Warning: this paper contains content that may be offensive or upsetting.\n Language has the power to reinforce stereotypes and project social biases\nonto others. At the core of the challenge is that it is rarely what is stated\nexplicitly, but rather the implied meanings, that frame people's judgments\nabout others. For example, given a statement that "we shouldn't lower our\nstandards to hire more women," most listeners will infer the implicature\nintended by the speaker -- that "women (candidates) are less qualified." Most\nsemantic formalisms, to date, do not capture such pragmatic implications in\nwhich people express social biases and power differentials in language.\n We introduce Social Bias Frames, a new conceptual formalism that aims to\nmodel the pragmatic frames in which people project social biases and\nstereotypes onto others. In addition, we introduce the Social Bias Inference\nCorpus to support large-scale modelling and evaluation with 150k structured\nannotations of social media posts, covering over 34k implications about a\nthousand demographic groups.\n We then establish baseline approaches that learn to recover Social Bias\nFrames from unstructured text. We find that while state-of-the-art neural\nmodels are effective at high-level categorization of whether a given statement\nprojects unwanted social bias (80% F1), they are not effective at spelling out\nmore detailed explanations in terms of Social Bias Frames. Our study motivates\nfuture work that combines structured pragmatic inference with commonsense\nreasoning on social implications.\n