XML is rapidly becoming technologies for information exchange and representation. It causes many research issues such as semantic modeling methods, conversion for interoperability with other models, and so on. Especially, the most important issue in practical area is how to achieve the interoperability between XML model and relational database model. Until now, many methods have been proposed to achieve it. However, several problems still remain. Most of all, existing methods do not consider implicit referential integrity relations, so it causes loss of information. This paper proposes an algorithm for extracting referential integrity relations during RDB to XML translation. The key point of our method is how to find implicit referential integrity relations among columns which have different names to represent the same semantic. To resolve it, we define an enhanced extraction algorithm which based on a widely used ontology, WordNet. The proposed algorithm can reduce an extraction time among comparison columns in RDB tables and prevent loss of information.