The IMS-Wrocław-Szeged-CIS Entry at the SPMRL 2014 Shared Task: Reranking and Morphosyntax Meet Unlabeled Data⇤

This paper describes our contribution to the SPMRL 2014 Shared Task. We participated in the predicted POS and morphology setting using full-size training data, and for all languages except Arabic. Our approach builds upon our contribution from last year (Björkelund et al., 2013), with additions that utilize unlabeled data. We observed that exploiting unlabeled data is challenging and we could benefit from it only moderately. We achieved best scores on all languages in the dependency track and on all languages except Polish in the constituency track.