Multi-Level Variation

The paper is a follow-up to a joint presentation by
Dino Buzzetti and myself at the Digital Humanities
conference in Oulu 2008[1]. There, we applied a case
study of a complex medieval text to generate a general
model towards dynamic editions. It lies in the nature
of such a conference that a 20 minutes presentation is
hardly enough to achieve both: an introduction to a project
whose textual variation is vital in understanding its
content and genesis, and to derive a general view of both
its markup and its structural information. One of the reviewers
commented at that time, already before the conference:
“I also feel far too much is being proposed here
for one twenty-minute talk; either part I or part II would
still be straining the limits of what can fit”. This paper,
proposed for the Digital Humanities conference 2009
will thus concentrate on the case study kundige bok, but
bring in new results from recent research on the text and
its variants. Kundige bok is a 15th century legal text manuscript,
characterised by many revisions where the different layers
of the text usually represent a new stage of the town
law and must thus be treated equally[4].
The digital edition of kundige bok tries to achieve a dynamic
approach, allowing the user to generate views of
the text in the status it had at a certain point in time.
Variation in the case of kundige bok is three-fold:
1. The textual variation, determined by the operations
performed on the text by the medieval
scribes.
2. The sequence of revisions / layers of the text as
a whole respectively the underlying law reflecting
its genesis. This can be quite linear but also
very complex as for instance Peter Robinson
has shown in his work on Chaucer’s Canterbury
Tales.
3. The editor’s or user’s interpretation of this sequence.
Owing to the fact that we often cannot
reveal the actual development of the text due
to our limited knowledge about its production,
this leads to an interpretational variation.
While we can be quite sure about the textual variation itself,
the more we try to group these operations on the text
to revisions and text layers, the more our understanding
of the text becomes inconclusive and uncertain (see figure
1). This is illustrated by an example which is also
used in the following to explain the three kinds of variation
that can be found in kundige bok and can be represented
using graph structures, similar to the works by
e.g. Huitfeldt and Sperberg-McQueen [2] or Schmidt[3]. The example, taken from kundige bok, is a statute about
beer brewery. What we find nowadays looking at the
manuscript, is the sentence “We ock vorschote 100
marck, de darf 3 warve bruwen” with the numbers ‘100’
and ‘3’ struck out and the numbers ‘150’ and ‘2’ written
above the line. This allows four different variants of the
regulations as shown in figure 2 using a graph representation.
Here, I use the vertices as a container for the portions
of the text and the edges to describe the variants.
Each variant (A-D) is thus represented as a path through
the graph.
However, we do not know per se which of these variants
really have existed as town law and in which order. In other words: the grouping of the operations on the text
(e.g. the replacement of ‘100’ by ‘150) to revisions and
the (chronological) ordering is ambiguous. As long as
we have no further information, we must live with three
different possibilities through the genesis of the text,
each of those representing a different grouping of the
operations on the text to revisions and different layering.
Comparable to the example of textual variants, this
kind of interpretational variation can be represented by
a graph structure as well (figure 3). The three paths (a)-
(c) through the graph represent the possible evolutions
of the text. Note, that there is no edge between B and C
since this would be possible only if one of the changes
was reversed. This paper discusses furthermore the question of whether
a graph representation like this and the underlying mathematical
model is indeed suitable to deal with the uncertainty
that forced its creation. Although we do not have
evidence that one of the three paths actually took place,
we might have a clue which makes one or two more
probable than the others. The extension of the directed
graph in to a weighted directed graph could be a way to
deal with this additional information. The weights, put
on the edges can indicate which variation is more likely
than the other and easily be changed should new information
be available (figure 4).
Figure 4: Interpretational variation as weighted graph.
This data can then be used for various purposes: firstly,
to create a dynamic visualisation of the different variants,
e.g. using colours for the probabilities, secondly to
automatically process the data, e.g. ignoring paths with
low probabilities. Finally it might be a good approach,
especially when you have a lot of such cases for which
you can give probabilities individually but not for the
evolution of the text as a whole. Applying weights to
all single variations can then lead to an overall view by
mathematical calculation, e.g. computing the “cheapest”
path through the graph, an algorithm related to the Travelling
Salesman (TSP) problem.
Other examples in kundige bok are, of course, more
complex and this was a major issue in analysing the
text and creating the edition and its user-interface. The
question of uncertainty arises generally in the creation
of genetic editions (and these thoughts are also inspired
by the working group on genetic editions within the TEI
Special Interest Group on Manuscripts[6]), thus, dealing
with the described issues might be a step forward towards
a model for encoding and formalisation of genetic
editions as well
References
[1] D. Buzzetti and M. Rehbein, Towards a Model for
Dynamic Editions, paper given at the Digital Humanities
conference, Oulu 2008.
[2] C. M. Sperberg-McQueen and Claus Huitfeldt,
GODDAG: A Data Structure for Overlapping Hierarchies,
in: DDEP-PODDP 2000, ed. P. King and E.V.
Munson, Lecture Notes in Computer Science 2023 (Berlin:
Springer, 2004), pp. 139-160, online: http://www.
w3.org/People/cmsmcq/2000/poddp2000.html.
[3] http://multiversiondocs.blogspot.com.
[4] M. Rehbein, Reconstructing the Textual Evolution of
a Medieval Manuscript, in: Literary & Linguistic Computing,
forthcoming.
[5] http://en.wikipedia.org/wiki/Traveling_salesman_
problem.
[6] M. Rehbein and J. Tonra, Encoding Genetic Editions
— Two Case Studies, paper given at the TEI Members
Meeting, London 2008.