The Logic of Indirect Speech

PERSPECTIVE

Steven Pinker*†, Martin A. Nowak‡, and James J. Lee* *Department of Psychology, and ‡Program for Evolutionary Dynamics, Departments of Mathematics and Organismic and Evolutionary Biology, Harvard University, Cambridge, MA 02138

Edited by Jeremy Nathans, Johns Hopkins University School of Medicine, Baltimore, MD, and approved December 11, 2007 (received for review July 31, 2007)

When people speak, they often insinuate their intent indirectly rather than stating it as a bald proposition. Examples include sexual come-ons, veiled threats, polite requests, and concealed bribes. We propose a three-part theory of indirect speech, based on the idea that human communication involves a mixture of cooperation and conflict. First, indirect requests allow for plausible deniability, in which a cooperative listener can accept the request, but an uncooperative one cannot react adversarially to it. This intuition is supported by a game-theoretic model that predicts the costs and benefits to a speaker of direct and indirect requests. Second, language has two functions: to convey information and to negotiate the type of relationship holding between speaker and hearer (in particular, dominance, communality, or reciprocity). The emotional costs of a mismatch in the assumed relationship type can create a need for plausible deniability and, thereby, select for indirectness even when there are no tangible costs. Third, people perceive language as a digital medium, which allows a sentence to generate common knowledge, to propagate a message with high fidelity, and to serve as a reference point in coordination games. This feature makes an indirect request qualitatively different from a direct one even when the speaker and listener can infer each other’s intentions with high confidence.

eople often don’t blurt out what cesses by which speakers veil their human communication. In cases of pure they mean in so many words but requests and hearers recover them have cooperation, one expects maximally effi- veil their intentions in innuendo, been well documented (1–3). However, cient conspiratorial whispers; in cases of euphemism, or doublespeak. Here the reason people engage in these maneu- pure conflict, one expects a shouting P vers in the first place (as opposed to say- are some familiar examples: match (8). The complex, coded communi- ing what they mean clearly and succinctly) que´s that characterize human language Y Would you like to come up and see is still largely unexplained. In this Perspec- bespeak a mixture of cooperation and my etchings? [a sexual come-on] tive, we apply ideas from the analysis of Y If you could pass the guacamole, that conflict. This conclusion is reinforced by signaling in evolutionary biology and from considering that most of the practical ap- would be awesome. [a polite request] evolutionary game theory to illuminate Y Nice store you got there. Would be a plications of indirect speech (diplomacy, possible advantages of indirect speech (4). extortion, bribery, and sexual harassment) real shame if something happened to Existing theories of indirect speech are take place in arenas of conflict. it. [a threat] based on the premise that human conver- Indirect speech takes many forms, in- Y We’re counting on you to show lead- sation is an exercise in pure cooperation, cluding gestures of sympathy and defer- ership in our Campaign for the Fu- in which conversational partners work ture. [a solicitation for a donation] together toward a common goal—the effi- ence (termed ‘‘positive politeness’’ and Y Gee, officer, is there some way we cient exchange of information, in the in- ‘‘negative politeness’’ in Politeness The- could take care of the ticket here? [a fluential theory of H. P. Grice (5), or the ory). The phenomenon we address here is bribe] maintenance of ‘‘face’’ (esteem and auton- sometimes called ‘‘off-record indirect omy) in Penelope Brown and Stephen speech acts.’’ We propose a theory in This phenomenon poses a theoretical puz- Levinson’s Politeness Theory (1). Yet a three parts, which apply to interactions zle. Indirect speech is inefficient, vulnera- fundamental insight from evolutionary with successively more subtle cost–benefit ble to being misunderstood, and seemingly biology is that most social relationships structures (4). The first part is the logic of unnecessary (because only a naı¨f could involve combinations of cooperation and plausible deniability. In a simple case like fail to see past the literal meaning). Yet conflict (6, 7). This insight applies to com- bribing a police officer, the appeal of a politeness and other forms of indirectness munication among organisms no less than veiled bribe is intuitively clear: If some in speech appear to be universal or nearly to physical actions, and indeed animal officers are corrupt and would accept the so (1). We all play this game and may be signaling has been found to involve ex- bribe, but others are honest and might offended at those who don’t, setting the ploitative manipulation as well as the co- arrest the driver for bribery, an indirect stage for the hypocrisy and taboo in social operative exchange of information (8). In bribe can be detected by the corrupt cop life that are ubiquitously decried, yet ubiq- the human case, one has to think only of while not being blatant enough for the uitously obeyed. threats (the proverbial ‘‘offer you can’t honest cop to prove it beyond a reason- Indirect speech also has considerable refuse’’), dangerous secrets (hence the able doubt. A simple game-theoretic practical importance. It figures in the de- need for witness protection programs), model can delineate the circumstances in sign of computer language understanding contaminating leakage (such in blind ref- systems, which need to be programmed which indirect speech is an optimal solu- ereeing, sealed bids, and clinical trials), tion to this problem. not to take indirect requests, such as ‘‘Can and incriminating questions (for which youtellme...’’or‘‘Doyouknow...,’’ one answer might be damaging, the other literally. It is also a major bone of conten- a lie, and a refusal to answer a de facto Author contributions: S.P. and M.A.N. designed research; tion in the framing and interpretation of confession that those are the respondent’s J.J.L. performed research; J.J.L. analyzed data; and S.P. diplomatic agreements, and in the prose- two options) (ref. 9; see also Gerd Giger- wrote the paper. cution of bribery, extortion, and sexual enzer’s Law of Indispensable Ignor- The authors declare no conﬂict of interest. harassment. ance, www.edge.org/q2004/page2.html# This article is a PNAS Direct Submission. For 50 years, indirect speech has been gigerenzer). The very existence of indi- †To whom correspondence should be addressed. E-mail: intensively studied by linguists, philoso- rectness in language suggests that such [email protected]. phers, and psycholinguists, and the pro- adversarial dynamics might be in play in © 2008 by The National Academy of Sciences of the USA

www.pnas.org͞cgi͞doi͞10.1073͞pnas.0707192105 PNAS ͉ January 22, 2008 ͉ vol. 105 ͉ no. 3 ͉ 833–838 Downloaded by guest on October 1, 2021 The second part of theory extends the would be to take care of it here’’). Sup- cide how to react to the proposition; this game-theoretic logic to social situations in pose he knows that the officer can recog- tendency can be captured by a decision which there are no fines or other tangible nize it as an intended bribe, and that the function, L, which monotonically relates costs and benefits, such as a diner without officer knows that he couldn’t make a the directness of the proposition to the a restaurant reservation who bribes a mai- bribery charge stick in court because the probability that the officer will treat it as tre d’ for a quick table, or a person who ambiguous wording would prevent a pros- an attempted bribe and act accordingly. tenders a sexual invitation to a friend af- ecutor from proving his guilt beyond a Putting these together, the expected cost ter a dinner. Unlike the driver and the reasonable doubt. The driver now has a to a driver facing a corrupt cop is yc ϭ officer, the speaker would incur no finan- third option: c0 p ϩ c1(1 Ϫ p); the cost when facing an cial or judicial penalty were the hearer to honest cop is yh ϭ c2 p ϩ c1(1 Ϫ p); alto- turn down a blatant proposition, so the Dishonest Honest gether, the driver’s expected cost is y ϭ officer officer question here is why speakers still resort qyh ϩ (1 Ϫ q)yc. to innuendo. Don’t Traffic Traffic Now, if the corrupt and honest cops The third part addresses scenarios in bribe ticket ticket share a single linear decision function which people use indirect speech even L and hence have the same p for any when the degree of uncertainty about the Bribe Go free Arrest for proposition, the optimal level of direct- other’s intentions is low—either because bribery ness will simply be determined by the variance among listeners’ values is low (so Implicate Go free Traffic fraction of honest officers. If q Ͼ (c1 speakers’ confidence in their values is bribe ticket Ϫ c0)/(c2 Ϫ c0), then the optimum high), or because the listener is astute strategy for the driver is not to make enough to understand the intent of a ϭ The payoffs in this third row com- any bribing attempt at all: d 0. If the speaker’s innuendo with high confidence. bine the very large advantage of brib- fraction of honest officers is less than Why, in such cases, is a thinly veiled prop- ing a dishonest cop with the relatively this critical value, then the optimum osition still more acceptable than a naked small penalty of failing to bribe an strategy for the driver is to make the one? The answer must pertain to some honest one. In these circumstances, most direct and unambiguous bribing property of overt language itself, as op- ϭ indirect speech is the rational option. attempt, d 1. In this model, so far, posed to the processes of social inference indirect speech is never an optimum that power the interpretation of innuendo. Note how this analysis is inconsistent with the traditional idea that indirect strategy. The reason is that the cost speech is an implementation of pure functions are linear in d (Fig. 1). Part 1: Plausible Deniability For an indirect bribe to be advanta- Consider a speaker whose speech obeys cooperation: The driver here is using indirect speech not to help the honest geous to the driver, his overall cost Grice’s maxims (5) of efficient communi- function must be nonlinear. This situa- cation and is thereby always succinct, officer attain his goal (viz., to enforce the law) but rather to confound that tion could come about if the honest truthful, direct, and relevant. He is pulled and corrupt officers employ nonlinear over for running a red light and is pon- goal. The intuition that indirect speech can decision functions Lh and Lc relating dering whether to bribe the officer. His the probability they will act (p)tothe choice is whether to remain silent or to be an optimal strategy can be confirmed in a simple model of a Rational Briber. directness of the bribe (d), and if the say, ‘‘If you let me go without a ticket, I’ll two decision functions are distinct. pay you $50.’’ The expected cost of a bribe y can be cal- culated from (i) the proportion of officers That is, even if honest and dishonest Unfortunately, he doesn’t know officers interpret indirect speech the whether the officer is corrupt and will that are honest, q;(ii) the cost of the bribe, c ;(iii) the cost of the ticket, c same way, and thus have identical as- accept the bribe or is honest and will ar- 0 1 sessments as whether an invitation to rest him for attempting to bribe an officer. (which must be greater than the cost of the bribe, or it would never pay to bribe); ‘‘settle it here’’ is an attempted bribe, The game-theoretic conundrum where the honest cop may be more hesitant (iv) the cost of an arrest for bribery, c one actor does not know the values of the 2 to arrest the driver than the dishonest (which must be greater than the cost of other has been explored by Thomas cop is to accept the implicit bribe, be- the ticket; otherwise, it would always pay Schelling, who calls it the Identification cause of the burden of proof in a pros- to bribe); and, the crucial psychological Problem (9). The payoffs are as follows: ecution. In general, the expected cost variable, (v) the probability p that an offi- for the driver is as follows: Dishonest Honest cer will treat a statement with a given de- officer officer gree of directness d as an attempted bribe. ϭ ͕ ͑ ͒ ϩ ͓ Ϫ ͑ ͔͖͒ ϩ ͑ Ϫ ͒ y q c Lh d c 1 Lh d 1 q Directness is a semantic variable that cor- 2 1 Don’t Traffic Traffic responds to the degree of vagueness of ⅐ ͕ ͑ ͒ ϩ ͓ Ϫ ͑ ͔͖͒ bribe ticket ticket c0Lc d c1 1 Lc d the proposition (the number of readings) Bribe Go free Arrest for and the proportion of those readings that In the simple case in which Lh and Lc bribery are consistent with its being a bribe rather are step functions, the scenario may be than an innocent remark. An if–then prop- displayed as in Fig. 2. If the driver doesn’t try to bribe the osition (‘‘If you let me go, I will give you The expected cost for the driver is as officer (first row), either way he gets a $50’’) is most direct; a leading question follows: ticket; if he does offer the bribe (second (‘‘Is there some way to take care of it ϭ Ͻ row), the stakes are much higher either here?’’) is less direct; and a generic re- y c1 if d dc way: going free with just the cost of the mark (‘‘I’ve learned my lesson; you don’t y ϭ qc ϩ ͑1 Ϫ q͒c if d Ͻ d Ͻ d bribe if he is facing a dishonest cop, or an have to worry about me doing this again’’) 1 0 c h is least direct. (In practice, p can be esti- ϭ ϩ ͑ Ϫ ͒ Ͻ arrest for bribery if he is facing an hon- y qc2 1 q c0 if dh d est one. mated empirically by asking people their Now consider a driver who knows how degree of confidence that a given sen- The intermediate region, dc Ͻ d Ͻ dh, has to use an ‘‘implicature’’ to convey an am- tence was intended as a bribe.) Finally, the lowest expected cost for the driver biguous bribe (‘‘So maybe the best thing the target of the indirect speech must de- and, therefore, represents the optimum

834 ͉ www.pnas.org͞cgi͞doi͞10.1073͞pnas.0707192105 Pinker et al. Downloaded by guest on October 1, 2021 cost arresting the driver will thus depend on relationships in all cultures fall into only the proportion of dishonest and honest three distinct types and that most of the arrest drivers who utter a remark with that level complexities of social life within and honest of directness, and on the professional across cultures may be explained in terms ticket rewards for successful arrests and the pen- of variation as to which relationship type alties for false ones. Conversely, for a dis- applies to a given dyad. (Fiske’s theory corrupt honest cop, the cost depends on the bribe also posits a fourth relationship type, amount of the bribe, the chances of his ‘‘market pricing,’’ but holds that it is spe- being arrested in a sting operation, and cific to industrial and postindustrial societ- subtle explicit the penalty for being convicted of accept- ies.) Each relationship type is character- ing a bribe. It is unlikely that the two de- ized by an ethos governing the directness of proposition cision functions would have the same distribution of resources between partici- shape. pants, and a straightforward evolutionary Fig. 1. The expected cost to a driver of tendering basis that specifies which kinds of dyads bribes to honest and corrupt police ofﬁcers with Part 2: Relationship Negotiation linear decision functions. fall into that type of relationship by de- The second puzzle of indirect speech is fault. However, each relationship type can why people use it in nonlegal situations, be extended to other dyads by negotiation where there are no financial or legal pay- level of directness. This result confirms and manipulation: offs and penalties. Consider bribery in that there are plausible circumstances in everyday life, such as bribing a maitre d’ Y which indirect speech is an optimal strat- The dominance or authority relation- at a popular restaurant to be seated im- egy and is a formal implementation of the ship is governed by the ethos, ‘‘Don’t mediately despite having no reservation. everyday concept of plausible deniability. mess with me.’’ It has a basis in the A restaurant critic, given the assignment The result does not depend on the of- dominance hierarchies common in the to write about such an experience for animal kingdom, although in humans, ficers’ decision rules Lh and Lc being step- Gourmet magazine (10), reported that the it is based not just on brawn or se- functions. They could also be sigmoid prospect of being turned down filled him functions such as logistic or normal-ogive. niority but on social recognition: how with anxiety and that he resorted to indi- much others are willing to defer to As long as corrupt officers have lower rect speech to tender the bribe, such as ‘‘I threshold parameters than honest officers you (12). It is generally communi- Ͼ hope you can fit us in’’ or ‘‘I was wonder- cated by psychophysical cues to in their sigmoid function, that is, Lh(d) ing if you might have a cancellation.’’ The Lc(d) over some appreciable interval, it is strength and resolve (13). second part of the theory thus seeks to Y The communality or communal shar- easy to show that a wide range of parame- explain why speech would be indirect in a ing relationship conforms to the ter values yields a minimum of y with nonlegal context such as a restaurant ethos, ‘‘What’s mine is thine; what’s respect to d between the boundaries of bribe or in a sexual overture among peers. thine is mine.’’ It naturally arises silence at one end and bald directness at A key to this puzzle comes from Polite- among kin, who are bound by shared the other. The results also hold when ness Theory (1), which proposes that lan- there are distributions of the threshold guage serves two purposes: to convey a genes (14, 15), within monogamous parameters in the two populations of po- proposition (e.g., a bribe, a command, an pair-bonds, who are bound by their lice officers. Yet another plausible exten- offer) and to negotiate and maintain a shared children (16), and by close sion to real life is the use of a sequence of relationship. People achieve these dual friends and allies, who are bound by statements escalating in directness (2), ends by using language at two levels. The shared interests (17). It can be ex- thereby probing the reaction of the officer literal form of a sentence is consistent tended to others by nonverbal cues of (‘‘What a beautiful morning. I’m very with the safest relationship between solidarity such as physical contact, sorry for speeding. I know I’ll have to pay speaker and hearer. At the same time, by communal meals, and coordinated for my mistake. I admire officers doing implicating a meaning between the lines, movements and experiences (13). their duty. Can I make a contribution to the speaker counts on the listener to infer Y The reciprocity or equality-matching the policeman’s benevolent association? Is its real intent, which may initiate a differ- relationship obeys the ethos, ‘‘You there some way we could avoid the paper- ent relationship. For example, in a polite scratch my back; I’ll scratch yours.’’ work and settle it here?’’). request, such as ‘‘If you could pass the It has an evolutionary basis in recip- How plausible is the key assumption salt, that would be awesome,’’ the literal rocal altruism (18, 19). It is usually that the decision functions of honest and content violates Grice’s maxims of signaled by tit-for-tat exchanges or dishonest officers differ? The answer de- efficient conversation (5) because it is ir- division into equal portions but, un- pends on the determinants of their cost relevant and untruthful (i.e., an overstate- like the other two relationship types, functions. Take the honest officer: why ment). The hearer implicitly reasons: ‘‘The can be negotiated by people via ex- wouldn’t he arrest anyone who offered a speaker is saying that an outcome of an plicit verbal contracts (13). veiled bribe, exactly as a dishonest officer action by me is good. Therefore, he must would accept all such bribes? The reason be requesting it.’’ The overall effect is that The assumed relationship type among a is that even if all dishonest drivers offer the intended content—an imperative— pair of individuals has dramatic effects on remarks that can be interpreted (cor- gets through, but without the presumption the behavior that is acceptable between rectly) as implicated bribes, some honest of dominance that would ordinarily them; not surprisingly, given that the rela- drivers make those remarks too, as inno- accompany an imperative, with its tacit tionship type governs the acceptable dis- cent observations (this is inherent to the assumption by the speaker that he can tribution of resources in a given social definition of indirectness), so any arrest expect the hearer’s compliance. setting (13, 17). Behavior that is accept- might be unsuccessful. An unsuccessful This reticence raises the question of able in one relationship type can thus be arrest might be costly to the officer, ex- what kind of relationships, other than anomalous in another. For example, at a posing him to a charge of false arrest and dominance, people are mindful of when party, one might help oneself to a shrimp the police department to punitive dam- choosing their words. Alan Fiske (11) has off the plate of one’s spouse or sibling or ages. The cost to the honest officer of advanced the strong claim that human close friend (communality), but not off

Pinker et al. PNAS ͉ January 22, 2008 ͉ vol. 105 ͉ no. 3 ͉ 835 Downloaded by guest on October 1, 2021 cost specific ways of allocating resources avail- that it is interpreted as intended is close able, with the consequence that a given to 1). That is, in many circumstances, arrest type of behavior can vary radically in its both parties know when an overture has honest adaptive value depending on which been made by innuendo. Can any adult ticket scheme is currently in effect. For example, really claim to be fooled by the etchings, helping yourself to a person’s food or the offer to ‘‘settle it here,’’ or the advi- other resources can be a prerogative in sory on the accidents that can befall a bribe corrupt the context of one relationship type but a store? Any ‘‘deniability’’ in these cases case of theft in another. Ordering some- is really not so plausible after all. The one around can be a requirement of your puzzle, then, is why, in cases of low or subtle explicit job in one setting but a case of extortion zero uncertainty about the agents’ val- corrupt honest in another. ues and intentions, an obvious indirect copʼs copʼs The fraught nature of divergent rela- overture should still feel less awkward threshold threshold tionships gives rise to a social identifi- than a direct overture that is sensed to directness of proposition cation problem that is isomorphic to be ‘‘out there’’ or ‘‘on the record.’’ the legal identification problem de- This circumstance is addressed by Fig. 2. The expected cost to a driver of tendering scribed by Schelling (9). In this case, the third part of the theory, which bribes to honest and corrupt police ofﬁcers with the emotional cost of a relationship holds that language is tacitly perceived nonlinear decision rules. mismatch duplicates the payoff matrix to be a digital medium. Discreteness is of other risky propositions. Take the in the very design of human language. ambiguity between the authority rela- Features, morphemes, words, and the plate of one’s boss (dominance). Also, tionship ordinarily commanded by a phrases are concatenated, not blended, a guest at a dinner party (communality) maitre d’ (in which he assigns tables as and each one’s contribution to the would be perceived as crass, not fair, if at he pleases) and the reciprocity rela- meaning of the whole is qualitative: A the end of the meal he pulled out his wal- tionship suggested by a diner offering a sound that is halfway between to bat let and offered to pay his host for the bribe (in which the maitre d’ would be and to pat does not refer to an action meal (reciprocity). bound to offer a table in exchange for that is halfway between batting and When relationships are ambiguous, a the bribe). The payoff matrix is identi- patting. Moreover, in all languages, divergent understanding between the par- cal in structure to the one for bribing a real-world continua of space, time, and ties can lead to the aversive emotion we police officer, but the payoffs are reck- matter are digitized into discrete cate- call ‘‘awkwardness.’’ There are awkward oned by matches or mismatches in the gories such as in versus on, past versus moments in a workplace or university assumed relationship between the two nonpast, and pebbles versus gravel when an underling or student makes a parties rather than in dollars and cents: (4, 20). Propositions with discrete truth transition from a subordinate (dominance) to something closer to a friend (communality). Good friends (communality) are Dishonest maitre d’ Honest maitre d’ advised not to engage in a business trans- action (reciprocity), like the sale of a car Don’t bribe Long wait (dominance/ Long wait (dominance/ or a house, which can endanger the dominance) dominance) friendship. The ambiguity between domi- Bribe Instant seating (reciprocity/ Awkwardness (reciprocity/ nance and sex (a kind of communal rela- reciprocity) dominance) tionship) is the battleground of sexual harassment conflicts, and the ambiguity Implicate bribe Instant seating (reciprocity/ Long wait (dominance/ between friendship and sex gives rise to reciprocity) dominance) the frisson of dating. The ultimate reason that relationship mismatches impose emotional costs (and An implicated bribe (third row), such as values, too, may be conveyed through hence motivate the parties to align their ‘‘I was hoping you might have a cancella- the choice and arrangement of words. behavior with one relationship type or tion,’’ combines the best of the first two This does not imply that in practice another) is the familiar principle from rows: the high payoff of an overt bribe (a linguistic information is transmitted biology that cooperation is evolutionarily quick table) with the low penalty of not with perfect fidelity (the phenomenon fragile (because it is vulnerable to cheat- bribing at all (a long wait). of indirect speech is an example to the ers), and that each variety of cooperation contrary), but the intuition that lan- depends on a particular set of circum- Part 3: Language as a Digital Medium guage is a reliable medium is wide- stances being in place for the cooperation There is one remaining problem, which spread as a folk theory of language to be adaptive (18, 19). Depending on arises in cases where speakers use indi- (21), and belief in it may affect how their ecological niche and evolutionary rect speech despite a lack of genuine people choose and interpret their antecedents, different organisms may co- uncertainty on the part of one or both words. This hypothesis has (at least) operate via nepotism, mutualism, or reci- of the conversational partners. One such three corollaries. procity, or they may not cooperate at all, case is when the Identification Problem First, overt propositions are perceived ceding resources via contests of domi- does not arise because the values of the as certain, as opposed to merely highly nance. Humans avail themselves of all of listener are known—for example, if all likely. The relevance to relationship nego- these options, facultatively choosing of the police officers or maitre d’s in a tiation is that the signals of many relation- among them on the basis of their current given city are known to be bribable. An- ship types may be analogue and highly social context rather than on a fixed, phy- other occurs when an indirect proposi- ambiguous. How close does a man have to logenetically typical one. The anxiety sur- tion is so conventional or transparent as sit to a woman, how lavishly can he com- rounding relationship mismatches is the to leave no doubt in the hearer’s mind pliment her looks, and how secluded are price we pay for having multiple, context- as to what was intended (the probability the locales he leads her to before she con-

836 ͉ www.pnas.org͞cgi͞doi͞10.1073͞pnas.0707192105 Pinker et al. Downloaded by guest on October 1, 2021 cludes that his intentions are sexual, and the authority of a dominant figure; yoke common knowledge, they cannot maintain must be accepted or rebuffed? When the interests of couples, families, and the fiction of a friendship, and they would faced with these ambiguities, people may other communal groups; or disseminate have the sense that ‘‘it’s out there,’’ and treat certainty as a ‘‘focal point’’ in Schell- information about the probity of recipro- that ‘‘he can’t take it back.’’ ing’s sense (9). A focal point is a salient cating partners (24). The ultimate reason that common location that two rational agents can agree The third corollary invokes a concept knowledge would mandate a relationship on when they would be better off coordi- that linguists, logicians, and economists change is that the alternative—maintain- nating their behavior than acting indepen- have called common knowledge, mutual ing the relationship in the presence of dently, and when there is no rational basis knowledge, and common ground (2, 9, common knowledge that contradicts it— for choosing one value over another. Ex- 25–30). In common knowledge, not only would entail the listener relinquishing her amples include a prominent landmark in a does A know x and B know x, but A claim to rationality, resolve, or honesty. city where separated spouses agree to knows that B knows x, and B knows that She would be admitting that, despite pos- meet, or a round number that a buyer and A knows x, and A knows that B knows sessing knowledge that is incompatible seller in a negotiation can settle on. Al- that A knows x, ad infinitum. As with with the presumptions of pure friendship, though the preference for a focal point other phenomena in linguistics in which a she is willing to pay the costs or exploit (when one exists) is an inherently rational person is said to ‘‘know’’ an infinite num- the perquisites that go with it (e.g., ac- strategy that does not depend on details ber of things, the knowledge is not enu- cepting or offering favors without expecta- of the agents’ psychology, the question of merated as an infinite list, of course, but tion of reciprocation). Similar tacit confes- whether a focal point exists, and where in is implicit in a finite recursive formula. In sions would be entailed by other failures the problem space it lies, depends on de- this case, it could be the formula y: to change a relationship following a com- tails of the local environment and on ‘‘Everyone knows x, and everyone knows monly known direct overture, as in bribes common properties of the agents’ percep- y’’ (2). Moreover, common knowledge can and threats. tual and cognitive systems that predictably be ascertained perceptually, by observing This three-part theory (plausible deni- single out the same point as psychologi- that x is perceptible or broadcasted in ability, relationship negotiation, and cally salient. Daniel Kahneman and Amos public circumstances. language as a digital medium) makes nu- Tversky, in their work on Prospect Theory The paradigm illustration of common merous empirical predictions about how (22), have shown in studies of people’s knowledge is the story of the Emperor’s people interpret the intent and perceive tolerance for risk that human psychology New Clothes. When the boy called out, the relationship between a speaker and a categorically distinguishes 100% certainty ‘‘The emperor is naked!’’ he was not tell- hearer depending on the level of indirect- from all other probability values. We sug- ing the onlookers anything they didn’t ness in the speaker’s words. The plausible- gest, then, that overt propositions are per- already know. Yet he was conveying deniability hypothesis predicts that the ceived as certain and act as focal points, knowledge nonetheless: Now everyone directness of speakers’ wording of a veiled whereas implicatures from indirect speech knew that everyone else knew, and that bribe or other overture (assessed on lin- are perceived as being some measure everyone else knew that they knew, and guistic grounds) is not an arbitrary social short of certainty (even 99%). With the so on, and that common knowledge li- ritual, like saying ‘‘Please’’ and ‘‘Thank lack of a focal point to trigger a change of censed the people to challenge the domi- you,’’ but is predictable from strategic fac- relationship, the speaker is given ‘‘the nance relationship commanded by the tors affecting its expected utility, such as benefit of the doubt,’’ and the relationship emperor. The moral for the present the- the proportion of honest and dishonest can remain unchanged. The deniability, ory is that language is an efficient way of officers in an area, the cost of a bribe, the then, doesn’t have to be plausible, only generating common knowledge. cost of a ticket, and the cost of a bribery possible. This corollary of the language-as- charge. For the listener’s part, the direct- A second corollary of language as a digital-medium hypothesis, then, is that ness of a speech act should predict their digital medium invokes Erving Goffman’s indirect speech merely provides shared subjective estimates of the likelihood that hypothesis (23) that people always behave individual knowledge, whereas direct the speaker intended the fraught proposi- as if they are playing to an audience. A speech provides common knowledge, and tion as opposed to making an innocent crucial feature of indirect speech is that it relationships are maintained or nullified remark. The relationship-negotiation hy- can be interpreted only in context. The by common knowledge of the relationship pothesis predicts that indirect speech hearer’s interpretation taps into many as- type. Imagine that Harry says, ‘‘Would should be judged as generating less awk- pects of the context, such as the lead-up you like to come up and see my etch- wardness and discomfort, as being more to the speech act, the speaker’s body lan- ings?’’ and Sally demurs. There is little or respectful, as better acknowledging the guage, or his tone of voice. Overt proposi- no uncertainty about Harry’s intent, and expected relationship with the hearer tions, in contrast, may be perceived as none about Sally’s: Sally knows that she (such as affection, deference, or collegial- context-free—their intent can be inter- has turned down an overture, and Harry ity), and as making it easier for the partic- preted by eavesdroppers—and lossless— knows that she has turned down an over- ipants to resume their normal relationship their intent can be transmitted perfectly ture. However, Sally doesn’t necessarily should the offer be rebuffed. The digital- along chains of gossipers (in the same way know that Harry knows; she might think medium hypothesis predicts that overt that other digital media, such as files of to herself, ‘‘Maybe Harry thinks I’m na- speech should be judged as certain in in- music or images, can be transmitted loss- ¨ve.’’ı In addition, Harry doesn’t necessarily tent—that listeners and overhearers lessly). According to this hypothesis, the know that Sally knows that he knows; he should assess the probability of a direct deniability is plausible to the virtual audi- might think to himself, ‘‘Maybe Sally proposition as 1.0, with negligible variance ence, even if it is not particularly plausible thinks I’m obtuse.’’ Although there is indi- among people, whereas even the most to the hearer, and people tacitly consult vidual knowledge, there is no common direct innuendo should be perceived as the reactions of a virtual audience in re- knowledge, and they can maintain the admitting of nonzero uncertainty, and taining or switching relationship types. fiction of friendship. In contrast, if Harry with substantial variance. Finally, overt The ultimate reason is that relationship were to have said, ‘‘Would you like to speech, compared with indirect speech, types are not solely dyadic but depend in come up and have sex?’’ then Harry should be perceived as lossless in chains part on the tacit ratification of a commu- knows that Sally knows that Harry knows of gossip (the last link in a chain should nity, which may be necessary to back up that Sally knows, and so on. With this be as confident in the speaker’s intent as

Pinker et al. PNAS ͉ January 22, 2008 ͉ vol. 105 ͉ no. 3 ͉ 837 Downloaded by guest on October 1, 2021 the first) and in recursive embeddings of If these analyses are correct, the phenom- use language to do so. In exploring the knowledge (the speaker and hearer should enon reveals a number of important facts boundaries of relationship types, humans be as confident in their interpretation of about social behavior in Homo sapiens. anticipate what other humans think about the other’s interpretation of their interpre- Humans employ several, mutually incom- the relationship: what the other party in tation as they are in the simple interpreta- patible, modes of cooperation and, as a the relationship thinks; what overhearers tion itself). result, are extremely touchy about their and gossipers think; and what the other The theory leads to further hypotheses relationships. With some of their fellows party thinks about what they think about what the other party thinks about what about even higher-order psychological (typically kin, lovers, and friends), they they think, and so on. The need to pre- processes that might be triggered by peo- freely share and do favors; with others, serve their relationships while transacting ple’s experience in tendering and inter- they jockey for dominance; with still oth- the business of their lives can thus explain preting indirect speech. For example, ers, they trade goods and favors. People humans’ tendency to fill their social life hearers may credit a speaker with consid- distinguish these relationships sharply, and with innuendo, hypocrisy, and taboo. eration for their dignity and feelings, or when one person breaches the logic of a with greater social intelligence, if he uses relationship with another, they both suffer ACKNOWLEDGMENTS. We thank Peter Ga¨rden- indirect speech in a skillful way. an emotional cost. Nonetheless, humans fors and two anonymous reviewers for comments on an earlier draft. This work was supported by Most generally, indirect speech is a often have to risk these breaches to get on National Institutes of Health Grant R01 HD-18381 unique window into human social life (4). with the business of life, and they often (to S.P.).

1. Brown P, Levinson SC (1987) Politeness: Some Univer- 12. Maynard Smith J (1982) Evolution and the Theory of 22. Kahneman D, Tversky A (1979) Prospect theory: An sals in Language Usage (Cambridge Univ Press, New Games (Cambridge Univ Press, New York). analysis of decisions under risk. Econometrica 47: York). 13. Fiske AP (2004) Four modes of constituting relation- 313–327. 2. Clark HH (1996) Using Language (Cambridge Univ ships: Consubstantial assimilation; space, magni- 23. Goffman E (1959) The Presentation of Self in Everyday Press, New York). tude, time, and force; concrete procedures; abstract Life (Doubleday, New York). 3. Holtgraves TM (2002) Language as Social Action (Erl- symbolism. Relational Models Theory: A Contempo- 24. Nowak MA, Sigmund K (1998) Evolution of indirect baum Associates, Mahwah, NJ). rary Overview, ed Haslam N (Erlbaum Associates, reciprocity by image scoring. Nature 393:573–577. 4. Pinker S (2007) The Stuff of Thought: Language as a Mahwah, NJ). 25. Vanderschraaf P, Sillari G (Winter 2005) Entry for ‘‘com- Window into Human Nature (Viking, New York). 14. Hamilton WD (1964) The genetical evolution of social mon knowledge.’’ The Stanford Encyclopedia of Phi- 5. Grice HP (1975) Logic and conversation. Syntax & Se- behaviour. I. J Theor Biol 7:1–16. losophy, ed Zalta EN (Metaphysics Research Lab, Center mantics, eds Cole P, Morgan JL (Academic, New York), 15. Hamilton WD (1964) The genetical evolution of social for the Study of Language and Information, Stanford Vol 3. behaviour. II. J Theor Biol 7:17–52. University, Stanford), http://plato.stanford.edu/ar- 6. Williams GC (1966) Adaptation and Natural Selection: 16. Alexander RD (1987) The Biology of Moral Systems chives/win2005/entries/common-knowledge/. A Critique of Some Current Evolutionary Thought (Aldine de Gruyter, Hawthorne, NY). 26. Schiffer SR (1972) Meaning (Oxford Univ Press, New (Princeton Univ Press, Princeton). 17. Tooby J, Cosmides L (1996) Friendship and the Banker’s York). 7. Trivers R (1985) Social Evolution (Benjamin/Cummings, Paradox: Other pathways to the evolution of adapta- 27. Lewis DK (1969) Convention: A Philosophical Study Reading, MA). tions for altruism. Proc Brit Acad 88:119–143. (Harvard Univ Press, Cambridge, MA). 8. Dawkins R, Krebs JR (1978) Animal signals: Information or manipulation? Behavioral Ecology, eds Krebs JR, 18. Trivers R (1971) The evolution of reciprocal altruism. Q 28. Smith NV, ed (1982) Mutual Knowledge (Academic, Davies NB (Blackwell, Oxford), pp 282–309. Rev Biol 46:35–57. New York). 9. Schelling TC (1960) The Strategy of Conflict (Harvard 19. Nowak MA (2006) Five rules for the evolution of coop- 29. Clark HH, Marshall CR (1991) Definite reference and Univ Press, Cambridge, MA). eration. Science 314:1560–1563. mutual knowledge. Elements of Discourse Under- 10. Feiler B (2000) Pocketful of dough. Gourmet Octo- 20. Pinker S (1994) The Language Instinct (HarperCollins, standing, eds Joshi AK, Webber BL, Sag IA (Cambridge ber:99–101. New York). Univ Press, New York), pp 10–63. 11. Fiske AP (1992) The four elementary forms of sociality: 21. Reddy M (1993) The conduit metaphor: A case of frame 30. Chwe MS-Y (2001) Rational Ritual: Culture, Coordina- Framework for a unified theory of social relations. conflict in our language about language. Metaphor and tion, and Common Knowledge (Princeton Univ Press, Psychol Rev 99:689–723. Thought, ed Ortony A (Cambridge Univ Press, New York). Princeton).

838 ͉ www.pnas.org͞cgi͞doi͞10.1073͞pnas.0707192105 Pinker et al. Downloaded by guest on October 1, 2021