Zero-Shot Semantic Parsing for Instructions

Proceedings of the 57th Annual Meeting of the Association for Computational Linguistics, pages 4454–4464 Florence, Italy, July 28 - August 2, 2019. c⃝2019 Association for Computational Linguistics 4454 Zero-Shot Semantic Parsing for Instructions Ofer Givoli1 and Roi Reichart2 1 Faculty of Computer Science, Technion, IIT 2 Faculty of Industrial Engineering and Management, Technion, IIT ogivoli@cs.technion.ac.il, roiri@technion.ac.il Abstract We consider a zero-shot semantic parsing task: parsing instructions into compositional logical forms, in domains that were not seen during training. We present a new dataset with 1,390 examples from 7 application domains (e.g. a calendar or a file manager), each example con- sisting of a triplet: (a) the application’s initial state, (b) an instruction, to be carried out in the context of that state, and (c) the state of the application after carrying out the instruc- tion. We introduce a new training algorithm that aims to train a semantic parser on exam- ples from a set of source domains, so that it can effectively parse instructions from an un- known target domain. We integrate our algo- rithm into the floating parser of Pasupat and Liang (2015), and further augment the parser with features and a logical form candidate fil- tering logic, to support zero-shot adaptation. Our experiments with various zero-shot adap- tation setups demonstrate substantial perfor- mance gains over a non-adapted parser.1 1 Introduction The idea of interacting with machines via natural language instructions and queries has fascinated researchers for decades (Winograd, 1971). Recent years have seen an increasing number of applica- tions that have a natural language interface, either in the form of chatbots or via “intelligent personal assistants” such as Alexa (Amazon), Google As- sistant, Siri (Apple), and Cortana (Microsoft). In the near future, we may find ourselves in a world where even more functionality could be accessed via a natural language user interface (NLUI). If so, we better seek answers to the fol- lowing questions: Will every developing team need to hire NLP experts to develop a NLUI for 1Our code and data are available at: https://github.com/givoli/TechnionNLI. their specific application? Can we hope for a gen- eral framework that once trained on annotated data from a set of domains, does not require annotated data from a newly presented domain? Previous work on tasks related to NLUI for applications mostly relied on in-domain data (e.g. Artzi and Zettlemoyer (2013); Long et al. (2016)), and pa- pers that did not rely on in-domain data did not at- tempt to parse instructions into compositional log- ical forms (Kim et al., 2016). To fill this gap, we address the task of zero- shot semantic parsing for instructions: training a parser so that it can parse instructions into compo- sitional logical forms, where the instructions are from domains that were not seen during training. Formally, our task assumes a set D = {d1, ..., dn} of source domains, each corresponding to a sim- ple application (e.g. a calendar or a file manager) and an application program interface (API) con- sisting of a set of …

Content truncated to 500 words