Annotated Facsimile Editions: Defi ning Macro-level Structure for Image-Based Electronic Editions

Facsimile images form a major component in many digital
editing projects. Well-known projects such as the Blake
Archive [Eaves 2007] and the Rossetti Archive [McGann 2007]
use facsimile images as the primary entry point to accessing
the visually rich texts in their collections. Even for projects
focused on transcribed electronic editions, it is now standard
practice to include high-resolution facsimile.
Encoding standards and text processing toolkits have been
the focus of signifi cant research. Tools, standards, and formal
models for encoding information in image-based editions have
only recently begun to receive attention. Most work in this
area has centered on the digitization and presentation of visual
materials [Viscomi 2002] or detailed markup and encoding
of information within a single image [Lecolinet 2002, Kiernan
2004, Dekhtyar 2006]. Comparatively little has been work has
been done on modeling the large-scale structure of facsimile
editions. Typically, the reading interface that presents a facsimile
determines its structure.
Separating the software used to model data from that used
to build user interfaces has well-known advantages for both
engineering and digital humanities practices. To achieve this
separation, it is necessary to develop a model of a facsimile
edition that is independent of the interface used to present
that edition.
In this paper, we present a unifi ed approach for representing
linguistic, structural, and graphical content of a text as an
Annotated Facsimile Edition (AFED). This model grows out of
our experience with several digital facsimile edition projects
over more than a decade, including the Cervantes Project
[Furuta 2001], the Digital Donne [Monroy 2007a], and the
Nautical Archaeology Digital Library [Monroy 2007b]. Our work
on these projects has emphasized the need for an intuitive
conceptual model of a digital facsimile. This model can then
serve as the basis for a core software module that can be
used across projects without requiring extensive modifi cation
by software developers. Drawing on our prior work we have
distilled fi ve primary goals for such a model:

• Openness: Scholars’ focused research needs are highly
specifi c, vary widely between disciplines, and change over
time. The model must accommodate new information needs
as they arise.
Figure 1: A simplifi ed diagram showing an edition
of collected poems (in two volumes) represented
as an annotated facsimile edition.
• Non-hierarchical: Facsimile editions contain some
information that should be presented hierarchically, but
they cannot be adequately represented as a single, properly
nested hierarchy.
• Restructuring: A facsimile is a representation of the
physical form of a document, but the model should enable
applications to restructure the original form to meet
specifi c needs.
• Alignment: Comparison between varying representations
of the same work is a fundamental task of humanities
research. The model must support alignment between
facsimiles of different copies of a work.
Annotated Facsimile Editions
The Annotated Facsimile Edition (AFED) models the macro
level structure of facsimile editions, representing them as a
stream of images with annotations over that stream. Figure 1
shows a simplifi ed diagram illustrating a two-volume edition
of collected poems. Annotations encode the structure of the
document and properties of the structural elements they
represent. Separate annotation streams encode multiple
analytical perspectives. For example, in fi gure 1, the annotations
shown below the image stream describe the physical structure
of the edition (volumes, pages, and lines) while the annotations
shown above the image stream describe the poetic structure
(poems, titles, epigraphs, stanzas). Annotations within a
single analytical perspective—but not those from different
perspectives—follow a hierarchical structure.
The Image Stream
The image stream intuitively corresponds to the sequential
ordering of page images in a traditional printed book. These
images, however, need not represent actual “pages.” An image
might show a variety of artifacts including an opening of a
book, a fragment of a scroll, or an unbound leaf of manuscript
notes. While it is natural to treat facsimile images sequentially,
any particular linear sequence represents an implementation
decision—a decision that may not be implied by the
physical document. For example, an editor may choose to
arrange an edition of letters according to the date written,
recipient, or thematic content. The image stream, therefore,
is an implementation detail of the model. The structure of the
edition is specifi ed explicitly by the annotations.
Annotation Management
Perspective Analytical perspective e.g., physical structure, narrative
elements, poetic.
Type The name of this type of annotation, e.g., page, volume,
chapter, poem, stanza
Start Index The index into the image stream where this annotation
starts.
Stop Index The index into the image stream where this annotation
ends.
Sequence A number for resolving the sequence of multiple
annotations on the same page.
Content
Canonical Name A canonical name that uniquely identifi es this content
relative to a domain specifi c classifi cation scheme.
Display Name The name to be displayed when referring to an instance
this annotation
Properties A set of key/value pairs providing domain specifi c
information about the annotation.
Transcriptions A set of transcriptions of the content that this annotation
specifi es.
Structural Information
Parent A reference to the parent of this annotation.
Children A list of references to the children of this annotation
Table 1: Information represented by an annotation.
Many historical texts exist only as fragments. Many more have
suffered damage that results in the lost of a portion of the
original text. Despite this damage, the general content and
characteristics of the text may be known or hypothesized based
on other sources. In other cases, while the original artifact may
exist, a digital representation of all or part of the artifact may
unavailable initially. To enable scholars to work with missing or
unavailable portions of a facsimile, we introduce the notion of
an abstract image. An abstract image is simply a placeholder
for a known or hypothesized artifact of the text for which no
image is available. Annotations attach to abstract images in the
same way they attach to existing images.
Annotations
Annotations are the primary means for representing structural
and linguistic content in the AFED. An annotation identifi es a
range of images and specifi es properties about those images.
Table 1 lists the information specifi ed by each annotation.
Properties in italics are optional. As shown in this table,
annotations support three main categories of information:
annotation management, content, and structural information.
49
The annotation management and structural information
categories contain record keeping information. Structural
information describes the hierarchical structure of annotation
within an analytical perspective. The annotation management
category specifi es the annotation type and identifi es the image
content referenced by the annotation. The sequence number is
an identifi er used by AFED to determine the relative ordering
of multiple annotations that have the same starting index.
AFED is agnostic to the precise semantics of this value. The
annotation type determines these semantics. For example,
a paragraph annotation may refer to the paragraph number
relative to a page, chapter, or other structural unit.
The content category describes the item referenced by the
annotation. Annotations support two naming conventions. To
facilitate comparison between documents, an annotation may
specify a canonical name according to a domain specifi c naming
convention. Canonical names usually do not match the name
given to the referenced item by the artifact itself and are rarely
appropriate for display to a general audience. Accordingly, the
annotation requires the specifi cation of a name suitable for
display.
Descriptive metadata can be specifi ed as a set of key/value
properties. In addition to descriptive metadata, annotations
support multiple transcriptions. Multiple transcriptions allow
alternate perspectives of the text; for example, a paleographic
transcription to support detailed linguistic analysis and a
normalized transcription to facilitate reading. Transcriptions
may also include translations.
AFED’s annotation mechanism defi nes a high-level syntactical
structure that is suffi cient to support the basic navigational
needs of most facsimile projects. By remaining agnostic
to semantic details, it allows for fl exible, project specifi c
customization. Where projects need to support user
interactions that go beyond typical navigation scenarios, these
interactions can be integrated into the user interface without
requiring changes to the lower-level tools used to access the
facsimile.
Discussion
AFED has proven to be a useful model in our work. We have
deployed a proof of concept prototype based on the AFED
model. Several of the facsimile editions constructed by the
Cervantes Project use this prototype behind the scenes. Given its
success in these reader’s interfaces, we are working to develop
a Web-based editing toolkit. This application will allow editors
to quickly defi ne annotations and use those annotations to
describe a facsimile edition. We anticipate completing this tool
by the summer of 2008.
By using multiple, hierarchical annotation streams, AFED’s
expressive power falls under the well-studied class of
document models, known as OHCO (ordered hierarchy of
content objects). Specifi cally, it is an instance of a revised form
of this generic model known as OHCO-3, [Renear 1996].
Whereas most prior research and development associated
with the OHCO model has focused on XML-based, transcribed
content, we have applied this model to the task of representing
macro-level structures in facsimile editions.
Focusing on macro-level document structure partially isolates
the AFED model from the non-hierarchical nature of documents
both in terms of the complexity of the required data structures,
and in terms of providing simplifi ed model to facilitate system
implementation. If warranted by future applications, we can
relax AFED’s hierarchical constraint. Relaxing this constraint
poses no problems with the current prototype; however,
further investigation is needed to determine potential benefi ts
and drawbacks.
In addition to macro-level structures, a document model
that strives to represent the visual content of a document
for scholarly purposes must also account for fi ne-grained
structures present in individual images and provide support
for encoded content at a higher level of detail. We envision
using the AFED model in conjunction with models tailored for
these low-level structures. We are working to develop a model
for representing fi ne-grained structure in visually complex
documents grounded in spatial hypermedia theory.
Acknowledgements
This material is based upon work support by National Science
Foundation under Grant No. IIS-0534314.
References
[Dekhtyar 2006] Dekhtyar, A., et al. Support for XML markup
of image-based electronic editions. International Journal on
Digital Libraries 6(1) 2006, pp. 55-69.
[Eaves 2007] Eaves, M., Essick, R.N., Viscomi, J., eds. The William
Blake Archive. http://www.blakearchive.org/ [24 November
2007]
[Furuta 2001] Furuta, R., et al. The Cervantes Project: Steps
to a Customizable and Interlinked On-Line Electronic
Variorum Edition Supporting Scholarship. In Proceedings of
ECDL 2001, LNCS, 2163. Springer-Verlag: Heidelberg, pp. 71-
82.
[Kiernan 2004] Kiernan K., et al. The ARCHway Project:
Architecture for Research in Computing for Humanities
through Research, Teaching, and Learning. Literary and
Linguistic Computing 2005 20(Suppl 1):69-88.
[Lecolinet 2002] Lecolinet, E., Robert, L. and Role. F. Textimage
Coupling for Editing Literary Sources. Computers and
the Humanities 36(1): 2002 pp 49-73.
[McGann 2007] McGann, J., The Complete Writings and Pictures
of Dante Gabriel Rossetti. Institute for Advanced Technology
in the Humanities, University of Virginia. http://www.
rossettiarchive.org/ [24 November 2007]
[Monroy 2007a] Monroy, C., Furuta, R., Stringer, G. Digital
Donne: Workfl ow, Editing Tools and the Reader’s Interface of
a Collection of 17th-century English Poetry. In Proceedings of
Joint Conference on Digital Libraries JCDL 2007 (Vancouver, BC,
June 2007), ACM Press: New York, NY, pp. 411-412.
[Monroy 2007b] Monroy, C., Furuta, R., Castro, F. Texts,
Illustrations, and Physical Objects: The Case of Ancient
Shipbuilding Treatises. In Proceedings of ECDL 2007, LNCS,
4675. Springer-Verlag: Heidelberg, pp. 198-209.
[Renear 1996] Renear, A., Mylonas, E., Durand, D. Refi ning our
Notion of What Text Really Is: The Problem of Overlapping
Hierarchies. In Ide, N., Hockey, S. Research in Humanities
Computing. Oxford: Oxford University Press, 1996.
[Viscomi 2002] Viscomi, J. (2002). ‘Digital Facsimiles: Reading
the William Blake Archive’. Kirschenbaum, M. (ed.) Imagebased
Humanities Computing. spec. issue of Computers and
the Humanities, 36(1): 27-48.