Cognitive complexity can provide insight into how people make decisions, ranging from the most minor to the most impactful. Here, we present a novel approach to inferring the complexity of processes associated with preference and decision making. We measured the complexity of participant-generated descriptive features of consumer products and the relationship to preference rankings. In order to measure cognitive complexity over a sparse set of features, we developed a natural language processing approach that compared the descriptive words generated by participants to those generated by a machine learning model; words that were more distinct from those generated by the model were rated more complex. We show preliminary evidence that cognitive complexity is related to preference for products, explaining unique variance in rankings and also capturing a new facet of the process through which preference is revealed through choice. We also show the value of participant-generated features for understanding choice processes.