Rethinking agency in language and society

The issue of agency is generally considered in terms of constraints on and outcomes of the practices of social actors. In language studies, this has taken the form of asking how actors make their worlds through language use, a prominent example being language policy and management. In this regard, the notion of agency – not to mention the related ones of creativity and competence in language use – is typically understood as stemming from the goals and desires of human actors. It is humans who have agency. Because of this, it is humans who are creative, and it is humans to whom competence in language use can be meaningfully attributed. This article argues for the need to go beyond such concerns to focus instead on more fundamental questions about the locus of agency.

Agency has long bedevilled social theorists, with various proposals claiming that its locus lies within the individual, in human created social structures, or that it emerges from some kind of dialectic between the two (Bourdieu 1977; Giddens 1984; Parson 1937; Simmel 1976). These proposals share the idea that agency must have some localizable human fount, an idea that the study of language in society has taken on board also.

For example, Tollefson (1991) distinguishes between neoclassical and historical-structural approaches to language policy. The former emphasizes the rational and individualistic nature of choices. Thus, individuals may choose to learn a new language because of certain perceived benefits such as access to better jobs. Or they may decide that the time and money spent on learning a new language may not be worth the potential benefits, and hence, may not make the effort to expand their linguistic repertoire. Whatever the outcome, the neoclassical approach treats these as decisions that are freely and rationally made. But Tollefson points out that we need to also ask questions like, “Why must that individual expend those particular costs? Why are those particular benefits rather than others available to that individual? What are the costs and benefits for other people in the community?” (1991: 32). These kinds of questions require attending to the socio-historical contexts and constraints inherited by individuals and, mutatis mutandis, communities. Tollefson’s position is that the neoclassical approach has been all too dominant. Countering this with the historical-structural approach would instead shift the focus to examining “the historical basis of policies and to make explicit the mechanisms by which policy decisions serve or undermine particular political and economic interests” (Wiley 1996: 32).

Though Tollefson’s distinction is important, problems still remain. Positioning the issue of agency as a series of dichotomies – between individual and group, between the ahistorical and the historical – presents the analyst with a forced choice where, for example, focussing on the individual would come at the expense of a focus on the group. There is an assumption that agency has an identifiable locus, either individual or group – with Tollefson coming down in favour of the latter.

In a similar vein, Spolsky (2009) has called for attention to be given to language management. It is certainly not unreasonable to talk about language management since there are clearly attempts by various authorities to influence the language practices of targeted populations. Where this becomes problematic, however, is when we assume that the agency of management can and should be isolated and unequivocally identified, as in Spolsky’s (2009: 6) assertion that “I will take the position that it is management only when we can identify the manager.”

The idea of a manager once again presumes an identifiable locus of human agency, and it returns us to the problems with Tollefson’s distinction between the neoclassical and the historical-structural. Ahearn (2001) summarizes the complicated questions about agency that arise:

Can agency only be the property of an individual? What types of supra-individual agency might exist? […] Similarly, we might also be able to talk about agency at the sub-individual level […] thereby shedding light on things like internal dialogues and fragmented subjectivities?

(Ahearn 2001: 8)

Such complications arise because even a body such as “the government”, “the ministry” or “the community” is really an abstraction over multiple sub-entities (themselves potentially recursively sub-dividable) so that “internal dialogues and fragmented subjectivities” apply no less to organizations and groups than they do to individuals (Wee 2018).

There is another problem in addition to the distributed nature of agency: the tendency to downplay if not dismiss the roles of non-human entities:

No one really knows what human agency is, or what humans are doing when they are said to perform as agents. In the face of every analysis, human agency remains something of a mystery. If we do not know just how it is that human agency operates, how can we be so sure that the processes through which nonhumans make their mark are qualitatively different?

(Bennett 2010: 34)

A good illustration of what Bennett means is provided by Latour:

According to Latour, the NRA’s [National Rifle Association] braying insistence that “guns don’t kill people, people kill people” is premised on more anthropocentric understandings of agency that treat material things such as guns as diligent instruments of human volition. Latour contends, on the contrary, that once a person picks up a gun, she or he is not quite the same person as before. Guns, among other things, when connected with humans, make up new networks or assemblages that embolden or enable certain kinds of actions, specifically killing. (One would not use the barrel of a gun to arrange a bouquet of roses, after all.) […] According to Latour, when a person kills with a gun, it is not only the person who kills. It is the larger assemblage that kills. Its murderous agency is distributed across its many parts including a finger, a trigger, a bullet, a human brain, violent films, and so on. Agency is always complex agency, unlocalizable and distributed across assemblages of both humans and things.

(Hazard 2013: 66, italics added)

Latour’s example illustrates the problem with restricting agency to human actors such that non-human items are seen as mere “instruments of human volition”. The person who is holding a gun has agency in a way that is different than a person not holding gun – even if both have the desire or intention to kill. The combination “person + gun + intention to kill” constitutes a new network or assemblage that allows for some types of actions over others. Taking this insight seriously means recognizing that agency is not only “unlocalizable”; it is also “distributed across assemblages of both humans and things”.

These points about agency carry serious implications for the study of language in society. It is undeniable that various technological advancements ranging from relatively simple computer programs to highly developed artificial intelligence (AI) are increasingly involved in our use of language for communication. Take, as one example, use of automated signs at carparks to indicate to drivers if a carpark is full and, if not, just how many empty lots are actually available. A driver who is approaching a carpark may encounter a sign that says “Carpark full” or one that says, for instance, “86 lots available”. The automated sign at the entrance to the carpark is obviously intended to be communicative, having been programmed to take note of the number of cars that are already present in the carpark and to convey in real time the relevant information (i.e. how many lots are still empty) to drivers who may be thinking of parking their cars there. In a multi-story carpark, drivers may even be told of how any empty lots available are actually distributed over the different levels (e.g. “Level 3, 24”, “Level 4, 35”).

This raises the question of how we are supposed to conceptualize the communicative act. Certainly, we would want to acknowledge that whoever programmed and installed the machine intended for it to be useful and that its usefulness includes conveying the relevant information. But already here there are at least two problems. First, we cannot blithely assume that the programmer is a human being since there may be cases where some form of communicative technology has been created by a computer program. This latter scenario is not as farfetched as it seems because machines are already capable of writing their own code.

Second, regardless of whether the communicative technology was created by another human being or by a computer program, the specific information that is being conveyed at any one time about the state of the carpark (e.g. the actual number of lots available or whether the carpark is full) is not something that the programmer would likely even be aware of. The information is gathered and conveyed via programmed sensors that are independent of the programmer (which is of course the whole point of the programming). As programs get more sophisticated and autonomous (think of Google Assist or Apple’s Siri), it becomes more difficult and less plausible to equate the messages communicated by a program with the goals and intentions of its programmer.

Another example involves the concept of an echoborg. An echoborg is a person whose utterances and gestures are determined to varying degrees by the communications that originate from an artificial intelligence program. The interactional goal is to give the illusion that one is communicating with a fellow human being when in fact the communication originates from an AI. The human with whom one is apparently communicating is really working at the behest of the AI. Echoborgs can be useful since some individuals might feel more comfortable if they think they are interacting with another human even though the kinds of information and advice they want is better and more efficiently provided by an AI. This “synching” of a human front with messages that are created by an AI raises conceptual issues such as the nature of speakerhood (Goffman 1981). Who exactly is speaking under such a condition where the activity of speaking is distributed over more than one entity? Is it the human extension or is it the AI, or is such a binary approach misguided?

Cases where the human element is increasingly distant from the use of language for communicative purposes will get harder to ignore. They pose conceptual challenges for the study of language in society. To address these challenges, a willingness of rethink the nature of agency would be an important start.