Knowledge Specification and Querying

Our colleagues from the Rules and Reasoning community have just reported interesting results. They built a system that achieves 100% accuracy on an extensive test suite of movie-related questions, e.g. “Who appears in a Steven Spielberg directed film?”. Their approach utilizes a Controlled Natural Language (CNL) that could enable domain experts, who are not trained logicians, to both create formal knowledge and query it. They contrasted their approach with a machine learning approach based on neural networks, which falls far short of this high mark. Link