Balisage Paper: Synthesis and Sustainability: The Evolution of Markup and Toolsets in the Women Writers Project

Introduction

The Women Writers Project (WWP) has been working with text markup for nearly 40 years, as part of its founding mission to recover and republish the works of early women writers. Founded at nearly the same time as the establishment of the SGML standard and of the Text Encoding Initiative (TEI), the WWP has by necessity treated its work as long-term active research undertaking that concerns not only these evolving markup systems but also the complex web of tools through which text markup can be created, assessed, and put into practical use for publication and analysis. The WWP has sought not only to build a stable, sustainable working system, but also to keep pace with new developments in both markup and tools, and explore their implications for research on early women’s writing. The intertwined history and co-evolution of these two sets of practices within the WWP’s many decades of work thus far offers a valuable perspective on the history of scholarly usage of markup in the humanities. This paper will explore the evolution of the WWP’s theory and practice of markup, from its pre-TEI origins to our present-day usage; as well as the evolution of the project’s publication systems, from its first forays into digital publication through DynaWeb to our planned implementation of a static-site framework that responds to the recommendations of the Endings Principles. We also consider the reciprocal pressure that tools and markup exert on each other, and the nuanced ways in which markup responds to the changing intellectual paradigms proposed by successive generations of tools and technologies (and vice versa).

Evolution of the WWP’s theory and practice of markup

The WWP’s earliest markup practices were anchored in typesetting tools: student encoders starting in 1988 used Waterloo Script to capture functional textual units and formatting details. This approach was not guided by an explicit theory of markup, but it did embody certain assumptions about the role that markup could play in representing primary source documents (still a somewhat novel role for markup at that time). The now fundamental idea that markup should emphasize semantics rather than solely formatting was already present in the ways that Waterloo Script supported the identification of textual components (headings, paragraphs, etc.) in a way that was separable from their appearance, and this was especially crucial for primary source materials whose actual appearance on the page might vary considerably from text to text. And although this first stage of the WWP’s practice did not at all anticipate what the TEI would come to be, it did lay a groundwork that was at least amenable to an eventual transition to the TEI once P1 and P2 were available to demonstrate what a more thoroughly semantic approach could look like.

We can see early traces of this shift from typesetting (and an initial orientation towards ideas of a “digital facsimile”) in the WWP’s initial debates about whether a glyph in the source text should be represented using a character that resembles it visually, or represents it informationally. For instance, the earliest WWP transcriptions used the letter I to represent the numeral 1 (which in older typography has horizontal serifs) on the theory of greater visual correspondence to the original. This practice, which was quickly discarded, yielded a valuable clarification that seems obvious in retrospect: the appearance of the original glyph is separable from (and for the WWP, less informationally important than) the meaning of that glyph within the overall notational system. Interestingly, early modern typesetters themselves sometimes used a character that resembled the one intended—either accidentally (if a piece of type had made its way into the wrong box), or deliberately if they had run out of the appropriate character: for instance, the letters “u” and “n” (which when inverted are often indistinguishable from each other), the substitution of zero for a capital O, or the use of “vv” for “w”. In these cases, the WWP chose to retain the original glyphs (even though they were in effect a “facsimile” of something else) while also providing a regularized transcription reflecting the intended meaning.

The WWP’s thinking about these kinds of cases was shaped by concerns among humanities scholars about loss of information about the physical source, and a resulting desire for visual fidelity to the text. Seen from that perspective (and in the absence of Unicode), the use of a capital I to represent a numeral 1 could appear defensible on visual grounds. But very early on, the project identified that resemblance (unsurprisingly) as a fallacy—dependent on typeface and likely to lead to all sorts of future practical and conceptual problems. The idea of markup as a “digital facsimile” gave way quickly to the idea of markup as a more informational representation of document structure—an approach which has served us well in the longer term because of its greater robustness of display and behavior. The project continued to be concerned with ways of representing the material aspects of documents, with each encoded transcription derived from an identifiable physical copy, and capturing aspects of formatting and material structure. But the TEI model demonstrated how that could be done explicitly through the markup itself: for example, through the @rend attribute or through mechanisms like <sic>:

                        
<sic corr="O">0</sic>

The WWP’s early work on rendition ladders provided an opportunity to explore how far the material and visual properties of texts could be formalized for descriptive (rather than output) purposes.

In the early days of the TEI in the 1990s, the WWP was a typical early adopter in two senses. With its mission of creating a large collection of early women’s writing in English, the project emphasized both comprehensiveness and the recovery of inaccessible materials. But these two goals proved over time to pull in different directions, as the evolution of the TEI as a scholarly technology revealed different ways for digital resources to function as research objects. The idea of creating reference corpora or comprehensive research collections was an early and significant driver for the TEI and is exemplified by projects like Perseus, ARTFL, and the early documentary editing projects that constituted the Model Editions Partnership. Such efforts signalled the importance of data interoperability and of markup systems that could work towards that end: schemas that could strongly reinforce consistency and “conformance” and provide a clear model for tools to operate on. The WWP’s work was certainly animated to some degree by motivations towards interoperability; the project explored collaborative approaches to the encoding of personal names with several other projects focused on women’s writing, and placed a high value on following emerging standard practices when possible. But an important early element of the project’s mission was to treat early women’s writing and early printed books as potential sources of complication and resistance to standardization.

With the release of TEI P5 in 2007, the function of customization changed fundamentally: no longer as a laborious way of departing from the standard, but as a more or less required aspect of using the Guidelines, a tool that was in everyone’s hands. As the “Design Principles” section of TEI Guidelines, “About These Guidelines” describes the shift:

Another crucial step in the evolution of the WWP’s use of markup—also connected with the release of TEI P5—was the availability of markup structures akin to linked open data: the ’ographies^[1] and their relationship to the emergence of RDF and technologies like XPath that enabled linking to precise locations within XML documents. For the WWP, this launched a shift away from a sole emphasis on documents, to representing a larger universe of entities in which those documents are one meaningful component. The WWP had already begun using internal authority control to manage a database of persons referenced in WWP texts, starting in 1996 with a formal keying system developed by Syd Bauman. In 2008 the WWP received an NEH Digital Humanities Start-up grant to explore the complexities of managing personographies in the context of early modern women’s writing (see Melson and Flanders 2010). Over time, the WWP has developed systematic data on people (with particular emphasis on authors featured in Women Writers in Review and Women Writers in Context), texts (featured in Women Writers: Intertextual Networks), and even to some extent events (which are featured in timelines as part of Women Writers in Context). This work diversifies the role of markup for the WWP, going beyond modeling documents to creating data structures that are independent of documents but constitute an important contextualization mechanism for them.

The final niche in the WWP’s evolving use of markup brings us full circle to the roots of SGML as a way of modeling data in which “text” faces Janus-like towards the domains of document and data and operates effectively in both spaces. Over time the WWP has found important uses for lightweight, systematic, highly consistent, “data-like” markup that anticipates interfaces that are focused on retrieval and analysis rather than on “reading” in the traditional sense (as in Women Writers Online’s presentation of primary sources). The project uses TEI for its encoding documentation, for the periodical review documents in Women Writers in Review, and for the essays in Women Writers in Context, and in all three cases the approach is minimalist and functionalist rather than being animated by a research-oriented philosophy of full textual representation (as in the WWP textbase).

Evolution of toolsets

The WWP has had an unusually long history that traverses many generations of SGML and XML publication tools, and the project’s data has presented some valuable challenges and edge cases that have tested and revealed the limits of the tools it has used. The project’s main publication, Women Writers Online, has used four different platforms since its first publication in 1999:

The initial digital publication of Women Writers Online in 1999 showcased Renaissance Women Online (RWO), a subset of WWO representing works originally published between 1500 and 1670. RWO consisted of two complementary forms of interface. At the time, WWP Electronic Publications Editor Paul Caton described these interfaces:

The initial WWO and RWO collections were further defined by robust search with keywords-in-context; tables of contents for user navigation within a given work; as well as contextual materials such as short summaries, scholarly introductions to works, and essays on topics and collections.^[2] The following screenshots show some of the details of these interface features:

Figure 2

The first version of WWO had a table of contents with expandable items showing the major sections of each work.

Figure 3

The search interface for the first version of WWO included context-sensitivity and wildcards.

Figure 4

The early WWO search results included a keyword-in-context option, with links to both the inidividual hit and the text as a whole.

By 2003 the WWP began seeking a successor to DynaWeb for both technical and administrative reasons. The Inso Corporation had been bought by another company, which altered both the level of support and also the license terms in ways that made it increasingly untenable to use DynaWeb for a small academic project. In 2004 the project began experimenting with Philologic, developed by Mark Olsen at the University of Chicago, and in 2005, the WWP released an updated WWO running on Philologic. Philologic placed its emphasis on speed and power; it was database-driven and selective in its indexing, so that it was able to provide very fast search results as well as context-sensitive searching. Its search also included fuzzy matching, which made it adept at handling the highly variable spelling in older WWO texts. This system served us well for several years, but by 2011 the project was seeking a more modern XML framework that would support more flexible experimentation with interface, responding to the significant expansion of interface paradigms as “digital humanities” became a widespread academic domain. The screenshots below illustrate some of these features:

Figure 5

The second iteration of the WWO search offered fuzzy matching and the ability to refine search results.

Figure 6

The second iteration of WWO also offered collocations.

Starting in 2012, Women Writers Online was served out of XTF (eXtensible Text Framework), a Java- and XSLT-based publication system created by the California Digital Library. The WWP staff heavily customized XTF’s stylesheets in order to produce indexable versions of the WWO documents, as well as HTML representations for display. On the front end, Women Writers Online was reimagined as a dynamic, three-pane interface, with users seamlessly flowing from browsing to searching to reading a document. Within this design, WWO offered users a sense that searching/filtering, visualizing, and reading are really three manifestations or framings of the same thing, at different levels of scale. This approach contrasted strongly with the traditional (at the time) search model in which users would first specify a set of search parameters and then execute a search to receive results—with the need to go back to the start of the process if the results were not what the user wanted. (A few years later, Stephen Ramsay’s essay on “The Hermeneutics of Screwing Around” [Ramsay 2010] captured the element of play and exploration that such “pipeline” systems foreclose.) By searching and filtering across WWO, one can “read” the corpus as a whole as a spread of dots on a timeline, or as a list of matching search results — keywords in context for multiple WWO documents. One can then focus in on a more specific area (e.g., a specific genre or time period) and finally on a specific text. Although this interface doesn’t fully implement this idea, it introduces it and makes it possible. And within this interactive paradigm of usage, markup has a very significant role to play: it makes all features of the text potentially discoverable and operable (addressable to use Michael Witmore’s term [Witmore 2012]), not only as hooks for formatting and presentation but also as informational filters and points of correlation. However, for the WWP, the challenge has been to anticipate and present those opportunities to users without requiring them, in effect, to write XPath expressions.

Around the same time, the WWP universe began to expand beyond Women Writers Online. The WWP produced Women Writers in Context (WWiC), a collection of modern scholarly essays providing crucial background for the documents, authors, and genres found in WWO. Later, Women Writers in Review (WWiR) made contemporary reviews of WWO authors’ works available, and referenceable from WWO itself. These two endeavors aim to flesh out the milieu of WWO authors. The sites link to each other to aid users’ exploration, but are not integrated directly into a single interface. Unlike WWO, the sister sites are public resources, available to anyone, regardless of WWO subscription. Their documents are lightly customized TEI, much more geared towards supporting collection-level metadata, presentation and quick publication turnaround than research artifacts. Their interfaces in turn prioritize findability and the ability of users to follow connections between resources.

Women Writers: Intertextual Networks (WW:IN) sought to compile significant amounts of data from TEI sources for exploration and research. The data was drawn from two primary sources: (1) the WWO documents, which were marked-up to add references to (2) entries within a separate, TEI-encoded bibliography. Unlike its sister sites, however, the WW:IN interface was designed to do much more work on the server-side, with Javascript limited to functions that would make the site more interactive. An EXPath application, housed in the XML database eXist-DB, would apply users’ desired filters to return customized datasets in full HTML responses. The same API that returned JSON or XML would also serve out the webpages, putting far less pressure on users’ devices.

This succession of different tools has been challenging, in that it has required periodic episodes of substantial redevelopment. However, those transitions have also given the project an important push at each point to rethink how the interface could best serve readers and best express the capacities of the data. And they also prevented the project from experiencing the publication platform as a fixed horizon of possibility to which the data should adapt. Instead, the project has developed an extensive set of pre-publication processes that take the WWP’s source data and transform it as necessary for the specific foibles and requirements of the current tool.

For example, like many TEI projects, the WWP makes frequent use of intra-word markup in Women Writers Online. One example of this is the WWP’s <vuji> tag, which is used to encode letterforms commonly substituted for each other in early typesetting, such as an “j” represented with an “i”. The <vuji> element is a useful shorthand for a TEI <choice>.

                        
in<choice><orig>i</orig><reg>j</reg></choice>urious

                        
in<vuji>i</vuji>urious

XML publishing software consistently treats element tags as distinct words. Our pre-publication stylesheets adapt our markup as needed to avoid this problem. Originally, the original letterform represented in <vuji> was converted to its modern equivalent, and the tag itself was removed.

More recently, when WWO introduced a feature allowing users to toggle between modern and original typography, the pre-publication stylesheets retained <vuji>. When the TEI was transformed into HTML, <vuji> became an HTML <span>, and its text content flipped to the modern version.

Figure 9

Jane Anger Her Protection for Women, as it first appears in Women Writers Online. The word “injurious” has been highlighted. The browser’s Inspector tool shows that the letter “j” is encoded in HTML as <span class="icor icor-reg">j</span>.

Figure 10

Jane Anger Her Protection for Women with the “Show original typographical features” preference toggled on. The highlighted word now reads “iniurious”, as it did in the 1589 publication. The browser’s Inspector tool shows that the letter “j” was changed to <span class="icor icor-reg" data-wwo-orig="i">i</span> in the Document Object Model.

Unfortunately, the new feature had the unfortunate side effect of reducing XTF’s ability to index and search older documents, because a single word would be broken up by tags.

in j urious

Importantly, while the code for the WWO site was updated and adapted for use in eXist, the WWO documents themselves stayed the same. The relationship between the markup and the publication tool is thus very indirect. The markup represents a durable, evolving expression of our analysis of the source texts, based on the project’s editorial principles and expectations about the kinds of research questions this data should support in the very long term. The publication platform, on the other hand, represents a particular manifestation of the potential of that data, under particular circumstances of resource constraints and technological capability. Even in the WWP’s other publications (see Appendix A) where the encoding is more deliberately aimed at producing a specific kind of outcome, the source markup remains tool-agnostic and could be published through varied systems to produce varied user experiences.

Read broadly, this history reveals several important points. First, the lifespan of monolithic tools was comparatively short: the first two iterations of Women Writers Online (DynaWeb and Philologic) each lasted about six years, whereas our current system has evolved more or less gracefully over the past 14 years with incremental modifications. Second, although the WWP never “coded to the tool” in the sense of altering our data to match what a given tool could or could not do, our publication tools have definitely put their stamp on what Women Writers Online as a publication could express about that data. The centrality of the keywords-in-context display in DynaWeb, the excellent division-level context-sensitive searching in Philologic, and the dynamic relationship between global and local reading views in the current interface all reflect, at a macroscopic level, the platform’s understanding of what “XML publishing” means (inflected as well by the WWP’s own intentions for the data).

Project sustainability

In July 2023, the Women Writers Project staff held a retreat to discuss plans for the next ten years of the WWP — what would the Project need in order to continue on indefinitely, in areas such as staffing, technology, and administration. The retreat focused on staff analysis of various aspects of the Project. Our markup and publication processes seemed stable for the moment, but analysis revealed that these systems were potentially at risk to events for which we had no contingency or recovery plan.

For example, many of the WWP’s web resources are dependent on eXist-DB. For years, the eXist database had its own divergent implementations of XQuery standards, such as XPath functions, XQuery Update, and output serialization.^[3] eXist-specific code requires eXist-specific knowledge to use and maintain it — a high barrier to entry and a high barrier to exiting. Northeastern’s Digital Scholarship Group — of which the Women Writers Project is a part — had already adopted BaseX as its XML database of choice for TEI and other XML projects, due to BaseX’s lightweight footprint, excellent documentation, and adherence to the XQuery standard. At this time, the WWP is the only DSG project which uses eXist. The WWP’s dependency on eXist is also a dependence on continued staffing with eXist-specific skills.

By the close of the retreat, we decided it would be prudent to familiarize ourselves with the outcomes of the Endings Project. Convened in 2016 by the University of Victoria’s Humanities Computing and Media Centre, the Endings Project’s goal was to develop and recommend practices that would aid the longevity of their digital artifacts. In 2018, the team released their first version of the “Endings Principles for Digital Longevity” (Endings Principles 2023). These principles describe the foundations of a digital project which can be used, cited, reproduced, and archived. By building for open use, the Endings Project argues, the digital products are far more likely to be reused and maintained, and thus, more likely to survive.

The Endings Project promotes the use of standalone, static websites, with no server-side processing or backend components. The HTML markup is not programmatically built out after the page loads; all content is available and all dependencies are satisfied simply by accessing the file (Endings Principles 2023). This greatly aids those who intentionally set out to archive the project materials, such as a digital repository or the Internet Archive’s Wayback Machine. The static site approach also makes it easy for researchers and other users to become casual archivists. It is not necessary to use extensions or software to generate a WARC file from the page; saving the page as an HTML file is enough to obtain your own working copy.

Crucially, the Endings Project does not intend for digital humanities projects to pursue archival solutions on their own. The survival of the original publication on the web is paramount. As Ending Project technical leads Martin Holmes and Joey Takeda explained, We have therefore focused on building digital editions which can run on any web server, anywhere, without any specific dependencies. If spinning up a new site requires nothing more than copying a collection of files to a server and circulating the URL, there is a far greater chance that functional copies of the products of your work will survive in a usable form. Every server-side dependency is a barrier to replication and therefore to survival (Holmes and Takeda 2023). It is important for the Internet Archive to be able to create WARC files from a digital humanities project’s site, but it is just as important for the project to be able to transfer ownership of the site to a repository or library for hosting and long-term maintenance.

Though the WWP puts much of our emphasis on the markup — without which, there would be no publication — the Endings Project counsels against considering our data alone as enough for project preservation. Dynamic products such as website interfaces are arguably the most recognizable and least preservable part of a digital project (Holmes, Jenstad, and Huculak 2023). A website is the most public expression of a project’s data, and an invitation for users to participate in the products’ use and reuse.

In our pursuit of maintainability, the WWP is working towards greater synthesis between markup, documentation, and website, especially for Women Writers Online. We intend to make it easier for users to download WWO’s TEI, packaging the files as part of the interface. With careful design, the WWO interface can demonstrate the capabilities of the markup, as well as catering to those who just want a clean, readable text. For instance, we plan to give users more control over how features of the markup are presented in the interface. The “original typographic features” toggle (Figure 10) is one example. Another toggle could show the lineation of the original publication, making <lb>s (line beginnings) and <pb>s (page beginnings) visible as HTML <br>s.

However, the highest priority for the WWP is refactoring our current publication processes for sustainability, and to produce static websites. Step one — currently in progress — is to migrate our apps from eXist-db to BaseX, as well as documenting and automating the publication processes. This step mostly consists of rewriting XQuery code so that a BaseX environment can produce the same JSON output that eXist currently does. The goal of this step is to ensure more people can publish new records and maintain the site, but where possible, we are also trying to make it easier to accomplish the work to come.

For Women Writers in Review and Women Writers in Context, the next step will be to shift away from single-page Javascript applications and towards giving the user a complete HTML page on request.

Step three is to generate static HTML. The current hope is that we will use the migrated XQuery code to generate HTML out of BaseX. Rather than serving the BaseX endpoint directly, we will save each page as a separate HTML file, in directories mapping onto the URL paths. When a user requests a web page, the server will respond with the static HTML file. We may still incorporate search through BaseX, but because it will no longer be required to respond to all requests, its workload (and perhaps response time) will be lessened.

Throughout the process, we are also working to make our web content more accessible. For example, the Women Writers Online interface includes a standard “Search” feature, implemented as a form. When a user clicked the “Search” button, the Backbone.js application would prevent the browser from navigating the user to a new page with their query parameters applied. Instead, the application requested the search results and populated the current page with them. From a sighted user’s standpoint, this was convenient and useful. From a blind user’s perspective, however, nothing would happen after they clicked the button; the WWO application did not alert them that the page had changed. If the request had been allowed to go through the browser, a screen reader would have informed the blind user that they had gone to a new page as a result of their request, with the search results found within the content of the page. By removing usability barriers, we extend our invitation for use and reuse to people of all abilities. And by preferring semantic HTML markup over CSS formatting, we bring the full semantics of the markup into focus.

It is worth situating this work against the Principles that we are using as our foundation. The Endings Principles are written in technical language: prescriptions for each component of a digital project. The Principles are also an enactment of a philosophical framework designed by the UVic Humanities Computing and Media Centre. As we considered the Endings Principles, we sought to understand the framework behind the Principles, and to map that framework onto the WWP’s own ethos. There are no accompanying Endings Principles for the accessibility of the interface, and the UVic team may have taken it for granted that the data should be embodied in the design and features of the interface, not just served alongside. These ideas are, however, natural extensions of Endings Project’s philosophical framework, and part of the WWP’s framework. As publications and user-facing tools, web interfaces are the most prominent manifestation of our markup.

Conclusions and Next Steps

Michael Sperberg-McQueen, in his 2012 keynote at the Knowledge Organization and Data Modeling conference, said that modeling is a way to make explicit our assumptions about the nature of a text/artefact (Sperberg-McQueen 2012). These assumptions are often complexly layered and may even represent divergent perspectives or possibilities in the text. The tools that express the potential of our modeling also participate in making those assumptions explicit, by enacting some of the practical or analytical consequences of those assumptions: if the modeling tells us that certain passages are endnotes, our page generation can gather them at the end of the text; if the modeling tells us that certain readings are typographical errors, our search tool can ignore them. We might thus think of tools as an intensification of the rhetorical aspect of markup: the space where the provisional, suspended potential of descriptive markup for action is realized, within a specific set of rhetorical conditions: audience, purpose, context. Because of its long history, the WWP has had an unusual opportunity to express the potential of its markup in a changing set of rhetorical contexts, and through tools that realize the basic activities of reading, searching, and analysis in somewhat different ways. What these experiments have revealed is the durability and adequacy of the TEI’s general approach, and of implementations like the WWP’s, which emphasize abundant, semantically rich markup that has a focused methodological orientation but is open-ended as to the specific activities through which reading and analysis will be practiced—and also a continued experimentation with how interfaces can support the evolution of those practices.

Interfaces serve as manifestations of markup and as an invitation to the user to interact: not only with patterns of tokens or with structures inferred from them but also with human readings. This is particularly significant for historical materials because the meaning of what they present to us is not transparent or unimportant: reading such texts is a process that has to remain frictional and critical. The markup retains that critical friction; it documents the process of observation; this is what makes such data valuable to scholarly readers. And for those readers, interfaces and analytical tools do their work best when they provide a view of that friction as well. In a sense, this is the strongest case we can make for markup, and we consider it important for researchers of the current generation to be reminded how important markup is, at a time when the ubiquity of machine-learning approaches may seem to eclipse this kind of durable annotation. The WWP came of age at a time when theorizing markup was a core digital humanities research concern, considering fundamental questions about how markup’s formalisms function in relation to human description and argumentation, how markup systems can be designed to balance interoperability with extensibility, how formal models exist in relation to material manifestations (such as book structures or typography), how formal models are instantiated through tools, and many more. We see a renewed opportunity for research that explores markup from a perspective of labor politics, agency, minimal computing, accessibility, and other areas in which it contrasts strongly with machine learning approaches, even as there may also be areas where the two are complementary. The potential of markup systems and markup-based tools for transparency and accountability remain among their enduring strengths.