Towards an Interpretation Support System For Reading Ancient Documents

Constructing readings of damaged and abraded ancient
documents is a difficult, complex, and timeconsuming
task, often involving reference to a variety
of linguistic and archaeological data sets, and the integration
of previous knowledge of similar documentary
material. Due to the involved and lengthy reading process,
it is often difficult to record and recall how the final
interpretation of the document was reached, and which
competing hypotheses were presented, adopted, or discarded
in the process of reading. This paper discusses the
development of an Interpretation Support System (ISS),
which aims to provide a system, which can aid the dayto-
day reading of ancient documents, and in future other
damaged documents, by keeping track of how these are
interpreted and read. Such a system will facilitate the
process of transcribing texts by providing a framework
in which experts can record, track, and trace their progress
when interpreting documentary material. Furthermore,
it will allow continuity between working sessions,
and the complete documentation of the reading process
that has hitherto been implicit in published editions.
Introduction
The process of reading ancient documents is traditionally
undertaken by an expert such as an epigrapher,
palaeographer or papyrologist. The expert uses accumulated knowledge combined with external resources
to piece together an interpretation of each ancient document.
Such interpretation is often prolonged, and it can
be difficult for experts to maintain a record of the interpretations
made whilst undertaking their reading (Youtie
1963). This is important when disputing interpretations
and sharing hypotheses with other experts, or pausing
the reading of an ancient text and hoping to continue the
same thought process at a later time.
The Image, Text, Interpretation: e-Science, Technology
and Documents project (also known as eSAD: e-Science
and Ancient Documents, http://esad.classics.ox.ac.uk),
aims to use computing technologies to aid experts in
reading ancient documents. The project is developing an
Interpretation Support System (ISS) that can support the
day-to-day reading and interpretation of ancient documents.
This involves advanced IT tools that can aid the
interpretation of damaged texts such as the stylus tablets
from Vindolanda (http://vindolanda.csad.ox.ac.uk) and
image processing algorithms to analyse detailed digital
images of the documents (Tarte et al. 2008).
Background
Although Classics as a subject has made much use of
information technology (see Crane 2008 for an overview),
the use of IT to aid in the actual reading process
of ancient documents is in its infancy. Terras (2006)
developed a prototype system which demonstrated that
it was possible to propagate plausible and useful interpretations
of ancient texts, in a realistic timeframe. This
used linguistic and palaeographic datasets to provide the
“knowledge base” which could inform a decision making
system to aid experts in reading texts.
Decision Support Systems (DSS) have previously been
developed in the Department of Engineering Science at
the University of Oxford to aid multi-disciplinary teams
working with cancer patients in making decisions about
their treatment (Austin et al. 2008). This system is based
on a set of rules and allows experts to analyse and interpret
digital images while recording decisions made
about diagnosis and treatment, and suggesting possible
next action steps.
Building the Interpretation Support System
The research presented here, though inspired by the
above-mentioned medical application, shifts the focus
from a Decision Support System to an Interpretation
Support System (ISS). In contrast with medical practitioners,
experts transcribing ancient documents do not
make decisions based on evidence but instead create interpretations
of the texts based on their perception. The
ISS relies upon the idea that an interpretation is made
up of a network of minor perceptions (percepts) ranging
from low level percepts such as “these three line fragments
are an incised stroke” to higher level percepts
such as “these five letters can make up the word ‘legio’”.
We want to make this otherwise implicit network of percepts
explicit in a human-readable format through a web
browser based application. To build an explicit network
of percepts leading to an interpretation, we define an elementary
percept as a region of an image that contains
what is perceived to be a grapheme. The image can then
be divided into cells where each cell is expected to contain
what is perceived as a character or a space. This division
of the image constitutes a tessellation. A single
document might be tessellated in various ways and each
of the tessellations might yield either an interpretation
or a dead-end, but in both cases, the explicit network of
percepts will document this.
The making of the tessellation, which in itself is an interpretive
process, marks the boundary between lower and
higher level percepts. The lower level percepts are based
on physical identification of the features of the document
(through the application of image analysis methods to
detect features such as strokes); the higher level percepts
(words, groups of words) work more towards gradually
adding meaning to the transcription in progress. Ultimately,
an interpretation can then be represented as a
network of substantiated percepts, which will be made
explicit through an ontology. Here an ontology is defined
as a model of the concepts found in a text such as the
concept of a word that contains several characters.
The ontology aims to make the rationale behind the network
of percepts visible and thus expose both: (a) some
of the cognitive processes involved in damaged texts interpretation;
and (b) a set of arguments supporting the
tentative interpretation. The system will use the ontology
as a framework to assist the expert through the different
levels of percepts ultimately yielding a final transcription.
The transcription is a part of the overall edition of
which there may be several and it will be formatted in
EpiDoc style XML (http://epidoc.sourceforge.net/) allowing
further interaction with other documents.
Building the Knowledge Sets
Much of the knowledge base that serves as justification
for the commitment to a given percept during the interpretation
process will come from the experts. However,
letter frequency, word-and character-lists from documents
such as the Vindolanda ink tablets will provide
an invaluable source of information which can be used
to generate the statistical likelihood of patterns in language and writing which may appear on the texts. We
have taken a new approach to the XML encoding of the
Vindolanda ink tablets based on contextual encoding
(Hippisley 2005). The Vindolanda ink tablets have been
encoded with EpiDoc standard XML to a very detailed
granularity. The contextual encoding which is then imposed
on the documents consists on encoding words,
person names, geographical place names, calendar references
and abbreviations. For example any instance of the
word pulli (=’chickens’) in a document will be encoded
<w lemma=”pullus” n=”1”>pulli</w>. This encoding
provides us with the information that the word pulli has
the lemma pullus under which we can index this instance
of the word and that this is the first instance of this lemma
in the document. This information has been used to
generate word frequency lists and is extremely useful as
a part of a knowledge base to build the ISS on. Further
knowledge bases will be generated from the marked up
dataset, to provide uncertainty and character frequency
lists. Additionally, further work will be undertaken with
the experts to generate lists of common percepts and
interpretation making processes. By encoding these in
XML, the knowledge sets for the system will be in place.
Conclusion
The construction of an Interpretation Support System
for ancient texts, although ambitious, will provide a
useful tool for those experts who work on developing
interpretations of damaged documents by facilitating
and recording the evolving interpretation process. Additionally,
by making explicit the percepts which trigger
such transcriptions of ancient documents, we will further
our knowledge of the reasoning process undertaken by
experts in propagating readings of ancient documents.
Furthermore, the successful development of an image
and language based Interpretation Support System will
provide a set of tools which can be adopted and adapted
by other domains which rely on detailed analysis and interpretation
of image based material.
References
Austin, M., Kelly, M., Brady M. (2008), “The benefits
of an ontological patient model in clinical decision-support”.
In Fox, D. and Gomes, C. P. (eds), AAAI, pages
1774–1775. AAAI Press.
Crane, G. (2008) “Classics and the Computer: An End
of the History”. In Schreibman, S., Siemens, R., and
Unsworth, J. (eds). A Companion to Digital Humanities.
Blackwell Companions to Literature and Culture. Blackwell.
http://www.digitalhumanities org/companion/vie
w?docId=blackwell/9781405103213/9781405103213.
xml &chunk.id=ss1-2-4&toc.depth=1&toc.id=ss1-2-
4&brand=default
Hippisley, D. (2005) “Encoding the Vindolanda tablets:
an investigation in contextual encoding using XML and
the EpiDoc standards.” MA Dissertation submitted for
the MA in Electronic Communication and Publishing,
School of Library, Archive and Information Studies,
UCL.
Tarte, S. M., Brady, J. M., Roued Olsen, H., Terras, M.,
Bowman, A. K. (2008), “Image Acquisition and Analysis
to Enhance the Legibility of Ancient Texts”. UK e-Science
Programme All Hands Meeting 2008 (AHM2008),
Edinburgh, September 2008.
Terras, M. (2006).”Image to Interpretation: Intelligent
Systems to Aid Historians in the Reading of the Vindolanda
Texts”. Oxford Studies in Ancient Documents. Oxford
University Press.
Youtie, H. C. (1963): “The Papyrologist: Artificer of
Fact”. GRBS 4 (1963), p. 19-32.