Many predictive tasks, such as diagnosing a patient based on their medical\nchart, are ultimately defined by the decisions of human experts. Unfortunately,\nencoding experts' knowledge is often time consuming and expensive. We propose a\nsimple way to use fuzzy and informal knowledge from experts to guide discovery\nof interpretable latent topics in text. The underlying intuition of our\napproach is that latent factors should be informative about both correlations\nin the data and a set of relevance variables specified by an expert.\nMathematically, this approach is a combination of the information bottleneck\nand Total Correlation Explanation (CorEx). We give a preliminary evaluation of\nAnchored CorEx, showing that it produces more coherent and interpretable topics\non two distinct corpora.\n