Knowledge reasoning of detective novels by abduction, SAT problem and BERT model

Reasoning the same as Sherlock Holmes ~the Second Knowledge Graph Reasoning Challenge 2019~

Abstract

Logic and its structure embedded in the sentence are intensively investigated to grasp sentence semantics in natural language processing. We focused on knowledge processing of reasoning in detective novels as an example of semantics analysis based on logics and reasoning with logics. We define the process of reasoning with predicate logics, and solved a problem to identify the criminals as a hypothetical reasoning and SAT problem. As logical conclusions of given natural sentences described in 5 detective novels and some external knowledge, we demonstrated our proposed method could identify the criminals and result in interpretable solution. Also, our method can show all possible solutions resulted due to lack of input information and what additional information is necessary for obtaining unique solution.