Mimetic Metadata: Linguistic Representations of Visual Objects in Image-Based Electronic Projects

Following a largely formalistic presentation on image description at
ACH-ALLC 2000, the author of that paper was pressed during the Q&A session
to talk at more length about the exact function of the descriptions at the
William Blake Archive. Though clearly operating within the same semantic
field as the Archive's indexing terms, the descriptions are, nonetheless,
distinct from them and, moreover, beg the question of why we need their
second-order mimesis at all given the availability of the digital images,
expertly edited by Joseph Viscomi. Theoretically at least the search
vocabulary could function perfectly well without them, and certainly the pride
of place high-quality digital facsimiles have in the WBA makes obsolete the
need for a prose description that functions as a surrogate for an absent image, in
much the way that descriptive catalogue raisonne entries did prior to
widespread photomechanical reproduction. Divested of any immediately
obvious praxis, then, the descriptions might understandably be construed as
an expendable stratum within the Archive's editorial matrix. Although
contestable in a variety of ways (see, for example, last year's ACH-ALLC
abstract for "Image Description at the William Blake Archive"), the
perception of redundancy and lingering doubts about the usefulness of image
description deserve further consideration, all the more so given the
increasing visibility of descriptive metadata in a growing number of
electronic visual resource collections.

Taking the question of function to heart, I propose this year to open a
window onto the more expansive vista of what I'll call _mimetic metadata_ as
it is practiced across multiple image-based electronic projects. Surveying
the metadata scene, one is struck by the relative lack of trenchant
theoretical scholarship addressing this category of image information.
Among other things it quickly becomes clear that the problem of defining the
relationship between descriptive and more strictly classificatory categories
of information is endemic to the rapidly expanding field of computerized
image indexing. In its recently published guidelines on the topic, for
example, the Digital Imaging Group (comprised of representatives from Kodak,
Canon, Hewlett-Packard, and Fuji, among a host of others) had this to say:
"once an image is retrieved, some data that describes the image but is not
useful when searching may be included. For example--'Craig is the guy
asleep on the lounge' is not all that useful when searching, but is useful
when describing the content." Now it is not my intention to impugn the
recommendations set forth in this ambitious document (a milestone
achievement that attempts to establish much-needed jurisdiction over the
largely makeshift world of metadata production; its discussion on image
capture metadata, for example, is admirably comprehensive and
authoritative), but the tautology underlying this quotation (keywords in
first sentence: "describes . . . useful . . . searching"; keywords in second
sentence: "useful . . . searching . . . describing") does little to help
clarify the nature of the interface between these respective data fields.

Consider another example furnished by the Visual Resource Association's
recently published inventory of subject classification systems for image
collections: its author, Colum Hourihane, profiles controlled vocabularies
in considerable detail (e.g., primary, secondary, and tertiary indexing
levels; thesaurus formats; cross-referencing capabilities, etc.), but
generally appends little more than a brief mention of whether or not
free-text constitutes part of the individual systems under consideration.
Consequently he glosses over the crucial question of how the prose
descriptions work in conjunction with the classificatory vocabulary, taking
recourse in the ambiguous terms "supports" and "accompanies" to suggest how
the one data type relates to the other. Largely through perfunctory
treatment, then, both the VRA inventory and the Digital Imaging Group's
guidelines tend to relegate description to an ill-defined secondary status
(and this despite its strong showing in the systems under review), with
controlled vocabulary more clearly identified as the metadata workhorse.

The foregoing examples suggest that while controlled vocabularies are a
favorite talking point among information managers, the free-text
descriptions that often accompany visual objects in a database haven't
always benefited from the same level or caliber of discussion. The
burden is on us to more ably address these issues in the interest of
advancing the art and science of metadata. What do humanists want to do
with mimetic metadata? What do information managers want to do with it?
What do we want the computer to do with it? In a presentation of this
length, where it is necessary to forego comprehensiveness in favor of
suggestiveness, my objective is to lay the groundwork for a lengthier study
in four main areas:

1. Definition: I'll provide an extended statement about what mimetic
metadata is and isn't, briefly describe how it fits into the superordinate
class of metadata known as free-text, and illustrate the definition with
examples. I propose the term "mimetic metadata" to designate a particular
class of so-called "added-value" information frequently attached to images
in digitized collections. Specifically, mimetic metadata refers to finely
grained free-text representations of visual objects--what I have elsewhere
more generically labeled "image description." The coinage consequently
excludes what W. J. T. Mitchell would call metapictures, i.e., pictures
about pictures (for example, image representations that occupy a curious
middle ground between mimetic and non-mimetic, such as Blobworld's
representations, developed within the framework of content-based search and
retrieval technology) in the interest of a narrower definition that
establishes an intermodal relationship between the primary and secondary
objects, between the visual data and its verbal metadata. Mimetic metadata
is a linguistic imitation of a pictorial work. As such, it participates in
the long tradition of ekphrasis--a poetic description of a work of art--and
indeed it is one objective of this paper (see below) to historicize the
practice of mimetic metadata, which has fared more or less indifferently in
the library science and art historical literature to date despite its
growing visibility in digital resource collections, including the William
Blake Archive, the Perseus Project, and various image databases being
developed under the tutelage of ICONCLASS evangelist Hans Brandhorst.

2. Function: I'll suggest how mimetic metadata might be deployed in the
research of two scholarly prototypes: the iconographer and the textual
critic.

3. History: I'll point to the catalogue raisonne as a venerable art
historical genre whose traditions of image description merit our sustained
attention, instruct us about past uses and future possibilities of mimetic
metadata, and lend continuity and historical weight to present efforts.

4. Futurity: I'll outline (necessarily cursorily) a speculative
interdisciplinary program that brings iconicity of syntax and the mapping of
visual performance to bear on the way we currently create and think about
mimetic metadata. Though it has
become de rigueur for a scholar working in image/word studies to
invoke the ineluctable gap between the sister arts (Magritte's famous _ceci
n'est pas une pipe_ often serves as a favorite rallying point), new media,
technologies, and disciplinary emphases are challenging this pervasive
assumption in ways that are relevant to the future of mimetic metadata. The
new SVG (Scalable Vector Graphics) image format, for example, illustrates
that images can be generated precisely and accurately from linguistic code.
Taking as my presupposition, then, a greater
degree of congruity between image and word than is perhaps customarily
assumed, I will conclude by looking at the possibilities of maximizing the
iconic (or non-arbitrary) relations that obtain between referent and code--
or in this case between visual object and its linguistic representation.

References

Crane, Gregory, ed. The Perseus Digital Library.
<http://perseus.tufts.edu>.

The Digital Imaging Group. DIG35 Specification: Metadata for Digital
Images. Version 1.0. August 2000.
<http://members.digitalimaging.org/dls/dig35v1.0-Sept00.pdf>.

Eaves, Morris, Robert N. Essick, and Joseph Viscomi, eds. The William Blake
Archive. <http://www.blakearchive.org>.

Hourihane, Colum. Subject Classification for Visual Collections: An
Inventory of Some of the Principal Systems Applied to Content Description in
Images. Columbus: Visual Resources Association, 1999.

Kraus, Kari. "Image Description at the William Blake Archive" ACH-ALLC
2000. University of Glasgow, Scotland. 22 July 2000.
<http://www.nyu.edu/its/humanities/ach_allc2001/>.