Coordination Annotation Extension in the Penn Tree Bank

Proceedings of the 54th Annual Meeting of the Association for Computational Linguistics, pages 834–842, Berlin, Germany, August 7-12, 2016. c⃝2016 Association for Computational Linguistics Coordination Annotation Extension in the Penn Tree Bank Jessica Ficler Computer Science Department Bar-Ilan University Israel jessica.ficler@gmail.com Yoav Goldberg Computer Science Department Bar-Ilan University Israel yoav.goldberg@gmail.com Abstract Coordination is an important and common syntactic construction which is not han- dled well by state of the art parsers. Co- ordinations in the Penn Treebank are miss- ing internal structure in many cases, do not include explicit marking of the conjuncts and contain various errors and inconsisten- cies. In this work, we initiated manual an- notation process for solving these issues. We identify the different elements in a co- ordination phrase and label each element with its function. We add phrase bound- aries when these are missing, unify incon- sistencies, and fix errors. The outcome is an extension of the PTB that includes con- sistent and detailed structures for coordi- nations. We make the coordination anno- tation publicly available, in hope that they will facilitate further research into coordi- nation disambiguation. 1 1 Introduction The Penn Treebank (PTB) (Marcus et al., 1993) is perhaps the most commonly used resource for training and evaluating syntax-based natural lan- guage processing systems. Despite its widespread adoption and undisputed usefulness, some of the annotations in PTB are not optimal, and could be improved. The work of Vadas and Curran (2007) identified and addressed one such annotation defi- ciency – the lack of internal structure in base NPs. In this work we focus on the annotation of coordi- nating conjunctions. Coordinating conjunctions (e.g. “John and Mary”, “to be or not to be”) are a very common syntactic construction, appearing in 38.8% of the 1The data is available in: https://github.com/Jess1ca/CoordinationExtPTB sentences in the PTB. As noted by Hogan (2007), coordination annotation in the PTB are not con- sistent, include errors, and lack internal structure in many cases (Hara et al., 2009; Hogan, 2007; Shimbo and Hara, 2007). Another issue is that PTB does not mark whether a punctuation is part of the coordination or not. This was resolved by Maier et al. (2012) which annotated punctuation in the PTB . These errors, inconsistencies, and in particular the lack of internal structural annotation turned re- searchers that were interested specifically in co- ordination disambiguation away from the PTB and towards much smaller, domain specific efforts such as the Genia Treebank (Kim et al., 2003) of biomedical texts (Hara et al., 2009; Shimbo and Hara, 2007). In addition, we also find that the PTB annota- tion make it hard, and often impossible, to cor- rectly identify the elements that are being coordi- nated, and tell them apart from other elements that may appear in a coordination construction. While most of the coordination phrases are simple and include only conjuncts and a coordinator, many cases include additional elements with other syn- tactic functions , such as markers (e.g. “Both Al- ice and Bob”), connectives (e.g. “Fast and …

Content truncated to 500 words