Scaling up: using the WWW to resolve PP attachment ambiguities
Martin Volk · Zurich Open Repository and Archive (University of Zurich) · 2000
We have developed a method to resolve ambiguities in prepositional phrase (PP) attachment in German. We measure on the one hand the cooccurrence strength between nouns (N) and prepositions (P) and on the other hand between verbs (V) and prepositions. The competing values of N+P versus V+P are used to decide whether to attach a prepositional phrase to the noun or to the verb. We calculate the cooccurrence strengths from frequencies given by a WWW search engine. We show that our method is easy to use and assigns a confidence level to the attachments depending on the distance between the N+P and the V+P cooccurrence value. 1 Introduction In a first approach these cooccurrence values were derived from text corpora [Mehl et al. 98]. There, we achieved a rate of around 75% correct attachments for a domain specific Computer Magazine corpus (2.7 million tokens). But many of our cooccurrence values were based on rather low frequencies. We saw that attachment precision increases if these low f...