Précis for an action-oriented theory of language

What is a word, or message, or utterance? It is a move by an agent (~superorganism) undertaken in order to transform the environmental state, in a desired direction, by means of altering another agent’s information state. This transformation may occur directly, if the targeted slice of environment happens to be that agent whose informational state is altered, and whose behavior now or at some future time can be made different by means of this alteration. Or it can occur indirectly, by inducing an agent to transform some third party, be it another agent or else a non-agentic aspect of the environment. (E.g. asking for assistance in moving a heavy object.) The move is an instantiation of certain strategies (“social heuristics”1) for action possessed by the information-emitting agent—henceforth our game’s Writer. These heuristics guide which sorts of moves “ought” to be made, or are preferred, given the “type” of game state (or “situation,”2 or “context”3) at-hand. Like all social heuristics, and all moves born of social heuristics, their efficacy is non-PIG-optimal4 and depends on the receiving agent, henceforth “Reader.” As interventions or solutions, they are efficacious within system states/problem contexts to which they are fitted as interventions and solutions. And this efficaciousness may only be judged with respect to goals, that is, a desired state of the world to which the actual, brought-about state of the world might be compared.

[1]  In the “bounded rationality” frame.
[2]  To Cold War conflict theorists, pragmatist philosophers, behavioral scientists
[3] To linguists, ethnomethodologists, interaction scientists
[4]  PIG-optimal as in “Player-Independent, Globally Optimal”—non-PIG-optimality implying the solution is player-dependent & locally optimal

Perhaps more fundamentally—one surrogate stripped back—a word (or message, or utterance) is a strategic move by an agent to alter its own internal state, either directly (e.g. through emotional self-palliation), or indirectly (e.g. through outsourced affirmation and validation), or doubly indirectly, altering an agent’s information state such that they alter the environmental state so as to ultimately alter the Writer’s internals (as is the case with ordering takeout food and pick-up artistry). Here there is room to talk about wireheading and degenerate play, base- versus mesa-optimizers, and the destabilization of base-mesa alignment—in other words, the solution-context decoupling—brought about by environmental drift. (Including environmental drifts and solution-decoupling caused by the deployment of the context-scoped heuristic to begin with—this being the realm of “reflexivity” and frequency-dependent selection.) To wirehead, or play degenerately—from the perspective of a base-optimizer—is to act, as mesa-optimizer, in a way which advances one’s own goals but not the goals of the base—a misalignment brought about by the decoupling of a shifting environment.

From the perspective of evolution, what is central is the defense against entropy and the preservation of boundaries, the preservation of bodily function and ultimately the reproduction of genes. To induce an apple seller to sell you his apples, using the phrase “a dozen reds,” is “ultimately,” from the evolutionary perspective, a matter of bringing in energy and resources, in the form of carbohydrates and vitamins or minerals, into the body so that it may maintain homeostasis. This is the base-optimizer’s perspective, natural selection being the great base-optimizer to evolved, mesa-optimizer, agentic organisms. From the perspective of the mesa-optimizer, the apples are purchased in large part to accomplish such internal state changes as (to name a few motivations) the pleasure brought on by sugar, the prevention of hunger’s and thirst’s displeasure, the pains of poor dental hygiene, etc. The exact status and phenomenology of these desires is interesting, and may be worth exploring in order to better understand the structure of motivation which drives action—but for now, it can be abstracted away. 

The picture we are approaching is this: Words are actions; actions are moves in games. (Strategy games, to be precise.) Frequency-dependence, the reputations born of history, the desire to deceive and the desire to communicate, the Schelling points provided by sensorily salient differences—these and more characterize the meta-level dynamics which govern the development of communicative concretia. A strategy lies one level of abstraction up from a move; it is a principle, philosophy, algorithm, or rule. The rule rests upon an ontology of pragmatic (goal-relevant) distinctions, predicated on sensorily salient information cues afforded by the environment (including words and utterances, distinct colors or smells) to determine which move to select and deploy from the space of possibles provided by the Writer’s habitus.

The optimal selection between possible messages, and formation of semantic (i.e. pragmatic) distinctions, is described by information theory, which takes into account context, expectation, receiver schema, etc. This optimality, however, does not take account of the cost of gathering information, and the impossibility of ascertaining or verifying certain crucial sources of information. In practice, just as agents do not act “optimally” (in the sense of a resource-unconstrained, omnipotent and omniscient being), but rather operate by deploying historically high-performing heuristics, similarly they do not communicate optimally, but rather through heuristics. In other words, we do not weigh all the relevant information anew each time we size up a situation and make a decision. Instead, we determine the kind or “type”5 of situation, according to a pragmatic, context-fitted, goal-oriented ontology6 which is characterized, on the one hand, by sensorily salient informational cues that allow us to make distinctions between types in the first place, and on the other side, by the actions we find appropriate to such a type of situation. The two sides of this ontology are typically bridged by what is referred to as a “concept,” which is often but not necessarily accompanied by a linguistic signifier conventionally understood to invoke or refer to it.  

[5] This being an opportunity to import Schutz’s theories of typification
[6] See also “words as decision rules”

Still, we do track rough frequency in the population of moves and strategies deployed, partially for mimetic reasons, preferentially mimicking (learning from) publicly successful, prestigious, popular, or undervalued players and strategies. Agents observe not just other players’ action ontologies, but also the relative payoffs of their resultant moves (behaviors, utterances, messages), all scoped contextually to different game or environmental states. Today’s payoff matrix—the costs and benefits of various actions, picked up by learners to build an intuition for the stimulus-response patterns of the social and physical world—form a training or “traumatizing” system which ultimately alters the heuristics or strategies agents possess in their repertoire (space of possibles7) and when players decide to deploy them. And this training system works both through firsthand experience of payoffs and also through the observation of others’ payoff experience. This system becomes a part of the “feel” of moves—of messages, words, utterances, behaviors—giving rise to disgust and aversion complexes, prestige halos and class signals, acting as the basis for signaling games between social agents, etc. 

[6] To Bourdieu

For instance, if a musician is asked to describe his music, he might be expected to represent the “feel” of the music, and present a cluster of floating signifiers which give the Reader some approximate structure of anticipation for what a firsthand experience of the music might be like. And there is no question that at some level this is an accurate description of one layer of the shared project which the interaction participants thereby embark on, and of one guiding force in the selection among heuristics, from the space of possibles, which the musician makes to describe his music. But there are also will likely be desires, on the musician’s part, to impress the listener, or persuade him to listen or purchase a copy of the musician’s record—that is, in addition to the cooperative motivations and aspects of play, where a standard of “honest” transmission is held, and the Reader and Writer are both aligned in their goals, there is also an adversarial “sales” aspect, insofar as the writing musician might prefer to reduce the absolute number of false negatives (cases where the Reader would, in fact, enjoy the musician’s music, but decided not to purchase it) even at the cost of increasing the proportional number of false positives (cases where the Reader purchases on the basis of the musician’s report, but does not, upon listening, enjoy it), because such false positives are much cheaper to the writer-musician than to the reader-listener, who bears the brunt of the cost in wasted time and money. In other words, it is a mixed-motives game, to use Schelling terms, as all real-life strategy games are. And then if we ask what “honest” or “accurate” means, even in the case of some hypothetical, purely cooperative play, we will struggle to find any grounding for this other than in the effect of creating an approximate structure of anticipation in the listener, whose “cash value” consists of the Reader making the “correct” (for his own purposes) decision to purchase the album in all cases he would enjoy it, and abstain from purchasing in all cases he would not. Again, it is in the behavioral outcome that words such as “jazzy,” “chaotic,” “mystical” can be said to “mean” anything at all. 

We come to see quickly that on the one hand, the Writer uses action-oriented ontologies in order to pick between information-producing moves, and also that the Reader uses action-oriented ontologies in order to pick between information-responding moves—in other words, that the interpretive strategy of the Reader is also the ideal schema by which the Writer writes—the ideal system of sensorily salient, and highly conventionalized, differences predicted to make a difference, in terms of different moves’ relative efficacy in advancing goals. The system of action-oriented, pragmatic perceptual ontologies we use to scope and deploy appropriate action and communication strategies and moves using our intelligence (aka the adaptive context-sensitivity of behavior) is also the basis and target for manipulation by the Writer. When an agent is both a fluent Reader and Writer in a given action-oriented ontology (or “language”), the relative schemas for interpreting and encoding begin to sync up, with individuals using their own interpretive schema to choose appropriate message complexes, and vice-versa, and bracketing cognitive modeling of semantic difference to some combination of subcultural and psychological differences. When environments and ontologies together are roughly stable temporally and spatially across a population of ecologically proximate players, it becomes easy to mistake the pragmatic, project-fitted, action-oriented character of categories and linguistic meaning for objective reality—nature carved at its joints, rather than the basis of ours. 

In this frame, signaling, fashion, and expression generally all work according to the same information-theoretic, socially strategic abstract principles as language does, because they as well as language are all instances of communication, this being a specific subset of action more broadly; moreover, all operate together inside a shared cultural sphere and environmental state, and are frequently enlisted synergistically to furthering the same set of player goals.

This paper will include a narrativization of the historical lineage in linguistics, ethology, and philosophy I believe this approach to communication stands upon, as well as a discussion of:

  • Language as a structure of highly conventional, common-knowledge focal points for coordination
  • Pragmatism and the “picture” or correspondence model of language
  • Baudrillard’s simulacra, and mimesis as free-riding players’ discovered strategies/social heuristics
  • The conceptual relationship between ethnomethods and Schelling points, with an emphasis on spontaneous coordination technologies
  • Bourdieu’s habitus, rules of the game, space of possibles
  • The place of David Lewis’s “signaling game”
  • Generalized reading, surrogation, PIG-optimality
  • Distinctions between the view of language I call “action-oriented” or “functional-pragmatic,” and the (also Wittgensteinian) view of language held by conceptual engineering 
  • The role of reflexivity, hyperstition, “magic”—whatever you might call it, the process whereby representation alters experience. E.g. the Reader-listener above will have his experience transformed by the Writer-musician’s description. This complicates notions of accuracy/honesty as well as notions of what cooperative behavior looks like.

One response to “Précis for an action-oriented theory of language”

  1. This is an enjoyable read, with a lot in it. Many of the references are beyond me. It is related to a problem that I have had to think about recently (misinformation in from a decision-theoretic POV), so I figured I would share my thoughts.

    Your opening thesis:

    “What is a word, or message, or utterance? It is a move by an agent (~superorganism) undertaken in order to transform the environmental state, in a desired direction, by means of altering another agent’s information state.”

    At its core, this is the decision-theoretic perspective on information, yes? You have bracketed the question of desires (expected utilities) to emphasize the action portion. From there, you eventually arrive at the following:

    “When environments and ontologies together are roughly stable temporally and spatially across a population of ecologically proximate players, it becomes easy to mistake the pragmatic, project-fitted, action-oriented character of categories and linguistic meaning for objective reality—nature carved at its joints, rather than the basis of ours.”

    I would expand upon this to include the fact that (model) utility functions have stabilized — even if they differ between individuals, these differences are known and accounted for, and an “objective reality” of the social good emerges. This is worth distinguishing from other conceptualizations, because it refers to the objective(s) (“desires”) by which the action-oriented application of other linguistic categories makes sense. This helps me understand, at least, why an outsider to a metastable culture can be perceived as dishonest by insiders for manipulating the encountered ontologies in a novel way to their end, which does not fall within the motivation, language, and action structures of the natives. Few people in our contemporary culture, for instance, would fault someone for a white lie used to politely avoid a social engagement. Certain kinds of excuses are even understood to be of ambiguous “truth” value. On the other hand, a lie regarding the health of one’s child, used to miss a week of work to go on a vacation, would be perceived very differently.

    The concept I have wanted to understand¸on my end, is “misinformation”, which is a real concern of the social managerial class, and for good reason. Unregulated information flows expose every human being on the planet to the possibility of manipulation by both human and machine agents. However, “misinformation” is clearly not used to exclusively mean “incorrect statements about objective phenomena”, nor would that be a particularly useful definition. Even when that description applies to what is labeled misinformation, it is clear the induced actions on the part of the misinformed is the harm that is feared. “COVID is a hoax” is dangerous because it might lead an individual to catch and spread COVID. Less importantly, this becomes clear in how “misinformation” is thrown around in the culture wars. Underneath it, there are usually people from separate regions of social stability yelling across a a chasm at each other (that image is imperfect: the `chasm’ only exists in certain projections of the agents’ respective worldviews).

    Like

Leave a Reply to Jake (@Suppose_Not) Cancel reply

Fill in your details below or click an icon to log in:

WordPress.com Logo

You are commenting using your WordPress.com account. Log Out /  Change )

Twitter picture

You are commenting using your Twitter account. Log Out /  Change )

Facebook photo

You are commenting using your Facebook account. Log Out /  Change )

Connecting to %s

%d bloggers like this: