Csaba Pléh asks “How would referential understanding in beings who do not produce the signs be different from simple CS in the Pavlovian sense?
Good question!
(1) It is already beyond Pavlovian (i.e., Skinnerian) when the dog fetches the “named” toy.
(2) And it is already beyond Skinnerian when the dog fetches new toys after 1-shot or few-shot “naming” of new toys.
(3) But it is not linguistic reference until the dog can name the toy, the fetching, and anything else you can put into and define in a dictionary or textbook. (Language is a “mirror” capacity.)
Too demanding? It’s the nature of the unique, universal, and omnipotent capacity called natural language that demands it, and makes it possible.
(And, by the way, computation, including Platonic mathematics [though perhaps not Fields-Medal-level mathematical creativity], is part of that unique universal capacity. [All algorithms are in Hungarian.”] Nor are our sensorimotor — and sensorimotor-learning — capacities part of language, rather than the reverse. And referential understanding is no more a CS signalling skill than mathematical understanding is [pace Turing] a symbol manipulation skill.)