Coordinate Constructions in English Enhanced Universal Dependencies: Analysis and Computational Modeling

Proceedings of the 16th Conference of the European Chapter of the Association for Computational Linguistics, pages 795–809 April 19 - 23, 2021. ©2021 Association for Computational Linguistics 795 Coordinate Constructions in English Enhanced Universal Dependencies: Analysis and Computational Modeling Stefan Grünewald1,2 Prisca Piccirilli1,2 Annemarie Friedrich2 1Institut für Maschinelle Sprachverarbeitung, University of Stuttgart 2Bosch Center for Artificial Intelligence, Renningen, Germany stefan.gruenewald|annemarie.friedrich@de.bosch.com piccirpa@ims.uni-stuttgart.de Abstract In this paper, we address the representation of coordinate constructions in Enhanced Univer- sal Dependencies (UD), where relevant depen- dency links are propagated from conjunction heads to other conjuncts. English treebanks for enhanced UD have been created from gold ba- sic dependencies using a heuristic rule-based converter, which propagates only core argu- ments. With the aim of determining which set of links should be propagated from a se- mantic perspective, we create a large-scale dataset of manually edited syntax graphs. We identify several systematic errors in the orig- inal data, and propose to also propagate ad- juncts. We observe high inter-annotator agree- ment for this semantic annotation task. Using our new manually verified dataset, we perform the first principled comparison of rule-based and (partially novel) machine-learning based methods for conjunction propagation for En- glish. We show that learning propagation rules is more effective than hand-designing heuris- tic rules. When using automatic parses, our neural graph-parser based edge predictor out- performs the currently predominant pipelines using a basic-layer tree parser plus converters. 1 Introduction The Universal Dependencies (UD) formalism (de Marneffe et al., 2014) is a framework for repre- senting syntactic dependencies between words, pri- oritizing links between content words. UD parses provide two levels of analysis. Basic dependencies form standard syntactic dependency trees in which each node has exactly one governor (black links on top in Figure 1). Enhanced dependencies (Schuster and Manning, 2016) are extensions of these trees including additional relations (blue links below sen- tence) with the aim of representing linguistic phe- nomena such as coordination, control, or relative clauses. They have been shown to provide valuable In 1594 PEREZ wrote and published a book . case obl nsubj conj cc det root obj obj nsubj obl Figure 1: UD basic (top) and enhanced (bottom) de- pendencies. Green dotted link: proposed addition. input for information extraction tasks (Schuster et al., 2017). One of the most frequent phenom- ena addressed by enhanced UD is coordination. In the English Web Treebank (EWT), more than 15% of all sentences contain conjoined verbs. Hence, a good representation of coordination clearly is crucial for downstream tasks. For example, in Fig- ure 1, the enhanced layer explicitly captures that the arguments of the predicate “wrote” also fill the corresponding slots of “published,” which is highly relevant for natural language understanding tasks. In many cases, enhanced representations can be derived from the gold basic layer in a rule-based fashion (Schuster and Manning, 2016). The cur- rently available English enhanced UD treebanks have been created by applying such a converter. However, we are not aware of a large study re- garding their correctness and …

Content truncated to 500 words