The study of economic narratives using text mining techniques increased recently with a dominance of unsupervised models built upon the bag-of-word assumption, to extract meaningful information for further econometric inferences. yet word independence remains a constraint when the message relies on specific and intentionally word usage to explicit a particular message that often comes embedded with an ideological purpose. Monetary policy, as an important field of economics, carries uncertainty in its decisionmaking process as a forward guidance tool and an important aspect of its communication strategy. Noticeable is the construction of uncertainty indices using uncertainty-related word counts from different text sources. These are often embedded with an information bias that may bifurcate the intended aim of central banks. This work tries to quantify monetary policy uncertainty at the source and identifies its key drivers using Latent Semantic Scaling, a semi-supervised positional approach that renders similar capabilities as for word embedding, based on selected seed words given as priors to an augmented matrix factorization applied at the sentence level. Results show a receding uncertainty level at the Federal Reserve (1996-2020), due to a significant change in the monetary discourse after the 2008 financial crisis, where the use of uncertainty and risk-related words was reduced as part of a communication strategy aimed at sending neutral policy signals to the public amid unclear economic outlooks.