Models using the Rational Speech Act (RSA) framework typically assume that speakers are certain about the meaning being communicated. In this work we note that there are contexts in which this assumption does not hold, and propose a method (um-RSA) to incorporate this meaning uncertainty within the RSA framework. As a case study, we explore two sources of meaning uncertainty: Counting-Uncertainty (from numerical cognition) and Discounting-Uncertainty (from behavioral economics). We generate predictions from these two hypotheses and test these predictions with two human experiments. The results show that um-RSA can account for differences in uncertainty expression usage that the standard RSA framework cannot account for, thus demonstrating the usefulness of modeling meaning uncertainty.