Applying Markov logics for controlling abox abduction

Manually annotating the multimedia documents is a time-consuming and cost-intensive task. In this work, we define a media interpretation agent for automatically generating annotations for multimedia documents. Observations of the agent are given as surface-level information extracted by state-of-the-art media analysis tools. Based on background knowledge the agent interprets observations by computing high-level explanations. Observations and their explanations constitute the annotations of a media document. For this purpose, we investigate an abduction algorithm which computes explanations using a logic-based knowledge representation formalism. Multiple explanations might be possible for certain media content. Since the agent's resources for computing explanations are limited, we need to control the abduction procedure in terms of branching of the computation process and „depth“ of computed results, while still producing acceptable annotations. To control the abduction procedure, we employ a first-order probabilistic formalism.