In this paper, we propose an approach to improve image captioning solution\nfor images with novel objects that do not have caption labels in the training\ndataset. We refer to our approach as Partially-Supervised Novel Object\nCaptioning (PS-NOC). PS-NOC is agnostic to model architecture, and primarily\nfocuses on the training approach that uses existing fully paired image-caption\ndata and the images with only the novel object detection labels (partially\npaired data). We create synthetic paired captioning data for novel objects by\nleveraging context from existing image-caption pairs. We then create\npseudo-label captions for partially paired images with novel objects, and use\nthis additional data to fine-tune the captioning model. We also propose a\nvariant of SCST within PS-NOC, called SCST-F1, that directly optimizes the\nF1-score of novel objects. Using a popular captioning model (Up-Down) as\nbaseline, PS-NOC sets new state-of-the-art results on held-out MS COCO\nout-of-domain test split, i.e., 85.9 F1-score and 103.8 CIDEr. This is an\nimprovement of 85.9 and 34.1 points respectively compared to baseline model\nthat does not use partially paired data during training. We also perform\ndetailed ablation studies to demonstrate the effectiveness of our approach.\n