Reasoning over Semantic Networks – A Typology of Axioms for Natural Language Inference

There are problems related to question answering which obviously call for semantically based logical inferences. For example, a question-answering system should produce the same answer if asked with logically equivalent paraphrases of a question. Moreover, an answer should not only be found for direct matches of the query with the knowledge base but also if the answer is entailed by the latter. A meaning representation of NL texts and logical axioms defined on this representation must account for these aspects. Starting with logical properties of axioms, we define a typology based on five main characteristics which affect the validity and efficiency of natural language inference.