Learning Joint Semantic Parsers from Disjoint Data

Proceedings of NAACL-HLT 2018, pages 1492–1502 New Orleans, Louisiana, June 1 - 6, 2018. c⃝2018 Association for Computational Linguistics Learning Joint Semantic Parsers from Disjoint Data Hao Peng♦ Sam Thomson♣ Swabha Swayamdipta♣ Noah A. Smith♦ ♦ Paul G. Allen School of Computer Science & Engineering, University of Washington ♣ School of Computer Science, Carnegie Mellon University {hapeng,nasmith}@cs.washington.edu, {sthomson,swabha}@cs.cmu.edu Abstract We present a new approach to learning se- mantic parsers from multiple datasets, even when the target semantic formalisms are dras- tically different, and the underlying corpora do not overlap. We handle such “disjoint” data by treating annotations for unobserved for- malisms as latent structured variables. Build- ing on state-of-the-art baselines, we show im- provements both in frame-semantic parsing and semantic dependency parsing by model- ing them jointly. Our code is open-source and available at https://github.com/ Noahs-ARK/NeurboParser. 1 Introduction Semantic parsing aims to automatically predict formal representations of meaning underlying nat- ural language, and has been useful in question an- swering (Shen and Lapata, 2007), text-to-scene generation (Coyne et al., 2012), dialog systems (Chen et al., 2013) and social-network extraction (Agarwal et al., 2014), among others. Various for- mal meaning representations have been developed corresponding to different semantic theories (Fill- more, 1982; Palmer et al., 2005; Flickinger et al., 2012; Banarescu et al., 2013). The distributed nature of these efforts results in a set of anno- tated resources that are similar in spirit, but not strictly compatible. A major axis of structural di- vergence in semantic formalisms is whether based on spans (Baker et al., 1998; Palmer et al., 2005) or dependencies (Surdeanu et al., 2008; Oepen et al., 2014; Banarescu et al., 2013; Copestake et al., 2005, inter alia). Depending on applica- tion requirements, either might be most useful in a given situation. Learning from a union of these resources seems promising, since more data almost always trans- lates into better performance. This is indeed the case for two prior techniques—parameter sharing Only a few books fell in the reading arg1 room . arg1 mwe arg1 arg1 arg2 BV compound top in.prep Locative Relation Figure Ground fall.v Motion Directional Theme Place a few.art Quantity Individuals Figure 1: An example sentence from the FrameNet 1.5 corpus, shown with an author-annotated DM semantic dependency graph (above) and frame- semantic annotation (below). Two more gold frames (and their arguments) have been omitted for space. (FitzGerald et al., 2015; Kshirsagar et al., 2015), and joint decoding across multiple formalisms us- ing cross-task factors that score combinations of substructures from each (Peng et al., 2017). Pa- rameter sharing can be used in a wide range of multitask scenarios, when there is no data overlap or even any similarity between the tasks (Collobert and Weston, 2008; Søgaard and Goldberg, 2016). But techniques involving joint decoding have so far only been shown to work for parallel annota- tions of dependency-based formalisms, which are structurally very similar to each other (Llu´ıs et al., 2013; Peng et al., 2017). Of particular interest is the approach of Peng et al., …

Content truncated to 500 words