Title: An ASP-based Approach to Answering Natural Language Questions for Texts

Abstract: An approach based on answer set programming (ASP) is proposed in this paper for representing knowledge generated from natural language texts. Knowledge in a text is modeled using a Neo Davidsonian-like formalism, which is then represented as an answer set program. Relevant commonsense knowledge is additionally imported from resources such as WordNet and represented in ASP. The resulting knowledge-base can then be used to perform reasoning with the help of an ASP system. This approach can facilitate many natural language tasks such as automated question answering, text summarization, and automated question generation. ASP-based representation of techniques such as default reasoning, hierarchical knowledge organization, preferences over defaults, etc., are used to model commonsense reasoning methods required to accomplish these tasks. In this paper, we describe the CASPR system that we have developed to automate the task of answering natural language questions given English text. CASPR can be regarded as a system that answers questions by "understanding" the text and has been tested on the SQuAD data set, with promising results.