The systems and methods disclosed herein combine a plurality of interpretations of a voice-based input. The systems and methods may receive the voice-based input, process it using one or more automatic speech recognition modules to obtain a plurality of interpretations, and identify an entity set for each of the plurality of interpretations. The systems and methods may further generate a combined interpretation based on a first interpretation and second interpretation selected form the plurality of interpretations and assign a semantic score to the combined interpretation based on the entity sets of the first and second interpretation.