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 a long-term active research undertaking, concerning 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).

Research and working relationships

It may be helpful at the outset to provide some context on the WWP’s research and working relationships, since those had a strong influence on how the project’s philosophy of markup and tools took shape. One important factor in that evolution is the project’s simultaneous focus on early modern women’s writing, digital text representation, and theories of markup. Scholars of early modern texts (and of women’s writing in particular) are particularly concerned with textual materiality, book history, and the ways in which texts circulate and are consumed and curated and discarded. As a result it is natural for them (and hence for the WWP) to make the connection from “tools” to “technologies” to “technologies of text and writing”: that is, to treat the book itself as a tool that has a very interesting potential relationship with the other “tools” we are concerned with. The idea that the text’s original embodiment and typography are expressive technologies (and hence that there is something “to be expressed” that is separate from those) clearly informs a lot of bibliographic theory, particularly some schools of thought that were still prominent when the WWP was getting started (for instance, the work of R.B. McKerrow and Thomas Tanselle) even as feminist and materialist approaches also contributed crucially to our thinking. The impact of those theoretical orientations on our markup is beyond the scope of this paper to explore, but it’s worth noting that the feminist angle on these ideas has been always present in some form in the WWP’s thinking, moving us to pay close attention to the significance of embodiment, to the role of patriarchal capital in mechanisms of dissemination, to the position of women in ecologies of textual circulation, etc.

The other important shaping factor is the WWP’s relationship with Brown’s Scholarly Technology Group, which was a notable aspect of the project’s incubation; the WWP was directly supported by the STG from 1994 until at least 2005 and was also an important contributor to STG’s research and development work. STG was an applied research and development group that also supported other projects, which presented WWP staff with opportunities for dialogue on central research questions and learning from other experiments. These included the Decameron Web, MonArch, and also a number of projects that used databases rather than XML markup, leading to debates about data modeling and knowledge organization that also informed the 2012 workshop on Knowledge Organization and Data Modeling. Starting in 2013, our relationship with Northeastern’s Digital Scholarship Group has been similar: the WWP is embedded in conversations that shed a varied theoretical and practical light on questions about tools and data.

Against the backdrop of that broad history, the topic of this paper was also shaped by some particular preoccupations that deserve mention. For a major early portion of the WWP’s history, tools felt optional. WWP staff remember Allen Renear in the early 1990’s saying “There will be tools…”, anticipating the ways in which the rich data we and other projects were creating would certainly prompt the creation of systems for using it. We also remember that a bit later, at the TEI conference in 1997, Jon Bosak anticipated the impact that XML would have in facilitating the creation of tools that would work across specific markup languages. But the WWP worked for a number of years in the absence of tools we now consider fundamental: an XML parser, XML validation, a formatted display, a means of electronic dissemination. The emphasis on the value of data itself in its “pure” state suggests interesting connections to post-Reformation aesthetic austerity: markup in the absence of tools feels interestingly akin to the “plain style,” to Puritanism. Obviously we were using some tools: the Brown University mainframe computer, a TN3270 terminal emulator, the Waterloo Script routines that formatted our texts for printing, the gigantic printer that printed out our texts for proofreading and circulation. But none of these really did anything to influence or even notice the modeling of our data: they were largely in support of the most basic aspects of the data creation process, not part of the expression or circulation of the data. So this “pre-tool” moment is useful as a way of scoping what we mean when we say “tools” in this context.

Tools also felt potentially at odds with our modeling purposes, and in a potentially conflictual relationship with the data, a concern we can also observe in early humanities computing discussions of data in relation to tools, perhaps especially in the markup community. By limiting what can be expressed about the data and by limiting users’ interactions with it, tools (from that perspective) create an impoverished experience of the data—especially problematic for a research group whose work is animated by a commitment to rich informational detail. Tools also risk pushing the project to make intellectual concessions in the design of the data, either because of assumptions the tool makes about the shape the data will take, or because of shortcuts and unwillingness to anticipate and handle truly challenging situations. This might take many forms, for instance:

These kinds of incentives and disincentives may have arisen especially vividly in the incunabular days when tools for working with SGML/XML data were in their infancy. Some of the solutions we would now consider (customizing the tool, choosing a different tool) were not really on the table in those earliest moments. And other options, such as creating an intermediate data format that aligns with the tools’ expectations, became much easier once the XML programming environment and XSLT became available. But overall, these issues raised concerns about “coding to the tool” and this is worth unpacking a bit further. The idea of coding to the tool implicitly conveys that the tool is structurally alien to our purposes in a way that “the data” is not. Data is an expression of research ideas while a tool is a prosthesis; data is nuanced while a tool is coarse-grained and clunky; data is nimble while a tool has inertia. Responding to this, the digital humanities markup community has gravitated towards what Julia Flanders and Fotis Jannidis call tool agnosticism:

In this context, if we consider a standard schema as a “tool” that is distinguishable from (and could be distant from) the core motivations of the data-creating researcher, this helps to illustrate why the TEI customization mechanism is so distinctive and so important within the DH intellectual ecology. From the outset, the TEI’s customizability acknowledged the intimate and detailed relationship between researchers and their data. But even beyond this, putting the customization mechanism within easy reach of all data creators (by baking it into the process of creating any schema at all, and by providing a web-based portal to do so) ensures a very close relationship indeed between the data and its creator. The schema is “your tool,” an extension of your ideas and of your data rather than separable from them or alien to them.

Once we get past this weird initial combination of skepticism and longing, there were also some very interesting questions to be asked, which have deeply informed the WWP’s thinking about markup and tools. One of these is the lines of research by Allen Renear and others (see, for example, Renear 2003) about how XML sits in relation to work in library and information science about the relation between more and less fully “materialized” representations of information, for instance, in the Functional Requirements for Bibliographic Records (FRBR; FRBR 1998). Renear’s analysis suggests that tools are integral to that materialization. In FRBR terms, a “manifestation” of a conventional textual work is the “physical embodiment of an expression of a work” and has properties such as formatting, physical carrier/medium, and pagination. In the world of XML data, you can have a FRBR “expression” without a tool set (because expressions are still abstractions, and their properties can be described in XML data). But a “manifestation” requires taking some material action on the basis of the data, via a stylesheet or program that uses the markup to generate some kind of output that can possess these kinds of material properties, a PDF or a CD-ROM or a printout or a web site. Such output acts on the potential of the markup (to mark textual alternatives or semantic categories or whatever) by producing some tangible outcome. So this line of thinking about FRBR helped (among other things) to demonstrate how tightly and interestingly coupled tools and data really are.

To come at this from another angle: markup can be understood as potential information that carries a lot of surplus. Elena Pierazzo pointed out in a conference paper (Pierazzo 2006) that in order to preserve a digital edition, we need to preserve not only the marked up data but also the tool set through which the data was expressed and consumed, since that is what realizes the potential of the markup and makes a practical selection from the choices it offers. She proposed that the XSLT used to generate the disseminated version of a digital edition should be considered:

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 the TEI P5 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). The markup of the textual content focuses on a selection of high-value features (such as element names or quotations) that are specifically relevant to the rhetorical context of the materials, and the metadata carries more of the burden of representing features of the content that will be valuable for retrieval. In the case of encoding documentation, that includes including topic keywords and an inventory of elements and attributes referenced. In the case of WWiR, it includes thematic keywords and an inventory of authors and texts mentioned in the review. Shifting these aspects of content into metadata (rather than tagging them in the running prose of the review or documentation entry) serves the pragmatic purpose of streamlining the encoding process, but it also acknowledges that these informational aspects are intended to function as retrieval hooks rather than as an activity of analysis or textual representation. In other words, it posits a different rhetorical framing for this information.

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 individual 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. 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.

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. 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. The following screenshots show some examples.

Figure 7

The entry point for the current version of Women Writers Online (active starting in 2005), showing the user’s access to search, filter, and browse panes simultaneously.

Figure 8

Once a search has been executed, the current version of WWO shows search, context, and reading interface simultaneously.

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 11

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 12

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

As an example, Figure 14 is a screenshot from 2022 which demonstrates that XTF would rarely find a match for the word “rose” in WWO texts that were originally published before 1775. The older the source material, the more likely it was that “rose” would have a long-s in it: roſe. Our pre-publication workflow would introduce a tag around long-s characters so they could be swapped with their modern equivalent, consequently preventing XTF from indexing the full word.

Figure 14

XTF showed 119 documents matching the word “rose,” with a vast majority of the results clustered between 1775 and 1855 on the timeline.

At that time, work had begun on WWO’s eXist implementation. By running the same search in the new database, we discovered that XTF was missing one hundred documents matching the word “rose”:

Figure 15

This early version of WWO in eXist showed 220 documents matching the word “rose.” Results are evenly spread between 1550 and 1800.

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 therefore also a dependence on continued staffing with eXist-specific skills.

Conclusions and Next Steps

C. 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.

In the spirit of that continued experimentation, there are some research questions which feel urgent:

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.