Abductive Generalization and Specialization

Abstract

Abduction and induction both generate hypotheses to explain observed phenomena in an incomplete knowledge base, while they are distinguished in the following aspects. Abduction conjectures specific facts accounting for some particular observation. Those assumptions of facts are extracted using causal relations in the background knowledge base. As there are generally many possible facts which may imply the observation, candidates for hypotheses are usually pre-specified as abducibles. Then, the task is finding the best explanations from those candidates. By contrast, induction seeks regularities underlying the observed phenomena. The goal is not only explaining the current observations but discovering new knowledge for future usage. Hence induced hypotheses are general rules rather than specific facts. In constructing general rules, some constraints called biases are often used but candidates for hypotheses are not usually given in advance. The task is then forming new hypotheses using information in the background knowledge base.