The origins of meaning – from pragmatic control signals to semantic representations

Abstract

The concept of representations is widely used across the cognitive sciences, but its meaning is highly contested. Representations are often thought of as “vehicles” with “content” – that is, internal physical patterns that are correlated with some state of affairs and that usefully convey that state of affairs to the rest of the neural system or the cognitive economy at large. This raises a number of problems: how does an internal pattern come to be correlated with something else? How does the rest of the system know what the pattern means? How does the system know what to do with that information? I suggest that thinking of representations as vehicles with content presents a stumbling block to answering these questions. Instead, thinking of them simply as meaningful patterns offers a more naturalistic framework for understanding their roles in perception, behavioural control, and cognition. Meaning is not contained within a vehicle – it is relational, contextual, and interpretive. Here, I explore the origins of meaningful patterns in simple behavioural control policies and the subsequent evolution of internalised representations, decoupled from obligate action. This transition from pragmatic to semantic meaning demystifies the notion of “content”, situates talk of grounding in referential, consequential, and relational terms, and highlights the dual roles of current activity and stored context.

Beyond mechanism – extending our concepts of causation in neuroscience

“…In more complex cases, perceptual systems generate internal, genuinely semantic representations, which are decoupled from obligate action, and which are simply reported or made available to other parts of the nervous system (Mitchell, 2023b). They are indicative, rather than imperative.…”

“…This kind of view can thus reconcile computational theories of mind (which involve operations over currently active states comprising "symbolic representations") with connectionist theories (which supply the stable background context that grounds the meaning of the active states). (Mitchell, 2023b;Piccinini 2022).…”

If engrams are the answer, what is the question?

“…To distinguish between latent information and vehicles is not to separate them and treat them as separate entities, but to look at the same phenomena from different perspectives. Nonetheless, it creates a dualistic divide that is not always desirable(Mitchell, 2023).…”