Image, Text, Interpretation: e-Science, Technology and Documents

Decipherment and interpretation of manuscripts and documents is a central and fundamental component of research in all periods of literature and history and in related specialised disciplines such as linguistics, epigraphy, palaeography and archaeology. Such texts are, however, often very difficult to read, not least because of physical damaged which tends to be more severe the older the document is. Advanced imaging and computational technology can significantly aid the process of deciphering, reading and interpreting documents. In order to develop an effective IT-based technology in an e-science framework, it is also crucially necessary to analyse and understand the visual and cognitive processes which are involved in deciphering damaged or deteriorated texts. This is seldom made explicit to the non-expert reader of the texts and its scholarly interpretation. There are as yet no computational facilities for scholars working on propagating readings of texts to share data, discuss procedures, make explicit their reasoning, and make use of relevant advances in technological methods. This research seeks to develop a system which will aid the scholarly humanities community in reading manuscripts and documents from a wide range of literatures and cultures / emphatically not 'automating' reading, but providing a tool for the expert reader or scholar to harness and document the reasoning process; share, distribute, and manipulate data; and utilise advanced computational techniques and processing power afforded by e-Science technologies.

Building on successful earlier research on image-processing and decipherment in dealing with the unique Roman ink and stylus tablets from Vindolanda, the proposed new research will develop, in a web-based environment, training methods and a suite of tools, in effect a palaeographer's workbench, to aid in the reading of degraded documents from a wide range of periods and subject-areas. After being built and tested on Greek and Roman written and incised material, which is well-known and readily available to the research team, the techniques and methods will be tested and further development carried out for other types of documents and manuscripts in other languages and periods, with adequate documentation, to encourage wider use. Moreover, the potential applications of such image processing algorithms are not limited to reading manuscripts and documents. Understanding issues such as local energy and phase in image processing will result in knowledge transfer to domains such as medical imaging, where application of these technologies is fundamental.

The development of the system demands novel computational techniques, both in image processing, and in computational reasoning architectures. The approach taken here depends crucially on recognising the indispensability of the accumulated knowledge and judgment of the experts and aims to assist them, using a combination of image processing and data mining of (e.g.) letter forms, letter frequency, and word lists, techniques which have been demonstrated as a proof of concept in our earlier research. Proposals for the investigation into the mobilization of knowledge for the reasoning and interpretation of image data, or across semantic levels, are few. Yet there is a requirement for the investigation and development of such architectures to aid in tasks where uncertainty and probability need to be made explicit when reasoning about information (such as in medical imaging). Novel techniques in engineering science have been developed to allow data to be compared and contrasted across different semantic boundaries; these have been demonstrated as a proof of concept in propagating readings of the Vindolanda writing-tablets. Further development of these techniques and technologies will refine the approach, increase its effectiveness, and add to our understanding of role uncertainty, constraints reasoning, and analysis in image data.