| Summary: | Early English Books Online contains facsimiles of virtually every English work printed between 1473 and 1700; some 125,000 publications. In September 2009, the Text Creation Partnership released the second instalment of transcriptions of the EEBO collection, bringing the total number of transcribed works to 25,000. It has been estimated that this transcribed portion contains 1 billion words of running text. With such large datasets and the increasing variety of historical corpora available from the Early Modern English period, the opportunities for historial corpus linguistic research have never been greater. However, it has been observed in prior research, and quantified on a large-scale for the first time in this thesis, that texts from this period contain significant amounts of spelling variation until the eventual standardisation of orthography in the 18th century. The problems caused by this historical spelling variation are the focus of this thesis. It will be shown that the high levels of spelling variation found have a significant impact on the accuracy of two widely used automatic corpus linguistic methods - Part-of-Speech annotation and key word analysis. The development of historical spelling normalisation methods which can alleviate these issues will then be presented. Methods will be based on techniques used in modern spellchecking, with various analyses of Early Modern English spelling variation dictating how the techniques are applied. With the methods combined into a single procedure, automatic normalisation can be performed on an entire corpus of any size. Evaluation of the normalisation performance shows that after training, 62% of required normalisations are made, with a precision rate of 95%.
|
|---|
