Till den historiska korpuslingvistiken - stavningsnormalisering och automatisk annotering av verbpartiklar i en historisk text

Advances in computational linguistics can provide new opportunities for historical linguistics, but considerable work remains before we have resources for older Swedish that meet the standards generally set-up by historical linguists. This paper uses a resource for automatic spelling normalization developed by Pettersson et al. (2013) on a play from the end of the 18th century. Both the original and the modernized text are lemmatized and tagged using resources developed for modern Swedish, and the results are compared. Finally, particle verbs are annotated using two different automatic methods; one lexical, using the Swedish Academy word list (2015), one with a morphosyntactic tagger. The results are evaluated in relation to manual annotation. We conclude that automatic spelling normalization has good effects on lemmatization and some effect on tagging. With a few crucial exceptions, errors are not systematic in a linguistically relevant sense. With some adjustments, normalization can therefore significantly improve the usefulness of historical corpora. Annotation of particles is more difficult, and both automatic methods yield results that are systematically skewed in a way that makes them difficult to use in a study of syntactic change