Low Resource Methods for Medieval Document Sections Analysis

17647

Authors

nan

Abstract

This paper describes a small but unique digitized collection of medieval Latin charters. This collection consists of 57 charters of 7 types illustrating various purposes of issuance by the Royal Chancellery. Sections in these documents were manually annotated for deeper analysis of the structure of issued charters. This paper also describes two baseline methods for an automatic and semi-automatic analysis and detection of sections of diplomatic documents. The first method is based on an information retrieval paradigm, and the second one is an adaptation of Hidden Markov Models. Both methods were proposed to work with respect to a small amount of available train data. Even though these methods were specifically proposed to work with medieval Latin charters, they can be applied to any documents with partially repetitive character. Keywords: medieval documents, document sections detection, information retrieval, Latin 1.

Keywords

nan

Full Content

Low Resource Methods for Medieval Document Sections Analysis Petra Galuˇsˇc´akov´aab and Lucie Neuˇzilov´ac aUMIACS, University of Maryland College Park, MD, USA petra@umiacs.umd.edu cCharles University, Faculty of Arts, Department of Auxiliary Sciences of History Prague, Czech Republic Abstract This paper describes a small but unique digitized collection of medieval Latin charters. This collection consists of 57 charters of 7 types illustrating various purposes of issuance by the Royal Chancellery. Sections in these documents were manually annotated for deeper analysis of the structure of issued charters. This paper also describes two baseline methods for an automatic and semi-automatic analysis and detection of sections of diplomatic documents. The first method is based on an information retrieval paradigm, and the second one is an adaptation of Hidden Markov Models. Both methods were proposed to work with respect to a small amount of available train data. Even though these methods were specifically proposed to work with medieval Latin charters, they can be applied to any documents with partially repetitive character. Keywords: medieval documents, document sections detection, information retrieval, Latin 1. Introduction Medieval charters issued by the Royal Chancellery were composed with respect to their typical structure (Guyot- jeannin et al., 1993). They often consisted of repetitive sections and phrases, what is frequent in diplomatic lan- guage. Type and section ordering depends on the document type, which is defined by its purpose, type of the issuer and period of issuance. The purpose of this work is to 1) cre- ate a digital collection of medieval diplomatic charters, 2) provide manual annotations of a common structure of these charters, and 3) propose general methods for automatic de- tection of sections of these charters. The proposed meth- ods should be trained and tested on this specific collection. As the number of diplomatic charters created using a given structure is often very limited and the size of the available train set is thus restricted, we aim at the methods which use a small amount of manually annotated documents for the training. We expect that providing an annotated collec- tion of medieval charters and tools for automatic or semi- automatic recognition of document sections and their clas- sification will significantly facilitate reading and process- ing of charters. It will allow further deeper analysis of in- dividual diplomatic sections of charters and provide tools which will enable history researchers to think about rela- tions and classification of material instead of spending time with slow and monotonous work and thus speeds up the whole research process. 1.1. Medieval Manuscript Collections Lately, there is a growing effort towards digitization of medieval manuscripts. A number of such collections are provided as online digital libraries, e.g. Manuscripto- bPreviously Charles University, Faculty of Mathematics and Physics, Institute of Formal and Applied Linguistics rium1, Digital Scriptorium2, Manuscripta.at3, Syriaca4, E- codices5, Saramusik6 and Monasterium7. Manuscriptorium is a project of Czech National library which provides access to manuscripts, incunabula, early printed books, maps and charters. Digital Scriptorium is a consortium of American libraries and museums which provides free online …

Content truncated to 500 words

Labeled Text

This paper describes a small but unique digitized collection of medieval Latin charters. This collection consists of 57 charters of 7 types illustrating various purposes of issuance by the Royal Chancellery. Sections in these documents were manually annotated for deeper analysis of the structure of issued charters. This paper also describes two baseline methods for an automatic and semi-automatic analysis and detection of sections of diplomatic documents. The first method is based on an information retrieval paradigm, and the second one is an adaptation of Hidden Markov Models. Both methods were proposed to work with respect to a small amount of available train data. Even though these methods were specifically proposed to work with medieval Latin charters, they can be applied to any documents with partially repetitive character. Keywords: medieval documents, document sections detection, information retrieval, Latin 1.

Publication Details

Paper ID:
17647
Year:
2018
Publication:
TIPSTER
Publication Type:
conf
DOI:
nan