Parsing the Past - Identification of Verb Constructions in Historical Text
Eva Pettersson, Beáta Megyesi, Joakim Nivre · 2012
Even though NLP tools are widely used for contemporary text today, there is a lack of tools that can handle historical documents. Such tools could greatly facilitate the work of researchers dealing with large volumes of historical texts. In this paper we propose a method for extracting verbs and their complements from historical Swedish text, using NLP tools and dictionaries developed for contemporary Swedish and a set of normalisation rules that are applied before tagging and parsing the text. When evaluated on a sample of texts from the period 1550– 1880, this method identifies verbs with an F-score of 77.2 % and finds a partially or completely correct set of complements for 55.6 % of the verbs. Although these results are in general lower than for contemporary Swedish, they are strong enough to make the approach useful for information extraction in historical research. Moreover, the exact match rate for complete verb constructions is in fact higher for historical texts than for contemporary texts (38.7% vs. 30.8%). 1