Der FoxTagger: Ein Part-of-Speech-Tagger für das Englische mit Mehrwortterm-Erkennung
Fuchs, Markus · Zenodo (CERN European Organization for Nuclear Research) · 2009
This paper describes a part-of-speech tagger for English which is able to detect, lemmatize, and tag multiword terms (MWTs) to determine the correct ratio of function and content words. The multiword term detection poses new problems for the tagging task: First, inflected variants of multiword terms have to be recognized and normalized. Second, we have to cope with additional kinds of ambiguities arising through overlapping MWTs. For the disambiguation module of the tagger we trained a neural network on a subset of the Brown Corpus. However, the corpus had to be adapted to our tagset first, because the Brown tagset contains tags that can stand for function and content words. The multiword term detection algorithm employs WordNet as lexicon and makes use of the collocation probabilities of the Leipziger Wortschatz to resolve overlapping MWTs. The evaluation shows that a tagger with multiword term recognition not only has positive effects on processes following the tagging but also improves the tagging accuracy itself.