From Index Locorum to Citation Network: an Approavch to the Automatic Extraction of Canonical Reeferences and its Applications to the Study of Classical Texts

摘要

My research focusses on the automatic extraction of canonical referencesfrom publications in Classics. Such references are the standardway of citing classical texts and are found in great numbers throughoutmonographs, journal articles and commentaries.In chapters 1 and 2 I argue for the importance of canonical citationsand for the need to capture them automatically. Their importance andfunction is to signal text passages that are studied and discussed, oftenin relation to one another as can be seen in parallel passages found inmodern commentaries. Scholars in the field have long been exploitingthis kind of information by manually creating indexes of cited passages,the so-called indices locorum. However, the challenge we now face isfind new ways of indexing and retrieving information contained in thegrowing volume of digital archives and libraries.Chapters 3 and 4 look at how this problem can be tackled by translatingthe extraction of canonical citations into a computationally solvableproblem. The approach I developed consists of treating the extractionof such citations as a problem of named entity extraction. This problemcan be solved with some degree of accuracy by applying and adaptingmethods of Natural Language Processing. In this part of the dissertationI discuss the implementation of this approach as a working prototypeand an evaluation of its performance.Once canonical references have been extracted from texts, the web ofrelations between documents that they create can be represented as anetwork. This network can then be searched, manipulated, visualisedand analysed in various ways. In chapter 5 I focus specifically on howthis network can be leveraged to search through bodies of secondaryliterature. Finally in chapter 6 I discuss how my work opens up newresearch perspectives in terms of visualisation, analysis and the applicationof such automatically extracted citation networks.