Robust Semantic Parsing with Adversarial Learning for Domain Generalization

Abstract

This paper addresses the issue of generalization for Semantic Parsing in an\nadversarial framework. Building models that are more robust to inter-document\nvariability is crucial for the integration of Semantic Parsing technologies in\nreal applications. The underlying question throughout this study is whether\nadversarial learning can be used to train models on a higher level of\nabstraction in order to increase their robustness to lexical and stylistic\nvariations.We propose to perform Semantic Parsing with a domain classification\nadversarial task without explicit knowledge of the domain. The strategy is\nfirst evaluated on a French corpus of encyclopedic documents, annotated with\nFrameNet, in an information retrieval perspective, then on PropBank Semantic\nRole Labeling task on the CoNLL-2005 benchmark. We show that adversarial\nlearning increases all models generalization capabilities both on in and\nout-of-domain data.\n