Application of a POS Tagger to a Novel Chronological Division of Early Modern German Text
Brendan Ferreri-Hanberry · Carolina Digital Repository (University of North Carolina at Chapel Hill) · 2019
This paper describes the application of a part-of-speech tagger to a particular configuration of historical German documents. Most natural language processing (NLP) is done on contemporary documents, and historical documents can present difficulties for these tools. I compared the performance of a single high-quality tagger on two stages of historical German (Early Modern German) materials. I used the TnT (Trigrams 'n' Tags) tagger, a probabilistic tagger developed by Thorsten Brants in a 2000 paper. I applied this tagger to two subcorpora which I derived from the University of Manchester's GerManC corpus, divided by date of creation of the original document, with each one used for both training and testing. I found that the earlier half, from a period with greater variability in the language, was significantly more difficult to tag correctly. The broader tag categories of punctuation and "other" were overrepresented in the errors.