Predicting State Changes in Procedural Text using Analogical Question Answering

Abstract

Many of the changes in the world that happen over time are characterized by processes. Creating programs that comprehend procedural text (e.g. the stages of photosynthesis) is a crucial task in natural language understanding. In this paper we present a novel approach that uses analogical question answering to predict what state changes affect entities in a paragraph describing a process. We start from the hypothesis that human level QA requires multiple layers of rich, relational representations. For this reason, our model is built on the Companion Cognitive Architecture, which has a large knowledge base and a general-purpose semantic parser. During training, the system uses the output of the semantic parser to automatically construct query cases, which link annotated answers to semantic interpretations (i.e. logical statements). When faced with unseen questions, the system retrieves relevant query cases by analogy and uses them to predict sentence level state changes. To obtain a globally consistent sequence of events, we apply common sense constraints over the whole paragraph via dynamic programming. We test our system on AI2’s ProPara dataset where our approach achieves results comparable to topperforming models.