Key notions in the study of communication

I am enthusiastic about Thom Scott-Phillips’ book. It integrates cutting-edge research in several fields, from biology to pragmatics, relevant to the study of the evolution of human communication and it redirects the whole enterprise in a new, much more promising direction. This, however, is not the place to wax lyrical about the book; so, let me focus on a single conceptual question of broad relevance where Thom makes a very valuable contribution, but one that, I believe, still needs more work on the part of all of us.

In chapter 2, table 2-1, Thom proposes a set of “definitions of key terms in communication” developing the idea that “if an action [of one organism] causes a reaction [of another organism], and both are designed for the purposes of playing that role in the interaction [between the two organisms], then we can term the action a signal and the reaction a response, and the overall interaction communicative. If however, one half or the other was not designed for these purposes, then the action is either a cue or a coercitive behaviour.”


The idea of defining key notions in the study of communication and related phenomena in term of the presence or absence of interfacing designs or functions is a brilliant one (suggested by John Maynard Smith and David Harper and developed in an original way by Thom), but one that raises a number of difficult issues. I’ll focus on just one such issue.

The issue comes, in good part, from the fact that this approach is rooted in ethological approaches to communication. There, what matters is the behaviour of the interacting organisms. Their internal cognitive states and processes, if considered at all, are just a part of the mechanism causing the expected behaviour. What matters is the behaviour.

The ethological approach, whatever its great merits, isn’t such a good source of insight when you study forms of communication the success of which cannot be identified in terms of any behavioural reaction, as in much of human communication. The basic reaction to communication is, in all cases, cognitive; it may, as generally in animal communication, be a trigger to a specific behavioural response. Even in the human case, the cognitive reaction may be aimed at causing a specific behavioural reaction, as when what is communicated is a request (“pass the sugar, please!”). Much of human communication, however, isn’t aimed at causing some specific behavioural response. The production of this post, for example, is an instance of communicative action designed to elicit in you, readers, just a cognitive reaction. This post might also elicit a behavioural reaction is some of you (yawning or buying Thom’s book, for example), but this is not what it is designed to do.

Thom’s definitions are more appropriate for communication aimed at a behavioural response. Even there, the cognitive component makes communication special, in a way that is not reflected in his definitions. Thom definitions, actually, would be appropriate to talk not just of communicaction but of all forms of interaction where there is a difference between action and reaction. Communication, however, is about providing a cognitive input and causing a cognitive reaction (which may but need not be a way of in turn causing a behavioural reaction). I would suggest that this should be reflected in the way it gets defined. Otherwise, we extend our notion in a way that I find unhelpful.

With Thom’s definitions, somewhat pardoxically,

1) Mating would provide in itself the perfect example of communication since the action of one organism and the reaction of the other are each designed for this interaction. This is true not only of mating in animals that commonly do communicate in, around, and through mating, but also of mating between, say, two haploid yeasts of opposite mating types the behaviour of which can be parsimoniously described without invoking at all the notion of communication.

2) The relationship between beetles feeding on elephant dung and elephants would count as one of ‘cue’: the dung being the cue that the elephants produce and from which beetles benefit, the production of the benefit being an undesigned action and the feeding being a designed reaction.

3) Any designed type of action of one organism on another with no designed reaction (for example the action of a parasite plant feeding on its host) should count as “coercion”—which it is, of course, in this example but in the ordinary sense of ‘coercion’ rather than the one intended by Thom. (Actually, Thom’s example in his book of a women being pushed from her chair is a case of coercion in the ordinary sense, so, I may be misunderstanding him.)

The word “coercion”, by the way, works much better for non-cognitive interactions than for cognitive ones. For the cognitive case, I much prefer “manipulation” for what Thom has (or should have) in mind. “Coercion” may be a well-established term in one strain of biological writing on the issue, but “manipulation” has a good pedigree too, including in biology, and well beyond.

Why, you might well ask, not use for communication a classification that works for all kind of interactions? Because communication has some quite specific features that would not be properly identified under such a general definition.

There is, in particular, a basic asymmetry, of major evolutionary significance between the “reaction”, which is primary, can stand on its own, and doesn’t need any manipulative or signalling “action” to be adaptive, and the “action” side, which is adaptive only if there are receiver with a cognitive capacity to react.

Take the “cue” situation. It is just the ordinary situation of any organism involved in any perception-based cognitive activity. Such an organism has mechanisms designed to take advantage of some specific features in its environment, features that indicate something of relevance to it and that are, therefore cues (in the ordinary sense at least). Cues in this sense may be provided not only by other organisms but by anything that can affect sensory receptors. Organisms pick cues of impeding changes of weather, cues of the presence in useful quantity of various important substances, water, minerals, and so on. Cues of biological origin are, of course, centrally important: much of cognition is in the business of recognising and processing cues of the presence of mates, or of prey, or of predators, i.e. other organisms (which also have an interest in being or not being recognized).

Insect-eating birds are equipped to detect insects, and more specifically, to discriminate types of insects according to the benefits and costs there may be in eating them. The distinctive features of various species of insects provide cues for this discrimination. It may be in the interest of some species of insects to evolve (1) features that make them bad prey, and (2) features that advertise the fact that they are bad preys, as is famously the case with Monarch butterflies, which are poisonous and highly recognizable. The distinctive pattern of their wings is a classic example of interspecies “signalling”, in Thom’s exact sense: it is designed to cause a reaction in birds and this reaction is itself designed. My point here being that you need cognition at the reaction end of the interaction, but not necessarily at the action end: the Monarch butterflies’ pattern is not an output of their cognitive processes. More generally, signalling, even in its broad biological sense, is an exploitation of cognitive capacities in the recipients. This signalling can be “designed” by natural selection. It needn’t be intended or even cognized in any way by the signaller. Signalling needn’t involve any cognitive capacity.

Describing “coercion” or, as I prefer to call it, manipulation, as involving a designed action but no designed reaction is not strictly incorrect, but it may be quite misleading when one sticks to cases of cognitive reaction. Manipulation works by exploiting a cognitive mechanism, and it does so by providing an input with features that the mechanism is designed to react to but that, in this case,—and this is what makes it manipulation rather than signalling—provide the wrong input for further inference. So there is a designed reaction that is being exploited, but its function isn’t to be so exploited, making the case different from that of signalling.

The Viceroy butterfly, whose wing pattern mimics that of the Monarch, is picked up by a cognitive mechanism in birds that evolved to pick such patterns—the mechanism isn’t precise enough however to distinguish the Viceroy from the Monarch exact pattern (see figure); you may have trouble doing so yourself—, but it leads in this case to the mistaken inference that the butterfly is poisonous and shouldn’t be eaten (actually, as Olivier Morin warns me, whether the Viceroy is genuinely palatable and hence whether this whole classical example of “Batesian mimicry” is genuine has become controversial, but even so, it serves my illustrative purpose). Unlike coercion as defined by Thom, manipulation in this sense is restricted to actions that take advantage of the limits of the receiver’s cognitive mechanism by producing, in the case of mimicry, false positive, and, in the case of camouflage, false negatives (I would describe all this in terms of the proper and actual domains of the cognitive mechanisms involved). In other terms, the possibility of coercion/manipulation is dependent on the existence of cue-picking mechanisms whereas the possibility of cue-picking (i.e. of cognition) isn’t symmetrically dependent on the existence of coercion/manipulation.

One of the effect of this dependence of manipulation on genuine cues and, often, on actual honest signals, is that there is a continuum of cases between manipulation and signalling. Think of the expression of genuine emotions that can also involve a good dose of manipulation.

Thom’s “definitions of key terms in communication” raise several other important issues having to do, for instance, with the relationship between biological and cultural functions (both well-exemplified in human language) and with the graded vs. categorical character of these or any such definition (as suggested by the signalling manipulation continuum I have just alluded to). So I tend to see these definitions as a good provisional tool for the use Thom makes of them, but also as a way to frame some very basic and not yet resolved conceptual issues in the study of communication.

2 Comments

  • comment-avatar
    Archibald Haddock 23 June 2015 (08:33)

    Before I respond to his comments, let me take this opportunity to express in a public forum my deep admiration for Dan’s work, which has had a very substantial and profound effect on my thinking. This is true both of his long-term collaboration with Deirdre Wilson, and of many of his other projects. Speaking Our Minds would be a very different book if not for these prior contributions, and it is my sincere hope that readers are able to see and appreciate this. It is, moreover, a delight for me that in his comments above Dan reverses the trend, by taking some of my work on defining communication, and building on it himself.

    As Dan observes, I have followed a profitable tradition, well-established in ethology and related disciplines, of defining communication in terms of functionality (rather than in terms of, say, mechanism). He goes on to suggest that while useful, my framework requires further refinement if it is to address some “very basic and not yet resolved conceptual issues in the study of communication”. By way of illustration, Dan proposes that my notion of “coercion”might be better characterised as “manipulation”, and that whatever we call this category, it may exist on a continuum with signalling proper.

    To respond to this, let me first clarify how I think of the term “coercion”. Table 2.1 of SOM, which Dan reprints above, effectively provides a gloss of my definition, but there is a nuance not captured by the table itself. In the table’s legend (and in the main text), I write that what matters for my definitions is not simply that the actions and reactions in question are designed, but whether they are designed “for the purposes of playing that role [action or reaction]”. This is cumbersome language, but I hope the substantive point is clear: reactions are coercive not only when there is no designed reaction of any sort (as in, say, the example of somebody pushing a work colleague off her chair; see figure 2.2 of SOM), but also in those cases where there is a designed reaction, albeit not one that is designed to be a reaction to the action in question.

    Dan proposes that the language of proper and actual domains would be useful here, and I suspect he is right. Replace in table 2.1 the heading “Reaction designed?”with “Is the reaction designed, and does the action fall within both its proper and actual domains?”(a corresponding change could be made to the first column too, if desired). Signalling describes only those cases where the action falls within both the proper and actual domains of the reaction; and coercion describes all other cases. Then the sort of cases that Dan points to, such as the wings of the Viceroy butterfly, do fall properly within the definition of coercion, as they should. I suspect that this nuance also handles the otherwise paradoxical cases (1) and (3) from Dan’s comments. (I am not entirely clear why case (2) is paradoxical. The presence of the dung seems to me to be a cue, and correspondingly satisfies the conditions for being so. Perhaps I have misunderstood the problem.)

    Dan proposes instead that signalling and coercion (or, as he would prefer it, manipulation) exist on a continuum. I am not certain if I have read him correctly, but the argument I infer from his comments is that actions might fall within the actual domains of more than one mechanism, and hence there is a continuum of cases, from, at one end, those cases where the action falls within the proper domain of all the mechanisms involved (this is the signalling end of the continuum), and, and at the other, those cases where the action falls only within the actual domains of the mechanisms, and not their proper domains (this is the coercion/manipulation end of the continuum). This is a very thought-provoking proposal, which strongly merits further development and reflection, not just from myself or Dan, but from all of us concerned with a cognitive approach to human communication.

  • comment-avatar
    Dan Sperber 23 June 2015 (19:36)

    Thom, you are right to stress that, in your definitions of signalling, cue, and coercion, what matters “is not simply that the actions and reactions in question are designed, but whether they are designed “for the purposes of playing that role [action or reaction]”. I was well aware of this point in my discussion and should have mentioned it explicitly. This however doesn’t answer my concern: your definitions are such that there is nothing to restrict them to the case of communication (and the related cases of ‘cue’ and ‘coercion’/’manipulation). They apply quite naturally to all interactions between organisms where an action triggers a reaction.

    (1) If both action and reaction are designed to mesh with the other, then it is a case of mutual benefit between actor and reactor (mutually beneficial communication being a special case). (Incidentally, this rules out as an example falling under your definition the “physical fight between members of two species with attack-defense dispositions that evolved complementarily” evoked by Paulo Sousa and Karolina Prochownik in their comment, since the attack isn’t designed to trigger the defense: if the prey doesn’t fight all the better for the predator).

    (2) If the action is designed to trigger a certain reaction but the reaction isn’t designed to react to that particular action, this is a case of the actor benefitting asymmetrically from the reactor or ‘coercion’ (cognitive manipulation is a special case).

    (3) If the action, without being designed to do so, does trigger a reaction designed to take advantage of the action, then this is a case of the reactor benefitting from the actor (benefitting from cognitive cues being a special case).

    Note that, in the general case, the difference between cases (2) and (3) is a matter of who benefits, the actor or the reactor. Quite often, the actor-reactor distinction is itself a matter of descriptive choice: free-swimming algae arrives inside an oyster that eats them: is the actor the alga that moves into the oyster triggering its feeding reaction (and then it is a case of ‘cue’, in your terminology) or is it a case of ‘coercion’, the oyster coercing the alga’s movement and causing its decomposition? Both definitions work, which isn’t as good as it sounds.

    Now, in the case of cognition and communication, the asymmetry between the role of the source and the receiver of information is basic. I was suggesting that good definitions of communication and related phenomena should reflect this and therefore not just be a special case of general definitions of action-reaction pairings.

    The point is illustrated by the case of the dung beetle that I used in my post as something that shouldn’t be considered a cue. So, the elephant produces dung on which the beetle feed: benefit to the reactor in this case, ‘cue’ in your sense. Where is the problem, you might say? A slightly more fine-grained description would distinguish two related reactions from the beetle, (1) a cognitive reaction: the beetle recognize the substance as food (and for this, properties of the dung that the beetle is equipped to perceive serve as cues), and (2) a more ‘practical’ reaction: the beetle eats and digest the dung; here to describe dung as a ‘cue’ seems to me unhelpful or even misleading. To make the point clearer, compare the beetle and a plant the growth of which also benefit from the elephant’s dung. Is dung a ‘cue’ to the plant too?

    Look, the issue — the stake if you prefer — is a huge one: since Aristotle, Port-Royal grammarians, Peirce, Saussure, Goodman, and many others, insightful proposals have been made to define a number of notions including communication and around it. None of them is without problems. I believe your approach may well be aimed in a promising, novel direction, but we are not there yet (assuming you are interested in this very ambitious aim – as I mentioned, for your goals in the book, your definitions do the job). Or should we go all the way to the Rubicon and just catch some fish?