2.1 Human communication about threats: From automatic to intentional communication

The automatic expression of emotion communicates information about environmental benefits and hazards to others. For example, Blair (Reference Blair, Frith and Wolpert2003) argues that emotional expressions serve as unconditioned stimuli that enable a conspecific to evaluate the eliciting stimulus as positive or negative; for example, if another animal expresses disgust for a food, then conspecifics can learn to avoid this food. Indeed, animals are even able to use the emotional expressions of non-conspecifics to learn about dangers in their environment. For example, some Diana monkeys in the wild are able to infer from a chimpanzee's alarm call for Leopard! that a leopard must be present, and utter the Diana-ese warning call for Leopard! to their own conspecifics (Zuberbühler, Reference Zuberbühler2000a; Reference Zuberbühler2000b). Whether emitted automatically or intentionally, the Diana monkeys’ warning calls to each other presumably enable the group to take early and appropriate action to avoid the impending threat.

In this way, animal expression of emotion serves as a means of communicating to their conspecifics and to other species (Darwin, 1888), and influences others’ behaviour in a way that is adaptive for the group (and therefore the species), and which is in this sense “rational”. There is also a sense in which the alarm call in the presence of a predator gives a “reason” for the group to take flight, and the fact that Diana monkeys are able to infer that a leopard is present from hearing the chimpanzees’ alarm call for leopards shows that they have causal inference abilities. However, the processing of these alarm signals does not have to be conscious and deliberate for a communication system to be rational and adaptive for the group concerned.

For example, imagine I am talking to you, when you see my startled expression and hear my cry of alarm. Although my reaction has been automatic, it communicates to you that something potentially dangerous has come to my attention that has upset me, and which could also be dangerous to you. You may interpret my expression as effectively saying Flee! and do so without further ado. However, you may want more justification before doing so and so orient your gaze to look at what I am looking at, and if it is something dangerous (like a leopard that has just come into view), then take evasive action. Here you effectively interpret my line of regard as giving a reason for my emotional expression and saying Flee! There's a leopard! Finally, it may not be something physically present that has alarmed me, but a thought that has just come to my mind. For example, I may just have realised that unknown to the authorities, we are in an area that has just become potentially unsafe (e.g., a part of a zoo that is used to exercise animals after 5 o’clock), and now that it has just turned 5 o’clock, a leopard may come into view at any moment. Here I would normally have to use language to explain (intentionally and co-operatively) the reason for my alarm to you, saying Flee! A leopard might come!

2.2 Anticipatory planning and linguistic valence

Whereas avoiding harm may have an automatic, reflexive nature that is shared with other animals, the ability to engage in co-operative planning seems specific to humans, who can use language to describe things that are not present, and to evoke past states and negotiate future ones. Although chimpanzees and bonobos have shown some ability to use language as a representational system (Premack, Reference Premack1976; Savage-Rumbaugh, Shanker & Taylor, Reference Savage-Rumbaugh, Shanker and Taylor1998) and to communicate effectively about things that are physically present, they have difficulty in evoking things that are not physically present (Gardenförs, Reference Gärdenfors, Oller and Griebel2004). Detached representations of what is not physically present can enable humans to discuss future projects and co-operate to achieve them. Language may also help co-operation through serving as a “second signal system” (Luria, Reference Luria1959) that enables us to respond to ideas that are for the moment just that, and so defer immediate individual gratification in order to reap the delayed benefits of mutual co-operation, something that is difficult for non-linguistic animals (Stevens & Stephens, Reference Stevens and Stephens2004).

Nevertheless, the language that would be used in this kind of co-operative anticipatory planning seems to have an inherently valenced nature. For example, imagine a group of farmers who get together to build a fort to defend their harvests during the winter from attack by brigands, and who discuss plans for the construction of that fort (cf. Gardenförs, Reference Gärdenfors, Oller and Griebel2004). Communication about planning seems to be inherently valenced, as in such co-ordination, it will be useful to encourage some actions and discourage others. For example, the farmers in our example could advise each other on profitable courses of action: “There are a few usable trees in that wood over there” sounds like an encouragement to work in that wood, whereas “There are few usable trees in that forest over there” sounds like a discouragement from doing so. These quantifiers do more than just denote proportions; they also signal what the speaker encourages and discourages (cf. Teigen & Brun, Reference Teigen and Brun1999) by suggesting reasons: “Work in that wood because there are a few usable trees” vs. “Don’t work in that wood because there are few usable trees”. Thus although a plan may be conscious in form (someone imagines a goal that he wishes to attain), its linguistic expression will necessarily be valenced.

In the examples used above, I have introduced quantifiers to give valence to conditional expressions. However, even without polarized quantifiers, conditional advice and instructions would also seem to have a similar property of conveying encouragement and discouragement. For example, if I say “If there are trees in that wood, then work over there” sounds more encouraging than saying “Work over there only if there are trees in that wood” or “If there are no trees in that wood, then don’t work over there” (cf. Hilton, Kemmelmeier & Bonnefon, Reference Hilton, Kemmelmeier and Bonnefon2005; see below). What is striking is that there seems to be no “disinterested” middle way of making a statement about the quantity of trees available without implicitly suggesting that the characteristic in question is desirable or undesirable. In life, as in The Lord of the Rings, trees are generally thought of as good eggs: thus “There are a few trees over there” sounds more positive than “there are few trees over there”.

2.3 Argumentation and logical vocabulary for dealing with uncertainty

I argue that the essential structure of human communication about decision-making is argumentative. Communication about decision-making therefore has essentially an action + reason structure where an action (e.g., flight from a danger, or work to be done) is recommended on the basis of a reason (e.g., the presence of a danger, or of an opportunity). Arguments have to be either for or against a course of action or a point of view, and hence assume a positive or negative aspect. They cannot be neutral, nor will the inherently polarised nature of human language permit this. For this reason, argumentation is inbred in the structure of language itself, and part of the meaning of conditionals, quantifiers and probability expressions comes from their polarity (positive or negative) which allows — even forces — the speaker to express his attitude to the action in question. This phenomenon, whereby key parts of the human logical vocabulary automatically and invariantly express the speaker's attitude to the proposition in question without changing its truth-conditions generates what are known as “conventional implicatures” (Levinson, Reference Levinson1983). These conventional implicatures generate automatic inferences about the speaker's attitude to a proposition, that are as inherently part of the meaning of these words as the quantities and probabilities described by them.Footnote ¹

4.1 Linguistic polarity of quantifiers

Extensive research has been done which shows that quantifiers (Moxey & Sanford, Reference Moxey and Sanford1993a, Reference Moxey and Sanfordb; Reference Moxey and Sanford2000) are polarized; that is, they focus on reasons for action, and these reasons are either positive or negative. For example, Moxey and Sanford (Reference Moxey and Sanford1993) deal with constatives of the form:

A few tourists went to the Blue Lagoon because….

Few tourists went to the Blue Lagoon because….

They asked participants to give sentence completions, and found very strong tendencies for participants to give reasons for going in the case of positive polarity quantifiers like “a few”, “some”, “many” (e.g., because it was very famous), and reasons against in the case of negative polarity quantifiers such as “few”, “not many”, “not all” (e.g., because it was too hot). Moxey and Sanford (Reference Moxey and Sanford1993) obtained these results even though they showed that “few” and “a few” are considered to denote approximately the same numerical proportion (approx. 15–25%).

4.2 Linguistic polarity of verbal probability expressions

In a series of studies, Teigen and Brun (Reference Teigen and Brun1995; Reference Teigen and Brun2000; Reference Teigen and Brun2003) have demonstrated polarity effects in verbal expressions of probability, and found comparable results using probability expressions in an advice setting (Teigen & Brun, Reference Teigen and Brun1999). Thus they found that the words “some possibility” and “quite uncertain” are interpreted to signify about 30–35% probability of a target event in a specific context happening. However, if the expression “some possibility” is used, hearers are likely to prefer the action in question (91%) over the case where the expression “quite uncertain” is used (32%). Teigen & Brun (Reference Teigen and Brun1999) also presented a third condition where a numerical expression (“30% chance”) was used, leading an intermediate number (58%) to prefer taking the action. However, as I show below, it would be premature to conclude from this result that numerical probability expressions are inherently neutral.

4.3 Framing of options and numerical probability expressions

In a study that extended Teigen and Brun's (Reference Teigen and Brun1999) design, Juanchich (Reference Juanchich2006) used indicative conditionals that are used to give indirect advice, such as “If you take this operation, there is a chance of it succeeding.” She used positive and negative verbal quantifiers that were perceived to convey an approximately 40% chance of an operation succeeding (a chance vs. not certain), and found that participants were more favourably disposed to taking the operation after receiving the positive polarity quantifier. However, the same numerical information can also be framed in a positive or negative way: thus the prediction can be expressed as “If you take this operation there is a 40% chance of succeeding” as opposed to the complement negative form “If you take this operation there is a 60% chance of it not succeeding.” Juanchich found that the positive and negative framing had similar effects with the numerical expressions as with the verbal expressions; participants were more likely to be disposed to take the action in question after the positive framing. Juanchich's results are consistent with the findings of Sher and McKenzie (Reference Sher and McKenzie2006) who found that a recommendation that focused on the number of successes obtained by a management team was interpreted as a more positive recommendation than a message that focused on the number of their failures.

4.4 Co-ordination through linguistic polarity of risk communications

Villejoubert and Hilton's (Reference Villejoubert and Hilton2007) research suggests that evaluations of the desirability of an action (as evaluated in terms of its expected costs and benefits) will often determine choice of the polarity of quantifiers and probability expressions. For example, a doctor might be more likely to recommend a treatment to a patient if the treatment is cheap and fast-acting than if it is long and expensive, even if there are the same chances of cure in both cases. In the first case, treatment failure would not be serious, whereas in the second case the prospect of failure should be seriously considered before taking a decision. Accordingly, the doctor might emphasise the chances of success in the first case by saying “there is some possibility of success,” whereas he might emphasise the chances of failure in the second case by saying “it is quite uncertain that it will succeed.” There is evidence that hearers will indeed pick up the implicit preferences of the speaker in these kinds of scenario. Thus Teigen and Brun (Reference Teigen and Brun1999, Expt. 1) found that judges who receive these predictions are much more likely to recommend treatment to a friend in the first case than in the second, even though both terms convey approximately the same probability.

Results such as these suggest that hearers (e.g., patients) are able to pick up the implicit preferences of the speaker (e.g., doctor) quite independently of perceptions of the probability of relevant outcomes. Indeed while both expressions may be equally “rational” in that they describe an objective probability equally well, the choice of “some possibility” over “quite uncertain” will be more adaptive if (say) it is in the patient's interest to persuade him to take the treatment in question. This alignment of preferences shows how the “emotional undertone” due to the polarity of probability expressions can achieve adaptive social co-ordination between speaker and hearer. Below, I further consider the issue of achieving social-co-ordination in the management of uncertainty through the use of conditional expressions.

4.5 Social co-ordination of planning

In the view that has been developed above, groups manage uncertainty either through emotional expressions or through performatives, which express intentions that are recognized and acted upon by others, and which serve as reasons for their acting (Strawson, Reference Strawson1964). These expressions can aim to avoid threats to well-being or to profit from opportunities. They thus concern actions that can either avoid a loss (prevention focus) or attain a gain (promotion focus) from one's present position (Higgins, Reference Higgins2000). Although avoiding losses and realizing gains both bring utility to the agent, alarm calls (that prevent undesired outcomes) may sometimes be automatic and reflexive in a way that the kind of conditional thinking that is needed to negotiate and promote outcomes is not. Nevertheless, conditional performatives incorporate an emotional tone that expresses the speaker's implicit attitude to the action in question.

In this section I review a series of studies which show how this emotional tone can be used to co-ordinate group action round a shared goal. I then review the implications of this co-ordination for models of rationality.

4.6 The conversational action planning model of conditional performatives

Hilton, Kemmelmeier and Bonnefon (Reference Hilton, Kemmelmeier and Bonnefon2005) presented a conversational action planning (CAP) model of how conditional directives are used to co-ordinate responses to uncertainty. In the case that they studied, a boss (or principal) gives instructions to a subordinate (agent) about what to do should certain situations arise. For example, a security chief at an airport may give instructions to agents manning a luggage X-ray machine about what baggage they should take out and search. In the language of signal detection theory, there are two kinds of correct decision in such a vigilance task: a “hit” (when a suspect piece of baggage is correctly identified and taken out) and a correct rejection (when a piece of baggage is correctly identified as safe and is not taken out and searched). There are also two kinds of error: a “miss” (an error of omission, whereby a suspect piece of baggage is not taken out and searched) and a “false alarm” (an error of commission, whereby a safe piece of baggage is incorrectly identified as suspect and taken out and searched).

Below we consider how situational utilities might affect what the chief would most worry about, and thus guide the formulation of conditional instructions. For example, if there is a heightened threat of terrorist attack, the potential costs of missing a suspect bag will be very high, and therefore is much to be avoided. However, in a context where commercial considerations are salient, such as the potential costs of losing busy and lucrative passengers through making them wait too long for baggage searches, then the costs of false alarms will be salient. In Experiment 3, Hilton, Kemmelmeier and Bonnefon (Reference Hilton, Kemmelmeier and Bonnefon2005) varied the contextual utilities in four scenarios involving conditional instructions by making either the potential costs of misses or false alarms salient. They asked participants who were placed in the role of the chief to rate how well each of four conditional instructions “expressed their intention” in the two types of context. When the potential cost of misses was salient (e.g., because there was a heightened risk of a terrorist alert), participants preferred to express their instruction through the “standard” If p then do q form of the conditional, that is, If there is a suspect piece of baggage then take it out and search it. However, when the potential costs of false alarms was salient (due to the risk of alienating busy and lucrative passengers) the other formulations of the conditional (If and only if p then do q, Do q only if p, If not p then do not do q) were preferred, i.e., If and only if there is a suspect piece of baggage then take it out and search it, Take out the piece of baggage and search it only if it is suspect, If there is not a suspect piece of baggage then do not take it out and search it.

Contextual utilities (the importance of avoiding a miss vs. avoiding a false alarm) thus clearly influence the chief's selection of conditional instructions, and what kinds of actions he would like to encourage and what kinds of mistakes he would like to discourage. But do these different conditional instructions encode sufficient meaning to convey to the agent what to do when the contextual utilities are not clear to him? This may be the case when the agent is inexperienced, and does not know what is important to his chief or the organization that she represents. In Experiment 4, Hilton, Kemmelmeier and Bonnefon (Reference Hilton, Kemmelmeier and Bonnefon2005) required participants to put themselves in the role of a subordinate (e.g., a shop assistant) who had received an instruction from a chief (e.g., a shop owner). Their aim was to see whether recipients would interpret the instructions consistent with the utility-structures that would have incited their chiefs to select these instructions in the first place. For example, were instructions of the form If p then do q such as If a client touches the clothes then offer him assistance interpreted as if misses (p but not doing q) were to be avoided? In fact, recipients of this instruction were indeed likely to interpret this instruction in the way intended; do q if p (thus avoiding a miss, p the case but q not performed), and to use their discretion if not p were the case. For the other instructions (If and only if p then do q, Do q only if p, If not p then do not do q) the recipients correctly perceived that they were expected not to do q when p was not the case (thus avoiding a false alarm, p not the case but q performed). There were also subtle differences between these conditional instructions about what to do when p was the case, which were coherent with the speaker's evaluations of each conditional instruction in Experiment 3. This indicates that each conditional formulation has a distinct effect on the recipient's perceptions of what he has to do (see Hilton, Kemmelmeier & Bonnefon, Reference Hilton, Kemmelmeier and Bonnefon2005 for details).

Do participants in such a conversational exchange need to reason “logically” in order to co-ordinate effectively? Not necessarily. If we regard the phrasing of each conditional as carrying a “pragmatic signal” about what is important in the circumstances, then this may be sufficient to achieve co-ordination. Thus use of the “standard” If p then do q form of the conditional expression does not appear to be value-neutral and will be used when the action (q) is to be encouraged because there may be costs to not doing q. The other forms tested, (If and only if p then do q, Do q only if p, If not p then do not do q), all suggest that there could be costs to doing q, and act to discourage doing q unnecessarily.

For present purposes, what is important is that social co-ordination may be achieved through the emotional tone conveyed, and not through the logical interpretation of the conditional rule in question. It is of course also possible that participants’ responses are driven by the different logical meanings of each of the four conditional formulations. This could be done if we imagine that each conditional formulation defines a different set of “utility-conditions” analogous to truth conditions for indicative conditionals (Over, Manktelow & Hadjichristis, Reference Over, Manktelow and Hadjichristidis2004). It is of course possible that some participants react to the emotional tone, while others engage in full-blown analysis of the logical meaning of these expressions. Although reacting to the emotional tone of a conditional seems to have the advantage of being automatic and relatively effortless, for the purposes of the present argument it is enough to note that the “emotional tone” is sufficient to trigger appropriate behaviour in the recipient, not that it is the only mechanism by which this can be achieved. Footnote ²

4.7 Social rationality and co-ordination: Explicit rules vs. implicit understandings

Successful co-ordination between the chief and the subordinate depends in part on the chief's ability to express his intentions clearly. This is not the same as being explicit, as the chief may want to leave some latitude to his subordinate to exercise his discretion appropriately depending on the circumstances. For example, in giving the instruction If the client touches the clothes then offer him assistance the shop owner makes explicit what the assistant is to do if the client touches the clothes, but leaves open the question about what to do if the client is not touching the clothes. Here the assistant may use her discretion, and offer assistance if she judges that the client would welcome it (e.g., the client is looking puzzled). However, the instruction If and only if the customer touches the clothes then offer him assistance is explicit about what should be done in both cases, and thus is highly directive. It is absolutely clear and specific, and leaves no discretion at all to the assistant.

However, it may well be that the shop owner intends his assistant to use her discretion in certain cases. In this case, fully effective co-ordination will depend on illocutionary pick-up by the shop assistant who understands what is expected of her in cases where the shop owner has deliberately left it implicit what the assistant is to do. For example, the instruction If the customer does not touch the clothes then do not offer him assistance leaves it to the assistant's discretion what should be done if the client is indeed touching the clothes. In giving this instruction the shop owner may wish to curb overzealous attentiveness to the shop's clients on the part of his assistant, but there would normally be an understanding that the assistant would at some point offer assistance to a client who is touching the clothes, for that is her role. However, if the assistant never approaches a client to offer her help, the shop owner could well reproach the assistant of not doing her job properly, even though she has respected the letter of his instruction not to approach the client if he is not touching the clothes. This is because the instruction in this case is explicit about what should not be done, but is implicit about what should be done, leaving the decision about when the client should be approached to the judgment of the assistant. If she exercises good judgment about when to approach clients when they are touching the clothesthen the owner will be happy with her, and if she exercises poor judgment he will be unhappy with her. However, if she violates the explicit instruction by offering help to a customer who is not touching the clothes her boss may be entitled to feel angry with her, and not just disappointed with her performance.

4.8 Kinds of rationality: Rules, reasons and social co-ordination

In sum, the studies reviewed above show how it is possible for human work groups to co-ordinate around shared goals. Conditional instructions given by authorities or experts can be nuanced so that subordinates have a sense about what it is important to do and what to avoid doing in certain situations. Importantly, “doing the right thing” is to be evaluated in terms of aligning one's actions with the intentions of the speaker. In particular, successful illocutionary uptake (Austin, Reference Austin1962; Strawson, Reference Strawson1964) is achieved when the agent (e.g., assistant) does what the principal (e.g., shop owner) would want in these circumstances, rather than by following some kind of logical rule or by calculating an individual's expected utility (Evans & Over, 1996). In this analysis, the agent does what she does for a reason, specifically as a result of the principal's instruction. Clearly, the linguistic polarity of the principal's instruction will give the agent important clues about what is required in the circumstances.