Tüpa at SemEval-2019 Task1: (Almost) feature-free Semantic Parsing

Proceedings of the 13th International Workshop on Semantic Evaluation (SemEval-2019), pages 113–118 Minneapolis, Minnesota, USA, June 6–7, 2019. ©2019 Association for Computational Linguistics 113 T¨upa at SemEval-2019 Task 1: (Almost) feature-free Semantic Parsing Tobias P¨utz Department of Linguistics University of T¨ubingen SFB 833 A3 {tobias.puetz,kevin.glocker}@student.uni-tuebingen.de Kevin Glocker Department of Linguistics University of T¨ubingen Abstract Our submission for Task 1 ‘Cross-lingual Se- mantic Parsing with UCCA’ at SemEval-2018 is a feed-forward neural network that builds upon an existing state-of-the-art transition- based directed acyclic graph parser. We re- place most of its features by deep contextu- alized word embeddings and introduce an ap- proximation to represent non-terminal nodes in the graph as an aggregation of their termi- nal children. We further demonstrate how aug- menting data using the baseline systems pro- vides a consistent advantage in all open sub- mission tracks. We submitted results to all open tracks (English, in- and out-of-domain, German in-domain and French in-domain, low-resource). Our system achieves compet- itive performance in all settings besides the French, where we did not augment the data. Post-evaluation experiments showed that data augmentation is especially crucial in this set- ting. 1 Introduction Semantic Parsing is the task of assigning an ut- terance a structured representation of its mean- ing. The goal is to assign similar structures to utterances with similar meanings, regardless of their syntactic realizations. In Syntactic Parsing, for instance, the sentence ‘John saw Paul.’ will have a different structure than ‘Paul was seen by John’. Semantic Parsing, in contrast, aims to solely encode the fact that John saw Paul. De- riving a semantic representation of an utterance has various applications. It can serve as a start- ing point for the evaluation of machine translation systems, as the structure of the semantic represen- tation should be similar across languages. Birch et al. (2016) use human annotated scores of indi- vidual UCCA semantic units in their HUME met- ric to provide a fine-grained analysis of transla- tion quality and improve scalability to longer sen- tences by approximating human judgement semi- automatically from the annotated scores of each unit. Explicit semantic representations could also provide the structured information necessary to alleviate recent issues in Natural Language In- ference (NLI) where McCoy and Linzen (2019) showed that state-of-the-art NLI systems fail to recognize that e.g. ‘Alice believes Mary is lying.’ does not entail ‘Alice believes Mary.’. Using pre- cise semantic representations of the sentences a theorem could be built on which various logical in- ferences can be performed with a theorem prover such as in Mart´ınez-G´omez et al. (2016). Universal Conceptual Cognitive Annotation (UCCA) (Abend and Rappoport, 2013) is a se- mantic grammar formalism where natural lan- guage expressions are analyzed as deep directed acyclic graph (DAG) structures, deep meaning the graphs feature non-terminal nodes. Due to it’s coarse-grained representation using cogni- tively motivated categories it is both domain and language independent and quickly learned even by annotators without a linguistic background (Abend and Rappoport, 2013). The goal of the SemEval-2018 Task 1 ‘Cross- …

Content truncated to 500 words