Sentences with Gapping: Parsing and Reconstructing Elided Predicates

Proceedings of NAACL-HLT 2018, pages 1156–1168 New Orleans, Louisiana, June 1 - 6, 2018. c⃝2018 Association for Computational Linguistics Sentences with Gapping: Parsing and Reconstructing Elided Predicates Sebastian Schuster Joakim Nivre‡ Christopher D. Manning Departments of Linguistics and Computer Science, Stanford University {sebschu,manning}@stanford.edu ‡Department of Linguistics and Philology, Uppsala University joakim.nivre@lingfil.uu.se Abstract Sentences with gapping, such as Paul likes cof- fee and Mary tea, lack an overt predicate to indicate the relation between two or more ar- guments. Surface syntax representations of such sentences are often produced poorly by parsers, and even if correct, not well suited to downstream natural language understanding tasks such as relation extraction that are typi- cally designed to extract information from sen- tences with canonical clause structure. In this paper, we present two methods for parsing to a Universal Dependencies graph representation that explicitly encodes the elided material with additional nodes and edges. We find that both methods can reconstruct elided material from dependency trees with high accuracy when the parser correctly predicts the existence of a gap. We further demonstrate that one of our meth- ods can be applied to other languages based on a case study on Swedish. 1 Introduction Sentences with gapping (Ross, 1970) such as Paul likes coffee and Mary tea are characterized by hav- ing one or more conjuncts that contain multiple arguments or modifiers of an elided predicate. In this example, the predicate likes is elided for the relation Mary likes tea. While these sentences ap- pear relatively infrequently in most written texts, they are often used to convey a lot of factual in- formation that is highly relevant for language un- derstanding (NLU) tasks such as open informa- tion extraction and semantic parsing. For example, consider the following sentence from the WSJ por- tion of the Penn Treebank (Marcus et al., 1993). (1) Unemployment has reached 27.6% in Azerbaijan, 25.7% in Tadzhikistan, 22.8% in Uzbekistan, 18.8% in Turkmenia, 18% in Armenia and 16.3% in Kirgizia, [...] Paul likes coffee and Mary tea Paul likes coffee and Mary tea Paul likes coffee and Mary tea nsubj obj conj>cc conj>nsubj conj>obj nsubj obj cc conj orphan Paul likes coffee and Mary likes′ tea nsubj obj cc nsubj conj obj Figure 1: Overview of our two approaches. Both methods first parse a sentence with gapping to one of two different dependency tree representations and then reconstruct the elided predicate from this tree. To extract the information about unemployment rates in the various countries, an NLU system has to identify that the percentages indicate unemploy- ment rates and the locational modifiers indicate the corresponding country. Given only this sentence, or this sentence and a strict surface syntax repre- sentation that does not indicate elided predicates, this is a challenging task. However, given a depen- dency graph that reconstructs the elided predicate for each conjunct, the problem becomes much eas- ier and methods developed to extract information from dependency trees of clauses with canonical structures are much more likely to extract the cor- …

Content truncated to 500 words