A twin-model is proposed for coreference resolution: a link component, modeling the coreferential relationship between an anaphor and a candidate antecedent, and a creation component modeling the possibility that a phrase is not coreferential with any candidate antecedent. The creation model depends on all candidate antecedents and is often expensive to compute; Therefore constraints are imposed on feature forms so that features in the creation model can be efficiently computed from feature values in the link model. The proposed twin-model is tested on the data from the 2005 Automatic Content Extraction (ACE) task and the proposed model performs better than a thresholding baseline without tuning free parameter.