In Mind and Language, vol 17, No 5, November 2002, pp 457-466

Jean-Baptiste Van der Henst,

Laure Carles, and Dan Sperber

Abstract. Someone asked "What time is it?" when her watch reads (for instance) 3:08 is likely to answer "It is 3:10." We argue that a fundamental factor that explains such rounding is a psychological disposition to give an answer that, while not necessarily strictly truthful or accurate, is an optimally relevant one (in the sense of relevance theory) i.e. an answer from which hearers can derive the consequences they care about with minimal effort. A rounded answer is easier to process and may carry the same consequences as one that is accurate to the minute. Hence rounding is often a way of optimising relevance. Three simple, near-"natural" experiments, which involve approaching people in public places and asking the time, give support and greater precision to the view that relevance is more important than strict truthfulness in verbal communication.

Someone approached in the street and asked "What time is it?" at a point when her watch reads (for instance) 3:08 is likely to answer "It is 3:10." Though not a lie, such rounding is a departure from strict truthfulness. The fact that it is extremely common seems to argue against the standard philosophical view that verbal communication is primarily governed by a rule, "maxim" (Grice, 1975) or "convention" (Lewis, 1983) of truthfulness from which it follows that speakers are expected to speak the truth. What explains rounding? What criterion are helpful speakers trying to satisfy in telling the time? Strict truthfulness, i.e. accuracy to the best of their knowledge? Truthfulness to some situationally acceptable degree? Or relevance? We want to argue that helpful speakers are always trying to achieve relevance to their hearer, and aim at accuracy and truthfulness in what they say only to the extent that these are necessary for relevance, which is by no means always the case.

The notion of relevance we use, and more generally the theoretical approach for which we provide experimental confirmation, is that of relevance theory. According to relevance theory, the greater the true consequences derivable from an utterance, and the smaller the effort required to derive them, the greater the relevance of the utterance for the hearer. A speaker, by the very fact of addressing a hearer, presents her utterance as being "optimally relevant", that is, as relevant to the hearer as is compatible with her abilities and preferences, and, in any case, relevant enough to be worth the hearer's processing effort. Most of relevance-theoretic pragmatics consists in showing in detail how this "Communicative Principle of Relevance" governs verbal communication and grounds the interpretation of utterances in context.¹

In asking a question, people indicate what information would be relevant to them. In answering a question, helpful speakers try to provide that information without causing the hearer any unmotivated processing effort. Genuine information is, of course, true information, and so helpful speakers aim at providing information that is true. However, this is not the same thing as saying what is true (in the usual sense of "saying" where what speaker is saying coincides - issues of reference aside - with the meaning of the sentence she utters). The information provided by an utterance must be sharply distinguished from the linguistic meaning of the sentence uttered. This linguistic meaning is an input to an inferential process of comprehension. The information provided by the utterance is not so much in the input as in the output of this inferential process. Thus, it is possible, and indeed common, to satisfy a hearer's expectation of relevance while saying something one knows to be inaccurate or even plainly false, provided that the consequences that the hearer derives from the utterance are true and sufficiently important to be worth deriving. For an utterance to achieve relevance requires truth of the consequences derived, but not of what is said.

Hyperboles such as "I will be ready in a second," or "Everybody likes chocolate" illustrate the fact that the goal of achieving greater relevance often justifies the utterance of strictly false approximations. The fact that these utterances are strictly false does not make them misleading and thus does not compromise their relevance. Hearers are likely to ignore the false consequences derivable from these utterances (e.g. that the speaker will be ready in less than two seconds, or that there does not exist a single individual who dislikes chocolate). They are likely, rather, to derive from such utterances only true consequences which are salient in the context (e.g. that the speaker will be ready well in time, or that a gift of chocolate is appropriate even if you know nothing of the receiver's tastes). Moreover, processing these utterances may require less effort than would their strictly true counterparts ("I will be ready in an extremely short time," "Almost everybody likes chocolate"). Since the lower the effort, the greater the relevance, these hyperboles may well be more relevant than their strictly true counterparts.

What is the case with patently false hyperboles is even more obviously so with utterances that are merely inaccurate: strictly true consequences are easily derived from inaccurate utterances. Someone in the street who asks a stranger for the time and is told "It is 3:10" when in fact, to the best of the speaker's knowledge, it is 3:08, is likely to derive only true consequences from this inaccurate answer: for instance, that he is in plenty of time for his 3:30 appointment. Most people's activities are commonly scheduled (if they are scheduled at all) to start on the hour or half hour, rarely at more specific times, hardly ever at times that, expressed in minutes, are not multiples of five. This is true, for instance, of almost all appointments, TV programs, or university courses. Only train schedules are regularly specific to the one-minute interval. In the absence of evidence to the contrary, then, it is likely that someone who asks a stranger in the street for the time, with no indication of his particular purpose in asking the question, will not be worse off with an answer rounded to a multiple of five. According to relevance theory, moreover, it is likely that such a person will actually be better off with a rounded answer, which requires less processing effort for the same cognitive benefit.

Even philosophers such as Grice or Lewis, who argue for some rule of truthfulness, do not claim that this rule should always be strictly adhered to. There may be degrees of truthfulness, and some situations may tolerate lower degrees. Being asked for the time by a stranger may be a case in point. Speakers may be helpful, but they are also understandably unwilling to make too much effort. You cannot, for instance, expect a stranger to tell you the time to the minute if it is easier for her to round it to the nearest multiple of five. Similarly, from a relevance theory point of view, speakers are not expected to produce a maximally relevant utterance, but only an optimally relevant one, where optimality is bounded by the abilities and preferences of the speaker. A speaker aiming at optimal relevance may prefer a formulation that requires less effort from herself and more from the hearer, provided that the resulting utterance is still relevant enough to be worth the hearer's attention. Even though both the truthfulness and the relevance approaches recognise the imperfections of speakers, they yield different predictions about verbal behaviour in general - and testably different predictions about the particular case of the behaviour of strangers asked for the time. If truthfulness is the paramount criterion, rounding is at best a tolerable weakness. If relevance is the paramount criterion, rounding may be the best thing for speakers to do. Here is a case, then, where experimental data may bear on a standard issue in the philosophy of language.

The goal of the first two experiments we report was to ascertain people's tendency to round, and to show that two trivial factors, laziness and prudence, which are likely to contribute to this tendency, are not sufficient to explain it. The disposition to produce an optimally relevant answer is an important and independent explanatory factor. The subtlety of this disposition and the fact that it may involve attending to unexpressed concerns of the audience is demonstrated by the third experiment.

In speaking, as in any other action, people are trying to minimise their own effort. For those with analog watches, where only multiples of five minutes are marked (if anything), it is generally easier to round their answer. Thus, in rounding, such speakers may be minimising their own effort (i.e. be guided by laziness), rather than trying to minimise the effort of their audience (i.e. be guided by considerations of relevance to their audience). The frequency of rounding might just reflect the prevalence of analog watches in the population. This can easily be tested by comparing people who wear analog and digital watches. For speakers with digital watches, just reading aloud the unrounded time shown on their watch involves fewer steps, and should be easier than silently reading this unrounded time, mentally rounding it, and telling the rounded time. If speakers rounded merely in order to minimise their own effort, then people with analog watches would round and people with digital watches would not. If, as we claim, speakers are disposed to try and minimise their audience's effort, then some people with digital watches might make the effort of rounding for the sake of their audience. These contrasting predictions are easily testable.

Experiment 1 consisted in approaching people in a public place at random times and asking them (in French): "Hello! Do you have the time please?" ("Bonjour! Est-ce que vous avez l'heure s'il vous plaît?"). We distinguished two groups: an analog group of participants with analog watches and a digital group of participants with digital watches. The experimenter took note of the type of watch worn, and the answer given. In order to test a population where digital and analog watches were equally popular, the experiment took place at the entrance to the scientific University of Jussieu in Paris, and only male science students were approached (in France, where the experiments were performed, digital watches are not very popular and male science students presented a welcome exception in this respect). There were fifty-two people in the analog group and forty in the digital group.

Responses (in this and the two other experiments) were sorted into two types: 5x responses, in which the number of minutes is a multiple of five (these include all rounded answers and also answers given accurately when the watch happened to indicate a multiple of five), and not-5x responses, in which the number of minutes is not a multiple of five (these include only non-rounded answers).² If there had been no rounded answers, then there should have been 20% of type 5x, and 80% of type not-5x responses (since out of a random sample of times expressed to the minute, one fifth will end in a multiple of five). However, this theoretical distribution was not found, or even approximated, in any of the three experiments we performed. The percentage of type 5x responses over and above 20% indicates the importance of rounding. More precisely, we calculate the "percentage of rounders" by means of the formula:

Percentage of rounders = (percentage of 5x responses - 20)/80

According to this formula, zero percent of rounders corresponds to 20% of type 5x responses, and a hundred percent of rounders to 100% of type 5x responses.

Figure 1 presents the percentages of rounders in the three experiments. With analog watches, the high percentage of rounders (97%) could be explained both by a tendency of speakers to minimise their own effort and by a tendency to minimise audience's effort. With digital watches, the percentage of rounders was significantly lower (57%) (c 2 = 16.43, p< .0001).³ People round less with digital watches than with analog watches, which is unsurprising given that it is easier for them not to round. What is surprising is that so many of them do round at all. This result falsifies the hypothesis that rounding is caused just by the concern of speakers to spare their own effort and that therefore people with digital watches should not round. It is well explained by the hypothesis that speakers spontaneously tend to minimise their audience's effort, even at the cost of an extra effort to themselves. Moreover, the fact that more than half of speakers with digital watches are disposed to make an extra effort in order to round their answer, thereby purposefully departing from accuracy, shows, contrary to standard philosophical views, that relevance is more important than strict truthfulness in ordinary verbal communication.

 Experiment 2: Do speakers round in order to limit their commitment?

Some people may round not out of laziness but out of prudence. They may be unsure that their watch is accurate to the minute, and may limit their commitment by using a time-scale expressed in five-minute intervals. In other words, they may, in fact, be as strictly truthful as their knowledge allows. If this prudence factor, together with a common tendency to minimise one's own effort, were enough to explain why people round, then changing the context of the request should not affect the percentage of rounders. If, on the other hand, speakers are guided by considerations of relevance to the hearer, then framing the request so as to suggest that an answer which is accurate to the minute would be more relevant than a rounded one should lower the percentage of rounders. This prediction was tested in Experiment 2 where, in the experimental group, the experimenter indicated to participants that he was asking for the time in order to set his own watch correctly. He ostentatiously held a wristwatch in his hand and said: "Hello! My watch is going wrong. Do you have the time please?". In the control group, the experimenter just asked: "Do you have the time please?" Ninety-one people, all with analog watches, were asked for the time in the shopping centre of Euralille (Lille, France). The experimental group was composed of forty-five, and the control group of forty-six participants.⁴

Someone setting his or her watch right is likely to find an answer that is accurate to the minute more relevant than a rounded one. As predicted, there were fewer rounders in the experimental than in control group (see Figure 1, c 2 = 16.11, p< .0001). When a time that was accurate to the minute was clearly relevant to their audience, 51% of the people in the population tested were confident enough of their watch and helpful enough to give it. The fact that, when merely asked for the time, 94% of people rounded is therefore not an effect just of prudence, or of laziness, or of prudence and laziness combined. It is, at least in part, an effect of people's disposition to attend to what may be relevant to their hearer.

In Experiment 2, participants were given explicit evidence of the fact that a time that was accurate to the minute would be useful. The greater accuracy of their responses might be explained as a conscious and deliberate attempt to help the experimenter set his watch right, rather than by a more general spontaneous disposition to be optimally relevant to one's audience. Our third experiment was devised to elicit responses attuned to much subtler considerations of relevance, of a kind that can be taken into account only by speakers who attend to the unexpressed mental states of their audience.

The main variable of interest in Experiment 3 was not the question asked (since all the participants were asked exactly the same thing) but the interval between the time at which the question was asked and the time of an appointment that the experimenter indicated he or she had. Our prediction was that the shorter the interval between the time of the question and the time of the appointment, the more accuracy would be seen as potentially contributing to relevance, yielding more accurate answers. There are two related reasons for this prediction, one mathematical, the other practical. It can be assumed that the person who is asking for the time will use the answer to compute the interval of time left before the appointment. The shorter this interval, the greater the proportional difference each minute makes to it. The greater this difference, the more likely it is to have practical consequences. Suppose you have an appointment at 3:30 p.m. and it is 3:08. Being told "It is 3:10" is likely to be optimally relevant: the two-minute departure from the exact time is unlikely to have any consequences, and the rounded answer is easier to process. On the other hand, if the time is 3:28, accuracy is more likely to contribute to relevance than in the previous case. Being told the accurate "It is 3:28" rather than the rounded "It is 3:30" may make the difference between relaxing for another couple of minutes and beginning to worry.

In Experiment 3, a hundred-and-forty-two people with analog watches, walking in the Luxembourg Gardens in Paris, were approached by the experimenter who said:: "Hello! Do you have the time please? I have an appointment at t".⁵ We distinguished two groups of participants: an earlier group of those who answered with a time between 30 and 16 minutes before the time of the appointment, and a later group of those who answered with a time between 14 minutes before the appointment and the time of appointment itself. As shown in Figure 1, the percentage of rounders went down from 97% in the earlier group to 75% in the later group (c 2 = 6.64, p = .01).⁶ We know from Experiment 2 that, when the need for an accurate answer is explicit, about half of people will provide it. What Experiment 3 suggests is that nearly half of such helpful people will go even further. Even when what may be relevant to their audience is not at all explicit, they will nevertheless try to infer it from indirect evidence, and will adjust the accuracy of their answer accordingly. In order to adjust their response in such a manner, these particularly helpful speakers must, in the few seconds it takes to answer a request for the time, make an extra inferential effort.

To determine what is relevant to someone, it is necessary to attend to his or her states of mind. There is an extensive psychological literature on "theory of mind" or "mindreading" abilities, exploring how humans are capable of attending to one another's states of mind (e.g. Baron-Cohen, Tager-Flusberg & Cohen, 2000; Carruthers & Smith 1996; Davis & Stone, 1995). There is also a rich literature on perspective taking in communication which has shown that speakers take into account the point of view of hearers in a way that facilitates comprehension (see Krauss & Fussell 1996 for a recent review). The present study shows that such a mindreading ability and attention to the point of view of the hearer is at work even in the simplest forms of everyday communication between strangers. Speakers tend to make the effort of inferring what information may be relevant, i.e. be both consequential and easy to process, for the hearer. In so doing, they go beyond facilitating comprehension, and attend to the interests that make comprehension desirable to the hearer in the first place. Helpful speakers aim at relevance rather than accuracy (and hence strict truthfulness) in what they say. They spontaneously adjust the level of accuracy of their utterances - up or down as the context requires - so as to optimise relevance.

1 - For a general presentation of relevance theory, see Sperber & Wilson 1995; for a detailed discussion of the respective roles of truthfulness and relevance, see Wilson & Sperber, submitted; for some experimental applications see Jorgensen, Miller & Sperber, 1984; Happé, 1993; Sperber, Cara & Girotto, 1995; Politzer, 1996; Gibbs & Moise, 1997; Hardman 1998; Nicolle & Clark 1999; Van der Henst, 1999; for an extensive bibliography, see http://www.ua.es/dfing/rt.htm. (back to text)

2 - Five hard-to-assign responses were ignored. (back to text)

3 - In the three experiments, the chi-square was calculated on the basis of the frequencies of 5x and not-5x responses. (back to text)

4 - Five hard-to-assign responses were ignored. (back to text)

5 - This was a double blind experiment: the two experimenters, one female, one male, were not told the purpose of the experiment or our predictions and were instructed to make their request in the same manner and tone throughout. They each worked in three intervals of half an hour starting on the hour or the half hour, e.g. from 2:00 to 2:30 p.m., and mentioned as the time t of their appointment the end of the interval, e.g. 2:30. (back to text)

6 - We eliminated twenty-six people who gave as a response a time 15 minutes before the appointment, so as to have two groups who could each round to just three multiples of five (30, 25, and 20 minutes before the appointment for the earlier group, 10, 5, and 0 minutes before the appointment for the later group). We also ignored seventeen hard-to-assign responses. (back to text)

Baron-Cohen, S., Tager-Flusberg, H. & Cohen D.J. (2000). Understanding other minds: Perspective from developmental cognitive neuroscience, Second edition. New York: Oxford University Press.

Carruthers, P. & Smith, P. (Eds.) (1996). Theories of Theories of Mind. Cambridge: Cambridge University Press.

Davies, M. & Stone T. (Eds.) (1995). Folk Psychology: The Theory of Mind Debate. Oxford: Blackwell.

Gibbs, R. & Moise, J. Pragmatics in understanding what is said. Cognition 62, 51-74 (1997).

Grice, H.P. (1989). Studies in the way of words. Cambridge, Mass.: Harvard University Press.

Happé, F.G.E. (1993). Communicative competence and theory of mind in autism: a test of relevance theory. Cognition, 48, 101-119.

Hardman, D. (1998). Does reasoning occur on the selection task? A comparison of relevance-based theories. Thinking and Reasoning 4 (4), 353-376.

Jorgensen, J., Miller, G.A. & Sperber, D. (1984). Test of the mention theory of irony. Journal of Experimental Psychology: General, 113, 112-120.

Krauss, R. M., & Fussell, S. R. (1996). Social psychological approaches to the study of communication. In E. T. Higgins & A. Kruglanski (Eds.), Social psychology: Handbook of basic principles. New York: Guilford Press.

Lewis, D. (1983). Philosophical papers, vol I. Oxford: Oxford University Press.

Nicolle, S. & Clark, B. (1999). Experimental pragmatics and what is said: A response to Gibbs and Moise. Cognition 66, 337-354.

Politzer, G. (1996). A pragmatic account of a presuppositional effect. Journal of PsycholinguisticResearch, 25, 543-551.

Sperber, D. & Wilson, D. (1995) Relevance: Communication and cognition, Second edition. Oxford: Blackwell.

Sperber, D., Cara, F. & Girotto, V. (1995). Relevance theory explains the selection task. Cognition, 57, 31-95.

Van der Henst, J.B. (1999). The mental model theory and spatial reasoning re-examined: the role of relevance in premise order. British Journal of Psychology, 90, 73-84.

Wilson, D. & Sperber, D. (submitted) Truthfulness and relevance.