Pervasive context-aware systems base their responses on information about the environment collected from ubiquitous sensors. The inevitable drawback of such systems is that raw data collected from sensors is often noisy, corrupted, and imprecise. Erroneous sensor readings create uncertainties and ambiguous interpretations. Thus creating an interpretation challenge for the context-aware system that needs to reason about possible states of only partially observable subjects. We propose a mechanism for pervasive context-aware systems to process the information gathered from sensors so to obtain knowledge about possible environment states. This includes both the ability to reason about a situation with incomplete knowledge and to cope with erroneous contexts. We present a probabilistic approach to reason about the likelihood of each particular situation, state of a variable, and variable interdependence. The evaluation shows that the proposed approach is applicable to real-time context inference problems.