Digitizing the Act of Papyrological Interpretation: Negotiating Spurious Exactitude and Genuine Uncertainty

1. Digitization is a sampling
process
The act of papyrological interpretation is
a continuous thought process that unravels
non-linearly (Youtie, 1963; Terras, 2006).
Throughout this sense-making process, ancient
and scarcely legible documents progress from
the status of pure physical objects to that of
meaningful historical artefacts. Within the e-
Science and Ancient Documents project,
1
we
aim to make explicit some of the implicit
mechanisms that contribute to the development
of hypotheses of interpretation by designing and
implementing a web-based software offering
digital support for the hermeneutic task. This
tool aims to record the intermediary hypotheses
of interpretation, thus keeping track of the
rationale and allowing easier and better revision
when required. The model we have adopted
(Roued Olsen et al., 2009) is that of a
network of percepts, where a percept is defined
as a minor interpretation that stems from
perception and cognition (Tarte, 2010). An
understanding of expert knowledge and of
how it is mobilised is required to identify the
crucial steps that allow us to reconstruct a
rationale. The level of the granularity at which
we choose to provide support also is essential
to the usability of the software. Further, each
percept, each intermediary interpretation, each
piece of evidence used either to support or to
invalidate a claim is potentially mutable. The
implementation of an Interpretation Support
System (ISS) taking these considerations into
account poses the question of how to digitize
or record a thought process; it is an epitome
of the ‘continuous-to-discrete’ (or ‘analogue-to-
digital’) problem. In the theoretical and life
sciences, measurement devices are developed to
sample the signals of interest. Then, based on
the discrete sampled signal, on an underlying
model of the behaviour of the signal, and on
more general knowledge of signal processing
and information theory (e.g. the Nyquist-
Shannon sampling theorem), the continuous
signal can be reconstructed with minimal
deviation from the original signal. Similarly, the
ambition of our ISS is, based on an appropriate
model of the papyrological hermeneutic task,
to allow the user to capture the information
necessary to the reconstruction of the rationale
that yielded a given interpretation. Two
difficulties in sampling the interpretive thought
process are: (1) to take advantage and to beware
of the sense of scientific rigour that digitization
conveys; and (2) to allow the digital expression
of uncertainty and mutability of percepts.
In this paper, we explain how, while attempting
to digitize the papyrological interpretation act,
we strive to avoid spurious exactitude and
accommodate genuine uncertainty.
2. Choosing what to digitize and
how
The papyrological model of reading developed
by Terras (Terras, 2006) identified ten levels
of reading, corresponding to ten levels of
granularity at which an interpretation in
progress is discussed. Tools stemming from
web-based technology (Bowman et al., 2010),
image processing (Tarte et al., 2009) and
artificial intelligence can help support the
digitization of the hermeneutic task (see fig. 1).
To illustrate how we negotiate between spurious
exactitude and genuine uncertainty, we focus
here on two specific stages of the digitization
of the papyrological interpretation process: how
artefact digitization is being performed; and
how, by identifying the mechanisms that trigger
the jumps between the ten levels of reading,
we propose to address the representation of
uncertainty.

2
Figure 1: Model of the act of interpretation detailing the
various levels of reading based on (Terras, 2006) and
the tools involved in the implementation of our ISS.
2.1. Digitizing the text-bearing
artefact
The problem of spurious exactitude is most
prevalent at the stage where the text-bearing
artefact is digitized. For stylus tablets, for
example, high-resolution pictures are not
enough. The materiality of the artefact needs
to be taken into account in a similar way
as the experts exploit it in the real-world.
The guiding principle we choose to follow
in this context is mimesis. Indeed, when the
papyrologists have physical access to such an
incised tablet, in order to see better the incised
text, they lay the tablet flat on their hand,
lift it at eye level and expose it to raking
light while applying pitch-and-yaw motions to
it. This signal enhancement strategy exploits
the shadow-stereo principle by which stronger
and more mobile shadows and highlights occur
at the incisions than they do on the bare
wood; the text is thereby accentuated. Digitally
imitating this process, we capture series of
images of incised tablets with varying light
positions (Brady et al., 2005), allowing users
to reproduce digitally the visual phenomenon
they naturally exploit in the real-world. Note
that, similarly to signal measurement devices,
we adopt a digitization process that is already
part of the interpretation process. And the
intention behind artefact digitization, as well
as the intention behind signal measurement,
is always an implicitly set variable that affects
downstream results.
2.2. Digitizing the thought process
When attempting to capture the milestones of
the thought process that builds an interpretation
of an ancient or damaged text, we need to
capture the uncertainty of the intermediary
percepts and their mutability. A numerical
approach to uncertainty such as bayesian
networks could have been adopted, but
such a quantification of uncertainty usually
presupposes that problems are complete, i.e.
that all the alternatives to a given situation
are known (Parsons & Hunter, 1998). Instead,
we have decided to turn to argumentation
theory (Parsons & Hunter, 1998) and theory
of justification (Haack, 1993), and combine
them to provide a formal, yet invisible,
epistemological framework that will allow us
to point out inconsistencies without forbidding
them. Indeed, inconsistencies in an unravelling
interpretation naturally occur and can be rooted
either in the implicit expectations of the user or
in the validity of the actual claims (see Tarte,
2010 for an account of an inconsistency due
to an implicit assumption and its resolution).
The balance to be found here (Shipman III &
Marshall, 1999) is between on the one hand the
usefulness of a formal system as a backbone
to support reasoning under uncertainty and
make implicit mechanisms explicit, and on
the other the excess of formalism and explicit
formulation that can become a hindrance by
creating for the user an overhead disruptive
to the interpretation process. Here again, our
design choice is based on the observation of the
experts at work. We allow both palaeographical
and philological approaches in combination,
through the possibility of tracing the letters
and of handling the text as a crossword
puzzle (Roued-Cunliffe, 2010); these are both
approaches that we have identified as the main
strategies experts develop when interpreting
documents (Tarte, 2010). The expression of
uncertainty is then made inherent to the
mode of interaction with the artefact, and the
transposition of the real-world tasks of drawing
and crossword puzzle solving allows us to keep
the interface intuitive while, in the background,
more formal mechanisms can run routines such
as consistency checks and consultation and
constitution of knowledge bases. In her doctoral
work, Roued-Cunliffe (Roued-Cunliffe, 2010) is
currently concentrating on the crossword puzzle
approach and combining it to consultation of
knowledge bases through web-services.