Region visual features enhance the generative capability of the machines\nbased on features, however they lack proper interaction attentional perceptions\nand thus ends up with biased or uncorrelated sentences or pieces of\nmisinformation. In this work, we propose Attribute Interaction-Tensor Product\nRepresentation (aiTPR) which is a convenient way of gathering more information\nthrough orthogonal combination and learning the interactions as physical\nentities (tensors) and improving the captions. Compared to previous works,\nwhere features are added up to undefined feature spaces, TPR helps in\nmaintaining sanity in combinations and orthogonality helps in defining familiar\nspaces. We have introduced a new concept layer that defines the objects and\nalso their interactions that can play a crucial role in determination of\ndifferent descriptions. The interaction portions have contributed heavily for\nbetter caption quality and has out-performed different previous works on this\ndomain and MSCOCO dataset. We introduced, for the first time, the notion of\ncombining regional image features and abstracted interaction likelihood\nembedding for image captioning.\n