Thanks, Paul. If I’ve understood him correctly Herschbach believes the people he puts into the phenomenological camp would rule out simulation, believing that we understand the behaviour of other people without running any kind of model. Whether this is plausible may depend on what you understand by a ‘simulation’. You say, pertinently, ‘every major theoretical approach is sorely lacking in formal detail’ – yes, indeed. The formal details are where it gets tough (but surely also where big rewards might be found). I salute what you’re doing in this respect.

Drawn at its broadest, I reckon the task of understanding each other is one of the hardest we face, and it’s likely that all our mental faculties are called for in dauntingly complicated combinations.

I think in the abstract, you are right — mental state attribution cannot be computationally characterized as purely theory or purely simulation. Given the sheer volume of simulation required to participate in even the simplest dialogue, or in recognizing intentions in the routine actions of your coworkers, friends and family, it seems that at the least, simulations must be stored and possibly reified in the service of theory-driven inferences. But attribution of default beliefs to agents we’ve never encountered before, and the profound evidence for shared self-other neural circuitry in service of empathy *and* recognition of intention-in-action (grasps, reaches, etc) speaks against a purely theory-based approach.