Soft constraint automata with memory

We revise soft constraint automata, wherein transitions are weighted and each action has an associated preference value. We first relax the underlying algebraic structure to allow bipolar preferences. We then equip automata with memory locations, that is, with an internal state to remember and update information from transition to transition. We furthermore revise automata operators, such as composition and hiding, providing examples on how such memory locations interact with preferences. We finally apply our framework to encode context-sensitive behaviour