Imaginative Networks: Tracing Connec tions Among Early Modern Book Dedi cations

This essay uses network metrics (centrality, density, clustering coefficients) to account for shifts in dedicatory practice resulting from political crises, religious turmoil, and changes in book production practices. It constructs a network from the names that appear in dedications of EEBO­TCP texts; names are detected using the linguistic markup from the EarlyPrint project. The essay argues that we learn more about early modern book history by constructing networks of all the names that appear in dedications, not just those of authors, printers, and patrons. The network includes a mixture of religious and political figures, literary personalities, fictional characters, and bookmaking professionals, because this is the full range of names that dedicatory practice covers in the period. By proceeding in this way, network metrics can account for a range of dedicatory phenomena, including Queen Elizabeth’s popularity on both sides of the political aisle long after her death and, especially, consolidation around non­contemporary names in dedicatory practice as a result of both the Civil War and the Restoration. The imaginative networks revealed by early modern dedications are organized mainly around untimely figures from the recent and distant past, but despite this the networks are sensitive to historical change, especially at moments of political and social crisis. In A Description of the Famous Kingdome of Macaria (1641), Samuel Hartlib begins his book with a dedication to the “High and Honorable Court of Parlia­ ment,” but in the text of the dedication he saw fit to name a few people who have little do with the Parliament of the early 1640s: TO THEHIGHANDHONORABLE COURTOF PARLIAMENT WHereas I am confident, that this Honorable Court will lay the Cor­ ner Stone of the worlds happinesse before the final recesse thereof, I have adventured to cast in my widowes mite into the Tresurie, not as an Instructer, or Counsellour, to this Honourable Assembly, but have delivered my conceptions in a Fiction, as a more man­ nerly way, having for my pattern Sir Thomas Moore, and Sir Fran­ Journal of Cultural Analytics 3 (2021): 64­101. doi: 10.22148/001c.21993 JOURNAL OF CULTURAL ANALYTICS cis Bacon once Lord Chancellour of England; and humbly desire that this honourable Assembly will be pleased to make use of any thing therein contained, if it may stand within their pleasures, and to laugh at the rest, as a solace to my minde, being enclined to doe good to the publick. So humbly craving leave, that I may take my leave, I rest this 25. of October 1641.1 This dedication follows the conventional form by including an elaborate header that addresses the dedicatee, in this case Parliament. But Hartlib names two other individuals, More and Bacon, in the body of the dedication, and he even labels them as his “pattern” or role models in this publishing venture. Neither More (who died in 1535) nor Bacon (who died in 1626) would show up in a network graph focused on reconstituting the lived early modern social world of 1641, nonetheless Hartlib mentions each of them at the book’s open­ ing. Since he is putting forth a new utopian text, it makes sense that Hartlib would want to cite these two men as his “pattern,” making this a reference to both the authors and their books simultaneously. Hartlib uses his dedication to address his dedicatee but also to stake out some authorial territory by invoking his key influences. In this essay, I take advantage of the additional names that authors added to their dedications to create an expanded network of print re­ lationships that includes contemporary social bonds alongside historical—and even fictional—associations. These networks show that (1) naming in dedi­ catory practice is driven as much by untimely references to political and reli­ gious figures as by contemporary authors and patrons, and (2) these references, citations, and addresses through naming are responsive to political and social change, e.g. when the mid­seventeenth­century political crises arrive, dedica­ tory naming shifts markedly toward untimely references. As a further illustra­ tion, Hartlib’s reference to his influences is far from the only kind of naming that takes place in early modern dedications. Samuel Drake’s dedication of his ser­ mon Totum Hominis to Margaret Armitage also includes a number of additional names that belong to neither the author nor the dedicatee.

In A Description of the Famous Kingdome of Macaria (1641), Samuel Hartlib begins his book with a dedication to the "High and Honorable Court of Parlia ment," but in the text of the dedication he saw fit to name a few people who have little do with the Parliament of the early 1640s: TO THE HIGH AND HONORABLE COURT OF PARLIAMENT WHereas I am confident, that this Honorable Court will lay the Cor ner Stone of the worlds happinesse before the final recesse thereof, I have adventured to cast in my widowes mite into the Tresurie, not as an Instructer, or Counsellour, to this Honourable Assembly, but have delivered my conceptions in a Fiction, as a more man nerly way, having for my pattern Sir Thomas Moore, and Sir Fran Neither More (who died in 1535) nor Bacon (who died in 1626) would show up in a network graph focused on reconstituting the lived early modern social world of 1641, nonetheless Hartlib mentions each of them at the book's open ing. Since he is putting forth a new utopian text, it makes sense that Hartlib would want to cite these two men as his "pattern," making this a reference to both the authors and their books simultaneously. Hartlib uses his dedication to address his dedicatee but also to stake out some authorial territory by invoking his key influences. In this essay, I take advantage of the additional names that authors added to their dedications to create an expanded network of print re lationships that includes contemporary social bonds alongside historical-and even fictional-associations. These networks show that (1) naming in dedi catory practice is driven as much by untimely references to political and reli gious figures as by contemporary authors and patrons, and (2) these references, citations, and addresses through naming are responsive to political and social change, e.g. when the midseventeenthcentury political crises arrive, dedica tory naming shifts markedly toward untimely references. As a further illustra tion, Hartlib's reference to his influences is far from the only kind of naming that takes place in early modern dedications. Samuel Drake's dedication of his ser mon Totum Hominis to Margaret Armitage also includes a number of additional names that belong to neither the author nor the dedicatee. Armitage's husband, John, is included in the header, but more importantly Drake includes his dedicatee's two daughters, Margaret and Catherine, holding them up as examples of piety. Margaret and her daughters are each associated with the content of Drake's sermon, as exemplars of the virtues (justice, mercy, and humility) on which he preaches. Drake takes this a step farther when he insists that Margaret and Catherine "deserve the stile of Jemima and Kesia, (two of Holy Jobs Daughters)." In addition to the contemporary figures that are invoked here, Drake associates Armitage's daughters with biblical figures, further incor porating them into the subject of his sermon and demonstrating his knowledge of scripture. For Drake, as for many writers of dedications, the contemporary and the biblical exist sidebyside.
The habit of naming noncontemporary figures alongside living ones is repeated in the texts of many early modern dedications: some of the individuals men tioned are living, some have died recently, others are writers, influences, and religious figures from another time entirely. The imaginative network of a text involves all of these figures together, and by considering them in a single net work, I use network metrics-density, degree, strength, betweenness-to bet ter understand the interlocking motivations that underlie text creation, includ ing patronage, social and cultural aspiration, anxieties of influence, and social relationships. More specifically, by including names from the dedication body rather than just its heading, this study's networks show that early modern dedica tory practice is not mainly organized around presentday figures, as it is usually described. Instead the networks are tied together by a range of figures from the recent and distant past. And though these imaginative networks are not focused on contemporary figures, they are nonetheless responsive to contemporary poli tics. Shifts in network structure and centrality measures from the 1630s through the 1660s account for changes in dedicatory practice that arise from the political crises of that period: the replacement of the monarchy with a republic followed by the Protectorate, the exile of large portions of the English nobility, sudden shifts in censorship laws and printing regulations, and everchanging political and religious discourses. Rather than flattening the effects of political and so cial change on dedicatory practice, the untimeliness of naming in dedicatory networks alerts us to key moments of discursive shift amid political change.

Paratext, Dedications, and Naming Practices
The two examples above emphasize the untimeliness of the print networks I've reconstructed: just as the landscape of reading is made up of recently published texts and texts that first appeared long before, the landscape of print is made up of actors who are not physically present in the social world but are nonethe less crucial to understanding how the relation between persons and texts was constructed. The relationship between an author or text and a name mentioned in a dedication is an imaginative relation. When an author does not personally know the individual they are invoking in a dedication, or even when they do, the appearance of that name stakes out an intellectual and imaginative space for the idea of that person which could, but need not necessarily, fall back on any firsthand relationship with the individual. These imaginative relations abound during the creation of a text-we all carry the ideas of many people we may not currently know: public figures, deceased relatives, people we would like to meet. A dedication is one place where these imaginative relations can take form through naming, direct reference, and allusion. By recovering these rela tions as networks, it is possible to both visualize and quantify the striking degree to which noncontemporary names play a key role in early modern dedicatory practice, overshadowing the role of contemporary patronage. Furthermore, dur ing times of political and social upheaval, dedicatory mentions of figures from the distant past change radically, along with changes in patronage, censorship, and other social and economic shifts that underlie early modern bookmaking.
My study of dedications comes at a moment of renewed interest in paratexts for both book historians and digital humanities scholars. Rather than being viewed as ancillary to the main text, paratextual material such as prefaces, tables of contents, indices, footnotes, and addresses to the reader have been studied as literary objects in their own right. 3 Book historians over the past twenty years have drawn on paratexts to deepen our understanding of early modern printing and reading practices, showing that paratexts perform functions often central to the exchange between author and reader. 4 And with the availability of EEBO TCP and other digital resources coming at the same moment as this critical resurgence of paratext, opportunities for digital paratextual study abound. 5 In my work I often use network analysis to explore the ways in which social relations and collaborative writing practices shape literary forms and genres. I turn to paratext as a site of early modern relational thinking, combining my interest in abstract social networks with an attention to material texts. Many network projects that attempt to understand the world of early modern print use forms of metadata-information extrinsic to the text itself like titles, subject headings, and publication information-rather than the contents of the text or any part of the text. In "Metadata, Surveillance, and the Tudor State," Ruth and Sebastian Ahnert's network analysis of metadata from Tudor State Papers, the authors argue that "surprisingly deep insights can be gleaned from metadata by applying a range of easily available network analysis algorithms to a body of metadata." 6 In fact, as the Ahnerts show, metadata has often been deployed by governments, from the Tudor era to the present day, to surveil and abuse their citizens. 7 Metadata is a potentially powerful source of information, especially when used as the basis for a network, and it must be handled carefully. But thanks to the efforts of the Ahnerts and others, there is a robust recent history of responsible uses of early modern metadata in network visualization and analysis projects.
In fact there are so many examples I am unlikely to have cataloged them all. The Shakeosphere project is made from metadata: the information in the English Short Title Catalog. 8 Another network project, Six Degrees of Francis Bacon, is derived from the metacommentary on social relationships available in the secondary sources of the Oxford Dictionary of National Biography-not quite a source of metadata per se, but a method that similarly deploys information about information. 9 Even projects that derive their networks from EEBOTCP texts directly, as I do, make use of metadata in various forms. Maciej Eder 10 's study of style variation in early modern Latin relies on metadata to label and categorize its networks' texts even as its connections are drawn from text anal ysis. 11 Michael Gavin 12 's Historical Text Networks are made from a combina tion of EEBOTCP metadata (specifically imprint information) and additional information from paratexts. And paratext is precisely where my study intersects with concerns about metadata.
Paratexts occupy a liminal position between data and metadata. As part of the primary source, they can be considered data itself, but for the ways they comment on and relate to the main text, they are also a form of metadata. My project takes a specific paratextual form, the dedication, as its main source of (meta)data, leveraging its complex relationship to the text as a source of network information. By locating paratexts as a source of metadata, I do not intend to reinstitute the notion that paratext is only interesting insofar as it can tell us something about the main text. On the contrary, I hope to show that the liminal status of paratexts allows them to make meaning independent of the "main" text and that these meanings, taken in the aggregate, can productively complement our understanding of the networks of composition and book production.
Given this background on the scholarship of paratext, it seems reasonable to as sume that the best imaginative network of a text might include names from many Going back to Genette's initial taxonomy of paratext, scholars have recognized that dedicatory epistles have a "direct (economic) social function" with respect to a patron or dedicatee, but that because they are textually expanded beyond the mention of a single individual, they can include many other messages, including "information about the sources of the work, or comments on the work's form or meaning." 13 Genette acknowledges that while this function may seem to over lap with the function of prefaces and introductions, these additional messages are an "inevitable" part of dedications. dedicatory relations remain engaged with political and social events while not being tied to contemporary figures alone. This is why the imaginative networks of dedications are informative: they highlight a broad, transhistorical set of relations among books that nonetheless remains sensitive to social change.
Even within dedications alone, naming practices vary widely. Some of the nu ances of the names are tied to the part of the dedication in which they appear. In the header, where the dedicatee is addressed, names are highly formalized, often with multiple titles for each individual. In the signature, if the author chooses to include their name, it is often in an abbreviated or initialized form, even if the author's full name appears on the title page or elsewhere in the book. Both dedicator and dedicatee names can appear in the body of the dedicatory epistle, but usually in a simplified form. I have kept track of where particular names appear in dedications in order to account for this nuance. And when possible I have disambiguated names, so that a mention of "Robert, Third Earl of Essex" in a header is recognized as the same person as "Robert Devereux" mentioned in the same dedication's body text. 17 Knowing where a name appears in a dedi cation is a crucial clue to the relationship of that person to the book (particularly it can tell us whether the name refers to an author or dedicatee), and retaining this information allows for selecting subsets of nodes to compare networks of dedicatees and authors alone to those with all possible names. I do this, below, to highlight how the inclusion of names from the dedication body changes the network.
But differences in naming practices aren't limited to these larger name cate gories. Oneword names for wellknown figures are common, and there's little question about the referent of "Aquinas" or "Wycliffe." But references simply to "John," though usually naming the evangelist, are more difficult to disam biguate (more on this in the section below on data collection). And some ref erences to people, like a mention of "Virgil," may actually refer to a text or set of texts, as a kind of citation. The "people" whose names I detect are not the people themselves but textual traces, sometimes associated with a specific text. The computational method for detecting names, outlined below, cannot differ entiate a purely citational mention of Virgil from a more complex invocation of the man: though we should treat the notion that there is a clear distinction between these two categories with poststructuralist caution. In fact in these net works I am never dealing with actual persons but always with textual proxies for those people. Ultimately this is a project that deploys textual evidence to take seriously the complex relation between people and texts, and the varied naming practices in dedications is a crucial, specialized vector for this relationship.

Textual Evidence and Data Collection
The data for this project comes from EarlyPrint, the digital project on which I'm currently a collaborator. EarlyPrint extends the Early English Books Online-Text Creation Partnership (EEBOTCP) markup of a book's features with lin guistic markup that makes natural language processing (NLP) tasks easier and invites widely varied forms of analysis using the improved texts as a base. I used the "dedication" type attribute-added by EEBOTCP at the earliest stages of markup-to locate the sections of the text from which I drew names. The Ear lyPrint corpus contains linguistic markup for 52,149 of the 60,331 texts in the full EEBOTCP corpus. The remaining texts are written in languages other than English or consist largely of tables and figures that are more difficult to anno tate. Work at EarlyPrint is currently underway to annotate the approximately 8,000 remaining texts, but this relatively small number of texts, if they have dedications at all, are unlikely to change the overall arc of the findings in this project. I collected data from the text sections marked up as dedications in the 52,418 EarlyPrint texts (see tbl. 1 for more details). 18 From here the traditional approach would be to run namedentity recognition over the texts to detect personal names. But because contemporary NER tools are designed for modern language corpora, they are unreliable for highly vari able early modern texts. Irregular orthography, nonstandardized capitalization, initialism, pseudonymy, and anonymity all present distinct early modern chal lenges to computerassisted name recognition. For example, the default English model for Python NLP library spaCy, en_web_core_sm, has a precision score of 85.43% and a recall score of 85.72% when run on modern English text. 19 However, even after training the model on early modern examples, I struggled to get precision or recall above 70%. In some cases the model also struggled to identify proper nouns.
The EarlyPrint XML provided a different way forward. In fig. 1 you can see the same names from Hartlib's Macaria that I showed in the quote above, with the wordlevel markup that EarlyPrint added to EEBOTCP texts. This markup provides lemma, part of speech, and regularized spellings for every word in a text. The linguistic markup was provided by MorphAdorner, a Java NLP library developed by Philip Burns and Martin Mueller specifically for historical texts, with extensive training data and specialized code for early modern English. 20 MorphAdorner's part of speech tagging allowed me to collect proper nouns more accurately. From there I used custom rules to eliminate proper nouns that were not personal names and to capture parts of names that were not tagged as proper nouns. In a review of 100 texts, this method achieved 82.74% preci sion and 94.84% recall, results comparable to spaCy's performance on modern English text. 21 Figure 1: A sample of EarlyPrint XML from the dedication to Macaria, with the same two names highlighted, as in the page image above.
While collecting names, I used other parts of the dedication markup to cate gorize them. I kept track of where the name appeared: whether in the header, indicating it was probably the name of the dedicatee; in the signature, indicating it was the name of the author; or in the dedication body, indicating it could be the name of either author, dedicatee, or some other kind of person. I also matched names against a text's metadata, to identify the name of the author no matter where it appeared in the dedication. As I will show in the next section, these categories are useful for understanding how including names from the dedica tion body entirely changes the network. Here is a summary of the data broken down by decade and by type of name identified:  Body   14701479  17  1  0  0  6  14801489  37  1  0  3  12  14901499  48  1  0  0  14  15001509  66  0  0  0  0  15101519  64  0  0  0  0  15201529  84  7  2  12  35  15301539  240  22  3  32  202  15401549  387  80  17  139  487  15501559  371  90  28  161  525  15601569  451  140  29  250  1366  15701579  609  298  80  472  2009  15801589  855  462  139  662  2519  15901599  1046  539  157  755  2591  16001609  1518  763  221  1014  3530  16101619  1639  910  273  1271  3405  16201629  1998  843  236 1314 3332 The names summarized above were cleaned and disambiguated by hand using OpenRefine. While doing this, I was also able to eliminate additional false pos itives where my script misidentified a place name or other word as a personal name. Some ambiguities remain. For example, it is impossible to know whether references to "King Charles" refer to Charles I, Charles II, or a Charles of an other nation. Some assumptions may be made before the accession of Charles II in 1660, but especially after that date, given that I will show how frequently names of dead public figures are mentioned, it is dangerous to assume that any mention of Charles is automatically of the living king. In these cases I have done my best to aggregate names as much as possible while remaining sensitive to possible ambiguities.

Decade Total Texts Dedications Author Header
I assembled the resulting names into a bipartite, or bimodal, network-a net work with two node classes where nodes of one class can connect only to nodes of the opposite class. In this case the two node classes are names of people and the texts in which those names appear. As Gavin 22 does, I could have represented texts as edges and left only one kind of node: people. However, the bipartite representation allows me to consider relations among both texts and people. This allows me to show how dedicatory practices shape both the groups of names mentioned and their relationships to one another, as well as how it both drives and reflects relationships among texts within early modern print culture. Which is to say, modeling the data as a bipartite network allows me to focus on the print relations intrinsic to this network rather than abstract those relations to a set of relationships among people alone.

Untimeliness and Dedicatory Mentions
The resulting full network, covering texts from the full EEBOTCP range of 1473 to 1700, has 54,679 nodes and 113,862 edges. Of those nodes, 42,687 are unique names and 11,992 are texts with dedications in which those names appear. As I mentioned above, a major takeaway from this study is that thesẽ 40,000 names are not only the names of contemporary figures but the names of people from the recent and distant past, as well as fictional or legendary fig ures. And further, the untimely names of historical figures account for far more connections in the network than presentday individuals. Past studies of dedica tions at scale have focused on names of contemporary dedicators and dedicatees alone, but these data will show that, from a network analysis perspective, the most prominent names in the network refer to figures not living at the time of publication.
This phenomenon is easiest to see when looking a discrete subset of the networka subgraph-for a specific period in time. Here I use the example of 1660, which is also a crucial year for understanding how these networks change over time, which I will cover in the next section. The 1660 network has 1,496 nodes and 1,972 edges. Of the nodes, 225 are texts and 1,271 are names. Below is a bi partite layout of the 1660 subgraph, with blue nodes on top representing people and orange nodes on the bottom representing texts. Nodes are ordered left to right by their degree, and sized by degree centrality (fig. 2). To see individual names more clearly, fig. 3 shows a subdivision of the above with only the top To quickly get a sense of which names are most important in the network, we can rank them by a few basic centrality measures. In tbl. 2, I show the top twenty names ranked by degree (or number of connections), which in this network corresponds to the number of texts in which each name appears. I also show strength (or weighted degree), which corresponds to the number of times each name is mentioned, and bipartite degree centrality, the degree normalized by the number of names in the network. Of the names above, only a few clearly refer to contemporary figures-General Monck, William Prynne-and only one in the top ten is contemporary-Charles II. 23 The rest are names of saints, biblical and classical figures, and past monar chs. Importantly for the year of the Restoration of the monarchy, names of bib lical kings David and Solomon appear prominently, and I will say more about this resonance in the next section. The other names, John, Paul, Christ, Sa tan, Caesar, Elizabeth, and God, are some of the most frequently named figures across the entire corpus. It should be noted that for seventeenth century re ligious thinkers, Christ and God are very much contemporary living figures, but we would nonetheless put them in a different category from, say, a cur rently reigning monarch. These highestfrequency, noncontemporary names also take up a disproportionate amount of the total edges in the network. This is most clear when we look at the network's degree distribution for names ( fig. 4).
A degree distribution shows the number of nodes that have certain degree val ues. We can see that the degree distribution for names follows a power law, with only a few nodes of very high degree, and the vast majority of nodes with extremely low degree. This degree distribution is to be expected: it is very common among networks dealing with people and was mostly famously used to characterize the "smallworld network." 24 Not only are the most common names not from the present, but the frequent untimely names make up a much larger proportion of the network's connections than contemporary names do. The degree distribution is also evidence that this graph exhibits preferential attachment, in which certain highly connected nodes tend to accrue more con nections. And these highly connected nodes, the ones to which preferential attachment apply, are almost all names from the past.
Of the 1,496 names in the 1660 network, only 429 refer to currently living people, less than a third. And these 429 names only account for 523 of the 1,972 mentions recorded in the network as edges, barely over a quarter. That doesn't mean that these names are insignificant or inconsequential. Not only does this list include the aforementioned Charles II, it also includes popular po To understand just how much these names from the past affect the network, I compare different ways of assembling the network using the parts of the printed dedication as a guide. By limiting the network to only those names which appear in the header or signature of the dedication, I can construct a subgraph that is much closer to a traditional picture of a dedicator/dedicatee graph. fig. 5 shows a visualization of that graph. In some respects this graph is similar to the full 1660 graph. For instance its density-the ratio of the number of edges to the number of all possible edgesis almost identical, approximately 0.006 for both graphs, which is about 0.002 more dense than random graphs of the same size. 25 Density tells us how con nected a network is, how close it is to being a complete graph, one in which all possible edges exist. In this case, both graphs are slightly more dense than would be expected for random graphs of their size, but neither is that much more dense than the other, even though the graphs are different sizes. This graph is a good deal smaller than the full 1660 graph: it has only 553 edges and 647 nodes, and of those nodes 444 are names and 203 are texts. Notice that the 444 name count in this graph is quite close to the count of total contemporary names in 1660: 429. And the difference is more apparent when you look at the top 20 names in this graph (tbl. 3).  most entirely people from the present. Prynne, for example, moved far up the list, even though he's mentioned one fewer time when only counting mentions in the header or signature. Overall, this network gives us a much clearer sense of the presentday print networks, but it loses the texture of other references that are equally relevant to the ways authors are presenting their books. Both forms of the network are valuable, but having access to the many layers of non contemporary naming highlights the extent to which early modern authors sit uated their books within networks that stretch far beyond the contemporary. 26 And we need not stop at categorizing the names as simply contemporary and noncontemporary. As I've already suggested, the names fit into many over lapping categories, particularly names of figures from the past: biblical names, classical figures, names of dead monarchs and politicians, dead authors and artists, etc. A detailed taxonomy of names is outside of the scope of this study but would likely be essential for any study that wishes to take up these methods for an analysis of names outside of dedications, in the full text. As an illus tration of what such a taxonomy might do, and to demonstrate the significant amount of information added by searching for names in the body of the dedi cation (rather than just the header or signature), I have created and labeled six basic categories for names in the 1660 graph: living people (L), historical people (H), religious/biblical figures (R), fictional characters (F), and unknown/unclear (U). These categories are by necessity fuzzy and only intended to give a rough overview of the terrain of naming. 27 But by breaking down names in this way, it is easier to see the ways that names in the body of the dedication allow us to get a fuller picture of the untimeliness of dedicatory networks. First, tbl. 4 shows the number of names in each category and the average degree of those names. Most of the names in the network are early modern, but only narrowly so. Both the historical and religious/biblical categories are quite large, and with average degrees larger than the "living" category-which here includes persons alive at any point during the early modern period, roughly 15001700. The average degree for the religious category is much larger, because of a few names (God, Jesus) with a disproportionately high number of mentions. And because the living category includes the names of anyone alive at any point in the early modern period (not just in 1660), some of the highest frequency names in this category are from the recent past: Martin Luther, Queen Elizabeth, and King James, to name a few. The next question to ask is where certain types of names are coming from: whether from the header of the dedication or its body (tbl. 5).  All but a handful of historical, religious, and fictional names appear only in the body of the dedication, as we might expect. Dedication body text is adding most of the information about relationships to the distant past, religion, or fictional abstractions, classes of relationships which add new valences to our understand ing of dedicatory practice. Simply put, the dedication was a place where these influences and values were set out, alongside personal and economic relation ships to living persons. But the body of the dedication also adds to our under standing of how contemporary or nearcontemporary names are deployed.
More names of the living and recently deceased appear in the header or signature of the dedication, meaning that a little over two thirds of the names likely come from authors or patrons. But that leaves a sizable number in the living or recent category that only appear in the body. The body of the dedication adds lots of early modern names to the network, information about connection and associ ation that wouldn't be available if we looked only at authors and patrons. The average degree of names in this category that appear in the header or signature of a dedication is 1.29, while the average for "living" names in the body is 1. 16. These figures being so close to each other (and so close to one) indicates that neither authors and patrons nor other individuals are disproportionately driving this part of the network: names that appear purely in the body of the dedication are only slightly less frequentlymentioned on average. Authors and patrons are only telling part of the story of connection among texts.
Using these categories, we can easily construct a network that focuses exclu sively on more immediate relationships, acknowledging that an author's rela tionship with God or with Cicero is different from a relationship to a living monarch or more contemporary writer. A visualization of the graph of names only from the "Living" category shows a network with fewer nodes and edges overall ( fig. 6). The nodes have been rescaled by degree, but you can see a sim ilar degree distribution to the other graphs. A few names received lots of men tions, and the rest just one or two. But the highest degree names in the graph are markedly changed. Without religious and historical figures, the highestdegree node is the newly restored monarch, Charles II. Previous monarchs (James,

J O U R N A L O F C U LT U R A L A N A LY T I C S
Charles I, Elizabeth) and prolific authors (William Prynne, Thomas Aleyn) also make the top twenty, as does Martin Luther. Though this list emphasizes the prominence of living figures, like Charles II and General Monck, even when fo cused on "real" relationships the network remains less focused on timely names then one might expect. Even with the distant past taken out, the recent past has a strong hold on dedicatory relations. There are other revealing ways of categorizing nodes and understanding their impact on dedications. Comparisons of nodelevel metrics, especially central ity measures, show how different kinds of names play different structural roles within the network. For example, the two lists above show both degree-the number of texts in which a name appears-and strength-the number of times a name appears. A ratio of these measures will show which names appear many times in just a few texts. 28  Unsurprisingly, there are a lot of names in this table that appear in just one text, but multiple times-a figure with whom a certain author may be preoccupied. But a few figures-Jeroboam, St. Ambrose-appear in multiple texts but still have high strengthtodegree ratios. These names, of biblical figures and saints, are the kind of people whose writings one might discuss at length or about whom one might tell a story, as opposed to a person for whom a single reference may be enough.
A different set of names can be discovered by examining ratios with between ness centrality-a relative measure of the number of shortest paths that pass through a particular node in the network. Betweenness centrality, when com bined with degree centrality, provides a way of seeing which nodes stand be tween sections or communities in the network without themselves having many connections. These "bridges" are crucial for network cohesion. tbl. 7 shows names with high betweenness centrality and low degree centrality. These names are much more political than the religious and biblical names from the strengthtodegree list. Maecenas, counselor to Augustus, is at the top, along with Roman emperor Vespasian. Medical thinkers Hippocrates and Galen also make the list. 29 The bridges in this network may be names that can be discussed differently depending on the context-an author on one end of the political spec trum, a Royalist, might invoke Maecenas as an example of how well monarchy functions, while a Republican might invoke Maecenas to make the opposite The 1660 example shows that noncontemporary names make up a vast, impor tant part of the dedicatory networks, which in turn emphasizes that dedications are about a lot more than simply who the patron was. In addition to these names being important within the context of a specific year, by tracking the network over time we can see the way noncontemporary names are consistently de ployed across the period. fig. 7 is a stacked time series of the most common names in the whole network, and their degree centrality over time in networks grouped by fiveyear spans. I use bipartite degree centrality rather than raw degree here because it is normalized to account for the different sizes of the fiveyear networks.

J O U R N A L O F C U LT U R A L A N A LY T I C S
Except for the spikes caused by sparse data in the early part of the period, the relatively high degree centrality of these top nodes stays fairly consistent over time. The final name in this graph, King James, spikes in 1604 when James VI of Scotland takes the throne as James I of England. But after that, his pat tern follows other popular names like Moses, Solomon and Augustine-James enters a rotation of popular names that authors use even though they are not contemporary figures.
The same phenomenon is apparent if we look at a time series for Elizabeth I ( fig. 8). Variations on her name appear throughout the corpus, beginning as Princess Elizabeth before her reign and hitting a high point of degree centrality (0.08) in the 158085 graph. Significantly, Elizabeth's influence does not de crease as much after her death as one might imagine: in the decade immediately following her death, her degree centrality stays at about the same levels as it did in the final years of her reign. Afterward, it expectedly drops off somewhat, but crucially Elizabeth's influence in the network does not shrink to nothing. Her name seems to be invoked slightly more during periods of political transition: during the tumultuous final years of Charles I's reign, during the first years of the Restoration of Charles II, and in the run up to the Glorious Revolution of 1688. As a beloved political figure from the recent past, it seems that there is a habit of invoking Elizabeth at moments of political uncertainty. Though these kinds of mentions aren't as frequent as the mentions of her name during her reign, they nonetheless speak to the ways that noncontemporary naming is an important mode of dedicatory mention, driven as much by political and social concerns as by economic ones. The social concerns of dedicatory practice and the economic ones are inextricably linked-tracking dedicatory mentions over time gives a better sense of the way these two drivers for dedicatory naming overlap.
Names from the distant past appear with more regularity across the corpus than contemporary figures almost as a rule. I see this as an important corrective to the practice of using dedications to tell us only about authorpatron relations. It is now a commonplace in studies of reception to recognize that readers do not choose only from a crop of books newly published in a given era, but from a variety of books published before then. Reading practices are often backward looking, and any one person's reading can include books published across a long span of time. 30 Likewise with dedicatory practice, while early modern books are almost always dedicated to a living person, the names that appear in dedications can include a range of influences, citations, and relations from across a very wide span. A dedication-whether written before or after the main text-works along with other paratexts to signal the reader to the book's wide imaginative network. The invocation of names in the dedication allows readers to situate a book and its author within a network of influences, citations, and J O U R N A L O F C U LT U R A L A N A LY T I C S allusions, and that network is far broader than only contemporary figures.
However, these findings do not suggest a flattening effect, where the same names are repeated over and over without differentiation. Though some very common names don't vary much, there is quite a lot of variation over time even with noncontemporary names. A name like "Maecenas" or "King David" might suddenly surface in the network at a particular time for a particular reason. We see this with Elizabeth as well-though hers is a popular name to invoke, there are specific moments in which her name is used more often. In the next section I explore some of the political causes of this kind of change in dedicatory networks over time.

Political Crisis and Change Over Time
From the evidence above we can conclude that dedicatory networks are not bound by the relationships of the present day. On the contrary, one of the clear est features of these networks is the high centrality of names from the distant past. However it would be incorrect to extend that conclusion to say that the historical context of a dedication doesn't matter. Even though dedicators are invoking names from a wide swathe of history, they are doing so in response to the economic, social, and political concerns at the moment of publication. This is clear from Genette's initial observation of dedicatory epistles-that they contain a justification for the book with respect to the dedicatee, the imagined reader, and a broader societal context. Though the networks in this study show high centrality for names from the past, the deployment of contemporary and historical names alike is very responsive to immediate social and political con cerns. As I will demonstrate in this section, the networks that span the 1630s through the 1660s show a series of changes in network structure and name use that track with the tumultuous political and social landscape of those decades. In response to civil war, multiple changes in government, and shifts in the reg ulation of print, dedication networks show shifts in name mention that respond to and comment on this period of political crisis. In testing I found this grouping to be more sensitive to changes over time than tenyear spans and without the artificial increases in degree for individual nodes caused by spans smaller than five years. To understand network size and structure over time, I calculated the number of nodes and edges in each graph, as well as the average degree for each node set, the number of components, bipartite density, and the average of the bipartite clustering coefficients.  In addition to the raw calculations in tbl. 8, I also included comparisons for den sity and average clustering coefficient against the average of five random graphs of the same size in tbl. 9. Because these metrics are dependent on network size, they can't be directly compared across graphs of different size. Instead it is preferable to compare the difference between the metric and the average of the same metrics from samesize random graphs, which is what I've done here. However, before I discuss density and clustering coefficients, I want to empha size changes in network size over time. of EEBO's corpus in the 1640s. However, there's no evident 1640s spike in this network. That is likely because most of the political pamphlets and other kinds of cheap print produced in the 1640s did not have dedications. Dedica tions tend to appear more often in longer, more expensivelyproduced volumes. This graph suggests that the number of texts with dedications stays relatively consistent across this fortyyear period.
In addition to understanding the change in size of these networks over time, I use bipartite density and clustering coefficients to observe structural change in the networks. Bipartite density, as in a unipartite graph, is a ratio of the total edges in the graph to the number of all possible edges; it tells us how close the graph is to being complete. As you can observe in tbl. 9, all of these graphs are fairly sparse, but they are also of expected density when compared to ran dom preferential attachment graphs. That is to say, these networks are about as dense as a typical graph with the same number of names and the same de gree distribution. A bipartite clustering coefficient is a more local measure of density. In a unipartite network, it measures how often the neighbors of a node are connected to one another. In a bipartite network the measure is similar: it shows how often the second order neighbors of a node are connected to the same node. A clustering coefficient tells you how dense the immediate area around a single node is. Looking at the average clustering coefficient of a graph gives a sense of how closed a network's loops are and how closely related its nodes. In networks with lots of disconnected components, density will register as sparse even if some individual components are more dense; average clustering coeffi cient will register this local density. For both of these measures, we can chart how different each metric is from the expected average, over time.
As I've already suggested, compared to average networks, the density of these networks does not change much over time ( fig. 10). They all have a fairly high number of components, and are therefore all relatively sparse. And again, this is typical in preferential attachment graphs-the densities are not far off from the average for their size. There may be a slight rise in density in the late 1630s and 1640s, but it's difficult to see from looking at density alone. However,  the local density expressed by average clustering coefficient tells a different story. Graphs have significantly higher average clustering coefficients, relative to random graphs of the same size, in the two fiveyear spans of the 1640s. What shows up as a slight difference in density is much more pronounced in average clustering coefficient. This suggests that the graphs of the Interregnum, at least in its early years, have tighterknit communities, with the same sets of names appearing together in the same texts more frequently. And while this change takes place during and immediately after the Civil War, it doesn't seem to last: during the Protectorate and the Restoration, density and average clustering coefficient return to prewar levels.
To explain some of these structural changes, especially the increased density and clustering coefficients of the graph during the early part of the Interregnum, I looked at some of the most prominent names from the graphs of this period. The table in fig. 12 shows the top 20 names by degree for each half decade, with the  Observations from these degree rank lists are consistent with assumptions we might make based on historical facts. Names of monarchs, especially Charles but James and Elizabeth as well, have higher rank when there is a monarch on the throne, before 1640 and after 1660. In the 164549 graph, the years directly preceding the execution of Charles I, references to "King Charles" drop to number 8, nearly tied with "Thomas Fairax," the leader of the Republican forces, at number 9. The lists are sensitive to changes in leadership, which is consistent with my findings about dedicatory practice in the previous section: whether or not an author dedicates their text to the leader of the nation, it is likely that leader will be mentioned at some point in the dedicatory epistle. Indeed, in the 165559, the final years of the Protectorate, the title "Protector" appears in the top 20 names at number 9, the only time it appears in any of these lists.
But in the Interregnum graphs, monarch names are not simply replaced by Re publican leaders. Even Fairfax and Cromwell never achieve the heights in these networks that monarchs do in the 1630s and 1660s. Instead, the networks further consolidate around the already prominent religious figures. Which is to say, the preferential attachment toward names like "St. Paul," "St. John," and "Moses" becomes more pronounced. This may explain why the Interregnum graphs see an increase in density and average clustering coefficient-references to the con temporary political figures fall away somewhat, and references to names of pop ular religious figures increase as a result. Where the split is more even before and after the Interregnum, the 1640s and 1650s see a moment where dedicatory practice strays even farther from the contemporary and clings to names from the distant past. With the court dissolved and many aristocrats living in exile during this period, such a shift makes sense. The normal social structures that govern dedicatory practice have been upended.
Changes in other names on these lists, which do not correspond directly to changes in leadership, suggest shifts in discourse which also stem from political crises. References to King David are popular across the corpus, as we saw in the previous section, but those references get more central to the graphs during and after the Revolution. The name "David" could signify several things, including increased quotation of and commentary on the Psalms. But it could also con note increasing interest in anointed kingship, taken up by both monarchy's pro ponents and its detractors. Indeed, the name David appears most frequently (in 97 texts) in the fiveyear period leading up to Charles II's Restoration, and the name has its highest degree rank (third, ahead of Sts. Paul and John, Solomon, Moses, and Charles II himself) in the fiveyear period immediately following the Restoration. Likewise, while "Satan" is a common name throughout the pe riod, the name has a modest rank of 12th and 14th in the 163034 and 163539 ranges, respectively. But from 1640 to 1660, "Satan" doesn't rank lower than 6th. Part of the name "Satan"'s popularity may be the consolidation around pop ular religious figures in the absence of political ones that I mentioned above. But it is notable that Satan in particular rises in degree rank, as opposed to Moses, Abraham, or Augustine. This indicates that a particular kind of religious lan guage, perhaps of an apocalyptic flavor, has taken hold during the period, and it will also no doubt intrigue Milton scholars to know of Satan's increasing cen trality in the years leading up to the publication of Paradise Lost.
density measures for two graphs of the same size. Instead, best practice is to generate random graphs of similar size, and compare the network to the average metrics of those graphs. I've done that here by generated two 10graph samples similar in size to the full 1660 graph and the header/signature 1660 graph, respectively, using NetworkX's bipartite preferential attachment model. 26 Degree centrality in the headers and signatures alone is strongly but not perfectly correlated with degree centrality in the network taken from the full dedication (r=.674). Though degree centrality in the full network is similar to that in the header/signature only network for nodes that appear in both, there are only 444 names in the header/signature network compared to 1271 names in the full 1660 network. Though many of these nodes are low degree, they represent a large amount of information about naming that is added by looking at names beyond the header and signature alone. And a few are quite high in degree: the biblical king David, for instance, has a degree of 26 and a strength of 57 in the full 1660 graph, making his the third most frequent name overall. But his name only appears in the body of dedications, never in the header or signature.
27 "Living people" includes anyone who lived during the early modern period, roughly 15001700, i.e. it includes more than simply people alive in 1660 and encompasses both the living and what we might call the "recent past." "Religious/biblical" includes deity names (God, Jesus) and saints and other figures that appear in the bible (Peter, Moses). (Phrases like "Grace of God" and "God willing" were filtered out at an earlier stage of the process.) Saints who do not appear in the bible but were real historical people (who usually wrote texts being cited in a dedication) appear alongside Greek and Roman philosophers and political leaders, as well as anyone who lived before~1500, in the "historical people" category (Augustine, Anselm, Cicero, Tertullian, etc.). "Fictional characters" includes mythological figures (Athena, Apollo) alongside characters from literary works. Any names that could not be identified with a reasonable degree of certainty were labeled with a U.
28 This is nearly identical to document frequency/term frequency ratios commonly used in text analysis. 29 This table includes one item, "Democracy," which is clearly not a personal name. I eliminated the majority of the highest frequency mistakes in name detection during hand curation, but a small number of lowfrequency names terms like this remain in the data. Eliminating the rest of these lowfrequency occurrences would take a great deal of time and wouldn't change the overall trends in the data very much. They are infrequent, and I do not believe that they skew the overall trends in the networks. I remain satisfied by the precision and recall scores I cited in the last section, which are consistent with stateoftheart namedentity recognition.  32 This study focuses on individual personal names rather than collective names. There are dedications to Par liament throughout the period, and it could be useful to track how mentions of Parliament and other organizations changes during this period. But such inquiry is outside the scope of this essay.