Speakers’ Control Over Leaking Private Information During Language Production

Speakers’ descriptions sometimes inappropriately refer to information known only to them, thereby ‘‘leaking’’ knowledge of that private information. We evaluated whether speakers can explicitly control such leakage in light of its communicative consequences. Speakers described mutually known objects (e.g., a triangle) that had size-contrasting matches that were privileged to the speakers (e.g., a larger triangle visible to the speakers only), so that use of a contrasting adjective (e.g., small) involved referring to the privileged information. Half the time, speakers were instructed to conceal the identity of the privileged object. If speakers can control their leaked references to privileged information, this conceal instruction should make such references less likely. Surprisingly, the conceal instruction caused speakers to refer to privileged objects more than they did in the baseline condition. Thus, not only do speakers have difficulty not leaking privileged information, but attempts to avoid such leakage only make it more likely. Though people nearly always bring their own perspectives to any given situation, sometimes they behave as though they fail to appreciate that fact. This egocentrism has been explored experimentally in tasks like that illustrated in Figure 1. In this example, four objects are positioned between two people. One person can see three of them: a triangle, circle, and heart. The other person can additionally see a fourth object—a larger triangle. If the second person is asked to identify the mutually visible triangle so that the first person can pick it out, he or she ought to say ‘‘triangle,’’ just as ‘‘circle’’ would describe the sole circle. Yet sometimes speakers in this circumstance say ‘‘small triangle’’ instead (Horton & Keysar, 1996; Nadig & Sedivy, 2002; Wardlow & Ferreira, 2003), as if they fail to appreciate their unique perspectives. Why might speakers produce such seemingly erroneous utterances, like ‘‘small triangle,’’ in these situations? One possibility is that low-level factors might compel speakers to pay more or less attention to the shape that only they can see (hereafter, the hidden shape). (For accounts of how factors like these might operate, see Horton & Keysar, 1996, and Nadig & Sedivy, 2002.) For example, too much attention to the hidden shape may boost its salience, overwhelming the knowledge that it is hidden, and leading speakers to refer to it when labeling the to-be-described (target) shape. To the extent that such low-level factors are influential, utterances like ‘‘small triangle’’ are like ‘‘Simon says’’ errors; undue attention to the hidden shape compels speakers to refer to it, even though they should not and may not intend to. Another factor that might affect the likelihood that speakers will disregard their knowledge of perspective differences is knowledge of the high-level communicative consequences of producing such errors (Clark, 1996; Schober & Brennan, 2003): When a speaker says ‘‘small triangle’’ instead of ‘‘triangle,’’ he or she not only has communicated which shape the addressee ought to pick out, but also has potentially leaked implicit information. In particular, the addressee can infer that the speaker can probably see another triangle, which is likely to be the hidden shape. In most situations, leaked information is unlikely to harm speakers’ communicative goals (Nadig & Sedivy, 2002); if speakers aim to convey which triangle addressees should select, ‘‘small triangle’’ works about as well as ‘‘triangle’’ (addressees can see only one triangle), and the leaked information is largely irrelevant. Indeed, by communicating more information with fewer words, use of utterances like ‘‘small triangle’’ might be generally adaptive. Address correspondence to Liane Wardlow Lane, Department of Psychology 0109, University of California, San Diego, La Jolla, CA 92093-0109, e-mail: [email protected]. PSYCHOLOGICAL SCIENCE Volume 17—Number 4 273 Copyright r 2006 Association for Psychological Science But what happens when leaked information conflicts with speakers’ goals? Assume that in the situation illustrated in Figure 1, speakers are instructed not only to name the target shape, but also to conceal the hidden shape. In this case, speakers should avoid describing the target as ‘‘small triangle,’’ because the leaked information might cue addressees to the identity of the hidden shape. Can speakers’ high-level communicative intentions (to name the target and conceal the hidden shape) overcome their basic tendency to sometimes violate their knowledge of perspective differences? Or are the low-level factors (e.g., salience) that compel speakers to produce utterances like ‘‘small triangle’’ not under speakers’ intentional control? Ironic-processes theory (Wegner, 1994) suggests another possibility: Speakers may be more, rather than less, likely to refer to a hidden object precisely because of an intention to conceal it. Ironic-processes theory is a dual-process account of performance according to which an operator process attempts to perform a desired action, while a monitor process checks for signs of failure. Critically, monitoring can bring failure conditions into awareness, thereby ironically causing ‘‘precisely counterintentional’’ (Wegner, 1994) behaviors, especially when task conditions are challenging. For example, subjects attempting to hold a pendulum steady while counting backward by 3s will swing it along a particular axis more when that axis is forbidden than when it is not (Wegner, Ansfield, & Pilloff, 1998). Analogously, instructions to conceal the hidden object in situations like the one illustrated in Figure 1 could engage an ironic-process monitor, making counterintentional behaviors (e.g., saying ‘‘small triangle’’) more likely. The present experiment used a referential communication task (Hanna, Tanenhaus, & Trueswell, 2003; Keysar, Barr, Balin, & Brauner, 2000) like that illustrated in Figure 1. Speakers described to addressees mutually visible shapes on target cards while trying to ignore hidden shapes on foil cards. On critical trials, the object on the target card was medium-sized (see Fig. 2). On half the critical trials (test trials), foils and targets were the same shape, but contrasted in size. Thus, test trials were designed to elicit utterances that included modifiers that contrasted the target with the hidden shape. On the other half of critical trials (control trials), the foil was a different shape from the target. Control trials thus assessed how often utterances included modifiers irrespective of the contrast to the hidden shape. Speakers were tested in two blocks that were presented in counterbalanced order. In baseline blocks, speakers were instructed to identify each target so that addressees could select it from the mutually visible set. Separate scores were kept for speakers and addressees, each receiving 1 point whenever addressees selected the target. In conceal blocks, participants were given additional instructions encouraging speakers to hide the foil’s identity when identifying the target. Specifically, after addressees selected a target, they were allowed to guess the identity of the foil. A point was added to addressees’ scores for each correct guess, and a point was subtracted for each incorrect guess. Speakers were instructed not to allow addressees to gain Fig. 1. Example of the experimental setup. Fig. 2. Examples of the stimuli used on test (top) and control (bottom) trials. 274 Volume 17—Number 4 Don’t Talk About Pink Elephants!