Together with various colleagues from Leiden (Yiya Chen, Claartje Levelt, etc.) I have investigated the representation and processing of (linguistic) stress. In 90% of all Dutch words, the stress is on the first syllable. We thought: only the 10% of irregular words will be stored, the rest will be calculated using rules. But that's not how it works.
Within the irregular words, there are words which behave in a regular way. Thus, it may well be a combination of storing and computing lexical stress information.