Language use is a type of behaviour, and just like any type of behaviour, it is embedded in an environment, the context of language use. The context, with its affordances, determines which actions can be taken; it shapes and constrains what things can be said and how they can be said. A crucial component of the environment in which humans take actions are other humans. Humans certainly talk to themselves, or use language to reason about possible states of the world and to form ideas (those, too, after all, are part of the state of the world; they are physical connections and activations in the human brain) but the primary form of language use is in interaction with other humans.
The reason why humans take actions is to change the state of the environment. In particular, we try to change the state of the environment towards new states that either coincide with or get us closer to our goals. Imagine a room containing a desk, a chair, a sofa, a window, and two humans, Peter and Sally. Peter is at the desk and Sally is sitting on the sofa, closer to the window. The window is open and Peter is cold and tired. He thinks that if the window was closed, he would be warmer and happier. Now, with the goal to change the state of the environment—in particular, the state of the room—into one in which the window is closed, Peter may ask Sally ‘Could you please close the window?’.
Goals are transformations of the state of the environment into states with favourable properties: for example, states in which we are happier, experience more pleasure, or make less effort. More generally, goals are states which produce positive social, cognitive, and physical effects. The positive effects that states of the environment generate are their utility. Peter is really cold, and asking Sally to close the window may have an immediate positive physical effect: Sally might close the window and, soon after, Peter would experience an increase in body temperature. A body temperature increase is the utility of this possible future state of the room. Actions, by extension, also have or generate utility, namely the utility corresponding to the environment states that result from taking those actions. If asking ‘Could you please close the window’ transforms the room into a state in which the window is closed, this will in turn generate a temperature increase and positive effects on Peter’s body.
Utility can be both positive and negative. Certain states of the world may have positive social effects but negative cognitive effects. For example, being kind and polite in interactions tends to have positive social effects—e.g., it makes us nicer people to talk or it can make us more convincing—but at the same time, more often than not, it requires more physical and cognitive effort. Peter is tired, and utters ‘Close the window’, which is a much less effortful piece of behaviour to produce. In other words, the negative utility of ‘Close the window’—in this case, its physical and cognitive cost—is lower in comparison to that of ‘Could you please close the window?’.
The interlocutor, or audience—here, Sally—perceives the state of the environment and the bit of linguistic behaviour produced by the speaker—Peter. Relying on her model of the environment and her ability to recognise other humans’ goals and plans (via models of the speaker and their utility), the audience uses the speaker’s behaviour as a set of instructions to reconstruct, or predict, the new state of the environment that the speaker intended to communicate. Linguistic interaction is successful when the audience’s reconstruction of the speaker’s goal is faithful to the originally intended new state of the environment. The act of comprehending an utterance also comes with efforts and positive utility. For example, the audience must pay attention to the speaker’s behaviour, it must take some time to process it, and it must interpret it within the environment. For all of these actions, the audience pays some cognitive cost, which may be recompensed, for example, by positive social utility. If Sally takes the time and effort to comprehend Peter’s request, and decides to indeed open the window, she might make everyone happier.
The audience may be physically co-present with the speaker as in the case of Peter and Sally, they might only perceive the linguistic act after one hundred years as it is the case with books, they might be purely hypothetical, or they might be the speaker themselves. In any case, a linguistic act is inseparable from its audience, just like it is inseparable from its context(s) of production and comprehension. This is as true for language production in face-to-face spoken dialogue as it is for a theatre monologue, a podcast, or a written text. Because the connection between linguistic behaviour and underlying communicative intent is purely arbitrary, and because intents are non-observable (at least until new groundbreaking scientific discoveries), a linguistic act is only complete when it is perceived and interpreted by an audience. In fact, if it is perceived multiple times, potentially by different audiences—song lyrics are an example of this—the same linguistic act can induce multiple changes of the state of the world. These changes can be different each time, they might vary for different audiences, and there is no guarantee that the new state of the world will be the one the speaker intended.
The reason why linguistic acts change the state of the world—that is, beyond their mere occurrence, which itself is an obvious world change—is that while linguistic behaviour is an arbitrary encoding of communicative intents, groups of humans converge on similar models of the world and develop shared systems of interpretation. They interpret the same behaviours in similar ways.
Humans continuously make and accept proposals about how language should be used, and the most successful of these proposals become conventionalised. Peter and Sally’s is a story about a single interaction, in a single environment, between two individuals. These are in a sense units on the temporal, spacial, and social dimensions of context that constitute the space of language interaction. The temporal, spacial, and social dimension jointly shape and constrain what things can be said and how they can be said. When we combine these three dimensions and observe them at once, we appreciate language as a complex adaptive system of interactions. The relationships between the system’s parts, i.e., among humans and between humans and the environment, give rise to ever-changing collective forms of linguistic behaviour.
What makes the emergence of such system of interactions possible are a few fundamental characteristics of linguistic behaviour—so to say, a few preconditions. The first condition is agency, i.e., the ability to behave non-randomly but rather as a function of one’s utility and model of the environment. Utility generates intents; the model of the environment prescribes what actions are more likely to lead to intended outcomes and what outcomes are plausible given the current environment state. The ability to speak a certain natural language can be subsumed under this model of the environment. Moreover, because other humans are part of the environment, the model of the environment is also a model of interlocutors. Without the socio-cognitive skill to entertain a model of the interlocutor, language interaction is hardly ever successful, and language cannot be learned in the first place.
The model of the interlocutor—whether a speaker or a comprehender—must prescribe (i) that the interlocutor’s actions can be assumed to have an intended utility, and (ii) that the interlocutor makes the same assumptions. If Sally did not ascribe any intent to Peter, she would not open the window—nor would she even try to interpret Peter’s utterance as a request. She would rather take it as an arbitrary piece of behaviour to observe, but with which she has nothing to do, like a thunder or a falling leaf. If Peter did not assume that Sally believes he has intents, there would be no point in requesting anything from her. This mutual and recursive recognition of agency (i.e., entertaining intents and performing behaviour as a function of those intents) is a prerequisite for joint action, and thus a second necessary condition for language interaction to occur.
To predict their joint utility, humans use mental models, but choosing which utterance to produce and how to interpret it towards joint utility maximisation are decision-making problems. Beyond models of the environment and interlocutors, speakers must thus possess a higher-level model of utility, either implicit or explicit, as a way to determine the relative importance of different lower-level utilities and to modulate producer and comprehender’s utilities; as well as the ability to deal with uncertainty: as humans do not possess a perfect model of the world, they cannot be certain about the transformations of the state of the world their actions will cause. If Peter had been wise, when choosing between the two alternative utterances above, he would have considered that not only does ‘Close the window’ have lower or negative social utility but also (and probably as a consequence) it creates more uncertainty about the future state of the environment. If he had been wise, he would have thought of the last times he and Sally worked in that room with the window open. A few times he used the utterance ‘Close the window’ and this resulted in a state of the room in which Sally told him to ask more politely, and a few more linguistic actions were needed to bring back their joint social utility to positive levels. Peter, however, is tired, and these considerations require an amount of cognitive effort that he is not in a position to expend. So he says ‘Close the window’ and, more or less knowingly, he takes a higher risk.
Luckily, Sally has an easier time making social utility calculations, she has a good model of Peter, and she knows he is tired. She opens the window anyways.
From Chapter 1 of my dissertation on Neural Models of Language Use.