A deeply annotated testbed for geographical text analysis

Abstract

This paper describes the development of an annotated corpus which forms a challenging testbed for geographical text analysis methods. This dataset, the Corpus of Lake District Writing (CLDW), consists of 80 manually digitised and annotated texts (comprising over 1.5 million word tokens). These texts were originally composed between 1622 and 1900, and they represent a range of different genres and authors. Collectively, the texts in the CLDW constitute an indicative sample of writing about the English Lake District during the early seventeenth century and the early twentieth century. The corpus is annotated more deeply than is currently possible with vanilla Named Entity Recognition, Disambiguation and geoparsing. This is especially true of the geographical information the corpus contains, since we have undertaken not only to link different historical and spelling variants of place-names, but also to identify and to differentiate geographical features such as waterfalls, woodlands, farms or inns. In addition, we illustrate the potential of the corpus as a gold standard by evaluating the results of three different NLP libraries and geoparsers on its contents. In the evaluation, the standard NER processing of the text by the different NLP libraries produces many false positive and false negative results, showing the strength of the gold standard.

Hearing the silence: finding the middle ground in the spatial humanities? Extracting and comparing perceived silence and tranquillity in the English Lake District

“…In a final exploration of our corpora, we set out to discover not only what was talked about, but where. Place names in the CLDW have been georeferenced using toponym recognition and resolution (Rayson et al 2017) and used to explore broader themes, including local variation in the use of aesthetic language ) and acoustic experience . In contrast, the Geograph corpus explicitly links descriptions to 1km grid squares, and texts can therefore be mapped without any additional processing.…”

Plague Dot Text: Text mining and annotation of outbreak reports of the Third Plague Pandemic (1894-1952)

“…To process the plague reports, we used the Edinburgh Geoparser [Grover et al, 2010], a text mining pipeline which has been previously applied to other types of historical text [Rupp et al, 2013, Clifford et al, 2016, Rayson et al, 2017, Alex et al, 2019. This tool is made up of a series of processing components.…”

An annotated dataset of literary entities

“…Much of this work relies on the ability to extract entities accurately, including work focused on modeling (Bamman et al, 2014;Iyyer et al, 2016;Chaturvedi et al, 2017). And yet, with notable exceptions (Vala et al, 2015;Brooke et al, 2016), nearly all of this work tends to use NER models that have been trained on non-literary data, for the simple reason that labeled data exists for domains like news through standard datasets like ACE (Walker et al, 2006), CoNLL (Tjong Kim Sang and De Meulder, 2003) and OntoNotes (Hovy et al, 2006)-and even historical non-fiction (De-Lozier et al, 2016;Rayson et al, 2017)-but not for literary texts. This is naturally problematic for several reasons: models trained on out-of-domain data surely degrade in performance when applied to a very different domain, and especially for NER, as Augenstein et al (2017) has shown; and without indomain test data, it is difficult to directly estimate the severity of this degradation.…”