source, but it offers a free 14-day trial (which I used to create this map), as well as 2 years of free use for students through GitHub. (However, the approval process for the latter is cumbersome — I’m still waiting to hear back from GitHub about this.) I don’t know whether my map will disappear from Carto after the free trial ends.

Compared to tools like QGIS, Carto has a much easier-to-use interface, which makes it more accessible (despite the closed-source platform) for a newbie like me to explore and represent geographic data. It also takes just 2 clicks to create a publicly shareable link to the map. 

Overall, the Carto interface allowed me to move quickly from local Excel data — full of latitude and longitude coordinates that don’t mean much without being plotted — to an intuitive and public map. Beyond representing the data, the map opens up new ways of thinking about the creative process — after plotting the poems, it becomes easier to consider how factors like location and routine (e.g., walking the dog) might influence creative output.

Unfortunately, Carto didn’t preserve the line breaks that I had originally placed in the Excel file to divide each of the three lines in the haikus. This undermines the presentation of the haiku form. Given more time, I would research ways to add the line breaks back in by using Carto’s HTML back-end or else by modifying the original CSV file.

Several of this week’s readings emphasize how the digital humanities must continually assert why they matter. On the one hand, this serves a practical purpose: doing so helps ensure DH’s survival against a backdrop of academic austerity cuts and the dominance of for-profit tech giants in the digital landscape. On the other, the imperative helps align DH projects with the field’s ideological aims — to participate, as Matthew K. Gold and Lauren F. Klein write, in a “larger technically and historically informed resistance” (“A DH That Matters”).

“Torn Apart / Separados” is a prime example of how the digital humanities can spur action and intervene in sociopolitical issues in real time. This project visualizes a series of relationships and networks that make up the US’ 2018 Zero Tolerance Policy, which cut off asylum seekers by prohibiting and prosecuting illegal border crossings. In the project contributors’ words, this is a “rapidly deployed critical data & visualization intervention” (the project’s welcome page), suggesting it is neither a retroactive analysis, nor a neutral or removed take.

With this project, the researchers aimed to demonstrate that it is possible “to respond quickly, yet carefully, in times of crisis” (the “Textures” tab) — an explicit nod to how DH can prove its value. “Torn Apart / Separados” plays out what Jacqueline Wernimont and Elizabeth Losh, citing scholar Jessica Marie Johnson, describe in their introduction to Bodies of Information: “As Johnson notes, this [the digital humanities] is not simply ‘academic’; the work and communities of Black, Native, Latinx, queer, trans, and intersectional digital scholars have ‘literally saved lives’” (“Introduction”). 

The project seamlessly moves back and forth between the academic and the public, between belief and action. For instance, its “Allies” tab — which aggregates and maps vetted organizations that are on-the-ground and able to assist — makes it abundantly clear that it brings value not only to academics, but also to the families and communities affected by the crisis.