Pseudolikelihood is based on the observation that conditional probabilities take a ratio-based form, and thus can be computed without knowledge of the partition function. For example, if a set of variables $x$ is partitioned into subsets $a$ (variables to find the conditional distribution over), $b$ (variables to condition on), and $c$ (variables outside the query), pseudolikelihood focuses on the required conditional distributions rather than a global normalization constant.