Experiments with an Annotation Scheme for a Knowledge-rich Noun Phrase Interpretation System

This paper presents observations on our experience with an annotation scheme that was used in the training of a state-of-the-art noun phrase semantic interpretation system. The system relies on cross-linguistic evidence from a set of five Romance languages: Spanish, Italian, French, Portuguese, and Romanian. Given a training set of English noun phrases in context along with their translations in the five Romance languages, our algorithm automatically learns a classification function that is later on applied to unseen test instances for semantic interpretation. As training and test data we used two text collections of different genre: Europarl and CLUVI. The training data was annotated with contextual features based on two stateof-the-art classification tag sets.