Finding Names in Trove: Named Entity Recognition for Australian Historical Newspapers

Finding Names in Trove: Named Entity Recognition for Australian Historical Newspapers Sunghwan Mac Kim∗ Data61, CSIRO, Sydney, Australia Mac.Kim@csiro.au Steve Cassidy Department of Computing Macquarie University Sydney, Australia Steve.Cassidy@mq.edu.au Abstract Historical newspapers are an important resource in humanities research, provid- ing the source materials about people and places in historical context. The Trove col- lection in the National Library of Australia holds a large collection of digitised news- papers dating back to 1803. This paper re- ports on some work to apply named-entity recognition (NER) to data from Trove with the aim of supplying useful data to Hu- manities researchers using the HuNI Vir- tual Laboratory. We present an evalua- tion of the Stanford NER system on this data and discuss the issues raised when ap- plying NER to the 155 million articles in the Trove archive. We then present some analysis of the results including a version published as Linked Data and an explo- ration of clustering the mentions of certain names in the archive to try to identify in- dividuals. 1 Introduction In recent years, digitised newspaper archives have appeared on the web; they make fascinat- ing reading but also provide important primary sources for historical research. The Trove (Hol- ley, 2010)1 Newspaper collection at the National Library of Australia (NLA) provides an inter- face for users to search and browse the collec- tions of scanned pages using an optical character recognition (OCR) based transcript of each article. While the OCR results contain errors, they provide enough detail to enable a full-text index to return relevant results to search terms. The documents ∗The work was done while the first author was a research associate at Macquarie University and was supported by a grant from NeCTAR. 1http://trove.nla.gov.au/ stored in the Trove archive are made freely avail- able for any purpose by the National Library of Australia. An abundance of natural language process- ing (NLP) tools have been developed for Digi- tal Humanities (Brooke et al., 2015; Scrivner and K¨ubler, 2015) and such tools can greatly facilitate the work of Humanities scholars by automatically extracting information relevant to their particular needs from large volumes of historical texts. This project explores the use of Named Entity Recog- nition on the Trove Newspaper text to provide a resource for Humanities scholars. Newspapers are an important repository for his- torical research. Digitisation of newspaper text via Optical Character Recognition (OCR) enhances access and allows full text search in the archive. It also supports more sophisticated document pro- cessing using Natural Language Processing (NLP) techniques. Europe and the United States have ac- tively participated in research on digitised histor- ical newspapers and developed web-based appli- cations using NLP to provide visualisation of use- ful information (Willems and Atanassova, 2015; Torget et al., 2011). The web-based applica- tions have empowered digital humanities schol- ars to efficiently exploit historical newspaper con- tent. The Europeana Newspapers project was per- formed to provide access to digitised historical newspapers from 23 European libraries (Willems and Atanassova, 2015). They used …

Content truncated to 500 words