We demonstrate that subjective creativity in sentence-writing can in part be predicted using computable quantities studied in Computer Science and Cognitive Psychology. We introduce a task in which a writer is asked to compose a sentence given a keyword. The sentence is then assigned a subjective creativity score by human judges. We build a linear regression model which, given the keyword and the sentence, predicts the creativity score. The model employs features on statistical language models from a large corpus, psychological word norms, and WordNet.