Robert Southey, a member of "the Lake School" of poetry—really, the first "avant garde" in the history of English literature—was as prolific a letter writer as he was of poetry and prose during the Romantic era in England, roughly 1780-1830: "to write to a dear friend," Southey says, "is to me like escaping from prison" (Southey 2009, I.141). A massive digital scholarly edition is underway, The Collected Letters of Robert Southey. It is divided into eight parts: I:1791-1797, II:1798-1803, III:1804-09, IV:1810-15, V:1816-21, VI:1822-27, VII:1828-33, VIII:1833-39. Currently Parts I and II have been completely edited and made available to the public. These letters capture a set of intellectual, amicable, and financial relationships established while Southey lived in and travelled away from his home base of Bristol in "the West Country" before he actually moved to the Lake District. Because each person and place name in the letters was encoded using TEI P5, we have been able to create a data set that indicates who is mentioned in the letters and to whom each letter is addressed. That data set was fed into a Directed and Undirected Graph to be visualised, which is available to view and manipulate online here: http://idhmc.tamu.edu/relate (a quick caveat: if you see nothing in your browser, please try using a different one. It works beautifully in Chrome 22, Safari 5.1.7, Firefox 9.0.1, and presumably anything higher).

Experts are now busy graphing social networks that have been created by people via e-mail or Facebook. Our tool called "Relate," built using the Java Universal Network/Graph Framework (as described below), allows us to graph early social networks created by letters. We used Relate to graph all of Southey’s letters. In the dynamic, malleable graph of these letters, uni-directional and bi-directional edges, representing relationships, connect nodes that represent the people to and about whom Southey wrote. Relate visualises the number and kinds of connections that were established among people by Southey’s letters. These 877 letters spanning the years 1791-1803 graph a social network of the era. The Relate Graph tells us a lot about Southey’s relationships as well as gestures toward what other kinds of data it might be useful to have. But it also prompted me to ask two questions that will be discussed in this essay: first, what are the protocols for reading graphs? The second question has two parts: a) what kinds of new knowledge can be generated by visualising textual data? b) If visualising information helps experts (literary experts in this case) perform deeper research, then what precisely is the new relationship between experts and tools once tools become, like the Relate dynamic graphing tool, "smart tools?"

In the process, I will argue here that visualising Southey’s letters suggests a new hypothesis about high Romantic literature, knowledge unavailable to the Romantics, from whose self-understanding we have perhaps taken too much instruction. These letters clearly demonstrate something "we"—i.e., disciplinary experts in Romanticism—have always known: that the friendships surrounding Southey, Coleridge, and Wordsworth constitute a "network of ambitious young writers" in the 1790s-1810s (Pratt 2000a, 317). But information visualisation leads us also to hypothesise that high romanticism or the publishing history of Romantic-era writers of the Lake School, may in fact be a "network effect." The network effect has been defined by economists as increasing the value of something through wide adoption of it. Facebook is an example, but so is the telephone or e-mail, two services that are only valuable if other people also have telephones and can get e-mail. Our capacity for seeing Romanticism as a network effect rather than the achievement of specific individuals is augmented not simply by graphs but by the disciplines that have grown up around reading them.

For those not yet familiar with the field of Information Visualisation, it "has emerged as a new research discipline" beginning in the 1980s (Keim et. al. 1998, 160). Despite an outpouring of publications in the field between the first and second editions of Chaomei Chen’s Information Visualization (1999, 2006), Ben Shneiderman called it in 2006 a "still emerging academic field" (qtd. in Chen 2006, xi, vii). Colin Ware defines Information Visualisation as "the use of interactive visual representations of abstract data to amplify cognition" (2004, xvii). Before discussing how precisely visual representations might do such a thing, I want to address the bugbear in this statement for literary scholars, viz. "abstract data."

Abstraction in general gives humanists some pause, especially the theoretically inclined who worry about universalising generalisations that hide power plays beneath the surface of abstraction: too often the "Man" starring in Enlightenment discourse is not humanity in general but white, male, Western, and even aristocratic. Normalising, according to Him, as a standard is one way of imposing His form of life upon hapless others who differ. But abstraction itself is a cognitive tool susceptible to being deployed in any kind of politics (Eagleton 1997, 1).

Abstraction via quantification is a process that is key to visualising information about or from texts. Qua "scientific reduction," it has been the focus of the "distant reading" debates surrounding new literary criticism that involves using data-mining techniques to analyse literature; techniques such as those used by the Stanford Literary Lab (http://litlab.stanford.edu/) or by Michael Witmore (http://winedarksea.org/), Robin Valenza (Dr. Valenza’s Visualizing English Poetry Project, supported by Mellon funding to the University of Wisconsin, has not yet been released, as of this writing), and Ted Underwood (http://tedunderwood.wordpress.com/). Whether you side with Franco Moretti or Katie Trumpener, however, one has to admit that reduction is a part of the process of thinking that is impossible to avoid. It is by now a truism that, unless the map is equivalent to the country, the map abstracts data from the country that it represents. But more than that, the country can only be thought—can only enter into anyone’s mind—as a reduction of itself. Although the mind may be wider than the sky, no portion at all of any physical landscape can be put inside a person’s head, forced into a synapse, if that’s where thinking takes place.

In his monograph Graphs, Maps, and Trees, where Moretti graphs the world-wide publication history of novels, he points out that his abstractions—dots on the graph—are indeed reductions, but so, he says, are the words "The" and "Novel" in the literary commonplace "The Rise of the Novel." I agree with this—it was a revelation to me—but I also agree with what literary scholar Andrew Stauffer said to me upon seeing one of my graphs: "I know what to do when I see words on page; when I look at this graph, I don’t know what to do, I don’t know how to read it." This candor is incredibly valuable; the mistakes that one can make without knowing how to read data visualisations are tremendous, and I will show some of them while introducing disciplinary instructions for reading visualisations of data. The benefits of rigorous interdisciplinarity, in which the advances made by one discipline are understood and appropriately adapted by another, are valuable both for archiving literary texts and better understanding literary movements.

1. What Are We Seeing?

One is tempted to read, without any training in the field of Information Visualisation, the Google Ngram viewer, showing us the use of the word "presumption" in the graph in Figure 1.

Figure 1: Use of the word "presumption" in a sampling of texts published between 1700 and 1900.

The word is suddenly used very frequently around 1815-20, right around the time that Mary Shelley first published the novel Frankenstein. It is obviously Victor who embodies presumption, presuming that a human can take on the powers of a god to give and take life. It is so obvious, in fact, that the first stage rendition of the novel was titled Presumption. But a major principle of Information Visualisation is that the first thing we will see when we look at a visualisation is "errors" in our data: "With an appropriate visualization," Colin Ware says, in one of the most influential textbooks in the field, "errors and artifacts in the data often jump out at you" (2004, 3). All the jokes about the Google Ngram viewer are by now commonplace; we know that it cannot "see" the long ‘s,’ and so, when we search for "prefumption" as well as "presumption," we get a graph that looks like the one in Figure 2.

Figure 2: "Prefumption" in red; "presumption" in blue.

But rather than calling what we see here "errors" in data, we should see it rather as information about how the data is structured. In this case, we are seeing information; it is not, however, information about the use of the word "presumption." Rather, it is information about the history of typography—and excellent information at that. If one overlays the graphs returned in the Google Ngram viewer for searches of presumption, prefumption, case, cafe, son, fon, curiosity, curiofity, one can see an amazing consistency in pattern (Figure 3).

Figure 3: Overlay of n-gram searches for prefumption/presumption; cafe/case; son;fon.

What we see here is that the long-‘s’ was no longer technically necessary from around 1790, and printers began to use type containing short ‘s’ around 1800; however, they still had a lot of older type left over that they wanted to use. Indeed, if you examine the 1798 edition of Wordsworth and Coleridge’s Lyrical Ballads, you can see a mixture of long- and short-‘s’ (Figure 4).

Figure 4: Mixture of long- and short-‘s’ in 1798 text.

However, by the 1800 edition, the long-‘s’ has been removed (Figure 5).

Figure 5: Long-‘s’ mostly removed in 1800.

Though publishers changed between the first and second editions, the printer (Nathaniel Biggs of Bristol) and the typeface remained the same (these snippets come from "The Female Vagrant," p. 75 of the 1798 editions [Bristol and London] and p. 73 of the 1800 "second edition" published by Longman. Wordsworth and Coleridge n.d.).

Colin Ware’s stricture that visualisations make data "errors" salient is one that we wish to keep in mind, therefore, as we look at the Southey Letters, we want to think first and foremost about how we might be seeing, not information about Southey’s network of friendships but, something about the data itself—not printing history, in the case of these letters that have been transcribed and coded by hand, but something about the structure of the archive of Southey’s letters. Another moral of the Ngram story is that we do want to follow disciplinary procedures for understanding what we see.

The protocols for reading graphs are laid out in graph/network theory. I will adumbrate these principles while illustrating them in the case of the Southey Letters as loaded into and manipulable within the Relate Tool. This tool was built by an undergraduate (at the time), Jon Jekeli, under the direction of Computer Science Professor Gerald Gannod of Miami University. Jon built it using JUNG (http://jung.sourceforge.net/), the Java Universal Network/Graph Framework. It was Dr. Gannod’s idea to map social networks in the letters—an idea then prominent in computer science research and prompting the building of tools such as Protovis, a JavaScript-based toolkit from Stanford. Travis Brown, Assistant Director of Research and Development at the Maryland Institute of Technology, made the tool viewable on the internet, pre-loaded with data from two Romantic Circles editions: The Letters of Robert Bloomfield and The Letters of Robert Southey. I (Laura Mandell) wrote the XSLT transforms to extract names from these TEI-encoded documents. XSLT, or eXtensible Stylesheet Language Transformations, work with XSLT processors to transform XML files into files of other sorts, HTML for viewing on the web, text files, database tables, etc. (see http://www.w3schools.com/xsl/default.asp). Travis Brown is working now to make the tool available on the web for use by others using any data set simply by putting it into a specific XML or plain-text format. Mandell will provide XSLTs to use on any TEI-encoded texts, but named-entity extractors could be used as well. This tool is currently available at http://idhmc.tamu.edu/relate, for use on Bloomfield and Southey, and information about how to load data will be available late 2013.

When one first goes to the Relate Tool, one sees, by default, information about Southey’s letters to his family. Because it reflects the letters ONLY to family members, the data set initially looks very small. After first arriving at the site, I will first switch to show non-family members, as you can see me doing in Figure 6, using the drop-down menu called Options > Show:

Figure 6: Default view of Robert Southey’s letters; names of family members written to and about in letters written between 1791-1803.

After I select "Non-Family Only" instead of "Family Only," I see the result pictured in Figure 7.

Figure 7: Non-family members written to and mentioned in letters written by Robert Southey, 1791-1803.

And it is here that I will begin to explain the protocols of reading graphs and demonstrate how to manipulate the views in order to "see" various kinds of information about Southey’s correspondence network.

In the Semiology of Graphics, Jacques Bertin tells us that reading a graph begins by identifying the invariant or common ground relating all the elements in the graph (1983, 5). Here, what you see are the people who are mentioned by Southey in letters, either through direct address (the letter is written to them) or through being mentioned in the body of the letter. Not every name mentioned in the letters appears, but only those chosen by the editors to include in the biographies of people associated with the edition. A name like "King George" for instance, would be described in a footnote instead of in this list of brief biographies, which really only contains "information about people with whom Southey is connected" (http://www.rc.umd.edu/editions/southey_letters/people.html). So the invariant or common ground linking all these spheres (people named) and lines (network connections) is this: Southey. It is crucial to remember that Southey himself is not pictured here. Thus, one will not see here how many connections Southey has with his correspondents and the people he mentions in letters. That Southey is connected to them all one or more times is presumed (he is "the invariant"). The strength of Southey’s connection to each person in the network IS visible however, in the number of times that any name appears in the oeuvre of letters; that is, a name that appears 113 times will have been someone written to and about a total of 113 times. One cannot tell from looking at the graph the breakdown. The person could have been written to 100 times, and mentioned in letters to others thirteen times. Alternately, the person could have been written to five times and about 108 times. In both cases, he or she was on Southey’s mind 113 times during the time-span that these letters were produced. The actual breakdown can be obtained by looking at the list of Southey’s correspondents online.

The variables, the things that one does see figured in the graph, show the connections among the people to and about whom Southey writes, the network of relationships established by and reflected in the letters. A connection is established between two people when one person is mentioned to another. There are two kinds of variables: spheres and edges. Spheres represent people, and they are bigger and a specific colour depending upon how many connections a person has to other people. Each sphere or person is opposed to other spheres or people in the perimeter of the circle (or later, wherever you drag them), and he or she is connected by an edge to one or more of those opposing spheres (people). The spheres vary in both size and colour, which are in fact correlated, except when colour is changed dynamically, by clicking on a sphere—that is, when you are using the tool, you may want to highlight one network, and when you click on a sphere, it and those connected to it change colour. The larger the size of the sphere, the more relationships that person has with other people in the network. The connections between people (spheres) are represented by straight lines called "edges." The edges are either solid—which means that the connection goes both ways ("bi-directional edges")—or dotted lines, with an arrow on one end ("directed edges"). These directed edges are counter-intuitive to a degree: the arrow points from a person who is addressed in a letter toward the person who is mentioned in it (pointer=person to whom letter is addressed; pointee=person mentioned). I labelled the spheres in the previous image according to number of connections (Figure 8).

Figure 8: Spheres labelled by number of connections.

Small turquoise spheres have one connection, larger gray ones have four to six connections, etc., up to the largest white sphere, which has eighty-two connections. In Southey’s network during the 1793-1802 time period, there are no people who have twelve or sixteen relationships, just by chance.

As mentioned earlier, this colour scheme changes dynamically: when you click on one of the spheres, it turns red and the name of the person it represents appears (Figure 9).

Figure 9: Detail of Figure 8.

Another new colour has appeared in the screen capture above: yellow. Yellow marks the spheres representing the people with whom the clicked red sphere is connected—here, there is only one connection. In Figure 9, Elizabeth Smith (red) is connected to the big yellow sphere (we would have to make the mouse hover over it to see who it represents). In Figure 10, one can see the same dynamic response to clicking on a sphere; in this case, the sphere representing a person has more than one connection.

Figure 10: Dynamic Response to selecting one sphere.

Grosvenor Charles Bedford—large and white before clicked, but red in Figure 10, after clicking—has eighty two connections, and, if you count the yellow spheres, you would find almost eighty two people to whom he is connected. Some of the eighty two might be family members, in which case they wouldn’t be pictured here because we have selected "non-family only" for viewing in the dynamic graph. Were the image to show both family and non-family, we could count eighty two yellow spheres, Bedford’s eighty two connections.

In both the case of Elizabeth Smith, with her one connection (Figure 9), and the case of Grosvenor Bedford, with his eighty two (he’s the largest; Figure 10), one cannot see the edges that connect them as the graph is currently configured. There is in the centre of the graph a massive mesh of gray lines, some solid, some dotted. We’ll shortly start manipulating the graph so that you can see the edges and get more information from them.

On the right-hand side of the screen, the lower box has the title "K-Cores." The number of K-Cores is the number of relations that any given person has.

Figure 11: K-Cores.

The K-Core, sphere size, and unclicked colour all represent the number of relationships with other people; Grosvenor has relationships with eighty two others (represented by spheres). Again, the relationship is constructed through Southey’s writing of the letters. A letter either indicates that there is in fact a real relationship between the person to whom it is addressed and the person mentioned in it ("remember me to Duppa," Southey will say to Grosvenor (2009, i.108), or the letter itself will establish a relationship between the person to whom it is addressed and the person mentioned, even if that relationship is only cognitive. So for instance, Southey writes to Horace Bedford that the second edition of his epic Joan of Arc ought "to be dedicated to Mary Wollstonecraft" (2009, i.160), thereby ensuring, presuming the letter was received and read, that Horace has thought about Mary whether in fact he ever physically met her or not. Southey also mentions his epic in connection with Charlotte Cordé (Corday), but the editors of the text have written a footnote explaining who she is rather than encoding her name and linking it to their biographies section. That editorial procedure tends to eliminate mere mentions of unrelated people from the network. Southey writes to Grosvenor Bedford about Plato in Letter 77, for instance, but thankfully Plato is not figured here as a person in his network. Some of the eighty two people have been mentioned to Grosvenor and thereby are connected to him many times. But the K-Core registers not the number of mentions of other people, but rather the number of people, i.e., the number of people to whom he or she is connected by all the letters in the edition, up through Part II (1803), and the living breathing people who are more or less in or out of this network.

The K-Core boxes are checkable and uncheckable, and we will use this feature to reduce the number of relationships viewed at any time so that we can see the edges. Beginning at K-Core 1, people who have only one connection with one other person and so are very inconsequential members of the network, we uncheck the boxes, moving up in number from only five relationships and up, to nine or more (Figures 12 and 13).

Figure 12: Unchecking K-Cores.

Figure 13: Unchecking K-Cores Continued.

Again, what we are seeing here is anyone (represented by a sphere) who has nine or more relationships as indicated by the K-Core boxes checked (nothing below nine), the colours (a subset of green, and then all pink, blue, and white), and the spherical sizes. To get a better look at the edges and see who is related to whom and how, we can begin to click on and move the spheres around the workspace (Figure 14).

Figure 14: Graph with 11+ K-Cores, manually manipulated.

Among the lower number of connections, blues and pinks (eleven to fifteen), one can see that the edges are most often directed, and I will explain why and what that means momentarily. It must be emphasised that this is a dynamic graph. Apart from the default positioning of things, spheres and edges can be moved all over to get a better look at the relationships. You will see many screen captures of the graph, and they will almost all differ slightly, freezing one moment in my work with the graph. And other things begin to happen as one moves spheres around on the screen. First, let me show what happens when you click on any one of the spheres in the dynamic graph below (Figure 15):

Figure 15: Bedford’s Relations.

When you click on Horace Bedford, all the people who have been mentioned to him and/or to whom he has been mentioned (i.e. all his relationships) turn yellow. We see only ten yellow spheres here, whereas the tag tells us that he has "29 Connections." Again, if all the family and non-family spheres were visible, we would see twenty-nine yellow spheres, but we have selected non-family K-Cores of eleven and over, and among that set, Horace has ten relationships. So Mary Wollstonecraft, to whom he is connected via Southey’s letter saying that he wished he had dedicated Joan of Arc to her, is not one of these yellow spheres because her K-Core is eight:

Figure 16: Mary Wollstonecraft’s 8 Connections.

To find a person’s K-Core, one can select all K-Cores, and then scroll through the list of "Families"—the box above K-Core—selecting first "none," and then checking only the surname wanted. The number of K-Cores corresponds to the number of people to whom Wollstonecraft is mentioned plus the number of people who are mentioned to her, if indeed Southey wrote her letters.

Being "written about" and "written to" are two variables that are partly visible in this dynamic graphic via the edges. Figure 17 is a graph of non-family people with K-Cores of ten or above, and I have clicked on an edge between Charles Danvers, Southey’s neighbour in Bristol and longtime friend with whom he stayed while writing his second epic, Madoc, and Charles Biddlecombe, a neighbour in Burton, where he lived temporarily.

Figure 17: Directed Edge.

Here, the edge is one way: Danvers has been mentioned in a letter to Biddlecombe, but not vice versa. There is one connection between the two. There is a difference between the meaning of the number of "connections" marked on spheres, which is the number of people with whom that person has connections, and the meaning of the term "connection" as it is used on edges, which is the actual number of mentions. As seen above, Danvers has only been mentioned to Biddlecombe once, and so I surmise that Southey asked Biddlecombe to send something of his to Danvers, if not simply mentioning him in conversation. I checked my hypothesis by going to the Correspondents List, the list of people who are mentioned in letters, and found a letter to Biddlecombe in which Danvers is mentioned (2011, ii.598); and indeed, Southey is telling Biddlecombe to send the money he makes from the sale of Southey’s furniture to Southey’s mother via Charles Danvers’s address. Now we’ll examine a bi-directional edge (see Figure 18).Some (not many) letters

Figure 18: Undirected Edge.

Here I have clicked on the solid line connecting William Taylor, the German translator whom Southey met at Great Yarmouth in 1798, and Humphry Davy, whom Southey knew from living in Bristol where Davy worked under Thomas Beddoes Sr. at the Pneumatic Institute. The solid line means that the edge is bi-directional, that they have both been mentioned in letters to each other, and that there are seventeen connections between them. This means that Taylor was mentioned to Davy X times, and Davy to Taylor Y times, where X + Y = 17, (in this case, as I know from the Correspondents List, Taylor is mentioned to Davy once, as "William Taylor, the all-knowing" [Letter 454, 12 Nov. 1799], and Davy to Taylor sixteen times). However, we cannot see in the edge itself the distribution—we cannot tell from looking at the graph the asymmetry of Taylor’s relationship to Davy that Davy is being discussed with Taylor, and not vice versa. Taylor is Southey’s trusted intellectual advisor, and he discusses with him Davy’s genius as a chemist and his potential as a poet. Apparently, both Taylor’s judgement of Davy’s early poetry and Davy’s own propensity to put his work at the Pneumatic Institute before all else, scuttled his poetic career.

Now that we know the invariant, the variables, and what we are seeing when we look at any particular symbol—the minimal requirements, according to Bertin, for reading a graph accurately—we can look at and begin actually reading the K-Core 10+ graph, where reading means discovering information that is prominently visible when the relationships are represented in graphical rather than textual form. It is important to be able to see the spheres and edges without clicking on any or all of the spheres (the only way to see the names of the people involved) so I have added numbers and a name key to a screen capture of the graph in order to expose the name labels that the Relate tool will give users as they manipulate it by clicking on the spheres (Figure 19).

Figure 19: Name Key added to screenshot of dynamic graph.

Here again the blue spheres are ten to eleven relationships, the pink thirteen to fifteen, and the white seventeen to eighty two. Among the lower number of connections, blues and pinks (K-Cores 11-15), one can see that the edges are most often directed. It may even be the first thing one notices when looking at this graph, that there are a few spheres that have many arrows pointing at them (1-5, 8). This visually salient feature is significant as it indicates that these people are being mentioned in letters often, but other members of Southey’s social network are not being mentioned to them. So the question is: why?

Except for Mrs. Danvers and Southey’s aunt, Mrs. Tyler, the names of those mentioned are illustrious: head of the Bristol Pneumatic Institute and father of the physician poet, Thomas Beddoes Sr.; William Wordsworth; publisher Thomas Longman; and George Burnett, the only relatively unknown male member. I moved the diagram around a bit to get a better look at Longman (Figure 20).

Figure 20: Detail of Figure 19 (achieved by moving Thomas Longman’s sphere).

My first inclination in looking at these results was to think that these directed arrows betoken elitism on Southey’s part. He drops the names of people like Longman frequently in letters, and when he writes to such people, says little to them that is personal, and little about his own social network, wanting them to pay attention only to himself. Such a reading of the graph is irresponsible and not supported by the protocols for reading graphs, no more justified than free-associating on what or whom Wordsworth might mean by "Lucy."

A major principle of visualisations is, again, that the first thing one sees when one looks at a visualisation could be "errors" in the data. Upon investigation, I noticed first that many of these people have something in common. Mrs. Tyler and Mrs. Danvers share the fact that members of the network lived with them—Southey and Charles Danvers respectively. In fact, somehow Mrs. Tyler has not been properly designated as "family," revealing either a coding mistake or a problem with the tool. It may not be properly separating family from non-family, for which one would need to submit a bug report. When one looks at the huge number of mentions of Mrs. Danvers by going to the "mentioned" section of The Correspondents List, one finds them indeed in letters to Danvers, and a cursory glance reveals why: Southey almost always ends his letters to his good friend by mentioning her—"Our love to Mrs. Danvers" (2011, ii.644). If Mrs. Danvers is always sent wishes through her son Charles, with whom she is living, then the letter is not so much establishing a relationship between Charles and his own mother as it is between Southey and Charles’s mother. Finally, neither Beddoes, Burnett, nor Longman, the other people who are mentioned to others but not addressed directly, appear on the list of correspondents in Pratt’s edition, which "only names correspondents where one or more letters from Southey to that individual survives" (2009, Correspondents). Southey’s letters to Beddoes, Burnett, and Longman during the period 1791 to 1803 do not survive. We have no record of whom he mentioned to them, only mentions of them in letters to others. From this dynamic graph then, we cannot necessarily conclude that the number of directed edges—arrows—pointing at their spheres, the absence of reciprocity in mentioning others to them, comes from undue deference on Southey’s part, as I had imagined. Others may be mentioned to them in letters that are missing.

Thomas Longman’s publishing house suffered a fire during the mid-nineteenth century, and unfortunately letters were lost, though letterbooks containing Longman’s letters to Robert Southey may indeed survive. It is worth here remembering exactly the invariant in this graph. It shows how people are connected in a network that has been created via letters WRITTEN BY Southey, NOT by any letters written TO him. The Longman Archive Online advertises itself as containing "autograph letters from Longman authors 1799-1900 (authors include . . . Robert Southey . . .)." Despite the implication, it is not true that the archive contains autograph letters from Robert Southey, as I have confirmed via e-mail on Friday 17 August 2012 with Editor Lynda Pratt. She writes,

Some (not many) letters [written by Southey] to Longman do survive for later periods – scattered amongst a number of archives and not always easy to identify. Some of these later letters were published in J.W. Warter’s Selections from the Letters of Robert Southey (1856). To get a sense of how Southey viewed Longman, see Letter 1939: "The people at that house know nothing about books except in the mere detail of trade."

(At the time of final editing, the letter mentioned in the above quotation is now available: http://www.rc.umd.edu/editions/southey_letters/Part_Four/HTML/letterEEd.26.1939.html.)

Noteworthy at this moment in my exposition is the emergence of one very important aspect of our first principle for reading visualisations, one that these three "errors" just discussed concerning this data set makes salient. I put "errors" in quotation marks because what it really means here is incorrectly categorized information. We did get one error and now know one thing that needs to be fixed: —the name Tyler needs to be designated "family," not "non-family," whether through data correction or fixing the tool. But the other two "errors" are not mistakes as much as information of another kind, information not about Southey’s network but about the data itself. Here follows the first part of Ware’s principle as well as the portion about errors once again:

A visualization commonly reveals things not only about the data itself, but about the way it is collected. With an appropriate visualization, errors and artifacts in the data often jump out at you.

"Errors" are one thing that pops to the fore, but so are "things about" data, "artifacts in the data," and information "about the way it is collected." Visualisations give information about content, but also about the structure of the data—its medium, genre, and form—and therefore need to be read both ways, attuned to the possibility that at any given moment one could be exposed to information about either or both. Moreover, visualisations can be used to investigate the data collection. We now have a way of looking for which letters of Southey’s might be missing and need to be gathered. For example, imagine a person whom Southey mentions frequently in letters to others but to whom he does not mention the names of other people in his social network (e.g., he writes to Coleridge and Cottle about Longman but, when he writes to Longman, he doesn’t mention Coleridge or Cottle). Such a person will appear, because of the directed edges going toward him, as a sphere surrounded by arrowheads. When one sees such a sphere, there are three possibilities: 1) Southey doesn’t mention other people in his social circle to this person for a potentially interesting reason; 2) he doesn’t write to this person (also interesting); or 3) the letters to that person are missing (interesting as a fact about data more than a fact about the relationship between Southey and the person in question). The latter is the first thing to check. The conventions of closing letters are features of the epistolary genre: we now know one feature—this one, "my best to your mother"—is common sense, but even if such features are noticed, we can now discern a way to mathematically formalise this feature and could then search huge data sets for connections among people while discounting those mentioned in closing. Or indeed, we could look for closings in an unstructured data set of texts using that algorithm in order to find among them texts that are letters or are written in the epistolary genre.

2. Stages of Reading

Bertin lists three stages in the reading process:

1. EXTERNAL IDENTIFICATION. In this stage, we need to ask, "What components are involved?" The components are the invariant plus the "variational concepts." We have identified the invariant (Southey writing letters). The other components are:
   • a. People named in the letters, either as addressees or as subjects discussed;
   • b. Number of relationships that have been forged by the letters between people named in them;
   • c. Type of connections within any given relationship, whether it is reciprocal or exclusive; and
   • d. The relative strength of any relationship among Southey’s acquaintance’s relative strength as evinced by the number of connections: in social networks, "stronger ties represent close friendships and greater frequency of interaction" (Easley and Kleinberg 2010, 48). Strong ties represent friends, and weak ties represent acquaintances;
   • e. And size of the spheres representing the centrality of that relationship to Southey himself, whether in thought (he mentions the person in a letter to someone) or deed (he actually writes to the person).
2. INTERNAL IDENTIFICATION. "By what variables are the components expressed?"
   • a. People are expressed as "vertices"—also called "nodes"— in graph theory (they are called "vertices" because they constitute the point at which edges or lines meet). In the Relate tool, however, each vertex or node is marked by a sphere.
   • b. The number of relationships is expressed by the size and colour of the spheres, as well as by the legend on each sphere when it is clicked ("82 connections" on Grosvenor Bedford means that he has relationships with 82 people) and by the total of yellow spheres that appear when it is clicked. Those yellow spheres show, in graph theoretic terms, the red (clicked) sphere’s "network neighbors" (Newman 2010, 7.2);
   • c. The type and number of connections within a relationship is expressed by solid or dotted lines, "edges" in graph theory. The solid lines, called "Bidirectional Edges," indicate that Southey wrote to both people and mentioned the other to each one, a reciprocal relationship. The dotted lines, called "Directed Edges," indicate that Southey wrote to the person designated by the sphere where the line originates and spoke of the person designated by the sphere to which the arrow points. The iconology here is "counterintuitive" insofar as the arrow points AWAY FROM the person written to by Southey and TOWARD the person mentioned in the letter. Solid lines show reciprocal, strong relationships. Clusters of arrowheads around a sphere show that the person designated by that sphere is more talked about than written to. Crucial to understanding this graph is realising that edges indicate not just connections between people but the flow of information from one about the other through Southey. If an arrow points toward a person, the information about him flows between Southey and the person represented by the sphere from which the arrow shoots out. However, if there is no arrow in a connection between two spheres, then we are indeed seeing an information flow insofar as Southey mentions one to the other and vice versa in letters to him or her. The flow of information between the spheres with bi-directional edges always goes through Southey.
   • d. Unfortunately, our Relate Tool needs to be tweaked so that the relative strength of edges is made immediately visible through the thickness of the lines, at least for undirected edges. Right now, however, one can at least see the number of connections by clicking on an edge (see Figure 20).
3. PERCEPTION OF PERTINENT CORRESPONDENCES. "This perception is always the result of a QUESTION, conscious or not. What are the questions which one can ask in approaching the information?" The two major questions made possible by this dynamic graph are:
   • a. Does Southey discuss X and Y with each other?
   • b. Does Southey mention X to Y (with the arrow direction directly contradicting the syntax)?

One can always ask questions prompted by looking at the graph—e.g., how many of the people whom Southey habitually mentioned to each other actually met each other face-to-face? But that question cannot be answered by looking at the graph itself.

Having worked through our first rule—that the most salient characteristics of any graph represents so-called errors in the data—and thereby having recognised some information as telling us something about the dataset rather than about Southey’s relationships, we’ll look again at Southey’s social network from 1791-1803, those who have ten relationships or more (K-Core 10+; Figure 21). Our ultimate goal, to be achieved by asking lots of specific questions, is to divide conceptually the visible features of this network that indicate, on the one hand, something about how a data set is constituted, and on the other hand, something about Southey’s relationships with various people, as well as the impact of his network building upon their relationships with each other (Figure 21).

Figure 21: K-Core 10+.

By looking at specific details concerning how people are viewed here, and by correlating those facts with what we know about Southey’s biography and/or the data set, can we answer the following question: Do the larger spheres, those to whom Southey is mentioning more people and who are being mentioned more often (when the lines are solid), represent greater intimacy or involvement with Southey? It seems so. All the largest, white spheres (numbers fourteen on) can be said to be part of Southey’s inner circle of close friends. The one exception is Biddlecombe, whose numbers are inflated a bit because Southey is writing to this former neighbour in Burton about selling his things. The pink group, from number six through thirteen, are people with whom Southey is doing intellectual work of some sort or another, with two exceptions: 7) Thomas Lamb, really a father figure for Southey, and 10) Charles Collins, a close school-friend with whom Southey stopped corresponding in 1794. And the blue group, one through five, except for Aunt Tyler and Mrs. Danvers, are mere acquaintances. But it turns out that none of Southey’s letters to them survive, according to the Correspondents List. Whereas the missing letters to Longman can be accounted for by the fire, the missing letters here might in fact be accounted for by the level of Southey’s importance to these people, and vice versa. According to Lynda Pratt, Southey had a break with his Aunt Tyler that perhaps prompted both of them to destroy any letters written during a time when their ties were stronger (e-mail message to the author, 17 August, 2012). We can see in this edition that some of Southey’s mentions of Wordsworth are derogatory, and we know that Wordsworth resisted seeing Southey and himself as in one "school" of poetry. These people are more mentioned than written to, and they did not save Southey’s early letters to them because of a relatively low level of intimacy.

Lynda Pratt and I have had a discussion about this argument that I would like to relay. Dr. Pratt points to the element of chance in the survival of letters. In some cases,

later generations who inherited letters that had been preserved did not care for them and therefore destroyed them, perhaps because they contained information thought to be unsavory, perhaps because they were perceived to have no interest or value.  We know, for example, that one of Southey’s granddaughters destroyed family papers.  Such a bonfire might account for the lack of Southey’s letters to Caroline Bowles from the period of their courtship (pre-courtship letters have survived – so the gap is an interesting one).  One thing that is clear is the element of ‘chance’ in the survival/ transmission of MSS – and the Southey letters edition is (like all such editions) a wonderful example of this.  A key example of ‘missing’ correspondence is that between Southey and Lovell – only two letters survive (and one of these is a fragment).  This makes a key, pre-Coleridge relationship very difficult to reconstruct. (E-mail to the author, 17 August, 2012)

One can see the element of chance factoring into how our current data set has been structured, but also causal elements: Southey’s courtship letters to Caroline Bowles were "unsavory" to a Victorian descendant. Dr. Pratt pointed out to me "the chaotic lifestyle of Coleridge, who discarded letters unread" (e-mail to the author 17 August, 2012), which seems to me to be NOT a shaping of the archive by chance but rather another indication of an argument that I will make below, that Coleridge’s capacity for intersubjective relationships has been injured, and that consequently he serves as a break in network relations, disrupting their ordinary flow. Clearly, though, expert knowledge about Southey, Lovell, Bowles, and Victorian grandchildren is necessary for understanding what we are seeing in this dynamic graph.

Similarly, only literary historians can tell us what kinds of people make up Southey’s social milieu. Among the pink spheres are two doctors, and indeed the presence of three doctors in the K-Core 10+ network—Beddoes (1), King (6), and Davy (9)—demonstrates that some of Southey’s thinking and writing is inflected by these literati of the Bristol Pneumatic Institute. Moreover, two artists make up this group: Mary Barker (11) and Richard Duppa (13), suggesting that, as he was writing and planning publications, Southey was thinking about illustrations of them. That turns as we see in the Betweenness somewhat validated by this set of letters in which Southey was trying to get travel recommendations from Duppa, and so often included his name in letters to friends, mentioning that Duppa was helping Southey himself plan a trip to Italy. Mary becomes, during this period in Southey’s life, a friend as well as an intellectual influence. Nonetheless, Southey’s close social network includes practitioners of medicine and the plastic and visual arts, and he was intensely interested—as his network reveals—in both.

One bit of information stands out if one reads the graph with the question: "What kinds of relationships did Southey have with his publishers?" Joseph Cottle, the Bristol Bookseller, published the first edition of Lyrical Ballads, Coleridge’s Poems of 1796, 1797, Southey’s John of Arc, Letters from Portugal, and the Annual Anthology of 1799, among other things. In this graph, one can see substantiated the argument of Cottle’s own Reminiscences; Southey is as close with Cottle as with some of his habitual correspondents, his close friends. Both Southey and Coleridge almost always address Cottle in their letters with the same kind of sobriquet, "My dear Cottle." Although Cottle may not be a typical eighteenth-century bookseller, he plays a role in Coleridge and Southey’s lives well beyond any business relationship.

Of the letters that survive from Southey to Longman, which are not yet published by this edition but will appear in forthcoming parts, and even from Southey to Byron’s publisher John Murray (forthcoming), are addressed "Dear Sir." Southey’s relationship to Longman can also be seen through Coleridge’s printed letters (1956). Despite the fire, we do have two letters from Coleridge to Longman, one of which was copied before being sent. The fact of a letter being copied suggests that its importance is legal, having to do with business. Coleridge addresses him as "Dear Sir," not "My dear Longman" (1956, i.654-55). In Wordsworth’s letters, too, we find "my dear Cottle," whereas Longman is "Mr Longman" and "Dear Sir" (Wordsworth and Wordsworth 1967, 306, 307, 309). The content of Coleridge’s letters to Longman differ also from those to Cottle. The latter are playful and often quick invitations sent to Cottle at short notice, asking him to send his servant out for food to cook dinner, requesting tobacco to aid the writing process, and the like (Coleridge 1956, i.156-57). What book history experts know about Longman is visible in this graph of Southey’s social network, despite data loss—viz., that Longman is the first of the truly modern publishers (Feather 2006, 76) who produced books for business. T. N. Longman took his "bookselling" shop from visible inheritance of a family dynasty to "headquarters of an organization" (Briggs 1974, 9). In other words, the structure of the archive to some extent participates in its content.

3. Expert Knowledge

The structured-graph reading adumbrated above—the reading given so far— offers no new expert knowledge. The graph has helped me become more of an expert on Southey than I might otherwise have become by directing my readings through his letters and the letters of Wordsworth and Coleridge, for the sake of comparison. Reading the graph also required consulting Lynda Pratt, the expert on Southey. I have proven in the first part of the essay that it is not possible to read this graph accurately without knowing the details of Southey’s life well enough to be able to distinguish when we are seeing facts about the dataset, facts about Southey and the people in the network, and facts that are really true about both at the same time. But so far the reading has produced no new knowledge for either literary or information visualisation specialists.

We want it to do more. I will in the remainder of this article prove that it is possible to obtain some knowledge from this dynamic graph of Southey’s Social Network, knowledge that is new and valuable in both literary history and information visualisation.

4. Network Principles

Graph theoretic notions have been boosted by studies of social networks, some of them digital, some of them print. I will now demonstrate how several of these concepts offer new knowledge to literary historians and the scientists who study social networks.

4.1. Cliques

In Networks: An Introduction, M. E. J. Newman defines these terms as they are used in analysing graphs. "A clique," he says,

is a maximal subset of the vertices in an undirected network such that every member of the set is connected by an edge to every other. The word ‘maximal’ here means that there is no other vertex in the network that can be added to the subset while preserving the property that every vertex is connected to every other. (2010, 7.8.1)

The graph of Southey’s social network reveals several cliques according to that definition of it. In order to find these cliques, however, we need to turn off the "Directed Edges" by unchecking that box at the top right of the Relate tool. A clique can only be found in "an undirected network"—that is, the relationships must be reciprocal. Figures 22 and 23 show two of the small cliques that are possible to see in the graph of Southey’s letters.

Figure 22: One Clique (arrows added to a screenshot).

Figure 23: Another Clique (blue arrows added).

In Figure 22, I have traced a clique linking Coleridge, Danvers, Wynn, and Cottle; in Figure 23, I have traced Danvers, Wynn, Bedford, and Cottle. Again, a clique is defined as a group of people, all of whom are connected to each other. All the spheres in both Figures 22 and 23 do not together constitute a larger clique because Coleridge is not connected to either Bedford or May. However, another small clique including May can be traced in this cluster represented in Figures 22 and 23. Figure 24 shows a "maximal" clique.

Figure 24: Maximal Clique.

Figure 24 displays the maximum set of relationships that can be found in which every member is connected to every other. The clique of Southey’s friends involves a Bristol wine merchant (Danvers), a benefactor (Wynn, who gave Southey an annuity of 160 pounds), a businessman (May), a statistician (Rickman), and a publisher (Cottle).

Cottle, Danvers, and Wynn are common to all the cliques illustrated here in Figures 22-24. It makes sense that Wynn would be; one can imagine that Southey writes regularly of his activities to his benefactor. Cottle and Danvers, then? I think it is possible to see them as Southey’s closest friends. In his famous letter of 13 or 14 November 1795, when Coleridge "breaks up" with Southey over his withdrawal from their "Pantisocracy" plan, Coleridge defends himself against accusations of spreading rumours about Southey. He says that he answered the inquiries of friends about Southey’s plans only in general terms, but "To Danvers indeed and to Cottle I spoke more particularly—for I knew their prudence, and their love for you" (1956, i.168). These are the people closely connected to Southey, and so they are those to whom Coleridge turns in a crisis, the crisis of Southey struggling to determine his future career. Coleridge’s letter confirms what the graph shows, that Danvers and Cottle are indeed the people with whom Southey is conversing with most at this moment (1791-1803, as represented by Volume 1 and 2 of the letters), although they are still not necessarily his closest friends. In an e-mail on 17 August 2012 in response to my query, Lynda Pratt writes about these two of Southey’s relationships:

Should we see Danvers and Cottle as Southey’s ‘closest friends’?  I wonder if it’s more complex than this.  I think it is clear that Danvers is one of Southey’s closest friends – in fact, even something of an older brother, confidante.  Cottle is, I think, a slightly different case.  He and Southey are close at this time but their relationship is inflected (infected?) by business and a sense of class difference that never enters into Southey’s relationship with Danvers.  Yes, Southey is closer to Cottle than to any of his later publishers.  He even has a personal relationship with him. However, they are not quite on the same level, and later events make that clear: Cottle’s Recollections and Reminiscences literally writes Cottle himself back into the lives of the Lake School.  Cottle’s structuring of his own archive is interesting here.  For example, he habitually numbers and renumbers letters sent to him by his poet-friends suggesting that he considers and reconsiders the order and significance of those letters.  That’s before we get onto his habit of cutting and pasting letters when he incorporated them into his published memoirs which more drastically restructures his archive.

But if Coleridge’s letter and Lynda Pratt’s communication already describe the importance of these relationships to Southey, then what the graph reveals is not news, but only a way of augmenting what we already know. Visualisations should confirm what experts can tell us, or they may be simply wrong. But if we need to check them via expertise, how can they ever show us anything new?

Let me push the graph’s meaning as far as I can before we give into that paradox. The person who is written to most often, Grosvenor Bedford, appears in only one of the three cliques pictured above and not in the maximal clique. Coleridge has no reciprocal connection to Bedford; Southey mentions Coleridge in letters to Bedford, but not vice versa. If one takes the common denominators of the most cliques, including the maximal clique, as defining a person’s closest friends, then the graph argues that Coleridge and Bedford are not as close to Southey as Danvers and Cottle. Coleridge breaks up with Southey—Danvers and Cottle never do. Bedford is a close friend but he is not there living in Bristol and is not part of Southey’s daily life in quite the same way as Danvers and Cottle are until 1803, when Southey moves away from Bristol to the Lake District. But the graph argues more: that one’s closest relationships are as much a matter of a person’s imbrication in one’s web of relationships as they are a matter of individual choice. Pratt’s analysis shows us that Southey’s feelings for Cottle are infected/inflected by business. Thus, the "love" of which Coleridge speaks is less a sentiment and more the power that Cottle and Danvers have to influence Southey as defined by their places in his social network; "love" is a network effect.

4.2. Degree Centrality and Betweenness Centrality

The number of connections is the default setting for what is shown by the spheres in the Relate Tool, but there are two other settings, Degree Centrality and Betweenness Centrality (Figure 25).

Figure 25: Detail view of Relate’s Drop-down Menu.

The two are also visible at the bottom of our tool for any individual when their sphere has been clicked (Figure 26).

Figure 26: Detail view of the table appearing at the bottom of Relate.

What do these measures mean? Degree Centrality reveals which "are the most important or central vertices in the network" (Newman 2010, 7.1). In the table in Figure 25, one sees a list of numbers corresponding to a sphere or vertex (here called a "node," but basically, a person). Calculated according to "the number of edges connected to" a person, the number given on Relate at the bottom of the screen for "Degree Centrality" is the percentage of all possible relationships that this person has. There are 133 vertices or spheres, and in the Relate tool, at the bottom, called "nodes"—133 people in Southey’s non-family social network. Using the drop-down box under "Options," at the top right of the Relate Tool, I click on "Degree Centrality" rather than number of relationships. In Figure 27, we can see what share of those relationships is had by each person in the clique.

Figure 27: Degree Centrality of Members of the Clique.

These decimal points are carried out to many places that hamper our view of percentages (and we will fix this view on the tool), but just looking at two place past the decimal gives us a percentage. At the highest number of relations, Bedford garners roughly 62 percent of the relationships to be had in this network.

Betweenness Centrality differs from Degree Centrality; it "measures the extent to which a vertex lies on paths between other vertices" (Newman 2010, 7.7). It measures, in other words, the shortest paths between people. Newman warns us that: "communications do not always take the shortest path. Nonetheless, betweenness centrality may still be an approximate guide to the influence vertices have over the flow of information between others (2010, 7.7)."

The higher the number, the greater a person’s Betweenness Centrality is, as we see in the Betweenness Centrality of Southey’s inner circle (Figure 28).

Figure 28: Betweenness Centrality of Clique Members.

Though Bedford has the highest number of relationships in Southey’s social network, he has lower Betweenness Centrality than Wynn, Cottle, Coleridge, and Danvers. Danvers indeed has the highest. Coleridge, Cottle, and Danvers seem to be the most important clique members in terms of the flow of information, Wynn less so, and Bedford and May, still less. Again, these three are living and working together regularly in Bristol; it makes sense that information would flow through and between them, and so this Social Network re-affirms the hypothesis that high Betweenness indicates information flow.

Notice that Wynn, Danvers, and Cottle hover at around 45 percent Degree Centrality. That may represent the degree of network imbrication necessary for "love." Too much imbrication in a network (Bedford’s 62 percent) as far as letters are concerned may show that the friend is a confessor but not a participant in the network. One cannot talk about everyone to someone who is close to all those other people, only to someone who is less likely to repeat what he or she hears. That Bedford’s Betweenness Centrality is lower than the others indicates that information flows through him less than through others: Southey writes to him about others, but not to others about him. Given the evidence of the Southey social network, one can propose a hypothesis for graph theory: a high number of relationships coupled with a Betweenness Centrality lower than those with lower numbers of relationships signals a person’s role as confessor—someone who listens to discussions about a network without participating in it.

5. Conclusion: Local Bridges, Giant Components, and Network Effects

Before Southey moved to the Lake District, he was dubbed in the Edinburgh Review part of that "new school of poetry" that came to be called "the Lake School" even though he hailed, of course, from Bristol. Southey is called the member of an unnamed "sect" or "school" of poetry in the inaugural issue of the Edinburgh Review by Francis Jeffrey. The ER here begins its "unrelenting attack upon the ‘Lake Poets,’" editor Donald Reiman tells us (1972, 2.a.415), in Jeffrey’s review of Southey’s Thalaba, the Destroyer: A Metrical Romance, but the school is not named in this 1802 review. Jeffrey only later designates Southey, Coleridge, and Wordsworth the "Lake School," when reviewing Wordsworth’s Poems of 1807 (2.a.429). In a letter of 1804, Wordsworth balked at the idea of a school of poetry comprised of "Coleridge, Southey, Lamb and myself": "it is scarcely possible that a greater difference should exist between any set of men or Authors, than between these four men with the exception of Coleridge and myself..." (1967, 434). Wordsworth’s poetic tribute to Southey at his death speaks not at all of his poetry, and thus was deeply resented by his second wife Caroline Bowles, who called Wordsworth "‘that other star of the lakes,’" implying of course that Wordsworth wanted no competition in leadership of his school (Pratt 2006a, 221, 233). New work has recently demonstrated Cottle’s leadership in "The Bristol School" or West Country Romanticism (Cronin 1992; Cheshire 1992). Beginning with Southey’s Joan of Arc, Cottle arguably served as a muse for the epic poems that Southey wrote. Cottle himself wrote epics with "unhappy" results. As Southey puts it in a letter to William Taylor: "From England nothing has reached me but the unhappy Alfred of poor Cottle. I laboured hard & honestly to suppress its birth – & am thrown into a cold sweat by recollecting it" (2011, ii.558), (a search for author "Southey, Robert," and search terms "Cottle" and "Alfred" in NINES [http://www.nines.org/search/browse] will return all the letters by Southey in which he mentions the poem). The account of Joan given by Southey in the Preface to the first edition published by Cottle in December 1795 suggests rapid writing, correcting, and printing, all going on in the same room under Cottle’s auspices; the epic poem was originally written in six weeks, he says, augmented by Coleridge, and completely revised (except for 1000 lines) during its first printing (Southey 2004, 4). A parody akin to Byron’s report of looking forward to "‘an epic from Bob Southey every spring’ (Don Juan III, 97.4)" appeared in the Edinburgh Review of 1808:

A correspondent wrote to us lately an account of a tea-drinking in the west of England, at which there assisted no fewer than six epic poets—a host of Parnassian strength, certainly equal to six-and-thirty bands. . . . How unreasonable then is it to complain, that poetry is on the decline among us! (qtd. in Curran 1986, 158)

The tea would have been served at Cottle’s, in Bristol, "in the west," even though by 1808 Cottle is no longer a bookseller, though still involved in provincial printing with Nathaniel Biggs.

This West-Country epic literature always did have an uneasy fit within the Lake School. Dissension in the ranks due to Coleridge’s policy of worshipping only one great mind at a time mirrors what is fundamentally a generic distinction. In a letter written to Cottle shortly after Coleridge met Wordsworth, he repeats approvingly Wordsworth’s criticism of Southey’s writings, objecting to the "ease" and "fluency" with which Southey wrote (Coleridge 1956, i.320, also qtd. in Pratt 1994). Pratt points out that the criticism is articulated by quoting a line from Joan of Arc, an act of irony that "would not have been lost on Cottle" (1994, 336). Coleridge is obviously concerned to distinguish Wordsworth’s lyric condensation from Southey’s expansive epic ambitions.

I have argued that injury to one psyche can block intersubjective exchanges and thus disrupt the smooth operations of network evolution. Whatever the reasons, though, strong triadic closure failed to join Southey and Wordsworth in friendship via their ties with Coleridge and thereby to link the two giant components of Wordsworth’s social network with our graph of relationships constituted by Southey’s letters. "Giant component" is another term found in network theory, and the two giant components mentioned here appear only in an as yet imaginary giant graph of Romantic-era writers’ correspondence networks, a graph that would include Wordsworth, Coleridge, and Southey’s social circles. The separation of the circles of Wordsworth and Southey into two giant components within this imaginary, giant graph—their social networks would be joined only by what are called "local bridges" constituted by Coleridge and a few other common friends—seems to result from, cause, instill, exacerbate, or indicate generic as well as personal differences.

Genre seems to me to be a key, unstated invariant associated with Wordsworth (lyric) and Southey (epic). Wordsworth’s introspective long poem The Prelude, pieces together small lyrical poems about biographically significant moments (the legendary "spots of time" being only one sort of lyric moment, among others). Southey could not write such a sustained introspection, whether in poetry or prose. A history of the growth of his own mind is an account that Southey longs to but never can write (Southey 1969, i.160). On the other side, Wordsworth cannot write the epic Recluse—which would after all be addressed to former sympathisers with revolutionary undertakings such as Pantiscocracy—while Southey’s epic passion continues throughout his writing life, reinvigorated in 1808 while visiting Bristol (Pratt 1994, xvii). There seems to be a generic division of labour between Wordsworth and Southey, and so the question becomes, how indebted is the production of Lake-School poetry to the evolutions and devolutions of social networks? Should we see the failure of Southey’s introspective and Wordsworth’s epic ones as "personal" problems, as "writer’s block" for Wordsworth that just by chance is accompanied by a weirdly inverse writer’s block on Southey’s part—both of them being otherwise incredibly prolific writers (Leader 1991)? Maybe these ostensibly personal problems are in fact network effects. Literary scholars of the Romantic era have undoubtedly been attuned to "literary circles" throughout the history of literary criticism of the period, most recently visible in the emergence of Romantic Circles in 1993 and the discussion of the "Southey Coleridge Circle" as a "network" in 2000 (Pratt and Denison 2000b). But questioning whether Wordsworth and Southey’s "failings" to write epics and lyrics, respectively, are intersubjective network effects? That question is new.