Graph-based and transition-based dependency parsers used to have different strengths and weaknesses. Therefore, combining the outputs of parsers from both paradigms used to be the standard approach to improve or analyze their performance. However, with the recent adoption of deep contextualized word representations, the chief weakness of graph-based models, i.e., their limited scope of features, has been mitigated. Through two popular combination techniques – blending and stacking – we demonstrate that the remaining diversity in the parsing models is reduced below the level of models trained with different random seeds. Thus, an integration no longer leads to increased accuracy. When both parsers depend on BiLSTMs, the graph-based architecture has a consistent advantage. This advantage stems from globally-trained BiLSTM representations, which capture more distant look-ahead syntactic relations. Such representations can be exploited through multi-task learning, which improves the transition-based parser, especially on treebanks with a high ratio of right-headed dependencies.

This paper presents the IMS contribution to the CoNLL 2017 Shared Task. In the preprocessing step we employed a CRF POS/morphological tagger and a neural tagger predicting supertags. On some languages, we also applied word segmentation with the CRF tagger and sentence segmentation with a perceptron-based parser. For parsing we took an ensemble approach by blending multiple instances of three parsers with very different architectures. Our system achieved the third place overall and the second place for the surprise languages.

DIRNDL is a spoken and written corpus based on German radio news, which features coreference and information-status annotation (including bridging anaphora and their antecedents), as well as prosodic information. We have recently extended DIRNDL with a fine-grained two-dimensional information status labeling scheme. We have also applied a state-of-the-art part-of-speech and morphology tagger to the corpus, as well as highly accurate constituency and dependency parsers. In the light of this development we believe that DIRNDL is an interesting resource for NLP researchers working on automatic coreference and bridging resolution. In order to enable and promote usage of the data, we make it available for download in an accessible tabular format, compatible with the formats used in the CoNLL and SemEval shared tasks on automatic coreference resolution.