Answering Why-Questions Using Probabilistic Logic Programming

Abstract

We present a novel architecture of a closed domain question answering system that learns to answer why-questions from a small number of example interpretations. We use a probabilistic logic programming framework that can learn probabilities for rules from positive and negative example interpretations. These rules are then used by a meta-interpreter to generate an explanation in the form of a proof for a why-question. The explanation is displayed as an answer to the question together with a probability. In certain contexts, follow-up questions can be asked that conditionally depend on these why-questions and have an effect on the probability of the subsequent answer. The presented approach is a contribution to explainable artificial intelligence that aims to take machine learning out of the black-box.