• The original Democratic Review printing and the American Sunday School Union’s “A Visit to the Celestial City”
• The original Democratic Review printing and the Adventists’ tract edition in the Bible Examiner series

I look forward to the continued development of Juxta’s web service. In particular, I hope they develop a way for scholars to embed the tool in their project sites. I would love for visitors to this website to access my comparison sets directly.

“Mapping the Antebellum Culture of Reprinting”

[slide 1]

Today I want to talk about how mapping using global information systems (GIS) software might help us better understand the dynamic world of print culture in the United States before the Civil War—what Meredith McGill calls “the antebellum culture of reprinting.”

[slide 2]

So let’s start with tedium and confusion. In his introductory essay to the Placing History volume, Richard White claims, “Relationships that jump out when presented in a spatial format such as a map tend to clog a narrative, choking its arteries, until—even if the narrative does not expire—the reader, overwhelmed by detail, is ready to die of tedium and confusion.” Let me illustrate White’s distinction. and In the past few years I’ve been working with Nathaniel Hawthorne’s short story, “The Celestial Railroad.” This satirical retelling of Pilgrim’s Progress was widely reprinted in the years following its first appearance in the May 1843 edition of the Democratic Review. I’ve uncovered more than 50 reprintings of this text—the majority unauthorized and printed by denominational newspapers and magazines—and several hundred paratexts—articles, sermons, stories, tracts, &c. reference the story, its scenes, or its characters.

[slide 3]

Here’s what the reprinting history of “The Celestial Railroad” looks like in bibliographic form for the years 1843-1861. This is itself, of course, a visualization of the story’s textual history. The details of this history—the way readers encountered, responded to, and even modified the tale—were the basis of my paper at last year’s MLA, which I’d be happy to share with anyone interested. Today, however, I want to zoom out a bit and think about how spatial data might help situate histories of textual transmission and reception within larger social, political, or even technological contexts.

[slide 4]

Here I’ve georeferenced, or aligned, Henry Tanner’s 1846 Traveller’s Guide using GIS. The blue triangles represent sites where the story was reprinted; yellow circles represent paratexts. Icons overlap in places where the story was reprinted (or referenced) more than once; GIS can also represent events by size, with large icons representing more events in a single location. Laid out on a map, the story resonates differently than it does in a sequential list or in prose. The reader is immediately struck by the story’s reach, for one, and other spatial questions begin to suggest themselves, particularly if the reader is familiar with the texts of “The Celestial Railroad.”

I will return to this map in a few minutes, but first I want to delve into a few specific questions GIS is helping me explore in relation to Hawthorne’s tale.

[slide 5]

Franco Moretti argues that “[Maps] are a good way to prepare a text for analysis. You choose a unit…find its occurrences, place them in space…you reduce the text to a few elements, and abstract them from the narrative flow, and construct a new, artificial object…with a little luck,” Moretti claims, “these maps will be more than the sum of their parts: they will possess ‘emerging’ qualities, which were not visible at the lower level.” Moretti is discussing maps that illustrate narratives—mapping the places visited within a novel, for instance—

[slide 6]

but I’m interested in how isolating particular units of spatial data can help us understand the journey of a text itself—its sites of reprinting, its potential readers, and the means of its distribution. Like Moretti’s intratextual maps, these extratextual maps can be “more than the sum of their parts” and enrich bibliographical or textual investigations.

Though the output of GIS research usually takes the form of a map, the end of GIS is not, strictly speaking, a pretty visual. As a friend and geography professor told me when I started learning GIS, “If all you need is a pretty map to illustrate your article or book, hire a geography student for a few hours.” Instead, Ian Gregory describes GIS as “a kind of database management system” that allows researchers to bring different kinds of data—events, topographical features, census data, etc.—into conversation. In my first example I had two kinds of data: my database of “Celestial Railroad” reprints and references and a historical map of the United States. These next examples bring three kinds of data into conversation: my database,

[slide 7]

the Newberry Library’s Atlas of Historical County Boundaries,

[slide 8]

and very detailed county-level census records from 1840 and 1850.

[slide 9]

By joining these latter two data sets, I was able to create an (imperfect but) historically-accurate map of the United States in both 1840 and 1850 that I can use to represent various aspects of the nation in those years and illuminate the world through which “The Celestial Railroad” circulated.

I’m going to walk quickly through a several maps now; in many ways I want to demonstrate the experimental—even playful—qualities of this research. There are lots of data points in these census reports, and I’ve been experimenting with different ways of representing them, looking for “patterns in data”—as Ian Gregory says—that might be historically or textually interesting.

Ad lib through slides 10-17

[slides 10-11 demonstrate potential sites for further research]

[slides 12-16 show changing face of American Christianity]

[slide 17]

While these visualizations might tell us something about the shape of American Protestantism in 1850, however, they don’t tell us all that much about “The Celestial Railroad” in the ecclesiastical press.

[slide 18]

If we use GIS to compare some of these data, however, that changes. Here I’ve normalized my data on Baptist churches using my data on Methodist churches—in short, the darker counties are those where Baptists predominate, and the lighter counties are those with more denominational mixture.

[slide 19]

Here’s the Methodist data normalized by the Baptist data. As you can see, “The Celestial Railroad” tends to show up in regions without a dominant faith. You can run these comparisons with nearly all the denominations census takers tallied in 1850 and get similar results. This evidence reinforces insights gleaned looking at the introductions that religious editors wrote to “The Celestial Railroad”—the text was frequently recommended to “innovators” who had “strayed from the good old way.” Religious editors tended see the tale’s satire as directed toward those outside their own denomination, and they printed the story for missionary purposes. This missionary role into which religious readers enlisted “The Celestial Railroad,” makes even more sense when one sees that the story was reprinted in places of religious heterogeneity.

[slide 20]

Because GIS is, ultimately, a tool for linking different kinds of data, the map itself can be only one step in a longer research process. The map is a vehicle for correlating data that would otherwise be hard to compare.

[slide 21]

Here I’ve created a 10 mile buffer around each site of “The Celestial Railroad’s” reprinting history. I chose 10 miles rather arbitrarily—it seemed like a reasonable travel distance—but the precise distance could be modified easily. I then joined these “circles of influence” with the county-level census data below them in order to get a picture of the population within 10 miles of a “Celestial Railroad” event.

[slide 22]

We can estimate that within 10 miles of “The Celestial Railroad” there were 4.3 million people; nearly 800,000 families; twice as many Sunday School libraries as public libraries;

[slide 23]

more than 1,000 Baptist churches; nearly 1500 Methodist churches; and so on. There are of course problems with such estimates. Large cities such as New York, Boston, and Philadelphia, where the story was printed many times, only count once. The data can’t tell us precisely who among those 4.3 million people actually read the story. But these numbers can give us a picture of the markets in which the story circulated.

[slide 24]

When added to the insights we can glean from the texts and paratexts themselves, GIS can flesh out our understanding of both where a given text appeared and, it turns out, how it got there.

[slide 25]

Let’s look again at the georeferenced 1846 traveller’s map. We can see that the history of this text seems to fall directly along the railroad, a theory we can explore more precisely using GIS.

[slide 26]

As part of his Railroads and the Making of Modern America project (well worth checking out), William Thomas mapped the U.S. railroad network at various points in history, and he makes much of his GIS data available for free on the project’s site. Overlaying Thomas’ data about the 1861 railroad network

[slides 27—>28]

with my textual data

[slides 29—>30]

we can see a close correspondence. Indeed, every single reprinting or reference to “The Celestial Railroad” was printed in a city or town on the railroad network. Many of these towns sit, literally, at the end of a line. The same holds true if we look further west.

[slide 31]

Perhaps it’s not shocking to hear that a nineteenth-century story likely spread to new readers by rail. Nevertheless, it’s a relationship I would not have noticed (I don’t think) had I not juxtaposed the Tanner map with the aggregate print history of “The Celestial Railroad.”

[slide 32]

Moretti, you will recall, calls this serendipity an “‘emerging’ qualit[y]” of maps. The relationship between press and rail in 1850, which seems so obvious from a God’s eye view, was “not visible at the lower level” of the texts themselves.

[slide 33]

I particularly love the irony of this map. In “The Celestial Railroad,” the titular railroad is Hawthorne’s central symbol of unthoughtful, triumphalist modernity. The story’s pilgrims, who board the train for an easier journey to the Celestial City and mock the “old fashioned pilgrims” they see walking Bunyan’s route, are instead deceived and bound, in the story’s final paragraph, for hell. That this story likely traveled via the U.S. railroad network is a fun irony to contemplate (and yes, I’m cognizant of the double irony of me using twenty-first century technology to uncover this  nineteenth-century technological irony).

[slide 34]

So where might I go from here? First, I would like to further explore ways to integrate my textual and geospatial research.

[slide 35]

I’ve been using Juxta to coallate witness of “The Celestial Railroad,” uncovering the changes—sometimes quite substantial—that editors made to Hawthorne’s story as it circulated. One major goal of my project—which should come to fruition later this year—is to publish an electronic, variorum edition of “The Celestial Railroad” that will allow readers to compare witnesses. I hope to map the paths of each “Celestial Railroad” variant, so that readers can track the story textually or spatially.

[slide 36]

My next goal is to bring more textual histories into GIS for spatial comparison. Here I’ve mapped Edgar Allan Poe’s “The Raven.” Scholars have long been familiar with the wide reprinting history of Poe’s poem, which appeared two years after “The Celestial Railroad,” in 1845.

[slide 37]

We can see that the texts shared many sites of reprinting or reference, and we can also spot some significant differences: Hawthorne was more popular in Western New York—a hotbed of religious fervor during the time—and Poe was more popular in the South. If we look at Poe and the railroads,

[slide 38]

we can see that “The Raven,” like “The Celestial Railroad,” likely spread, at least in part, by rail. Comparing these two texts brings us back to a few ideas I suggested earlier.

[slide 39]

Poe’s poem was reprinted almost exactly the same number of times as Hawthorne’s story, and both follow similar geographic patterns, though the two were printed in only two of the same publications. Those places where Poe appears but Hawthorne does not might suggest differences in the way the works were received. These divergences might also suggest cities whose periodicals should be more carefully researched. Perhaps there are printings of Hawthorne in Charleston, Augusta, Mobile, and New Orleans that I’ve yet to uncover.

[slide 40]

The Southern writer Moncure Conway certainly remembered “The Celestial Railroad” as a popular story in the South, so perhaps Poe’s textual history could help guide my archival research into Hawthorne.

When we map multiple textual histories, GIS allows us to make direct comparisons between them.

[slide 41]

We can, for instance, compare the “circles of influence” of both stories. Though scholars know the print history of “The Raven” much better than that of “The Celestial Railroad,” we can see—with all the caveats I offered earlier—that the latter had the potential to reach 400,000 more Americans during this time period. Perhaps we know  the history of “The Raven” because it appeared primarily in literary magazines and secular newspapers, while “The Celestial Railroad” appear primarily in religious publications. We can see, however, that both texts had a similarly wide reach across the American landscape.

[slide 42 – timelapse video]

To really flesh out our picture of print culture in the 1840s and 50s, we need even more textual histories to compare. Right now I’m working with colleagues in the ARTFL project at the University of Chicago, who have developed a tool for automatically discovering reused snippets of text in digital archives. We plan to use this tool with nineteenth-century periodicals archives to rapidly discover histories of reprinting like that of “The Celestial Railroad” or “The Raven.” If we can compare not two print histories, but two hundred, we can really begin to define the network of nineteenth-century print culture. We can see what kinds of texts most frequently moved through the network (or moved the farthest). We can see which publications shaped the network—whose articles were most frequently reprinted. We can ask whether religious and secular texts circulated differently—where did religious and secular print networks converge? Where did they diverge? And we can use census and other data to evaluate how the print culture network aligned with other political and social forces.

I’m just beginning my GIS work, and I welcome questions and comments that will help me refine the questions I’m asking with it and develop new ones. I am excited about the potential for GIS in investigating antebellum print culture. Geospatial research can complement close study of texts and help scholars grapple with the big, sometimes unwieldy history of the antebellum print market—a market that was for its time as expansive, exciting, unchecked, and unruly as the internet is today. I look forward to delving deeper into the opportunities of GIS research, and I welcome your ideas about how I might proceed.

So I was thrilled this past week to read William G. Thomas’ talk, “What We Think We Will Build and What We Build in Digital Humanities,” from this year’s Nebraska Digital Workshop, and to learn from the talk about Thomas’ project, Railroads and the Making of Modern America. The project itself is fascinating, and I immediately wondered if some of their data might help me investigate the circulation of “The Celestial Railroad.” I’ve suspected for awhile that Hawthorne’s tale—which satirizes uncritical modernizing through the central image of a railroad—ironically may have spread around the country through the railroad system.

The historical map that I georeferenced at DHSI seemed to bear this conclusion out. On the Railroads and the Making of Modern America site, however, I was able to download a KML that more precisely charts the 1861 railroad system in America. I used ArcGIS to convert this KML to a shapefile, and then imported that shapefile into my “Celestial Railroad” map. The results were exciting:

The blue circles represent reprintings of the story; the yellow triangles represent paratexts. Larger icons mark places with multiple reprints or paratexts.

With only one exception—Louisville, Kentucky, which sits beside the Ohio River—the entire textual history I’ve so far uncovered for “The Celestial Railroad” seems to unfold along the nineteenth-century railroad network.

Of course, these results point to more work that needs to be done. The “Railroads” project claims they will soon be releasing their data for the American railroad system in 1840, 1845, 1850, and 1870. With that data, I could more finely tune my own investigation—correlate reprintings and paratexts from each time period with the exact railroad system that might have ferried them. That would allow me to see whether Hawthorne’s tale grew with the railroads. If it did—well, that would be interesting to say the least.

Which leads me to the question, “What’s next?” I’ve been thinking quite a bit about this over the past year, and have explored a few possible directions for this research. Exploring the extensive reprinting history I’ve uncovered for this one story—a non-canonical story by a hyper-canonical author—has convinced me that similar textual narratives must exist for many stories and poems—both by canonical and by forgotten authors.

Once I publish my Hawthorne research this summer, I want to start working on something much bigger: a reprint discovery engine for nineteenth-century periodicals archives. I imagine a tool not unlike the Google Ngram Viewer, but focused on textual reprint and reference. This project would likely start by investigating a database like the Library of Congress’ “Chronicling America” collection, which is open and includes “an extensive application programming interface (API) which you can use to explore all of our data in many ways.”

I imagine the reprint discovery tool developing in two stages:

1. In its first stage, the tool likely would require base texts for each inquiry. Users would enter, say, the text of Poe’s “Purloined Letter” and the tool would automatically break the short story into n-grams—sequences of words or letters. Then, the tool would automatically query a periodical archive for each n-gram sequence. Why so many queries? As I found with the Hawthorne project, simple title searches are insufficient, as reprints were often untitled or retitled by newspaper and magazine editors. I addition, title searches won’t return quotations from or references to the base text in other kinds of articles: such as the sermons or religious articles I found that quoted just a line or two from “The Celestial Railroad.” The tool should allow readers to tweak the length of the n-gram sequences on the fly—in my OS X-bound imagination, I see a slider—so that an inquiry could be broadened or narrowed based on the results returned. Such a tool would allow users to discover not only reprints of their chosen text, but also the paratexts essential to understanding the reception history of the story or poem.
2. In the tool’s second stage, I would hope to automate the first part of the reprint discovery process: the discovery of base texts. The problem with the tool I’ve outlined in stage 1 is that it would likely only be used for texts scholars already find interesting—stories or poems that scholars suspect are worth searching periodicals archives for, because they have some sense of an existing history of widespread reprinting and/or reference. If the tool itself could dig into the archive in search of base texts, however, then we might discover texts that were widely reprinted and referenced but have since fallen out of our cultural memory. Such a tool could generate significant new scholarship, as important new texts and authors resurfaced and demanded further study. How might this work technically? I’m not certain. Perhaps the tool would crawl through the entire archive database, breaking the archive itself into n-grams and then looking for matches. I’ll need a programmer to tell me whether that’s in the realm of possibilities, or whether there’s another approach that would be more fruitful.

This all leads me to three questions for the digital humanities community:

1. First, am I missing an existing tool that would enable this sort of discovery? I don’t want to spend time figuring out how to reinvent a tool that already exists (or nearly exists, and merely wants tweaking).
2. Second, does the tool I’ve described sound useful and compelling? Does this meet a need for scholars in literary history and/or periodical studies? If you were reviewing this grant proposal, what would you say about the tool’s “potential contributions to the field?”
3. Third, would you be interested in collaborating to build such a tool? I will, of course, list this project on DHCommons, but if you’re reading this and thinking either “this idea perfectly dovetails with my own research project” or “I could write an algorithm to do that in an afternoon,” please send me an email!

There are, of course, many more possibilities growing from such a tool. As I mentioned in my last post, thinking about nineteenth-century reprinting culture immediately leads to geospatial questions. Perhaps this reprint discovery tool could map search results, so that users could navigate results geographically. Indeed, such a tool might help untangle the complicated web of nineteenth-century reprinting culture, visualizing relationships between publications that frequently borrowed from one another and suggesting relationships scholars had not previously spotted. Perhaps I will speculate on geospatial possibilities in another post. For now, if you have ideas or suggestions for a C19 reprint discovery engine, please share them in the comments.

Keeping with this spirit, I attended the Geographical Information Systems (GIS) class at DHSI with a genuine question in mind: “do I actually need GIS for my research?” My Celestial Railroad project does include a geographic component. I’m tracing the spread of one Hawthorne story through the United States in the 1840s and 50s, tracking editorial changes made to various witnesses, as well as the larger cultural response to the story found in introductions, editorials, and references to the text. I’ve already mapped the story’s spread using David Rumsey’s historical maps in Google Earth. When I described my project to a friend in a geography department, he wondered why I needed to spend a week learning GIS at all. He pointed out that Google Earth was sufficient for creating most visualizations. If I wasn’t planning to use ArcGIS’s more advanced analytical tools—if my research question didn’t include issues such as topography, population density, or other census data, for instance—then learning GIS might be a waste of time. Why bring a jack hammer to a project when a hammer will do the job?

We spent the first three days of DHSI working through lessons and practicals that taught us the basics of the ArcGIS software. On the fourth day of DHSI, I started working on my own project with Henry S. Tanner’s 1846 “traveller’s map” of the United States, which is available through the wonderful, freely available David Rumsey Map Collection. I wanted to use Tanner’s map because it includes “the roads, canals, and railroads of the United States.” Though “The Celestial Railroad” satirizes antebellum American optimism in technology—including the railroad in its title—I suspected that the story owed its popularity to the transportation system of the 1840s and 50s. That’s not a surprising hunch, perhaps, but I hoped ArcGIS might help me verify it.

I spent a while georeferencing the Tanner map: aligning major points on the historical map with those same points on a modern basemap. This process can distort the historical map, depending on how precise it is by modern standards. You can see this distortion on the edges of the map below. Once I finished this process, I added my spreadsheet listing nineteenth-century reprintings, references, and reinterpretations of “The Celestial Railroad” to the map. In a few steps, I was able to separately map reprints, references, and reinterpretations on the map, using larger markers to indicate cities where multiple witnesses appeared. I must say: when those markers first appeared on the Tanner map, falling almost exclusively along his road and railroad network, I felt quite a thrill.

Click image for a high-resolution version.

That’s as far as I got at DHSI. Though fun, did I make use of ArcGIS’s full analytical powers? No, not quite. As I reflected on the week’s exercises and my map, however, I did think of some possibilities. I sent out a tweet wishing for datasets of nineteenth-century U.S. counties and nineteenth-century population data. Bethany Nowviskie responded with a link to the Newberry Library’s Atlas of Historical County Boundaries, which is freely available. She also sent information about historical census data, but that data is unfortunately not available at my institution.

If I can get hold of that data, though, I think I could do more. For instance, I could analyze the population that lived within certain distances of publication sites. I could determine—within a generous margin of error—how many Americans lived within 5, 10, or 20 miles of a “Celestial Railroad” publication. How many Americans had local access to Hawthorne’s story?

To return to my original question: do I really need ArcGIS for my work? Maybe. I see potential geospatial questions that would require the analytical power of GIS. So I’ll keep tinkering, and I’ll keep reporting on that tinkering here.

This week, however, I discovered that a number of the maps in the collection can be downloaded as a .kmz file to be viewed in Google Earth. Importing this file into Google Earth allows you to lay maps from the Rumsey collection over the Google Earth globe. These maps are georectified, meaning that the features on the maps have been lined up with their precise places on the more precise modern globe.

After playing with these maps for a few minutes, I quickly decided to overlay an 1839 map from the Rumsey collection with a .kmz I created in Google Maps of the towns in which “The Celestial Railroad” was republished between 1843 and 1860. Within minutes I had this visualization—a “historical” map of the story’s reprintings—

I made larger the markers for those cities where the story was more frequently reprinted. So New York and Philadelphia are the largest, as the story ran many, many times in both cities. That resizing was, for this quick project, entirely subjective—I hand-sized each city’s pin.

This isn’t, of course, the best visualization one could create of this, but I was impressed that I could put something that looks this good together in only a few minutes. I plan to keep experimenting with the Rumsey maps in Google Earth as I think through how best to tell the geospatial aspects of this story about Hawthorne and 19th Century publishing.

In addition to importing UTF-8 encoded plain text files, this new version of Juxta now supports direct import of XML source files in any well-formed schema, include TEI p4 and p5. No more preparing specialized versions of your witnesses for import into Juxta. Just import them and instantly start collating and learning things about your texts! You can configure how Juxta parses the tags it encounters. It can either include them in the reading copy, exclude them, or collate the tag type. For example if <b> changes to <i> for the same word across different witnesses, Juxta can help you detect this move. Complete details are in the online documentation on this website.

In other words, I can now compare TEI versions of “The Celestial Railroad.” I’ve been working with my research assistant to clean up the TEI on my most important versions so that I can build some comparison sets in the new version of Juxta. Exciting times!

In my talk I discuss not only the technology I’ve used in building this edition, but also the literary and historical discoveries the project has helped me make about Hawthorne, his audience, and his career. The Scholars’ Lab has posted the talk as a podcast (clicking the link will open iTunes). My talk starts at 28:08, but please listen to Alex’s talk first. The Q&A at the end addresses both talks.