Overview

§ 1 Digital Mappaemundi (DM) is a resource under development to generate open source tools for scholars to edit and annotate image data that are networked together, and for other users to search within this rich content. For the purposes of development, our data have been medieval mappaemundi ("maps of the world") and their geographical source texts. With funding from a 2009-2010 NEH Digital Startup in the Humanities grant, as well as a 2010 Scholarly Communications grant from the Andrew W. Mellon Foundation, the DM project team has finished the first stage of development, and worked through foundational issues of functionality, data management and interface. At the time of this publication DM has undergone significant evolution during its phase three beta development and is now a broadly applicable environment for the creation of annotation and linked data across a range of digital image/text collections. The use-case referred to "Digital Mappaemundi" in this essay is now the "Virtual Mappa Project" (VMP), and is in partnership with the British Library. For current functionality and features of the DM environment, as well as a list of medievalist projects using it, see http://ada.drew.edu/dmproject/.

Background and history

§ 2 In 2002, after finishing The Bayeux Tapestry Digital Edition (BTDE), Martin Foys began to look for another project to continue his work in applying digital technology to the study of early medieval artifacts. One of the things that drew Foys to the Tapestry in the first place was its hybrid nature of expression, and the way that its monumental materiality (over 230 feet long and containing four simultaneous registers of graphic and textual information) resisted traditional modes of printed scholarship (Foys 2007, 102-109). Medieval maps, particularly the Cotton Map, an Anglo-Saxon mappamundi from around the year 1000, became an ideal subject. Not art, not literature, not history, and from a long-standing and biased view of modern cartographic scholarship, not even a very good map, the Cotton Map had easily fallen in between disciplinary cracks. The initial approach was a remediating and docucentric one: create an edition of a single artifact in the same template as the BTDE, as a standalone edition published in a physical and closed format (CD-ROM) (Figure 1). Less than a decade later, the impulse to create such a publication reads as quaint, if not antique.

Figure 1: Early attempts at a docucentric edition of the Cotton Anglo-Saxon mappamundi (2004-2005)

§ 3 During this time, XML emerged as the dominant standard for data markup. Foys partnered with Asa Mittman and a group of student researchers to mark up in XML the Cotton Map and several source and analogue geographic texts, including the Latin and Old English versions of Orosius's description of the world from Adversus Paganos Historiarum Libri Septem ("Seven Books Against the Pagans"), the Bible (selected content specific to Cotton Map), the twelfth-century Expositio mappe mundi, St. Jerome's De situ et nominibus locorum hebraicorum liber (selected), Isidore of Seville's De natura rerum (selected), Pliny the Elder's Naturalis historia (selected) and Solinus's Collectanea rerum memorabilium (selected). This team used a modified version of the Electronic Production and Presentation Technology (EPPT) software that Kevin Kiernan and his team had developed to mark up transcriptions of manuscript folia (see Figure 2). In EPPT, the project was able to tag coordinate bounding boxes on an image of the Cotton Map in XML, and relate these coordinates not only to a textual transcription of the map's 147 Latin inscriptions, but also to specific moments in one or more of the source/analogue texts. In the midst of the process, it became clear that this kind of electronic editing could be used not in the service of one map, but rather to build a database network of medieval maps and texts that could be continuously expanded in the future by other users. Once other maps were edited, their coordinate-based XML tags could be used to cross-reference details within maps to their analogues on other maps, allowing for an innovative architecture and scale of comparative study. This realization served as the genesis of Digital Mappaemundi. In 2008, Foys moved to Drew University and Shannon Bradshaw, a computer scientist, joined the project as co-director, and work developing the data architecture and end-user interface began in earnest.

Figure 2: Early use of Electronic Presentation and Production (EPPT) software to link coordinate points and texts through XML (2005-2006)

Background: Geospatial bias and the nature of maps

§ 4 In Using Digital Primary Resources to Produce Scholarship in Print, John Unsworth argues that new possibilities for print scholarship are presented by born-digital information and the tools one uses with that information—to take just one example, consider geographic information systems [...] that map all kinds of social information onto geographic space. When people hear the DM project is working on technologies and maps, there is often a frustrating, if understandable response that assumes work with GIS (Geographical Information Systems) or maybe Google Earth. Such assumptions often go hand in hand with a curious tendency towards geo-rectification – that is, to take maps of the past world and align them to the mensurable reality of today's world's physical topography. Over the past five years, common approaches to historical geography and mapping have rapidly become synonymous with modern, measured, plotted and/or georectified spaces. For example, Mapping Medieval Townscapes: A Digital Atlas of the New Towns of Edward I (Lilley et al. 2005) creates a modern, accurate map on which medieval historical information and topography can be plotted. As such, it subordinates all content to current notions of geospatial representation and reality. Another common technological trope for the cartographic past is to take an old map and fill it with new knowledge. In the Map of Early Modern London project (Jenstad et al. 2010), for instance, students and scholars have linked "encyclopedia-style articles, scholarly work, student work, editions, and literary texts" to ‘hotspots’ on the 1578 Agas map of London. In this case, the resulting, rich resource seems less focused on the map itself than it does on gathering all kinds of information that can be added to it in digital form.

§ 5 This is in no way to discount the immense value, utility, and promise, of such resources as those described above. But they do unwittingly represent a tacit, largely unexamined assumption about what maps of the past are to us today and they explicitly measure the past views of the world against the present. The result is often a distortion of our critical understanding of them as we attempt to remake the older map into a repository for some other kind of information (Figure 3). What remains less fulfilled, technologically, is the excavation of the content already there on such maps. For digital projects focused on historical material to be fully realized, they must also grapple with material on its own terms, and acknowledge its particular concerns. Medieval maps are far more similar to literature or art in that they are representations that are not necessarily grounded in the specifics of our reality. They were not designed to correspond point for point with the globe—precision of distance and detail, a foundation of modern cartography, was irrelevant. Rather, medieval maps were aimed at helping their largely monastic audience understand their place in the world. As suggestive indices of the world, these documents work from a complex, referential network of texts and other maps, and digital tools need to be developed to study these networks. As J.B. Harley and David Woodward put it:

§ 6 Cartographic history [has now become] the study of needs and wants rather than of just the ability to make maps in the technical sense [...] it follows that the capacity of cartography to influence actions or to mold mental worlds must depend not only on the extent to which maps were actually seen but also on the way they, or their messages, were understood. (508)

§ 7 All maps do not represent physical or culturally total reality. Instead, maps are selective and highly inaccurate representations of both explicit and tacit ideological ends (Wood and Krygier). With regards to the study of medieval maps, it is time for the recognition of such differences to follow in technological application.

Figure 3: Geo-spatial bias in digital approaches to cartography of the past, from the "Google Earth" feature of the David Rumsey Historical Map Collection (2009 interface)

DM phase I development (2009-10)

§ 8 To realize a technological model for the genuine study of medieval maps, DM has proceeded on two fronts over its first two years of development: the generation of analyzable data from medieval maps and geographic texts, and the development of the architecture to make such generation and subsequent analysis possible. Using an early prototype of the DM annotation tool, the editing team under Foys annotated seven medieval maps and two geographic texts – the foundation for a small but representative sandbox of data.

§ 9 Maps:

• The Anglo-Saxon Cotton Map (British Library MS Cotton Tiberius B.V. f. 56v)
• A Higden Polychronicon Map (Corpus Christi College Cambridge [Parker Library], MS 21, f. 9r)
• The Thorney Map (Oxford- St. John’s College MS 17, f. 6r); the Sawley Map (Corpus Christi College Cambridge [Parker Library], MS 66, p.2)
• Matthew of Paris's Chronica Maiora Map of the Middle East (Corpus Christi College Cambridge [Parker Library], MS 26, f. 3v-4r)
• An unfinished late Anglo-Saxon mappamundi (Corpus Christi College Cambridge [Parker Library]; MS 265, p.210
• A mappamundi from Gunther Zainer’s 1472 printing of Isidore’s Etymologiae at section III ("De Asia") of Book XIV ("De Terra")

§ 10 Texts:

• Book One, Chapter Two of Orosius's Adversus Paganos Historiarum Libri Septem ("Seven Books Against the Pagans")
• Book One, Chapter One of the Old English Translation of Orosius's Adversus Paganos Historiarum Libri Septem ("Seven Books Against the Pagans")

§ 11 In addition, several other texts from the earlier EPPT phase (see section 2, above) await porting into the DM environment. All textual inscriptions and descriptions of locations on maps and in texts were recorded in XML (with coordinate locations for images and word placement for texts), along with accompanying descriptive data. These datasets, while relatively modest in size, provide excellent material through which to work out major issues of design and functionality. In addition, we now have robust data on which to build and through which to test our interface as the resource develops. The data may also serve as a framework model for subsequent users.

§ 12 During this phase, the programming team led by Bradshaw continued to developed early prototypes of editing and annotation resources. The earliest pre-alpha prototype was implemented in Adobe Flash and employed single coordinate-point visual tagging. Annotations were constrained to a custom set of XML descriptor tags and keywords (Figure 4). This phase I implementation of DM contained much of the basic functionality desired in such a resource (i.e. the ability to navigate and magnify an image of a map, tag and annotate specific areas, integrate annotations and include texts into a continuously extensible and searchable database, link directly from search results to precise locations on maps and text, and aid searches through incremental prompts). But a few months into development, the severe limitations of our approach became clear, in part due to the markedly variable and hybrid nature of the maps we were editing. While single-point coordinate editing worked well with limited graphical or textual detail on a map, it could not accommodate with necessary precision larger textual blocks that are found in documents like the Matthew of Paris map. Figure 5 shows a single-coordinate link from a search for a specific word in the Matthew of Paris map, which occurs within a much larger block of text. Short of creating a single-coordinate annotation for every single word in this target block (which would be cumbersome to both edit and view), editorial precision in this model was impossible. Additionally, mandating an exclusive one-to-one correspondence of target area to annotation denied the ability to link multiple moments of the map to the same annotation (without clunky redundancies in the data), or to precisely demark larger graphic or textual moments within the text. To paraphrase what one art historian said in response to an early presentation of DM: "That's great that you can tag that single-word inscription or a small iconic moment, but what if I wanted to annotate the whole oceanic circle that encompasses a map? How would I do that?"

Figure 4: Composite graphic demonstrating many of the annotation features of the phase 1 DM prototype

Figure 5: The limits of coordinate editing: Asa Mittman's use of DM to edit Matthew of Paris's map of the Middle East

§ 13 Our initial instinct to set fixed descriptive categories for annotations (following the traditional logic of taxonomic structures for metadata) was also problematic. At the beginning of our work, we approached annotation by imagining discrete sets of tags that aimed to collectively cover all possible content that medieval maps and related texts might contain (e.g. fields for inscription, translation, keyword, figure, continent, quality categories [city, people, river, mountain, etc.], and so forth), along with a search function that generated results by polling one or more of these selected fields. Such a taxonomic approach is perhaps most readily understood as analogous to the patterns of data annotation demanded by such protocols as TEI, which imagines a prescriptive set of tag fields that can comprehensively cover, in theory, any aspect of a text being marked up in XML. We realized relatively early in development that there are considerable limitations to such an approach, especially given the considerable variety of content found within medieval maps. We also had no way to anticipate and then to accommodate the ways in which future users might wish to annotate the materials we were editing, or how we might keep our own prescriptive categories for annotation free from our own heuristic bias. Such issues in annotation and editing led us to an acute back-to-formula moment, and to a radical reworking of almost every aspect of the resource to date, starting with a conceptual review of how scholars do the work they do.

DM phase II development (2010-11): Scholarly primitives

§ 14 In re-designing a full-featured system to support scholarship on digital maps and manuscripts the DM-team first reconsidered the research practices of its user base. By this we mean the activities in which scholars engage when carrying a piece of work through from research question to publication. John Unsworth has identified a useful set of "scholarly primitives" that researchers commonly employ in their work: discovery, navigation, annotation, referencing, organizing, and comparison (Unsworth 2000). We have adapted these primitives as a basis for designing our new DM methods and tools, and as we engage with scholars about their research processes we have used this model to frame conversations about the full range of capabilities DM must embody. We address each primitive defined in this model below. Our intention is to provide a sense of the challenges inherent in each, rather than a comprehensive treatment. As such, this discussion reflects lessons learned to date and thoughts on future development for DM.

Discovery

§ 15 Scholars engage in a variety of discovery tasks for the purpose of finding material necessary to their research. The example that usually leaps to mind is a Google-like search of a manuscript repository. Another is paging through manuscripts in search of folios of interest. But discovery processes may also include those that are less obvious in this context, such as encountering a previously unknown resource at the suggestion of a colleague. In light of this, we have identified at least two system requirements. The first is that DM must enable scholars both to access and annotate images available on-line to them, and to individually import images into the DM resource, as needed. This requirement causes a cascade of additional requirements to manage access, storage, and sharing of images (and will be the focus of a subsequent phase of our work). The second requirement is that in cases where scholars are unable to acquire electronic access to a map or manuscript existing in a digital repository, they must be able to incorporate that item into their annotations and overall work product as it develops within DM. This requirement has implications for most of the other scholarly primitives in our model.

§ 16 More generally, DM must support "social discovery" in a variety of forms. For example, scholars may wish to receive alerts about annotations others may make within DM on certain items or subjects. Likewise, annotations and other forms of content should each have a public URI so that they may be easily emailed, tweeted, and otherwise distributed. This raises the issue of public, private, and limited group annotations. For DM (or any other annotation resource deploying any social discovery tools) to develop to its full potential such concerns must be addressed, and will be the subject of future development and funding.

DM phase II development (2010-11): Interoperability and breakaway tools

§ 23 With these primitives in mind, we have begun to explore how the tools being developed within DM might be applied to help break open the largely siloed nature of on-line repositories of digitized manuscripts. With funding from the Andrew W. Mellon Foundation, we worked with Stanford University's project team for the Parker Library on the Web project (PW) to study ways in which annotations manuscript images stored in PW could be created using DM tools, and then accessed and indicated in the PW interface through an interoperating protocol (Figures 6-7). The resulting successful test, where we were able to open up a Parker manuscript folio from PW inside DM, annotate it, and then have PW dynamically create an active link for those annotations on its resource display for that page, was an exciting proof of concept, pointing the way forward for how archival resources and scholarly tools in the digital realm can be designed to interoperate.

Figure 6: A demonstration of interoperability proof of concept (part 1): A link from Parker resource opens folio page called from Parker in Phase 1 DM for editing.

Figure 7: A demonstration of interoperability proof of concept (part 2): After editing, Parker interface dynamically shows that edits in DM resource now exist and may be viewed.

§ 24 In a move away from coordinate-only editing (see section 12, above), we developed a new architecture for visual tagging, based on a tripartite approach of Dot-Line-Shape (D-LISH). D-LISH was designed to provide maximum flexibility with regards to annotators' graphic desires, providing them with the ability to more generally indicate annotated areas on a map through single coordinate points, joining points of significance lineally (either through traditional lines of text, or by connecting horizontally disparate points [i.e. by crooked lines]), or by demarcating areas of annotation within user-generated polygonal spaces (Figures 8-9). Further, a single annotation could then be grouped with any combination of multiple dots, lines and/or shapes. The result was an image-selection tool that allowed for considerably greater choice in determining what, exactly, was to be selected and annotated on an image than in pre-existing tools for digital scholarship.

Figure 8: DM Phase 2 prototype; new interface, D-LISH precise shape, line and dot annotation-area selection (for the Cotton Map's figure and inscription for "Babilonia" in green), user layers, and open-form annotation entry.

Figure 9: DM Phase 2 prototype; D-LISH annotation, segmented line selection for annotation of the Cotton Map's representation of the Orkney Islands ("Orcades Insulae," spelled across thirteen islands), and related visual analogue feature.

§ 25 At the same time that D-LISH was being developed, we radically shifted the direction of development with respect to models of annotation and search. Searches not only returned textual links, but visual correlatives and suggestions as well (Figure 9). We then focused on creating a completely open mode of annotation, where users simply enter annotations as untagged text, with the option to designate individual terms as prioritized keywords, through the use of a hash (#) tag. Such annotation cannot really be considered purely folksonomic, though it does retain the capacity for such tagging if desired by the user. In response to such a simpler, open-form annotation model, the development of our search algorithm has become substantially more complicated, as the work of parsing annotated material will have to be handled by the search, not the annotation function. In the end, however, this model will result in a substantially more agile and dynamic response for end users wishing to search a growing set of both materials and annotation.

§ 26 To support this new functionality and future interoperability, the backend database was redesigned and rebuilt with a more general data model. This model enables us to easily accommodate data in the Open Annotation Collaboration (OAC) protocol, which is emerging as an interoperability standard among large repository and tools initiatives underway at Stanford University, Johns Hopkins University, the University of Toronto, and the University of Maryland, among others. As a second step toward a broadly usable resource, we have exposed DM's annotation repository through a web services framework. The web service enables client systems (including our own) to search our annotation repository using an expressive query language and sophisticated search metrics. The search system is based on Solr from Apache's Lucene project. Clients may search the repository using raw search terms, faceted search by user-defined field, or more complex techniques such as more-like-this queries. Client systems may also mine or store annotation data by pulling or pushing XML feeds of layered annotations for any map in our dataset. Clients may also constrain annotation data feeds by proximity to a location on an image, by annotating user, by timestamp, and a variety of other metrics. Additionally, external URI links to specific coordinates and annotations may be generated.

§ 27 In rebuilding the front end, we moved from a Flash-based architecture to one that is based on Javascript. In phase 2 development of the user interface, we were careful to develop toward an open-source library that enables widespread adoption. At the end of Phase 2 development, DM had an alpha version of a set of tools that enabled scholars to mark regions-of-interest within images and associate textual annotations with those regions. Scholars may mark images with individual points, segmented lines, or custom polygonal shapes. Significantly, a scholar may identify any number of markers on any number of images as the targets for textual annotation. Additionally, a given marker may serve as the target for any number of textual annotations. Finally, scholars may organize their annotations into groups called layers so that different research questions involving a single image may be addressed separately through annotation. Scholars may choose to view a single layer of annotation or view multiple layers of annotation overlaid on one another.

§ 28 In 2012, the DM environment is in beta development, and working with a series of projects covering a wide range of digitized medieval materials. The Dictionary of Old English (DOE) project, for example, is using DM to annotate and link examples of Old English words from on-line manuscript collections to individual DOE entries. The Parker’s Scribes project at the University of Toronto and Oxford University annotating examples of sixteenth-century marginal notations from the Parker on the Web digital repository of medieval manuscripts and creating precisely linked data to be rolled out as an on-line resource. Individual scholars are also using DM: Lisa Fagin Davis is digitally annotating all of the content on a Chronique anonyme universelle -- a 50-foot-long historical and genealogical scroll with over 1,500 individual figures and events depicted for creation of an electronic edition to accompany a forthcoming academic publication for Brepols. And in his Insular and Anglo-Saxon Illuminated Manuscripts: An Iconographic Database project, Asa Mittman and his team is using DM to take Thomas Ohlgren's 1986 print "database" of 229 codices accounting for nearly all visual iconography in Anglo-Saxon manuscripts and turn it in to a linked, digital environment that will dynamically import images from on-line digital manuscript repositories. Finally, the original medieval maps project has grown to an institutional partnership, and become the Virtual Mappa Project (VMP), by which British Library and Foys will use DM to design and publish an on-line collection of historic maps that multiple groups may use to create linked environments of annotations for individual, collaborative, and instructional purposes. Once VMP is implemented, the British Library will then be able to partner with other institutions to augment this resource, fashioning a virtual collection that allows for true repository interoperability and the potential for deep public, educational and scholarly interaction. For a full overview of current DM functionality, as well as of projects using DM, please see http://ada.drew.edu/dmproject/