Evaluating hybrid versus data-driven coreference resolution
In this paper, we present a systematic evaluation of a hy- brid approach of combined rule-based ﬁltering and machine learning to Dutch coreference resolution. Through the application of a selection of linguistically-motivated negative and positive ﬁlters, which we apply in isolation and combined, we study the eﬀect of these ﬁlters on precision and recall using two diﬀerent learning techniques: memory-based learn- ing and maximum entropy modeling. Our results show that by using the hybrid approach, we can reduce up to 92 % of the training material with- out performance loss. We also show that the ﬁlters improve the overall precision of the classiﬁers leading to higher F-scores on the test set.