Description Logic-based knowledge bases (KBs), i.e., ontologies, are often greatly incomplete, necessitating a demand for ontology completion. Promising approaches to this aim are to embed ontology elements such as concept and role names, and logical axioms into a low-dimensional vector space and find missing elements by statistically inferencing on the latent representation. These approaches make inference based solely on structural relationship in the ontology; thus, the likelihood of its completion with implicit (duplicated) facts could be high, making the performance of embedding algorithms questionable. Therefore, it is essential for the ontology completion’s procedure that leverages statistical inference to prevent the completion with implicit facts. In this paper, we present a new strategy for detecting such duplication (defined as “garbage”) based on the logical constructs in description logics and a prototype system called KBCOps. Our experiments reveal that garbage could exist even using the state-of-the-art embedding algorithms.