HPE: Answering Complex Questions over Text by Hybrid Question Parsing and Execution

Abstract

The dominant paradigm of textual question answering with end-to-end neural models excels at answering simple questions but falls short on explainability and dealing with more complex questions. This stands in contrast to the broad adaptation of semantic parsing approaches over structured data sources (e.g., relational database), that convert questions to logical forms and execute them with query engines. Towards the goal of combining the strengths of neural and symbolic methods, we propose a framework of question parsing and execution for textual QA. It comprises two central pillars: (1) parsing a question of varying complexity into an intermediate representation, named H-expression, which symbolically represents how an answer to the question can be reached by hierarchically combining answers from the primitive simple questions; (2) to execute the resulting expression, we design a hybrid executor, which integrates deterministic rules to translate the symbolic operations with a drop-in neural reader to answer each simple question. The proposed framework can be viewed as a top-down question parsing followed by a bottom-up answer backtracking. H-expressions closely guide the execution process, offering higher precision besides better interpretability while still preserving the advantages of the neural readers for resolving primitive elements. Our extensive experiments on four different QA datasets show that the proposed framework outperforms existing approaches in supervised, few-shot, and zero-shot settings, while also effectively exposing the underlying reasoning process.