The understanding of novel compounds is a special case in which we can explore the deep generativity of natural language understanding. We report a model, PUNC, which captures the comprehension of novel noun-noun compounds. The model constructs multiple interpretations for a given compound, ranking these by their overall acceptability, using the key constraints of diagnosticity, plausibility and informativeness. We present a sensitivity analysis of the model's key variables, demonstrating a graceful degradation in performance as the weightings to these variables are altered, thus vindicating the model's underlying theoretical principles.