THE QUARTERLY JOURNAL OF EXPERIMENTAL PSYCHOLOGY 2006, 59 (3), 597–624

Inference-driven attention in symbolic and perceptual tasks: Biases toward expected and unexpected inputs Paolo Cherubini University of Milan-Bicocca, Milan, Italy

Michele Burigo Centre for Thinking and Language, University of Plymouth, Plymouth, UK

Emanuela Bricolo University of Milan-Bicocca, Milan, Italy

The aims of this paper are (a) to gather support for the hypothesis that some basic mechanisms of attentional deployment (i.e., its high efficiency in dealing with expected and unexpected inputs) meet the requirements of the inferential system and have possibly evolved to support its functioning, and (b) to show that these orienting mechanisms function in very similar ways in two perceptual tasks and in a symbolic task. The general hypothesis and its predictions are sketched in the Introduction, after a discussion of current findings concerning visual attention and the generalities of the inferential system. In the empirical section, three experiments are presented where participants tracked visual trajectories (Experiments 1 and 3) or arithmetic series (Experiments 2 and 3), responding to the onset of a target event (e.g., to a specific number) and to the repetition of an event (e.g., to a number appearing twice consecutively). Target events could be anticipated when they were embedded in regular series/trajectories; they could be anticipated, with the anticipation later disconfirmed, when a regular series/trajectory was abruptly interrupted before the target event occurred; and they could not be anticipated when the series/trajectory was random. Repeated events could not be anticipated. Results show a very similar pattern of allocation in tracking visual trajectories and arithmetic series: Attention is focused on anticipated events; it is defocused and redistributed when an anticipation is not confirmed by ensuing events; however, performance decreases when dealing with random series/trajectory—that is, in the absence of anticipations. In our view, this is due to the fact that confirmed and disconfirmed anticipations are crucial events for “knowledge revision”—that is, the fine tuning of the inferential system to the environment; attentional mechanisms have developed so as to enhance detection of these events, possibly at all levels of inferential processing.

Correspondence should be addressed to Paolo Cherubini, Department of Psychology, University of Milan-Bicocca, 1, Piazza Ateneo Nuovo, 20126, Milan, Italy. Email: [email protected] We thank Carlo Umilta`, Kenny Coventry, Timothy Hubbard, Brian Scholl, David Fencsik, Jim McAuliffe, and an unknown reviewer for their helpful comments on previous drafts of this paper, and Sally Couchman for language revision. # 2006 The Experimental Psychology Society http://www.psypress.com/qjep

597

DOI:10.1080/02724980443000863

CHERUBINI, BURIGO, BRICOLO

Visual attention and new stimuli Visual attention refers to the process of selecting a portion of the available visual information for object identification and localization. It can be deployed on different sources of information. The deployment of attention may sometimes depend almost exclusively on the properties of the stimuli (stimulus-driven attention; e.g., abrupt onsets in the periphery of the visual field). At other times it may be under strict supervision according to the observer’s expectations, beliefs, and goals (goal-directed attention). In most instances, some combination of these two influences determines how attention is distributed. For example, some feature singletons1 can capture attention even when they are task irrelevant (Joseph & Optican, 1996), but this occurs only if the observer is in a particular attentional set (singleton detection mode; cf. Bacon & Egeth, 1994; Folk, Remington, & Johnston, 1992). The main features of the attentional system have been adaptively shaped by evolution. Both stimulus-driven mechanisms and goal-directed mechanisms serve the purpose of enhancing detection of potentially important environmental information. For example, Yantis and colleagues (Enns, Austen, Di Lollo, Rauschenberger, & Yantis, 2001; Hillstrom & Yantis; 1994; Rauschenberger & Yantis, 2001; Yantis & Hillstrom, 1994; Yantis & Jonides, 1996) showed that neither movements of local features, nor changes in luminance, capture attention unless they allow the segregation of a new object from the background. Rauschenberger and Yantis (2001) and Hillstrom and Yantis (1994) argue that this is a fundamentally adaptive feature of the attentional system: For example, having attention automatically captured by all moving spots in a forest would yield an endless and useless processing of tree leaves; but having attention automatically captured by a pattern of moving spots that segregates the shape of a leopard could save one’s life. From the

evolutionary vantage point, stimulus-driven attention has evolved to single out environmental features that are highly relevant for survival: that is, those features for which the evolutionary cost of a false positive (i.e., attending to something that, later, proves irrelevant) is, on average, less than the cost of a miss (i.e., not attending to something that was relevant). Goal-directed attention should act the other way round, helping to single out features that are, according to the observer’s previous knowledge, important in a specific situation, but can be missed with no danger in most situations. Novelties are probably in the former class (Johnston & Hawley, 1994): Although most of them could turn out to be irrelevant, missing some of them could be costly. The belief that new things capture attention is old. In 1867, Hermann von Helmoltz wrote: “The natural tendency of attention when left to itself is to wander to ever new things: and so soon as the interest of its object is over, so soon as nothing new is to be noticed there, it passes, in spite of our will, to something else” (p. 372). Some empirical results corroborate this view. Important attractors of stimulus-driven attention are abrupt onsets—that is, the sudden appearance of new objects in previously empty locations (Hillstrom & Yantis, 1994; Rauschenberger & Yantis, 2001; Remington, Johnston, & Yantis, 1992). W.A. Johnston and colleagues (Hawley, Johnston, & Farnham, 1994; Johnston, Hawley, & Farnham, 1993; Johnston, Hawley, Plewe, Elliott, & DeWitt, 1990; Johnston & Schwarting, 1996, 1997; Johnston, Schwarting, & Hawley, 1996) found evidence for novel popout—that is, the tendency of novel items to be more localizable when they appear as feature singletons in otherwise familiar arrays than when they appear with other novel items. The adaptive importance of novelty detection is mirrored by the functional explanations of attentional phenomena such as inhibition of return (Posner & Cohen, 1984; Posner, Snyder, & Davidson, 1980), the tendency to inhibit previously attended

1

A feature singleton is an element that has a feature value that is locally unique within a dimension (e.g., a red element in a background of blue elements; Pashler, 1988).

598

THE QUARTERLY JOURNAL OF EXPERIMENTAL PSYCHOLOGY, 2006, 59 (3)

ATTENTION IN SYMBOLIC AND PERCEPTUAL TASKS

locations, which is conceived of as a mechanism that facilitates visual searches by biasing attention toward new locations (Klein, 1988; Klein & MacInnes, 1999; Posner, Rafal, Choate, & Vaughan, 1985; Pratt, O’Donnell, & Morgan, 2000; Pratt & McAuliffe, 2002). Although the concept of a processing priority for novelties is appealing, it raises some problems (Johnston & Schwarting, 1997). First, a novelty is, by definition, an unexpected event. The tools used by the cognitive system to build expectations are the inferential processes. Hence, detecting a novelty requires inferential processing of its background in order to know which events were expected. Sometimes this processing can be fast and automatic, as in expecting that in a static display an empty location at t0 should remain empty at t1 to preserve spatiotemporal continuity (the basic mechanism why abrupt onsets capture attention, according to Yantis & Gibson, 1994). But, in many other cases, inferential processes are postattentive. Given that attention is the resource for selecting elements for further processing, how is it that orienting of attention itself, when driven by unexpected events (i.e., novelties), by definition involves complex, postattentive processing (i.e., the processing needed to establish which events were expected)? Second, there is ample evidence that the perceptual system is advantaged in perceiving and localizing items that are expected in a given context (e.g., Biederman, Mezzanotte, & Rabinowitz, 1982; Jacoby & Dallas, 1981). This implies that “the mind appears to be biased simultaneously toward both expected and unexpected inputs” (Johnston & Hawley, 1994, p. 56, italics added). Does it follow that the mind is biased toward all inputs, or are there inputs that are neither expected nor unexpected? Do the two biases for expected and unexpected inputs work in the same way, or are they different?

In this study we sketch a functional interpretation of how attentional processes deal with expected and unexpected inputs. We assume that in the course of evolution attentional mechanisms have adapted so as to serve the workings of the inferential processes—that is, they were shaped in order to give priority to classes of events that were likely to be relevant for postattentive processing (which, in turn, is relevant for survival).

Generalities of inferential processes The cognitive system is a tool to build and maintain representations isomorphic to the environment (Newell, 1990). Within it, inferential processes expand representations of the environment beyond direct sensorial information. Inferential processes are almost ubiquitous, being involved both in perceptual tasks, like those used by a tennis player to predict where the opponent’s shot will land, and in symbolic tasks, like those used by a scientist trying to sort out the laws underlying a set of phenomena. Inferences in different domains can have different processing times, levels of consciousness, and precision (Berry, 1997; Holyoak & Spellman, 1993; Rips, 2001; Seger, 1994). However, at a computational level of analysis (that is, considering what inferential processes do rather than how they do it; Marr, 1982), inferential processes in all domains perform three main functions (see also the “perceptual cycle” by Neisser, 1976): 1. They gather knowledge about predictive regularities between events—that is, they build “rules” 2 about the environment (inductive function; cf. Holland, Holyoak, Nisbett, & Thagard, 1986; Rips, 2001). 2. They use the rules in (1) to anticipate states of affairs that are beyond the sensorial horizon (deductive function; cf. Anderson & Lebiere, 1998; Braine & O’Brien, 1998;

2

We do not commit ourselves to any specific algorithmic theory of reasoning or knowledge representation here: By using the term “rules”, we mean any possible way by which predictive environmental regularities are detected and stored, with no assumptions regarding their format, be it propositional, or based on models, or embedded in clusters of weighted associations within a network, or any other type of format. THE QUARTERLY JOURNAL OF EXPERIMENTAL PSYCHOLOGY, 2006, 59 (3)

599

CHERUBINI, BURIGO, BRICOLO

Johnson-Laird & Byrne, 1991; Nisbett, 1993; Rips, 1994, 2001).3 3. They compare the anticipations in (2) with ensuing states of affairs, in order to strengthen the rules that generated them if anticipations are confirmed, or to weaken or abandon the rules that generated them if anticipations are not confirmed. This knowledge revision function is also known as hypotheses testing when it occurs under attentional control (cf. Anderson & Lebiere, 1998; Elio & Pelletier, 1997; Holland et al., 1986; Klayman & Ha, 1987; Oaksford & Chater, 1994; Wason, 1960). When inductive processes are unable to sort out predictive regularities in a local environment, the resulting perception is that of randomness: Psychologically, the judgement of randomness results from the repeated failure of attempts to find regularities (Bar-Hillel & Wagenaar, 1991; Diener & Thompson, 1985; Falk & Konold, 1997). Random environments are those where no deductive anticipations of ensuing states are possible. Hence, expected and unexpected events do not amount to all events. Events can be classified in three classes: (a) expected events occur after having been anticipated; (b) unexpected events are those contrasting with an existing expectation; (c) random events occur in the absence of any expectation—that is, they do not conform to an expectation, but neither do they contrast with existing expectations. For example, the abrupt onset of an object in a previously empty location in a static display is an unexpected event, because spatiotemporal continuity enforced the expectation that the location would remain empty (cf. Yantis & Gibson, 1994): For example, if we are looking at a dry sidewalk, the first drop of rain can capture attention. But, if the abrupt onset of the same object occurred in a dynamic display

where other objects kept appearing in random positions (e.g., a drop of rain amid other drops of rain), it would not be an unexpected event, but a random event, because its location would not contrast with any previous expectation.

Knowledge revision and the attentional system: A general hypothesis Knowledge revision (Function 3 in the schema above) is of paramount importance in order to preserve the isomorphism between the environment and its mental representations. Without it the cognitive system would soon cease to be adaptive (Neisser, 1976; Newell, 1990). The two clues that trigger knowledge revision are confirmed and disconfirmed anticipations—namely, expected and unexpected events. They are a sort of litmus test to check whether our knowledge is fine-tuned to the environment, important both for automatic, weight-adjustment type of knowledge revision (cf. Anderson & Lebiere, 1998; Holland et al., 1986) and for voluntary processes of hypotheses testing (Oaksford & Chater, 1994; Wason, 1960). Accordingly, attention should be biased toward expected and unexpected events, because of their importance as a test of the validity/ invalidity of our knowledge about the environment (which, in turn, is directly relevant for survival). The attentional system should be at a loss, not “knowing” what to select, when confronted with random environments. In other words, the attentional system should have evolved so as to be oriented by inferences, spontaneously focusing on anticipated events and rapidly detecting their confirmations and disconfirmations, so as to allow keeping an up-to-date knowledge base. This general pattern of attentional orienting should be present in perceptual tasks (cf. Experiments 1 and 3), as well as in symbolic tasks (cf. Experiments 2 and 3).

3 Some scholars will not approve of considering “deductive” the generation of probabilistic anticipations; however, an inferential process is deductive as long as the falsity of its conclusions implies the falsity of at least one of the premises (i.e., pieces of previous knowledge) on which the conclusions were based. In this perspective, most knowledge-based anticipations are probabilistic only because they are grounded on contingent, uncertain knowledge; but, they are nevertheless deductive because discovery of their falsity implies the revision of the knowledge on which they were grounded.

600

THE QUARTERLY JOURNAL OF EXPERIMENTAL PSYCHOLOGY, 2006, 59 (3)

ATTENTION IN SYMBOLIC AND PERCEPTUAL TASKS

In Experiments 1 and 2 we test five specific hypotheses derived from our general hypothesis: (a) Attention focuses on anticipated states of affairs, so as to enhance processing of a correctly anticipated event; (b) disconfirmed anticipations trigger defocusing and redistribution of attentional resources in order to gather new information about the environment; (c) in the absence of anticipations (i.e., in a random environment), attention should lack proper guidance, resulting in decreased performance; (d) the above predictions, a –c are domain independent; inference-driven attention should operate in similar ways in perceptual and in symbolic tasks; and (e) attention-driving inferences do not depend exclusively on the detection of statistically predictive associations between environmental clues and target events, but rather on the inner mechanisms of the inferential system (i.e., detection, abstraction, and generalization of structural regularities; see Pena, Bonatti, Nespor, & Mehler, 2002, for implicit processes of structure abstraction, and Cherubini, Castelvecchio, & Cherubini, 2005, for explicit processes). Experiment 3 tests and rules out an alternative, nonattentional interpretation of the results of Experiments 1 and 2.

EXPERIMENT 1 In this experiment we tested the hypotheses that, in tracking a visual trajectory, anticipated events are processed more efficiently than unexpected events, that unexpected events trigger the redistribution of attention, that performance is less efficient when no regularities are available in the stimuli, and, finally, that all of these effects do not depend on the statistical validity of expectations. Many previous studies of visual attention have used mostly synchronic tasks: For example, classical search tasks involving feature singletons (for a review, cf. Yantis, 1996) usually display an array of similar elements with one element (or a few) differing in one dimension. In those studies the feature singleton is “unexpected”, because it contrasts with an expectation grounded on the coherence of the background elements. However,

elements in support of the expectation and elements that disconfirm it are simultaneously available, an unusual occurrence in most inferential processes. Since most inferences are diachronic (i.e., individuals predict an event on the ground of something else that occurred, or was perceived, before it), this sort of task is ill suited to studying the role of inferences in the orienting of attention. In many other studies—namely, those using cues—an element appears on a display for a brief period of time before a target probe. If the cue is statistically predictive of the target, participants are told so. This sort of diachronic task is more suitable for studying the role of inferences in attention. However, it is extremely impoverished: Instead of perceiving and extrapolating regularities assumed to be predictive, the participants are a priori told whether there are such regularities, or not; the inductive inferential function is completely omitted. Therefore we devised a new task, which required diachronic inferences based on regularities that the participants had to perceive and extrapolate by themselves from a continuous flow of events (i.e., without clearly marked intervals between different experimental trials). The task required the participants to follow a small round dot intermittently flashing on the screen and to respond to two events, one predictable (in some conditions) on the ground of regularities in the previous locations of the dot, and one unpredictable (in all conditions). The dot could be arranged to appear in (a) locations forming an uninterrupted regular trajectory, (b) locations forming a previously regular trajectory, later interrupted by an unexpected dot, and (c) random locations.

Method Participants A total of 12 participants (3 males, 9 females, mean age 24.6 years) volunteered to take part in the experiment. A total of 7 participants were graduate and undergraduate students from the University of Milan (1 male, 6 females, mean age 23.8 years), and 5 participants were graduate students from the University of Plymouth

THE QUARTERLY JOURNAL OF EXPERIMENTAL PSYCHOLOGY, 2006, 59 (3)

601

CHERUBINI, BURIGO, BRICOLO

(2 males, 3 females, mean age 25.4 years). All were right-handed and had normal or correctedto-normal vision. Stimuli and design The task was to follow a small round dot intermittently flashing on the screen (diameter, 0.488; duration, 650 ms; interstimulus interval, ISI, between dots, 50 ms; colour, light grey on black background). Each trial was composed of 12 dots, which could or could not be arranged to form a regular trajectory. Responses were to be given to two events by pressing two different keys: (a) “In-target” responses were required whenever the dot appeared inside a static outlined octagon at the centre of the screen (diameter of the target, 0.768; colour, light grey on black background); and (b) “repeated-dot” responses were required whenever the dot appeared twice consecutively in the same location on the screen. From now on, any dot requiring a response will be referred to as the imperative dot in a series. There were nine experimental conditions obtained by crossing two factors: series (regular trajectory vs. interrupted trajectory vs. random series) and required response (in-target response vs. repeated-dot response vs. no required response). Each condition comprised 32 trials (8 per block), for a total of 288 trials in the whole experiment (96 for each type of series). Examples of the trials in each condition are reported in Figure 1. In the regular series the dots appeared sequentially, each 0.958 apart from the previous one, forming a regular trajectory pointing at the target. Trajectories were either straight or slightly curved. Their starting point could be in the upper left, upper right, lower left, or lower right sector of the screen (the starting positions and the type of trajectory were counterbalanced within the experiment). Each dot could be anticipated as the regular continuation of the trajectory. The interrupted series were the same as the regular series, except for a dot that was displaced

orthogonally 0.958 away from the place that it should have occupied if the trajectory had been regular (either to the left or to the right). The displaced dot abruptly disconfirmed anticipations based on the regularity of the previous dots in the series. Since each trial followed the preceding one as a continuous flow, and participants did not know when one trial ended and the following began, the displaced dot could be interpreted as the beginning of a random series, inducing redistribution of attention. In the trials that required a response (66% of the total trials) the displaced dot appeared immediately before the imperative dot (be it a dot inside the target, or a dot that appeared on the same location as the displaced dot). The locations of the dots in the random series were randomly generated, so that no regular trajectories were perceived: The position of each dot could not be anticipated. No predictive statistical association between type of series and type of responses was possible: Each type of series was followed in one third of the trials by in-target responses and in one third by repeated-dot responses, and the remaining third did not require any response. The dot in a location immediately preceding the target (from now on, t-1) in the regular series and the displaced dot in interrupted series were statistical predictors, because they signalled imperative dots 66% of the trials,4 but nothing could be anticipated about which response was likely to be required. Therefore, anticipations of in-target responses could exclusively follow from perceiving the trajectories. The statistically predictive association was not possible in random series, where dots preceding imperative dots were indistinguishable from all other dots in the series. In noncatch trials, the imperative dots (both in-target dots and repeated dots) were always in one of the last four positions of the series; 7 to 11 dots were available prior to them, enabling the participants to comfortably perceive the trajectories and to build anticipations about them.

4

Since a participant could not know whether a t-1 dot was part of a regular or a random series, the actual association of t-1 dots with imperative dots in the experiment was 33%.

602

THE QUARTERLY JOURNAL OF EXPERIMENTAL PSYCHOLOGY, 2006, 59 (3)

ATTENTION IN SYMBOLIC AND PERCEPTUAL TASKS

Figure 1. Simplified examples of regular, interrupted, and random series, in Experiments 1 (upper left), 2/3-symbolic (upper right), and 3visual (bottom; without the example of random series). All the examples shown require a target response, but in the real experiments each series was equally associated with target responses, responses to repeated dots/numbers/hours, or no responses. In Experiments 2 and 3-symbolic, all the numbers appeared at the centre of the screen. The small numbers over the dots in Experiment 1 and near the clock hands in Experiment 3-visual were not present in the real task: They are displayed here to show the sequential order of presentation of the dots/hands.

There were no discrete interruptions between trials: All trials were presented one after the other as a continuous flow on the screen, so that participants could not distinctly tell when one trial ended and another began (actually, they

were never told that the experiment was divided into “trials”). Since the imperative dots were in one of the last four positions of the series, when two consecutive trials required a response the distance between the two responses ranged

THE QUARTERLY JOURNAL OF EXPERIMENTAL PSYCHOLOGY, 2006, 59 (3)

603

CHERUBINI, BURIGO, BRICOLO

unpredictably from 8 to 16 dots. Considering that one third of the trials did not require responses, variability in the cyclic presentation of response events was even higher: The only distinctive rhythm was that after a response event, no other response event was to be expected too soon (the actual minimal distance between response-requiring events was 8 dots, but no participants reported such a precise estimation of it). This “continuous flow” methodology is unusual in attentional research and has raised some concerns by some discussants. Admittedly, it increases risk of lapses of attention relative to usual methods in which observers trigger the beginning of each trial (and, accordingly, know exactly when to be vigilant). However, continuous flow of stimuli is what attention has to deal with in most real-world tasks. For example, before the radar age a fighter pilot scanning the skies for enemy planes did not benefit from “pauses between trials” and did not know exactly when to pay attention: If he wished to enhance his survival chances, he had to allocate attention on predictive regularities emerging from the environment (e.g., the direction of far dots moving) and unexpected events (e.g., a sudden burst from behind), and properly react to them. Although the laboratory task that we devised is streamlined and oversimplified with respect to real environments, we thought that preserving the “continuous-flow” quality of real environments enhanced the ecological validity of the results. In order to improve their performance, participants had to adapt their reactions to the dynamically mutating properties of the task environment. Any resulting change in the allocation of attention can then be viewed as a spontaneous adaptation to the task environment, with no possible parasite effects caused by intertrial pauses that tell the participants exactly when to expect a possible change in the task environment and exactly when to pay maximum attention.

Procedure The experiment was programmed with the E-PrimeTM experimental programming package and was run on a PC with 1700 colour monitor. Participants sat at 60 cm from the computer screen in a dimly lit quiet room. Response keys were counterbalanced between blocks. Participants were encouraged to follow the dot with their eyes,5 and they were asked to respond as fast as possible and as accurately as possible to both target events. After reading the instructions, 24 practice trials were run to ensure that the task had been correctly understood. During this period participants could talk to the experimenter in order to clarify their doubts. After training, the first experimental block began. There were four experimental blocks, each one comprising 72 trials and lasting on average 10 minutes. No feedback was given. After each block participants could have some rest. The target octagon remained visible at the centre of the screen for the whole block. Latencies and correctness of the responses were registered. Predictions Since the present task has never been used in previous studies of visual attention, predictions for all the conditions must be carefully assessed and contrasted with alternative predictions endorsed by some recent models of the orienting of attention. Figure 2 is a diagram of the main predictions endorsed by various approaches to the orienting of spatial attention. Regular-trajectory condition. Predictions in the domain of trajectories are so fast and effortless that some authors assume that they are hard wired in the human brain. The phenomenon of representational momentum supports this hypothesis: In the first 300 ms after the disappearance of a moving object6 the memory of its position

5 The experiment used a central rectangular portion of the screen of size 14.368
10.528. Therefore, asking the participants to maintain fixation on the centre would have caused many errors when the dot appeared twice consecutively in a location far from fixation. Experiment 3 uses a comparable task, modified so as to allow fixed gaze. 6 The effect has been observed for continuous movement and for inferred movement; in our experiment, the movement was inferred.

604

THE QUARTERLY JOURNAL OF EXPERIMENTAL PSYCHOLOGY, 2006, 59 (3)

ATTENTION IN SYMBOLIC AND PERCEPTUAL TASKS

Figure 2. Predictions about the RTs to the six conditions in Experiment 1, as endorsed by different orienting mechanisms. Read the arrows as “faster than”. a: inference driven attention; b: attentional momentum; c: inhibition of return; d: mismatch theory; e: abrupt onsets.

is translated ahead along the direction of its expected movement (Freyd & Finke, 1984, 1985; Hubbard, 1995; Hubbard & Bharucha, 1988; Hubbard, Blessum, & Ruppel, 2001; Hubbard & Ruppel, 1999), even though the observer was not trying to predict the trajectory. Even though the relationship between attention and representational momentum is not entirely clear, Kerzel, Jordan, and Muesseler (2001) found that judgement of a target (i.e., whether a circle had a gap at the top or bottom) was facilitated when the target was displayed in a location slightly beyond (i.e., in the direction of representational momentum) where a moving target had vanished, possibly suggesting that representational momentum shifted attention forward (T. L. Hubbard, personal communication, 10/12/2003). In the regular-trajectory condition of Experiment 1 the trajectory approaching the target allows anticipation of the exact instant in which the dot is to appear inside the target. According to the conjecture that attention focuses on anticipated events, participants should respond faster to “in-target” predicted events than to “repeated-dot” unpredicted events. Pratt, Spalek, and Bradshaw’s (1999) hypothesis of

attentional momentum, according to which attention keeps moving along the direction of its last movement, endorses the same prediction. Literature about the inhibition of return (IOR) phenomenon is consistent with the same prediction. The stimulus onset asynchrony between dots is 700 ms, compatible with the insurgence of IOR. The ISIs, 50 ms, are very short, and Pratt and McAuliffe (2005) showed that IOR is not found with short ISIs, suggesting that IOR is not a valid predictor for the present experiment. However, McAuliffe and Pratt (2004) found IOR with a temporal pattern very similar to that in the present experiment (SOA 800 ms and ISI 50 ms), but only when there was complete spatial overlap between the cues and targets. In the repeateddot responses of the current experiment the t-1 dot (acting as a cue) and the repeated dot (acting as a target) were completely overlapped, and accordingly IOR should affect the repeated dot. In the in-target responses, the t-1 dot and the in-target dot did not overlap, and IOR should not affect the in-target dot, or affect it to a lesser extent (because the surroundings of an inhibited area are partially inhibited; see Bennett & Pratt, 2001 and Collie, Maruff, Yucel,

THE QUARTERLY JOURNAL OF EXPERIMENTAL PSYCHOLOGY, 2006, 59 (3)

605

CHERUBINI, BURIGO, BRICOLO

Danckert, & Currie, 2000). Accordingly, IOR is either an invalid predictor for this experiment, or, if it is a valid predictor, it entails the fasttarget versus slow-repeated prediction. However, other approaches do not endorse the same prediction. Mismatch theory (Johnston & Hawley, 1994; Johnston, Hawley, & Farnham, 1993) maintains that data-driven processing of expected inputs is suppressed after their fast topdown recognition, thus enhancing data-driven processing of unexpected inputs. Accordingly, mismatch theory agrees with the previous approaches in predicting that the in-target responses should be fast (because of the fast topdown recognition of expected events); however, it disagrees with them in predicting that the repeated-dot responses should also be fast (because they are unexpected inputs with enhanced bottom-up processing). According to Yantis and colleagues’ analyses of automatic attractors of attention (e.g., Rauschenberger & Yantis, 2001; Yantis & Jonides, 1996), each abrupt onset should automatically capture attention (unless attention was preventively and voluntarily focused on a different location; Yantis & Jonides, 1990): Therefore, participants’ attention in the regular-trajectory conditions should be captured by each and every dot. When a dot at location t-1 appears, participants’ attention should be captured by it; accordingly, a new dot appearing at the same location should be detected faster than a dot appearing at a different location, yielding the prediction that in-target responses should be slower than repeated-dot responses. Interrupted-trajectory condition. In the interrupted-trajectory condition, participants can correctly anticipate the dot in all locations up to t-1. When the dot appears at t-1, they can anticipate the following dot inside the target. This expectation is abruptly disconfirmed, because the following dot is displaced (Figure 1). According to our conjecture, unexpected events are a clue that current knowledge is not reliable; the “attentional gatekeeper” (a nice metaphor by Johnston & Schwarting, 1997) should accordingly redistribute its resources to allow in as much information as

606

possible. As a consequence, an ensuing repeateddot event should be responded to faster than a repeated-dot event in the regular-trajectory conditions, and an ensuing in-target event should be responded to slower than an in-target event in the regular-trajectory condition. Pratt et al.’s (1999) attentional momentum hypothesis endorses a different prediction: According to that model attention should continue in the new path set by dot t-1 together with the displaced dot and thus focus somewhere that is neither the target location nor the repeated-dot location: Attentional momentum supports the prediction that in-target events should be responded to slower than in-target events in the regulartrajectory condition, but does not support the prediction that repeated-dot events should be responded to faster than repeated-dot events in the regular-trajectory condition. Accordingly, although our predictions cannot be distinguished from attentional momentum’s predictions in the regular-trajectory condition, the two approaches endorse different predictions when comparing repeated-dot responses in the regular- and interrupted-trajectory conditions. Inhibition of return predicts that repeated-dot responses should be slower than in-target responses, exactly as it predicted for the regular trajectories: However, it does not predict that repeated-dot responses should be faster in this condition than in the regular condition, and that in-target responses should be slower in this condition than in the regular-trajectory condition. Mismatch theory predicts that both responses should be fast: the repeated-dot response, because attention was previously driven by an unexpected event (the displaced dot) to the location where the repeated dot will appear; and the in-target response, because, after the displaced dot has appeared, the in-target dot is an unexpected event. However, mismatch theory does not predict any difference between this condition and the regulartrajectory condition. Yantis and colleagues’ (e.g., Rauschenberger & Yantis, 2001; Yantis & Jonides, 1996) hypothesis of capture of attention by abrupt onsets allows the same prediction here that it allowed in the regular condition: fast

THE QUARTERLY JOURNAL OF EXPERIMENTAL PSYCHOLOGY, 2006, 59 (3)

ATTENTION IN SYMBOLIC AND PERCEPTUAL TASKS

repeated-dot responses and slow in-target responses, with no differences from regulartrajectory condition. Random condition. According to our hypothesis, the orienting mechanism should be at a loss in this condition: There are no clues to make anticipations, and therefore there are neither expected nor unexpected events. Participants could adopt different strategies—for example, they could distribute attention to the entire display (with the resulting slowing of all response times in respect to the other conditions), or they could try to focus on each dot, one after the other (resulting in faster repeated-dot responses and slower intarget dot responses). However, the distinctive mark of this condition should be an increased inefficiency in performance. Both types of response in the random condition should be slower than intarget responses in the regular-trajectory condition (because the latter are enhanced by focused attention) and possibly slower than the corresponding responses in the interrupted-trajectory condition (because in the interrupted-trajectory condition attention is distributed on a smaller area than in the random condition). Attentional momentum hypothesis yields similar predictions as far as the in-target responses in the regular-trajectory conditions are concerned: In random series attention drifts to remote display locations, depending on the direction of movement inferred on the ground of the last two dots; it follows that all response times to repeated and in-target dots should be slower than in-target responses in the regular-trajectory condition. Inhibition of return predicts that responses to in-target dots should be fast, because in-target dots are on average quite far from the inhibited area where the previous dot appeared. Repeated-dot responses should be slow, because the second dot appears in an inhibited area. Mismatch theory has, here,

the same prediction as our hypothesis: In the absence of expectations, no suppression of expected events occurs, and thus no disinhibition of processing of unexpected events occurs. It follows that all dots should be processed inefficiently. Differently from our hypothesis, it further predicts that unexpected, repeated-dot responses in the regular condition should be faster than repeated-dot responses in the random condition. Finally, the capture of attention by abrupt onsets predicts that in-target responses should be particularly slow (because the mean distance of the dot preceding the in-target dot from the target is larger in this condition than in the previous conditions), and repeated-dot responses should be fast.7 The previous predictions and the corresponding diagram in Figure 2 clarify that a crucial prediction discriminates the inference-driven attention conjecture from all the other orienting mechanisms that we discussed, and not simply from one or some of them: It is the prediction that repeated-dot responses should be enhanced in the interrupted-trajectory condition with respect to the regular-trajectory condition. We call this crucial prediction the surprise effect, because it follows from redistributing attention after being “surprised” by an unexpected event.

Results and analyses Latencies of correct responses Latencies of correct responses are shown in Table 1 and Figure 3. Raw latencies of correct responses were filtered in order to eliminate outliers. As a criterion, we pruned the latencies more than 2 standard deviations away from the mean of the experimental condition for that participant (1.8% of all responses). Mean latencies in Table 1 were calculated after discarding outliers. The distribution of

7 Yantis and Hillstrom (1994) suggested that it is not merely an abrupt onset that captures attention, but it is the presence of new objects in the visual fields. If our current inference-based distinction between expected, unexpected, and random events is used, and new objects are equated to unexpected objects, then the predictions by the abrupt-onsets account are similar to those by our current account, except for an important detail: Since in-target dots in regular trajectories are expected objects, they should not capture attention; therefore, responses to them should not be advantaged.

THE QUARTERLY JOURNAL OF EXPERIMENTAL PSYCHOLOGY, 2006, 59 (3)

607

CHERUBINI, BURIGO, BRICOLO

Table 1. Latencies of correct responses in Experiment 1 Series Response

Regular

Random

Interrupted

In target Repeated

389 516

536 528

448 451

filtered latencies approximated normality. Accordingly, latencies were analysed by a 3
2 repeated measures analysis of variance (ANOVA), factors being type of series (regular vs. interrupted vs. random) and type of required response (in-target responses vs. repeated-dot responses; no-response trials were not included in the analyses). There was a reliable main effect of the type of series, with random series having slower responses than regular and interrupted series, F(2, 22) ¼ 99.7; p , .0001; regular, 450 ms; random, 532 ms; interrupted, 449 ms. The main effect of type of response was also significant, with responses to the repeated dots being slower than responses to the in-target dots, F(1, 11) ¼ 5.9; p , .05; in target, 456 ms; repeated, 497 ms. More important, there was a significant interaction between the two factors, F(2, 22) ¼ 43.1; p , .0001; means in Table 1. In the regular series, in-target responses were faster than repeated-dot responses (confidence intervals at 95%: in-target responses, 359 –420 ms; repeated-dot responses, 480 – 554 ms). In the

interrupted series, latencies of the two responses did not differ reliably (CI at 95%: in-target responses, 413 –484 ms; repeated-dot responses, 419 –483 ms). In-target responses in interrupted series were reliably slower than in-target responses in regular series (CI at 95%: 416 – 484 ms vs. 359 –420 ms), and repeated-dot responses in interrupted series were reliably faster than repeated-dot responses in regular series (CI at 95%: 419– 483 ms vs. 480 –554 ms). There were no differences between straight and curved trajectories, nor between trajectories starting from different sectors of the screen; all the analyses reported above were calculated collapsing across those levels. The above general pattern could be observed in all participants: All participants responded faster to in-target dots than to repeated dots in the regular series (Wilcoxon test: p , .0001), and all of them responded faster to the in-target dots in the regular series than in the interrupted series (Wilcoxon test: p , .0001), and slower to the repeated dots in the regular series than in the interrupted series (Wilcoxon test: p , .0001). Frequencies of errors Proportions of wrong responses (responding with the wrong key: i.e., responding with the repeated-dot key to an in-target dot, or viceversa), misses (not responding to an imperative dot), and false alarms (responding when no imperative dot had appeared on the screen) are reported in Table 2. Proportions of false alarms are computed collapsing the response factor, Table 2. Errorsa in Experiment 1 Series Response

Figure 3. Latencies of correct responses in Experiment 1.

608

Regular

In target

Wrong Miss

2 0

Repeated

Wrong Miss False alarms

3.3 7 4.2

a

In percentages.

THE QUARTERLY JOURNAL OF EXPERIMENTAL PSYCHOLOGY, 2006, 59 (3)

Random

Interrupted

4.9 2.8

3.4 0.2

6.8 12.5 2

5.2 3.6 3.7

ATTENTION IN SYMBOLIC AND PERCEPTUAL TASKS

because they were produced either before the imperative dot in response-requiring trials, or in no-response trials. Since error data were not normally distributed, nonparametric testing was used: Wilcoxon signed-rank test for two related samples was used to compare the two types of required response; its generalized form (i.e., Friedman’s ANOVA-by-rank test for K related samples) was used to compare the three different series; and in order to test nonparametrically the interaction between the two factors we used the test of stratum by treatment interaction for two ordered multinomials, as implemented in the StatXact 5w statistical software (Mehta, Patel, & Senchaudhuri, 2000; see also Kelley & Sawilowsky, 1997; Sawilowsky, 1990).

Discussion

Wrong responses. There were no reliable differences in the frequencies of wrong responses across conditions.

The surprise effect: Disconfirmed anticipations trigger the redistribution of attentional resources. In the interrupted series, where a displaced dot abruptly disconfirmed the set of anticipations generated by the trajectory, participants were equally fast in responding to the in-target dots and to the repeated dots. With respect to the regular series, in-target responses slowed down, whereas repeated-dot responses grew faster. These findings suggest that participants defocused from the intarget response and redistributed attention to both responses, and they are consistent with our hypothesis concerning inference-driven attentional mechanisms. The surprise effect is a crucial finding, because it was not predicted by any alternative model of the orienting of visual attention (Figure 2).

Missed responses. There were reliably more missed responses to repeated dots than to in-target dots (Wilcoxon signed rank test: two-sided exact p ¼ .001; in-target, 1%; repeated, 7.7%). There were reliably more missed responses in the random series than in the interrupted and regular series (Friedman exact test: p ¼ .007; random, 7.7%; interrupted, 2%; regular, 3.5%). Type of response and type of series interacted reliably, with a pattern qualitatively similar to that observed for latencies (test of stratum by treatment interaction for two ordered multinomials: exact p ¼ .04; frequencies in Table 2). False alarms. There were no reliable differences in the frequencies of false alarms across conditions. Sample homogeneity We retested all the main findings on latencies to detect interactions with geographical location of the participants (University of Plymouth vs. University of Milan) and gender. Location and gender had no reliable main effects. More important, they did not interact with the other factors, suggesting that the sample was homogeneous as far as responses to this task are concerned.

Processing of anticipated events is advantaged over processing of unexpected events. This is shown by the fact that in-target responses in regular series (which were very easily anticipated as the continuation of the trajectory) were faster than repeateddot responses in the same condition (which could not be anticipated). This result is consistent both with our general hypothesis and with the alternative attentional momentum (Pratt et al., 1999) and inhibition of return (e.g., Posner & Cohen, 1984) hypotheses, but it is not consistent with mismatch theory (Johnston & Hawley, 1994; Johnston et al., 1993), and with the hypothesis that each and every abrupt onset captures attention (e.g., Yantis & Jonides, 1996).

Decreased performance in random environments. In the random condition, participants could not perceive regularities that allowed anticipation of target events. Moreover, there was no perceivable statistical association that allowed anticipation of the imperative dot. As a result, responses to both events were slow with respect to the other conditions, except for repeated-dot responses in the regular condition, and the highest rate of missed responses was observed. These findings are consistent with our hypothesis that attentional mechanisms are fine tuned to the requirements

THE QUARTERLY JOURNAL OF EXPERIMENTAL PSYCHOLOGY, 2006, 59 (3)

609

CHERUBINI, BURIGO, BRICOLO

of the inferential system and do not work efficiently when the inferential processes are “switched off” by a random environment. The results are also consistent with the attentional momentum hypothesis and partially consistent with mismatch theory (with the exception that, according to the latter, responses in the random condition should be slower than repeated-dot responses in the regular condition). The findings are not consistent with the inhibition-of-return interpretation of the task and with the hypothesis that each abrupt onset captures attention. The results show that abrupt onsets do not capture attention in an environment with many random abrupt onsets (probably because each abrupt onset was not perceived as a new object). The effect of anticipations over attention is not limited to statistically predictive anticipations. There were no statistically predictive associations between the type of series and the responses that were to be given—that is, regular series were followed in an equal number of trials by in-target responses, repeated-dot responses, and no responses. However, as described in the Stimuli and design section, the t-1 dot in regular series and the displaced dot in interrupted series were statistical predictors, inasmuch as they signalled imperative dots in 66% of the trials (33% for t-1 dots, from the participants’ perspective; see Footnote 4). The finding that responses to interrupted series were not faster than responses to regular series shows that the statistically predictive association, which was stronger for interrupted series, was not the main determinant of responses. Furthermore, the finding that repeated-dot responses in regular series were not faster than responses to random events (where no statistically predictive associations were present) shows that all the predictive power of the t-1 dot in regular series was focused on in-target events. Since there was no statistical ground whatsoever to expect an in-target event rather than a repeated dot, this effect is exclusively caused by the regularity of the trajectory. On the other hand, responses to interrupted series were faster than responses to random series, showing that the predictive power of the displaced dot

610

affected both responses; this result corroborates the idea that an event that disconfirms an expectation (i.e., the displaced dot) causes redistribution of attention. These findings suggest that inferential anticipations grounded on perceived regularities affect the orienting of attention even when those anticipations have no statistical validity in the local environment. The alternative hypotheses that we considered did not make specific predictions about this result: However, Yantis and Egeth (1999) showed that perceptually salient singletons orient attention in a goal-driven fashion only when they are statistically task relevant. The present finding weakens that claim: Even though statistical regularities are undoubtedly important for the inferential system, inferential anticipations can affect attentional orienting even when they are not statistically sound. From a broad perspective, this is a minor but theoretically important point: In fact, the human cognitive system is sometimes very good at detecting statistical correlations (e.g., McKenzie, 1994), but it is also prone to illusory correlations (e.g., Chapman, 1967; Fiedler, 2000), allegedly one of the cognitive roots of social problems such as negative stereotyping of outgroup minorities (Hamilton, Dugan, & Trolier, 1985; Hamilton & Gifford, 1976). For example, if a person illusorily believes that a member of a minority group X is more likely than a member of a majority group Y to commit a crime, and wishes to avoid becoming a victim, he or she will probably orient attention so as to detect any nearby member of X more efficiently: In so doing, he or she would be orienting attention on the ground of an inferential expectation, but not one that is statistically sound. Unpredicted results The main effect of the type of response over missed responses was unpredicted and shows that the repeated dots were more difficult to detect than the in-target dots, especially in the random condition. We think that this effect can have two causes: (a) possibly, participants sometimes grew

THE QUARTERLY JOURNAL OF EXPERIMENTAL PSYCHOLOGY, 2006, 59 (3)

ATTENTION IN SYMBOLIC AND PERCEPTUAL TASKS

tired of reorienting their eyes on each dot in the random series, resulting in repeated dots being presented in peripheral vision (cf. Footnote 3); and, (b) although the presentation times and the high rate of hits suggest that attentional blink was not an issue in this task, some repeated dots could have been missed due to attentional blink. However, in our view, these unpredicted effects are irrelevant as far as the main hypotheses of this study are concerned.

EXPERIMENT 2 In 1980, Posner wrote: “There are a number of promising avenues open for linking evidence on spatial orienting to the mechanisms used for orienting to internal mental structures” (Posner, 1980, p. 22). He then enumerated a few places where concepts arising from work in spatial orienting and those from work in orienting to higher level structures seemed to be related, including studies of mental rotations, scanning of mental images, and distances in semantic spaces. Almost 25 years later, Wolfe similarly wrote “If we understand visual search, perhaps we can make progress in other attentional realms” (Wolfe, 2003, p. 75). Notwithstanding these iterated suggestions, we are aware of no direct attempts to link studies on spatial attention and other attentional processes—for example, orienting of attention in thinking. Meanwhile, different fields of cognitive research have successfully linked higher level structures (i.e., mental representations outside the domain of perception) to perceptual concepts: For example, there is intriguing evidence that numbers are represented in a continuous, quantity-based analogical format functionally isomorphic to physical lines (“mental number line”; cf. Dehaene, 1992; Gallistel & Gelman, 2000; Moyer & Landauer, 1967; Whalen, Gallistel, & Gelman, 1999). This suggestion is likely to entail more than a “spatial metaphor”: Evidence from patients affected by “neglect”, a neuropsychological disorder of spatial attention, supports the view that representation of space and numbers are grounded on the same, or very

similar, neurological mechanisms (cf. Zorzi, Priftis, & Umilta`, 2002). Other fields of research have gathered convincing evidence that some high-level processes (e.g., sentential reasoning and quantified reasoning) are grounded on analogical representations similar to spatial representations (cf. Johnson-Laird, 2001; Knauff, Fangmeier, Ruff, & Johnson-Laird, 2003). The hypothesis that mechanisms of attentional orienting can be similar in spatial tasks and in some symbolic tasks follows straightforwardly. The very general inference-driven mechanisms discussed in this paper are good candidates for a preliminary test of this intriguing conjecture. As an example of a symbolic task, we chose an arithmetic task that was likely to have a mental representation analogous to spatial representations. We ran an almost exact analogue of Experiment 1, requiring participants to track arithmetical series (with each number presented in the centre of the screen) instead of visual trajectories (Figure 1). In other words, in order to infer anticipations participants had to do mental calculations, instead of relying on visual pattern detection processes as in Experiment 1. The assumption that the two sorts of inference are at different levels of abstraction is supported both by the fact that pattern-detection is an innate or very early acquired ability, whereas mental computation must be learned at school; and by the suggestion by Freyd (1987, 1992, 1993) that representational momentum does not exist for series of integers.

Method Participants A total of 18 graduate and undergraduate students from the Universities of Milan and Padua (6 males, 12 females, mean age 26 years) volunteered to take part in the experiment. All but 2 were right-handed, and all had normal or correctedto-normal vision. Stimuli and design Each trial was composed of 12 numbers, which could or could not be arranged so as to form a simple arithmetical series. Participants were asked

THE QUARTERLY JOURNAL OF EXPERIMENTAL PSYCHOLOGY, 2006, 59 (3)

611

CHERUBINI, BURIGO, BRICOLO

to respond to two events, pressing a key on a response box whenever a specific number8 appeared (“target-number” responses) and a different key whenever any number appeared twice consecutively (“repeated-number” responses). For example, if the target number was 267, and the participants saw. . . 261, 263, 265, 267, they had to give a target-number response; if they saw . . . 261, 263, 265, 265, they had to give a repeated-number response. There were nine experimental conditions (see Figure 1) obtained by crossing two factors: series (regular series vs. interrupted series vs. random numbers) and required responses (targetnumber response vs. repeated-number response vs. no response). Each condition comprised 32 trials (16 per block), for a total of 288 trials in the whole experiment (96 for each type of series). Series were composed of three-digit numbers, with care being taken that no numbers similar to the target number appeared in the random series, and that each regular series began and ended within the same hundred. Each number appeared at the centre of the screen, was written in light grey on black background in Courier New 18-point characters (average dimension, 2.28
0.958) and remained on the screen for 600 ms; the ISI before the next number was 200 ms. Regular series could be either increasing or decreasing (to emulate the different starting points of the trajectories in Experiment 1), and the numbers could be spaced either by twos or by threes (to allow some variability among the series; in Experiment 1, variability was obtained by using straight trajectories and curves, but curvilinear arithmetical functions would have been quite difficult to follow and predict; accordingly, we used linear series, but with different intervals). In regular series, each number (comprising the target number) could be anticipated as the arithmetical continuation of the series (e.g., . . . 242, 239, 236, 233, 230 . . .). The interrupted series were the same as the regular series, except for a number that was substituted with another one incongruous with the series

(either larger or smaller than the expected number and within a different hundred, in order to minimize its similarity with the target number; e.g., 242, 239, 236, 475, 230 . . .). In all trials requiring a response, the incongruous number immediately preceded the imperative number (be it the target number, or the repetition of the incongruous number). The incongruous number abruptly disconfirmed anticipations based on the regularity of the series, perhaps suggesting that a random series was beginning and inducing redistribution of attention. Random series were composed of 12 pseudorandom three-digit numbers, taking care that no visible regularities emerged from them. Numbers in them could not be anticipated. Since each series was followed in an equal number of trials by target responses, repeated responses, and no responses, no predictive statistical association between type of series and type of responses was possible: As in Experiment 1, the only predictive association available in stimuli was that between the t-1 number and the appearance of an imperative number (66% of the trials), but it did not allow anticipation of which type of response was likely to be required. Accordingly, anticipations about the response to be given could exclusively follow from perceiving regularities in the series. In noncatch trials, imperative numbers always occurred within the last 4 numbers in the series, so that participants had 7 to 11 numbers prior to the imperative number, which comfortably enabled them to detect the simple regularities in the series. Examples are in Figure 1 (the spatial dislocation of numbers in Figure 1 is for explanation purposes only: In the experiment, all numbers appeared one at a time in the centre of the screen). Procedure The procedure was the same as that in Experiment 1, with the following changes: 1. There was no visual central target on the screen: Stimuli were numbers appearing in

8

To minimize memory problems, at the beginning of each block participants were asked to read aloud and repeat to the experimenter the target number for the ensuing block.

612

THE QUARTERLY JOURNAL OF EXPERIMENTAL PSYCHOLOGY, 2006, 59 (3)

ATTENTION IN SYMBOLIC AND PERCEPTUAL TASKS

the centre of the screen rather than dots appearing in different locations. 2. Accordingly, participants were asked to maintain fixation at the centre of the screen, reading mentally the numbers appearing there. 3. Trials were divided into two blocks rather than four blocks.

Results Latencies of correct responses Latencies of correct responses are shown in Table 3 and Figure 4. Raw latencies of correct responses were filtered in order to eliminate outliers. We pruned the latencies more than 2 standard deviations away from the mean of the experimental condition for that participant (1.9% of all responses). The distribution of the remaining latencies approached normality. The remaining latencies were analysed by a 3
2 repeated measures ANOVA, factors being the type of series (regular vs. interrupted vs. random) and the type of required response (target-number vs. repeated number; of course, no-response trials were not included in the analyses). There was a reliable main effect of the type of series, F(2, 34) ¼ 4.2; p , .05, with random series (682 ms) being slower than regular series (638 ms; pairwise comparison: p , .01) and interrupted series (658 ms; pairwise comparison: p , .05). Interrupted series were not reliably slower than regular series, although the pairwise comparison approached significance (p ¼ .06). The main effect of type of response was significant, with responses to repeated numbers being slower than responses to target numbers, F(1, 17) ¼ 22.4; p , .0001; target, 632 ms; repeated, 687 ms. More important, there was a significant Table 3. Latencies of correct responses in Experiment 2 Series Response Target number Repeated number

Regular

Random

Interrupted

550 726

681 683

665 652

Figure 4. Latencies of correct responses in Experiment 2.

interaction between the two factors, F(2, 34) ¼ 36.1; p , .0001; means in Table 3. In the regular series, target-number responses (which could be anticipated) were faster than nonanticipated repeated-number responses. In the interrupted series, latencies of the two responses were not reliably different (confidence intervals at 95% were widely overlapped: target-number responses, 616 –713 ms; repeated-number responses, 589– 715 ms). Target-number responses in the interrupted series were slower than target-number responses in the regular series (CI at 95%: 615 –713 ms vs. 487– 613 ms), and repeatednumber responses in the interrupted series were marginally faster than repeated-number responses in the regular series (CI at 95%: 589– 702 ms vs. 686 –781 ms). The above general pattern could be observed in most participants: In the regular series, all participants responded to target numbers faster than to repeated numbers (Wilcoxon test; exact p , .0001); 15 out of 18 participants responded faster to the repeated numbers in the interrupted series than to the repeated numbers in the regular series (Wilcoxon test; exact p , .001); 17 out of 18 participants responded faster to target numbers in the regular series than to target numbers in the interrupted series (Wilcoxon test; exact p , .001); in the interrupted series, participants were evenly split between those who responded faster to target numbers and those

THE QUARTERLY JOURNAL OF EXPERIMENTAL PSYCHOLOGY, 2006, 59 (3)

613

CHERUBINI, BURIGO, BRICOLO

who responded faster to repeated numbers (9 vs. 9; Wilcoxon test; exact p ¼ .74). Frequencies of errors Distribution of wrong responses (producing a target-number response when a repeated number was presented, or vice versa), misses (not responding when a response was required), and false alarms (producing responses when no imperative numbers had appeared) are reported in Table 4. Wrong responses. There were no reliable differences in the frequencies of wrong responses. Missed responses. There were reliably more missed responses to repeated numbers than to target numbers (Wilcoxon signed rank test: two-sided exact p , .0001; repeated numbers, 13.4%; target number, 4%). Missed responses were not significantly different across type of series (Friedman test), and the interaction Series
Response was not significant (test of stratum by treatment interaction for two ordered multinomials). False alarms. The rate of false alarms was not reliably different across the conditions. Sample homogeneity We retested all the main findings on latencies for interactions with geographical location of the participants (University of Padua vs. University of Milan), and gender. Location and gender had no reliable main effects. More important, they did Table 4. Errorsa in Experiment 2 Series Response

Regular

Random

Interrupted .07 2.6

Target number

Wrong Miss

1.7 3.8

1.6 5.6

Repeated number False alarms

Wrong Miss

2.3 10.9 2.3

1.7 17.4 1.8

a

In percentages.

614

2.3 11.8 2.7

not interact with other factors, suggesting that the sample was homogeneous as far as responses to this task are concerned.

Discussion The main findings of Experiment 2 mirror those of Experiment 1: a. Processing of anticipated events is advantaged over processing of unexpected events. This is shown by the fact that target-number responses in regular series (which were anticipated as the continuation of the series) were faster than repeated-number responses in the same condition (which could not be anticipated). b. Surprise effect: Disconfirmed anticipations trigger the redistribution of attentional resources. In the interrupted series, where a number inconsistent with the series abruptly disconfirmed the set of anticipations generated by the regularities in the series, participants were equally fast in responding to the target numbers as to repeated numbers. With respect to the regular series, target-number responses slowed down, whereas repeated-dot responses grew faster. c. Decreased performance in random environments. In the random condition, where participants could not perceive regularities in the series, both target-number responses and repeated-number responses slowed down with respect to responses in the other conditions, except the repeatednumber response in the regular condition; moreover, participants produced most errors in this condition; d. The effect of anticipations over attention is not limited to statistically predictive anticipations. There were no statistically predictive associations between the type of series and the type of response that was to be given. The only predictive associations were between numbers preceding the response event and the occurrence of a response event in regular and interrupted series, exactly as described in Experiment 1. This notwithstanding, most

THE QUARTERLY JOURNAL OF EXPERIMENTAL PSYCHOLOGY, 2006, 59 (3)

ATTENTION IN SYMBOLIC AND PERCEPTUAL TASKS

participants systematically focused on targetnumber responses in regular series, only to defocus and redistribute attention if the series was interrupted. Here, as in Experiment 1, inferential anticipations affect the orienting of attention even when those anticipations have no statistical validity in the local environment.

Cross-experimental analyses To further explore the apparent similarity between the findings of Experiments 1 and 2, we pooled together the latencies of Experiments 1 and 2 and analysed them with an Experiment (Experiment 1 vs. Experiment 2; between-groups factor)
Type of Series (regular vs. interrupted vs. random; within-participants factor)
Type of Response (target vs. repeated; withinparticipants factor) mixed ANOVA design (results in Tables 3 and 5). The type of experiment had a reliable main effect, with Experiment 1 having shorter latencies than Experiment 2, F(1, 28) ¼ 38.4; p , .0001; Experiment 1, 479 ms; Experiment 2, 659 ms. The main effect of type of response was significant, F(1, 28) ¼ 23.5; p , .0001; responses to targets, 545 ms; responses to repeated dots or numbers, 593 ms. The main effect of type of series was significant, F(2, 56) ¼ 36.6; p , .0001; regular, 546 ms; interrupted, 554 ms; random, 608 ms. There was a reliable interaction between type of response and type of series, F(2, 56) ¼ 66.6; p , .0001; regular-target, 470 ms; regular-repeated, 621 ms; interrupted-target, 556 ms; interrupted-repeated, 551 ms; random-target, 609 ms; random-repeated, 607 ms. Experiment interacted reliably with type of series, F(2, 56) ¼ 7.4; p , .005; regularExperiment 1, 453 ms; regular-Experiment 2, 638 ms; interrupted-Experiment 1, 450 ms; interrupted-Experiment 2, 658 ms; randomExperiment 1, 533 ms; random-Experiment 2, 682 ms. No other interactions were significant. Mean latencies suggest that the significant effect of type of series and the significant interaction Experiment
Series are probably due to random series alone. Random series were relatively faster in Experiment 2. We reanalysed the data

dropping the random series (2
2
2 mixed design). In this analysis, the main effect of experiments was significant, with Experiment 1 having shorter latencies than Experiment 2, F(1, 28) ¼ 36.9; p , .0001; Experiment 1, 451 ms; Experiment 2, 648 ms. The type of response still had a reliable main effect, F(1, 28) ¼ 45.1; p , .0001; responses to targets, 513 ms; responses to repeated dots or numbers, 586 ms, and there was a reliable interaction between type of response and type of series, F(1, 28) ¼ 77; p , .0001; regular-target, 470 ms; regular-repeated, 556 ms; interrupted-target, 621 ms; interrupted-repeated, 551 ms. However, the type of series had no reliable effect, F(1, 28) ¼ 1.48, and, more importantly, the interactions Series
Experiment and Series
Response
Experiment were not significant, F(1, 28) ¼ 3.08 and 3.21, respectively. However, because these latter results hinge on the acceptance of a null hypothesis, it is reasonable to wonder whether the design afforded adequate statistical power: Standard power analyses were actually rather low (Series
Experiment, .39; Series
Response
Experiment, .40). Discussion Cross-experimental analyses show that the pattern of latencies in the two experiments was not reliably different, except for the behaviour of random series. With respect to regular and interrupted series, random series were reliably slower in Experiment 1 than in Experiment 2. However, this unpredicted finding does not affect the main finding, that the predicted interaction between type of series and type of response occurs in both the perceptual and the symbolic task, with very similar patterns: That is, even though absolute response latencies were different, and the absolute values of differences between conditions were different, the directions of all significant differences between conditions were the same. In both experiments attention was focused on anticipations, and was defocused and redistributed when an anticipation was abruptly disconfirmed; in both experiments performance decreased in random conditions, except with respect to repeated-event responses in regular conditions.

THE QUARTERLY JOURNAL OF EXPERIMENTAL PSYCHOLOGY, 2006, 59 (3)

615

CHERUBINI, BURIGO, BRICOLO

In our view, the results suggest that inferencedriven attentional orienting involves similar mechanisms disregarding the domain upon which the inferential system is operating. Of course, these similarities are at a functional and behavioural level, and they do not necessarily imply similarities of the underlying neural substrate. In other words, since reacting to expected and unexpected events might be very useful for survival and adaptation, orienting mechanisms possibly grounded on different neural substrates could have evolved so as to show similar macroscopic behaviours.

EXPERIMENT 3 Experiments 1 and 2 showed a fairly consistent pattern of results. However, expectation-based motor preparation of the response, instead of enhanced or impaired detection caused by attentional orienting, could have caused the same pattern of responses. For example, target numbers and repeated numbers in regular series in Experiment 2 could have been detected equally fast, but participants could have preactivated the finger associated with the targetnumber, resulting in faster response latencies to target numbers than to repeated numbers. Enhanced responses to repeated dots and repeated numbers in the interrupted series could result from deactivation of a previously activated response to target events, resulting in increased availability of the alternative response. Experiment 3 consists of two slightly modified versions of Experiments 1 and 2, the main modification being that both responses require pressing the same response key. If motor preparation, instead of attentional orienting, is the main cause of the trends observed in Experiments 1 and 2, then responding with the same finger should result in equally fast responses to both response events within each series. Another minor variation consists of having a perceptual task so devised that participants could maintain fixation at the centre of the screen. This was done to check whether the pattern observed in Experiment 1 could be replicated in the absence of eye movements.

616

Method Participants A total of 26 undergraduate students from the University of Milan (16 females, 10 males, mean age 21 years) and 4 graduate students from the University of Plymouth (2 females, 2 males, mean age 28 years) took part in the experiment in exchange for course credits (Milan) or GB £5 (Plymouth). All were right-handed, and all had normal or corrected-to-normal vision.

Stimuli, design, and procedure There were 18 experimental conditions obtained by crossing three factors: task (visual vs. symbolic), series (regular series vs. interrupted series vs. random series), and required response (targetevent response vs. repeated-event response vs. no response). Each participant was randomly assigned either to the visual or to the symbolic task. Each task was divided into four blocks and comprised 32 trials (8 per block), for a total of 288 trials in the whole task (96 for each type of series). Examples of regular and interrupted series in the visual task are reported in Figure 1. The visual task, or “clock task”, was analogous to the task in Experiment 1, except for the following: 1. A light-grey static circle outline (“clock face”; diameter, 4.88) with a light-grey small fixation dot in the centre was present on the display throughout the task; the circle had one small placeholder corresponding to 6 o’clock or 12 o’clock (balanced across blocks). 2. A short segment intermittently flashed within the circle, orthogonal to its circumference (“clock hand”; length, 0.88; width, 0.18; duration, 650 ms; ISI, 50 ms), in the positions corresponding to the 12 hours. 3. Each trial consisted of the clock hand appearing in eight positions, which could or could not be arranged so as to form a regular series (clockwise or counterclockwise); we used 8-long series instead than the original 12-long series used in Experiment 1 because of the limited available positions.

THE QUARTERLY JOURNAL OF EXPERIMENTAL PSYCHOLOGY, 2006, 59 (3)

ATTENTION IN SYMBOLIC AND PERCEPTUAL TASKS

4. In interrupted series, the clock hand that interrupted the series could be anywhere on the clock face, instead of necessarily by the target (as it occurred for the displaced dots in Experiment 1). 5. The two response events were “target-hour” responses, required when the clock hand indicated a specific hour (12 o’clock or 6 o’clock, balanced across blocks), and “repeated-hour” responses, required when the hand indicated the same hour twice consecutively. 6. Responses to both events required pressing the same response key with the index finger of the right hand. 7. Participants were requested to maintain fixation on the dot at the centre of the clock face.

The imperative clock hands always appeared in one of the last four positions of the series; 3 to 7 hands were available prior to them to perceive the trajectories and build anticipations about them. The symbolic task involved arithmetic series as in Experiment 2, except for the following adjustments made for consistency with the clock task: 1. Each series comprised 8 numbers instead than the original 12. 2. Since the perceptual task involved only two possible types of regular series (clockwise and counterclockwise), we used two types of regular number sequences (increasing by twos and decreasing by twos). 3. Trials were divided into four blocks rather than two blocks. Procedure was the same as that in Experiments 1 and 2.

Results Latencies of correct responses Latencies of correct responses are shown in Table 5 and Figure 5. Raw latencies of correct responses were filtered and pruned as in the preceding experiments (2.1% of total correct responses were discarded). The distribution of the remaining latencies approached normality. We analysed them with a Task (Task 1 vs. Task 2; between-groups factor)
Type of Series (regular vs. interrupted vs. random; within participants factor)
Type of Response (target vs. repeated; within-participants factor) Table 5. Latencies of correct responses in Experiment 3 Series Task Arithmetic task

Clock task

Response

Regular

Random

Interrupted

Target number Repeated number

571

719

655

719

680

649

Target hour Repeated hour

421 538

520 536

505 503

Figure 5. Latencies of correct responses in Experiment 3. THE QUARTERLY JOURNAL OF EXPERIMENTAL PSYCHOLOGY, 2006, 59 (3)

617

CHERUBINI, BURIGO, BRICOLO

mixed ANOVA design. There was a reliable main effect of the type of task, F(1, 28 ¼ 85.2; p , .0001, with the clock task being faster than the arithmetic task (504 ms vs. 666 ms). The main effect of the type of series was reliable, F(2, 56) ¼ 21.8; p , .0001, with random series (614 ms) being slower than regular series (562 ms; pairwise comparison, p , .0001) and interrupted series (578 ms; pairwise comparison, p , .05). Interrupted series were not reliably slower than regular series (p ¼ .27). The main effect of type of response was significant, with responses to repeated events being slower than responses to target events, F(1, 28) ¼ 16.5; p , .001; target, 565 ms; repeated, 604 ms. More important, there was a significant interaction between the two factors, F(2, 56) ¼ 46.9; p , .0001; regular-target, 496 ms; regularrepeated, 628 ms; interrupted-target, 580 ms; interrupted-repeated, 576 ms; random-target, 620 ms; random-repeated, 608 ms. In the regular series, responses to target events were faster than responses to repeated events (CI at 95%: 465 –528 ms vs. 608– 649 ms). In the interrupted series, latencies of the two responses were not reliably different (CI at 95%: target responses, 557 –604 ms; repeated responses, 554 – 598 ms). Responses to target events were slower in the interrupted series than in the regular series (CI at 95%: 556– 604 ms vs. 465– 527 ms), and responses to repeated events were faster in the interrupted series than in the regular series (CI at 95%: 554 – 598 ms vs. 607 –649 ms). Contrary to predictions, the three-way interaction Type of Task
Response
Series was reliable, F(2, 56) ¼ 3.4; p , .05, implying that the crucial Response
Series interaction had a different trend in the symbolic and in the visual task. No other interactions were significant. Mean latencies suggest that the significant three-way interaction Task
Response
Series is probably due to random series alone. Responses to random series show a different pattern in the two tasks (Figure 5). We reanalysed the data dropping the random series (2
2
2 mixed ANOVA design). In this analysis, the main effects of the type of task and of the type

618

of response were reliable: task, F(1, 28) ¼ 76.4; p , .0001; clock task, 492 ms; arithmetic task, 649 ms; response, F(1, 28) ¼ 38.3; p , .0001; responses to targets, 538 ms; responses to repeated events, 602 ms. There was a reliable interaction between type of response and type of series, F(1, 28) ¼ 46.6; p , .0001; regular-target, 496 ms; regular-repeated, 628 ms; interruptedtarget, 580 ms; interrupted-repeated, 576 ms. However, the type of series had no reliable effect, F(1, 28) ¼ 3.2, and, more importantly, the three-way interaction Task
Response
Series was not significant, F(1, 28) ¼ 0.8, suggesting that the trends of responses in the clock task and in the arithmetic task were similar. However, once again the design did not afford adequate statistical power to confidently accept a null hypothesis (observed power, .53); that is, the similarity between the trends in the two tasks, although theoretically relevant, does not necessarily imply equality between them. The above Series
Response pattern was observed in most participants: in the regular series; 28 out of 30 participants responded to target events faster than to repeated events (Wilcoxon test; exact p , .0001; all 15 participants in the clock task, plus 13 out 15 participants in the arithmetic task); 26 out of 30 participants responded faster to the repeated events in the interrupted series than to the repeated events in the regular series (Wilcoxon test; exact p , .0001; 12 out of 15 in the clock task, plus 14 out of 15 in the arithmetic task); 26 out of 30 participants responded faster to target events in the regular series than to target events in the interrupted series (Wilcoxon test; exact p , .0001; 13 out of 15 in both tasks). In the interrupted series, participants were more or less evenly split between those who responded faster to target events (16 out of 30; 7 out of 15 in the clock task, with one tie, and 9 out of 15 in the arithmetic task) and those who responded faster to repeated events (13 out of 30; 7 out of 15 in the clock task, and 6 out of 15 in the arithmetic task; all differences are unreliable). Responses in the random series mirrored the reliable interaction of latencies

THE QUARTERLY JOURNAL OF EXPERIMENTAL PSYCHOLOGY, 2006, 59 (3)

ATTENTION IN SYMBOLIC AND PERCEPTUAL TASKS

with the type of task: In the clock task, 10 out of 15 participants responded faster to target events than to repeated events (Wilcoxon test; exact p ¼ .08), whereas in the arithmetic task the opposite result was observed, with 10 out of 15 participants responding faster to repeated events than to target events (Wilcoxon test; exact p , .05).

repeated-event misses and target-event misses was larger in the regular series than in the interrupted series.

Frequencies of errors Distribution of misses and false alarms are reported in Table 6. Since both responses required pressing the same response key, wrong responses could not be detected. We used the same nonparametric tests as those used in Experiments 1 and 2. As far as we know, there are no available nonparametric methods to test for three-way factorial interactions, and so we tested the errors of the two tasks separately.

Discussion

Missed responses. There were reliably more missed responses to repeated events than to target events, for both the clock task (Wilcoxon signed rank test: two-sided exact p , .01; repeated, 7.0%; target, 3.0%) and the arithmetic task (two-sided exact p , .001; repeated, 14.3%; target, 4.3%). For both tasks, missed responses were not significantly different across type of series (Friedman test). The interaction Series
Response was significant for the arithmetic task (test of stratum by treatment interaction for two ordered multinomials; exact p , .01) and marginally significant in the clock task (exact p ¼ .07), showing that the difference between Table 6. Errorsa in Experiment 3 Series Task

Response

Regular

Random

Interrupted

Arithmetic task

Miss target Miss repeated False alarms

3.7 15.4 2.5

4.9 15.0 1.4

4.3 10.6 2.8

Clock task

Miss target Miss repeated False alarms

3.0 9.7 3.1

3.0 6.0 2.1

2.9 5.3 4.4

a

In percentages.

False alarms. For both tasks, the rate of false alarms was not reliably different across the conditions.

a. Replication of the main findings of Experiments 1 and 2. Results in Experiment 3 replicate the main findings of the previous experiments. The surprise effect is present in both tasks, with responses to repeated events being faster in the interrupted series than in the regular series. The effect of anticipation is shown by responses to target events being faster and better detected than responses to repeated events in regular series. The visual versus symbolic nature of the task causes minor variations in performance, mostly limited to random series, as previously observed in the cross-experimental analyses of Experiments 1 and 2; but, regular and interrupted series show very similar patterns of responses in the two tasks. b. Inferential anticipations affect attention. Response preparation cannot explain the results of Experiment 3. The response preparation account claims that anticipating a target event activates the finger associated with the response, and having the anticipation disconfirmed by an interrupted series deactivates it and shifts activation to the finger associated with the alternative response (i.e., the repeated-event response). However, in Experiment 3 the target events and the repeated events were associated with the same finger. Accordingly, no differences in response latencies within a series could have occurred because of response preparation. Admittedly, response preparation could be a component of the response patterns observed in Experiments 1 and 2 and can possibly account for the minor differences in performance between Experiment 3 and the preceding experiments. However, response preparation is not the whole

THE QUARTERLY JOURNAL OF EXPERIMENTAL PSYCHOLOGY, 2006, 59 (3)

619

CHERUBINI, BURIGO, BRICOLO

story: The present results strongly suggest that the observed effect of anticipations is, at least partially, an attentional effect. c. The pattern of responses in visual tasks does not depend (exclusively) on eye movements. There are many differences between the clock task in Experiment 3 and the “dots task” in Experiment 1. First, of course, the response procedure was different, as previously discussed. Second, the random series in the clock task required distribution of attention to a much smaller area of the visual field than did the random series in the dots task, possibly resulting in the decreased advantage of interrupted series over random series observed in the clock task with respect to the dots task. Third, participants in the clock task were told to keep their eyes fixed. As a consequence of the presence of multiple manipulations, the results cannot be used to tell whether eye movements interact with the magnitude or quality of the observed effects. However, they corroborate the view that at least part of the effect of inferential anticipations on the orienting of attention in visual tasks does not depend on eye movements. d. About the “strange” behaviour of the random series: Does it denote the intervention of voluntary attentional sets? The three-way interaction Task
Response
Series is mainly caused by the behaviour of random series: In the clock task, target responses in random series are faster than repeated responses; in the arithmetic task, the opposite is true. The same trend can be observed in the odds of participants that responded faster to target events than to repeated events (2:1 in the clock task, 1:2 in the arithmetic task). In Experiments 1 and 2, responses to random series had a still different behaviour. The wide variability of responses in random series suggests that there is some undetected factor affecting them, causing what appears as erratic variations from experiment to experiment. In our view, this factor might be the involvement of voluntary attentional sets. People could either choose to pay more attention to target events, or to

620

repeated events, or try and keep attention distributed. The choice between the three sets can be idiosyncratically affected by individual differences and/or slight complexity differences between the various tasks. As a result, responses to random series, although generally consistent with the original predictions (i.e., slower responses with respect to interrupted series and to target events in regular series), show great variability from task to task. This interpretation suggests that when a structure is perceived in the environment (i.e., in regular and interrupted series), attention is strongly affected by the inferences supported by that structure; but, when no structure is perceivable in the environment (i.e., random series), attention is centrally directed by choosing a voluntary attentional set. This is a conjecture that could be worth some future research effort.

GENERAL DISCUSSION Summary According to Newell (1990), in order to maintain its adaptive function the inferential system must develop and preserve an isomorphism to the environment: That is, it must be as correct as possible in anticipating future states of the environment. Confirmed and disconfirmed anticipations are crucial events for the fine tuning of the system: Confirmed anticipations strengthen our confidence toward our current knowledge, whereas disconfirmed anticipations trigger knowledge revision and search for novel information. The human attentional system has developed a general orienting mechanism that gives priority to these crucial events. As Experiments 1, 2, and 3 showed, attention is focused on anticipated events, and it is defocused and redistributed in the case where the anticipation is later disconfirmed by a contrasting state of affairs. When the inferential processes cannot compute any anticipation—that is, when the local environment is random (or perceived as such)—the two clues are not available: As a consequence, performance is less efficient, as shown by increased reaction

THE QUARTERLY JOURNAL OF EXPERIMENTAL PSYCHOLOGY, 2006, 59 (3)

ATTENTION IN SYMBOLIC AND PERCEPTUAL TASKS

times and rates of errors in random conditions. Some results in Experiment 3 suggest that in these conditions participants are more likely to adopt a centrally directed attentional set, privileging either one or the other of the two target events. However, further research is needed to clarify this point.

Orienting by inferential clues: Automatic capture or voluntary control? The human inferential system is highly sensitive to statistical covariations in the environment (Holland et al., 1986; Holyoak & Spellman, 1993; Nisbett, 1993), but de facto it uses regularities to detect them: The inductive use of perceived, apparently informative, regularities is so compelling that it is not suppressed even when perceived regularities are not statistically valid predictors of the responses (as occurred when the participants focused on the target event in regular series in our experiments). In other words, inferences are based on perceived regularities in the environment; attention is driven by inferences; therefore, attention is driven by perceived regularities in the environment, even uninformative ones. This finding corroborates our introductory claim that giving priority to confirmed and disconfirmed events is a spontaneous tendency of the attentional system. However, is it entirely automatic, or can it be voluntarily inhibited? One easy way to avoid being misled by noninformative regularities would be not attending to the stimuli in which the regularities are embedded. However, our experimental tasks were devised in such a way that participants could not do that: Since one of the two possible responses required detection of the repetition of a stimulus, and since it could not be predicted which stimulus it was, participants had to attend to all stimuli. Possibly, in this situation perception of simple arithmetical or spatial regularities between the stimuli cannot be voluntarily inhibited. However, further research using explicit instructions not to use regularities in the stimuli is needed to clarify this point.

Concluding remarks The illustrated mechanisms are likely to be very general, because they function in a very similar way for two perceptual tasks and for a symbolic task. This is substantially consistent with the hypothesis by Posner and Petersen (1990), according to which the orienting of attention is a modular function whose characteristics do not depend on the type of input stimuli, and with Posner’s (1980) previous proposal that understanding the mechanisms of spatial orienting could help to understand the mechanisms of orienting to higher level structures. Further work in this direction, possibly expanding research to nonnumerical symbolic tasks, could help in bridging the gap between perception and higher level cognition. However, a cautionary remark is needed: Having shown that there are similarities in the orienting of attention in the two domains does not imply that the orienting mechanisms are strictly the same. Our methodology was sufficient to capture general, “strategic” similarities, but did not afford enough power to establish whether subtle differences were present. Accordingly, although our inference-driven attention proposal allowed for some distinctive predictions, it must be regarded as a model at a functional level of explanation. We have no corresponding algorithmic model to illustrate in detail how inferential expectations interact with the deployment of attention in perceptual and symbolic spaces. Developing such a model will probably require a great deal of cooperation between two research fields that have remained separated for a long time—that is, the study of spatial attention and the study of inferential processes. Original manuscript received 5 February 2004 Accepted revision received 22 July 2004 PrEview proof published online 29 August 2005

REFERENCES Anderson, J. R., & Lebiere, C. (1998). The atomic components of thought. Mahwah, NJ: Lawrence Erlbaum Associates, Inc.

THE QUARTERLY JOURNAL OF EXPERIMENTAL PSYCHOLOGY, 2006, 59 (3)

621

CHERUBINI, BURIGO, BRICOLO

Bacon, W. F., & Egeth, H. E. (1994). Overriding stimulus-driven attentional capture. Perception & Psychophysics, 55, 485– 496. Bar-Hillel, M., & Wagenaar, W. A. (1991). The perception of randomness. Advances in Applied Mathematics, 12, 428– 454. Bennett, P. J, & Pratt, J. (2001). The spatial distribution of inhibition of return. Psychological Science, 12, 76–80. Berry, D. (Ed.). (1997). How implicit is implicit learning? Debates in psychology. New York: Oxford University Press. Biederman, I., Mezzanotte, R. J., & Rabinowitz, J. (1982). Scene perception: Detecting and judging objects undergoing relational violations. Cognitive Psychology, 14, 143–177. Braine, M. D., & O’Brien, D. P. (Eds.). (1998). Mental logic. Cambridge, MA: MIT Press. Chapman, L. J. (1967). Illusory correlation in observational report. Journal of Verbal Learning and Verbal Behavior, 6, 151– 155. Cherubini, P., Castelvecchio, E., & Cherubini, A. M. (2005). Generation of hypotheses in Wason’s 2 – 4 – 6 task: An information theory approach. Quarterly Journal of Experimental Psychology, 58A, 309– 332. Collie, A., Maruff, P., Yucel, M., Danckert, J., & Currie, J. (2000). Spatiotemporal distribution of facilitation and inhibition of return arising from the reflexive orienting of covert attention. Journal of Experimental Psychology: Human Perception and Performance, 26, 1733– 1745. Dehaene, S. (1992). Varieties of numerical abilities. Cognition, 44, 1 – 42. Diener, D., & Thompson, W. B. (1985). Recognizing randomness. American Journal of Psychology, 98, 433–447. Elio, R., & Pelletier, F. J. (1997). Belief change as propositional update. Cognitive Science, 21, 419– 460. Enns, J. T., Austen, E. L., Di Lollo, V., Rauschenberger, R., & Yantis, S. (2001). New objects dominate luminance transients in setting attentional priority. Journal of Experimental Psychology: Human Perception and Performance, 27, 1287–1302. Falk, R., & Konold, C. (1997). Making sense of randomness: Implicit encoding as a basis for judgment. Psychological Review, 104, 301–318. Fiedler, K. (2000). Beware of samples! A cognitiveecological sampling theory of judgement biases. Psychological Review, 107, 659–676. Folk, C. L., Remington, R. W., & Johnston, J. C. (1992). Involuntary covert orienting is contingent on attentional control settings. Journal of

622

Experimental Psychology: Human Perception and Performance, 18, 1030– 1044. Freyd, J. J. (1987). Dynamic mental representations. Psychological Review, 94, 427– 438. Freyd, J. J. (1992). Dynamic representations guiding adaptive behavior. In F. Macar, V. Pouthas, & W. J. Friedman (Eds.), Time, action and cognition: Towards bridging the gap (pp. 309–323). Dordrecht, The Netherlands: Kluwer. Freyd, J. J. (1993). Five hunches about perceptual processes and dynamic representations. In D. Meyer & S. Kornblum (Eds.), Attention and performance XIV: Synergies in experimental psychology, artificial intelligence, and cognitive neuroscience (pp. 99 – 119). Cambridge, MA: MIT Press. Freyd, J. J., & Finke, R. A. (1984). Representational momentum. Journal of Experimental Psychology: Learning, Memory, and Cognition, 10, 126– 132. Freyd, J. J., & Finke, R. A. (1985). A velocity effect for representational momentum. Bulletin of the Psychonomic Society, 23, 443– 446. Gallistel, C. R., & Gelman, R. (2000). Nonverbal numerical cognition: From reals to integers. Trends in Cognitive Sciences, 4, 59 – 65. Hamilton, D. L., Dugan, P., & Trolier, T. K. (1985). The formation of stereotypic beliefs: Further evidence for distinctiveness-based illusory correlation. Journal of Personality and Social Psychology, 48, 5– 17. Hamilton, D. L., & Gifford, R. K. (1976). Illusory correlation in interpersonal perception: A cognitive basis of stereotypic judgments. Journal of Experimental Social Psychology, 12, 392– 407. Hawley, K. J., Johnston, W. A., & Farnham, J. M. (1994). Novel popout with nonsense strings: Effects of predictability of string length and spatial location. Perception & Psychophysics, 55, 261– 268. Helmholtz, H. von (1867). Optique physiologique. Paris: Victor Masson et fils. (Translated from Handbuch der physiologische Optik, 1856– 1867.) Hillstrom, A. P., & Yantis, S. (1994). Visual motion and attentional capture. Perception & Psychophysics, 55, 399– 411. Holland, J. H., Holyoak, K. J., Nisbett, R. E., & Thagard, P. R. (1986). Induction: Processes of inference, learning, and discovery. Cambridge, MA: MIT Press. Holyoak, K. J., & Spellman, B. A. (1993). Thinking. Annual Review of Psychology, 44, 265– 315. Hubbard, T. L. (1995). Environmental invariants in the representation of motion: Implied dynamics and representational momentum, gravity, friction, and centripetal force. Psychonomic Bulletin & Review, 2, 322–338.

THE QUARTERLY JOURNAL OF EXPERIMENTAL PSYCHOLOGY, 2006, 59 (3)

ATTENTION IN SYMBOLIC AND PERCEPTUAL TASKS

Hubbard, T. L., & Bharucha, J. J. (1988). Judged displacement in apparent vertical and horizontal motion. Perception & Psychophysics, 44, 211– 221. Hubbard, T. L., Blessum, J. A., & Ruppel, S. E. (2001). Representational momentum and Michotte’s “Launching Effect” Paradigm. Journal of Experimental Psychology: Learning, Memory, and Cognition, 27, 294–301. Hubbard, T. L., & Ruppel, S. E. (1999). Representational momentum and the landmark attraction effect. Canadian Journal of Experimental Psychology, 53, 242– 255. Jacoby, L. L., & Dallas, M. (1981). On the relationship between autobiographical memory and perceptual learning. Journal of Experimental Psychology: General, 110, 306– 340. Johnson-Laird, P. N. (2001). Mental models and deduction. Trends in Cognitive Science, 5, 434– 442 Johnson-Laird, P. N., & Byrne, R. M. (1991). Deduction. Hove, UK: Lawrence Erlbaum Associates Ltd. Johnston, W. A., & Hawley, K. J. (1994). Perceptual inhibition of expected inputs: The key that opens closed minds. Psychonomic Bulletin & Review, 1, 56 – 72. Johnston, W. A., Hawley, K. J., & Farnham, J. M. (1993). Novel popout: Empirical boundaries and tentative theory. Journal of Experimental Psychology: Human Perception and Performance, 19, 140– 153. Johnston, W. A., Hawley, K. J., Plewe, S. H., Elliott, J. M. G., & DeWitt, M. J. (1990). Attention capture by novel stimuli. Journal of Experimental Psychology: General, 119, 397– 411. Johnston, W. A., & Schwarting, I. S. (1996). Reassessing the evidence for novel popout. Journal of Experimental Psychology: General, 125, 208– 212. Johnston, W. A., & Schwarting, I. S. (1997). Novel popout: An enigma for conventional theories of attention. Journal of Experimental Psychology: Human Perception & Performance, 23, 622– 631. Johnston, W. A., Schwarting, I. S., & Hawley, K. J. (1996). Novel popout, perceptual inhibition, and the stability/plasticity dilemma. In A. F. Kramer, M. G. H. Coles, & G. D. Logan (Eds.), Converging operations in the study of visual selective attention (pp. 315–335). Washington, DC: American Psychological Association. Joseph, J. S., & Optican, L. M. (1996). Involuntary attentional shifts due to orientation differences. Perception & Psychophysics, 58, 651– 665.

Kelley, D. L., & Sawilowsky, S. S. (1997). Non parametric alternatives to the F statistic in analysis of variance. The Journal of Statistical Computation and Simulation, 58, 343– 360. Kerzel, D., Jordan, J. S., & Muesseler, J. (2001). The role of perception in the mislocalization of the final position of a moving target. Journal of Experimental Psychology: Human Perception and Performance, 27, 829– 840. Klayman, J., & Ha, Y. W. (1987). Confirmation, disconfirmation, and information in hypothesis testing. Psychological Review, 94, 211– 228. Klein, R. (1988). Inhibitory tagging system facilitates visual search. Nature, 334, 430–431. Klein, R., & MacInnes, W. J. (1999). Inhibition of return is a foraging facilitator in visual search. Psychological Science, 10, 346– 352. Knauff, M., Fangmeier, T., Ruff, C. C. & JohnsonLaird, P. N. (2003). Reasoning, models, and images: Behavioral measures and cortical activity. Journal of Cognitive Neuroscience, 4, 559– 573. Marr, D. (1982). Vision. New York: Freeman and Co. McAuliffe, J., & Pratt, J. (2005). The role of temporal and spatial factors in covert orienting of visual attention tasks. Psychological Research, 69, 285– 291. McKenzie, C. R. M. (1994). The accuracy of intuitive judgment strategies: Covariation assessment and Bayesian inference. Cognitive Psychology, 26, 209– 239. Mehta, C. R., Patel, N. R., & Senchaudhuri, P. (2000). Efficient Monte Carlo methods for conditional logistic regression. Journal of the American Statistical Association, 95, 99 – 108. Moyer, R., & Landauer, T. (1967). Time required for judgments of numerical inequality. Nature, 215, 1519– 1520. Neisser, U. (1976). Cognition and reality. San Francisco: Freeman. Newell, A. (1990). Unified theories of cognition. Cambridge, MA: Harvard University Press. Nisbett, R. E. (1993). Rules for reasoning. Hillsdale, NJ: Lawrence Erlbaum Associates, Inc. Oaksford, M. R., & Chater, N. (1994). Another look at eliminative and enumerative behaviour in a conceptual task. European Journal of Cognitive Psychology, 6, 149– 169. Pashler, H. (1988). Familiarity and visual change detection. Perception & Psychophysics, 44, 369– 378. Pena, M., Bonatti, L., Nespor, M., & Mehler, J. (2002). Signal driving computation in speech processing. Science, 298, 604– 607.

THE QUARTERLY JOURNAL OF EXPERIMENTAL PSYCHOLOGY, 2006, 59 (3)

623

CHERUBINI, BURIGO, BRICOLO

Posner, M. I. (1980). Orienting of attention. The VIIth Sir Frederic Bartlett Lecture, Quarterly Journal of Experimental Psychology, 32, 3 – 25. Posner, M. I., & Cohen, Y. (1984). Components of visual orienting. In H. Bouma & D. G. Bouwhuis (Eds.), Attention and performance X (pp. 531–536). Hove, UK: Lawrence Erlbaum Associates Ltd. Posner, M. I., & Petersen, S. E. (1990). The attention system of the human brain. Annual Review of Neurosciences, 13, 25 – 42. Posner, M. I., Rafal, R. D., Choate, L. S., & Vaughan, J. (1985). Inhibition of return: Neural basis and function. Cognitive Neuropsychology, 2, 211– 228. Posner, M. I., Snyder, R. R. C., & Davidson, B. J. (1980). Attention and the detection of signals. Journal of Experimental Psychology: General, 109, 160– 174. Pratt, J., & McAuliffe, J. (2002). Inhibition of return in visual marking? The importance of the interstimulus interval and the type of search task. Visual Cognition, 9, 869– 888. Pratt, J., O’Donnell, C., & Morgan, A. (2000). The role of fixation location in inhibition of return. Canadian Journal of Experimental Psychology, 54, 186– 195. Pratt, J., Spalek, T. M., & Bradshaw, F. (1999). The time to detect targets at inhibited and noninhibited locations: Preliminary evidence for attentional momentum. Journal of Experimental Psychology: Human Perception and Performance, 25, 730– 746. Rauschenberger, R., & Yantis, S. (2001). Attentional capture by globally defined objects. Perception & Psychophysics, 63, 1250– 1261. Remington, R. W., Johnston, J. C., & Yantis, S. (1992). Involuntary attentional capture by abrupt onsets. Perception & Psychophysics, 51, 279– 290. Rips, L. J. (1994). The psychology of proof. Cambridge, MA: MIT Press. Rips, L. J. (2001). Two kinds of reasoning. Psychological Science, 12, 129– 134 Sawilowsky, S. S. (1990). Nonparametric tests of interaction in experimental design. Review of Educational Research, 60, 911– 926.

624

Seger, C. A. (1994). Implicit learning. Psychological Bulletin, 115, 163– 196. Wason, P. C. (1960). On the failure to eliminate hypotheses in a conceptual task. Quarterly Journal of Experimental Psychology, 12, 129– 140. Whalen, J., Gallistel, C.R., & Gelman, R. (1999). Nonverbal counting in humans: The psychophysics of number representation. Psychological Science, 10, 130– 137. Wolfe, J. M. (2003). Moving towards solutions to some enduring controversies in visual search. Trends in Cognitive Sciences, 7, 70 – 76. Yantis, S. (1996). Attentional capture in vision. In A. F. Kramer, M. G. H. Coles, & G. D. Logan (Eds.), Converging operations in the study of visual selective attention (pp. 45 – 76). Washington, DC: American Psychological Association. Yantis, S., & Egeth, H. E. (1999). On the distinction between visual salience and stimulus-driven attentional capture. Journal of Experimental Psychology: Human Perception and Performance, 25, 661– 676 Yantis, S., & Gibson, B. S. (1994). Object continuity in apparent motion and attention. Canadian Journal of Experimental Psychology, 48, 182– 204. Yantis, S., & Hillstrom, A. P. (1994). Stimulus-driven attentional capture: Evidence from visual search. Journal of Experimental Psychology: Human Perception and Performance, 20, 95 – 107. Yantis, S., & Jonides, J. (1990). Abrupt visual onsets and selective attention: Voluntary versus automatic allocation. Journal of Experimental Psychology: Human Perception and Performance, 16, 121– 134. Yantis, S., & Jonides, J. (1996). Attentional capture by abrupt visual onsets: New perceptual objects or visual masking? Journal of Experimental Psychology: Human Perception and Performance, 22, 1505– 1513. Zorzi, M., Priftis, K., & Umilta`, C. (2002). Neglect disrupts the mental number line. Nature, 417, 138– 139.

THE QUARTERLY JOURNAL OF EXPERIMENTAL PSYCHOLOGY, 2006, 59 (3)