Language in English Civil War Title Pages

The CONCLUSIONS drawn Therefrom

Mark-Up and Schema: Structural Commonalities and Variations

We began our analysis of the title page corpus through coordinated XML mark up of the basic structural and literary elements of each text. After thorough mark up, we planned to run XQuery searches to determine any correlations between a title page's political advocacy and its literary conventions (such as epigraphs, allusions and Latinity), presentational choices (such as the use of 'KING' as opposed to 'king'), and structural components (such as subtitles and authorship attributions). Ultimately, we succeeded in creating a sufficiently consistent mark up of the documents for these purposes. In the process of coordinating the mark up with an effectively constrained schema, however, we reached several interesting conclusions on the nature of the title page as a genre.

We concluded that the conventions of the title page allowed ample variability for the author and printer, with very little uniformity across documents' structural arrangements. We explain this by the dramatic increase in the number and category of printed title pages in such a short period of time, which would strain any existing conventions. For example, we found that while authors are always attributed on title pages of books in contemporary publishing, it was much less common in mid-seventeenth century Britain - only 21 of the 37 documents (57%) in our corpus contained any authorial attribution. For those that did contain an attribution, only 13 of the 21 (62% of attributions, 35% of title pages) placed the attribution within a by-line distinct from the title itself - as is standard in contemporary printing. We encountered similar irregularities with the structural elements of 'Titles,' which in many cases proved so extensive that they required internal division. We decided that the components of a title could be best categorized by their function, that is, what they express to a reader. We divided them within a &type attribute as "subtitle" which gives a secondary introduction to the text, "summary" which describes the contents in more depth, and "purpose" which states the intention of the text. The commonality of these types varied drastically, and their categorization proved inherently subjective. Needless to say, our schema allowed a considerable flexibility, and therefore could not constrain markup.

By noting the complexity and variability of these documents' structures, we also concluded that they cannot satisfactorily conform to the TEI standards for title pages . These standards parallel our standards in their division of title components by function, and in their treatment of particular literary devices - specifically 'epigraphs.' Their applicability, however, falters in their protocol for publication data, specifically, the date of print, the location of print, the printer and the bookseller. The first three of these elements fall within the 'tei:docImprint,' yet it does not distinguish them within the element. This disallows more sophisticated treatment of continuities amongst a particular printer's output, or of a geographic analysis of print-shop locations and political persuasion, for instance. Moreover, it disregards the common convention of including the designated bookseller for the work - usually introduced with the phrase 'printed for'. This is not conducive to studies of the complex economic conditions of printing and bookselling in an age of rampant literary piracy.

Topic Modelling: Pamphlet Categories and Genres

In order to map any correlations between title page styles and structures and the ideologies they represent, we required a way to rationally categorize those title pages within certain ideological and thematic headings. That is, if we wanted to study whether 'parliamentarian' texts contained more epigraphs than 'royalist' texts, we would need a way to make that distinction without looking into the document following the title page. Moreover, the possibility of anachronistic readings of literary categories necessitated a more objective approach to this type of interpretation. We decided that the best method for this approach was in 'Topic Modelling software, specifically the 'MALLET' Program.

By statistically analyzing all words in a plain text, this program can conclude the prevalence of words and their relationships to other words within texts. Specifically, it can generate 'topics,' composed of words, which best express the consistency of those words mutually occurring within a document.

We inputed all 37 title pages in this program to generate an output of ten topics. We determined that ten topics could provide the variability to capture a corpus of very diverse texts, but constrained enough to respond to those texts' brevity. We based this decision on the 40 topics generated in the 'Mining the Dispatch' program, which analyzed a considerably larger corpus of 112,000 documents of comparable dissimilarity. We decided to maintain the case-sensitivity, since we already that the different forms of presentation were semantically meaningful in a document used for largely advertising purposes. We generated the following ten topics, which when clicked display a graph of pamphlets by percentage in a new window:

With this information, we then applied that subjective, humanistic interpretation to note any meaningful similarities amongst the words within each topic. These similarities could correlate either to a particular subject matter or political persuasion. We noted the following:

Topic 0: Subject - "Political"
Topic 3: Persuasion -"Royalist"
Topic 6: Subject - "Religion"
Topic 8: Subject - "Academic"

The remaining topics evinced no meaningful similarities amongst their words.

The MALLET program then produces the results of how strongly each topic composes each document. The results yielded this chart of each document, and the strength of each topic in descending order:

Evidently, there is only marginal variation in word -selection from document to document, and from category to category. We determined this by the minimal variations in in topic strength within title pages and between title pages. The strongest topic correlation is 25% (topic 5 in pamphlet 15), and the most considerable difference between the strongest topic and the weakest is 19 percentage points (pamphlet 15). The average difference between strongest and weakest stood at 7.1 points.

We can offer two possible explanations for these results, neither of which exclude the other. First, our very limited sample taken from such a considerable corpus of text limits the ability to extract meaningful differences. When dealing with only 37 title pages taken from a corpus that contains 8864 documents (that is, every document published between 1642 and 1649 within the Thomason Tracts collection), and when those documents contain minimal text, the results of such a statistical analysis will produce not only skewed results, but marginally differing results. Such a method could have skewed results by producing topics which simply recreated particular texts which supplied the most words to the entire corpus - a possible explanation for the topic distribution of pamphlet 15, a comparatively lengthy outlier.

Second, the words on title pages themselves cannot inform us of a particular title page's political persuasion or subject matter. What is more, there may not have been clearly defined differences in subject materials within the texts themselves. While some may tend to differentiate, say 'religious' from 'political' texts as distinct categories, such a distinction may simply not have existed in the mid-seventeenth century. If this were the case, we could explain it by means of the earlier explanation proffered for the lack of structural continuities; that is, the sudden expansion of literary output in such a short period of time and the inability for established distinctions to accommodate so many new title pages.

We yielded the second conclusion from this data by measuring the topic strength of those topics which we had previously interpreted as categorically significant. We predicted that these topics' salience as common categories, rather than arbitrary assortments, would yield a greater number of documents most strongly represented by these categories, and possibly a greater topic strength within each document. We determined this by locating those documents whose most informative topic was one of these. We then averaged the topic strength of each topic within that document, and the results are as follows:

From these conclusions, it is clear that these topics do not represent documents more strongly than those with no discernible consistency. We conclude, then, that these interpretations do not meaninfully represent the topics, or that no topics could satisfactorily differentiate English title pages of the 1640s