Open Collections

UBC Theses and Dissertations

UBC Theses Logo

UBC Theses and Dissertations

Goal-driven and stimulus-driven control of visual attention in a multiple-cue paradigm Richard, Christian M. 1999

Your browser doesn't seem to have a PDF viewer, please download the PDF to view this item.

Item Metadata

Download

Media
831-ubc_1999-464091.pdf [ 7.3MB ]
Metadata
JSON: 831-1.0089346.json
JSON-LD: 831-1.0089346-ld.json
RDF/XML (Pretty): 831-1.0089346-rdf.xml
RDF/JSON: 831-1.0089346-rdf.json
Turtle: 831-1.0089346-turtle.txt
N-Triples: 831-1.0089346-rdf-ntriples.txt
Original Record: 831-1.0089346-source.json
Full Text
831-1.0089346-fulltext.txt
Citation
831-1.0089346.ris

Full Text

GOAL-DRIVEN AND STIMULUS-DRIVEN CONTROL OF VISUAL ATTENTION IN A MULTIPLE-CUE PARADIGM By CHRISTIAN M . RICHARD B.A. , Simon Fraser University, 1994 M . A . , Simon Fraser University, 1995 A THESIS SUBMITTED IN PARTIAL F U L F I L M E N T OF T H E REQUIREMENTS FOR THE DEGREE OF DOCTOR OF PHILOSOPHY in THE F A C U L T Y OF G R A D U A T E STUDIES (Department of Psychology) We accept this thesis as conforming to the required standard THE UNIVERSITY OF BRITISH C O L U M B I A September 1999 © Christian Michel Richard, 1999 In presenting this thesis in partial fulfilment of the requirements for an advanced degree at the University of British Columbia, I agree that the Library shall make it freely available for reference and study. I further agree that permission for extensive copying of this thesis for scholarly purposes may be granted by the head of my department or by his or her representatives. It is understood that copying or publication of this thesis for financial gain shall not be allowed without my written permission. The University of British Columbia Vancouver, Canada Department of DE-6 (2/88) ABSTRACT Twelve spatial-cueing experiments examined stimulus-driven and goal-driven control of visual attention orienting under multiple-cue conditions. Spatial cueing involves presenting a cue at a potential target location before a target appears in a display, and measuring the cue's effect on responses to the target stimulus. Under certain conditions, a cue that appears abruptly in a display (direct cue) can speed responses to a target appearing at the previously cued location relative to other uncued locations (called the cue effect). The experiments in this dissertation used a new multiple-cue procedure to decouple the effects of stimulus-driven and goal-driven processes on the control of attention. This technique involved simultaneously presenting a red direct cue (Unique Cue) that was highly predictive of the target location along with multiple grey direct cues (Standard Cues) that were not predictive of the target location. The basic finding was that while cue effects occurred at all cued locations, they were significantly larger at the Unique-Cue location. This finding was interpreted as evidence for stimulus-driven cue effects at all cued locations with additional goal-driven cue effects at the Unique-Cue location. Further experiments showed that Standard-Cue effects could occur independently at multiple locations, that they seemed to involve a sensory-based interaction between the cues and the target, and that they were mediated by a limited-capacity tracking mechanism. In addition, Unique-Cue effects were found to be the product of goal-driven operations, to interact with Standard-Cue effects, and to involve inhibited processing at unattended locations. These results were explained in terms of a filter-based model of attention control that assigns priority to potential attention-shift destinations. According to this model, stimulus-driven and goal-driven factors generate signals (activity distributions) that drive a filter to open an attention channel at the highest priority location by suppressing the signals at other locations. The final experiments confirmed the central assumptions of this model by providing evidence that the priority-destination process was sufficient to produce cue effects independent of attention, and that attending to a location involved a suppression of processing at unattended locations. The implications of this model for the larger visual attention literature were also discussed. TABLE OF CONTENTS A B S T R A C T T A B L E O F C O N T E N T S L I S T O F T A B L E S X -L I S T O F F I G U R E S A C K N O W L E D G E M E N T S & • 1. INTRODUCTION 1.1 BASIC SYMBOLIC- A N D DIRECT- C U E DIFFERENCES 5 1.2 MODELS OF ATTENTION ORIENTING 10 1.2.1 Single-Mechanism View 10 1.2.2 Dual-Mechanism View: 1-1 1.3 C U E EFFECTS AT DIFFERENT STAGES OF PROCESSING 12 1.4 C U E EFFECTS AT MULTIPLE LOCATIONS 13 1.5 SIMULTANEOUS G O A L - D R I V E N A N D STIMULUS-DRIVEN C U E EFFECTS 15 1.6 SELECTIVE ATTENTION MODELS 15 1.7 GOALS OF THIS RESEARCH 23 2. ESTABLISHING THE PHENOMENON 27 2.1 GENERAL METHODS -28. 2.1.1 Apparatus: 2& 2.1.2 Stimuli: 28. 2.1.3 Procedure: 29. 2.1.4 Design: 31. 2.1.5 Subjective Luminance Matching:. 32 2.1.6 Eye-Movement Monitoring: 32 2.1.7 Data Analysis: 3.4 2.2 EXPERIMENT 1 34. 2.2.1 Methods 35. 2.2.2 Results and Discussion 35 2.3 EXPERIMENT 2 36. 2.3.1 Methods .38. 2.3.2 Results and Discussion 3 9 2.4 EXPERIMENT 3 39. 2.4.1 Methods .41. 2.4.2 Results and Discussion .41 2.5 EXPERIMENT 4 47. 2.5.1 Methods .48. 2.5.2 Results and Discussion .49 3. STIMULUS-DRIVEN EFFECTS 54 3.1 EXPERIMENT 5 54. 3.1.1 Methods 56. 3.1.2 Results and Discussion 56 3.2 EXPERIMENT 6 59. 3.2.1. Methods 61. 3.2.2 Results and Discussion 62 iv 4. GOAL-DRIVEN EFFECTS 62 4.1 EXPERIMENT 7A 68. 4.1.1 Methods 69-4.1.2 Results and Discussion 20 4.2 EXPERIMENT 7B -7-4-4.2.1 Methods 36. 4.2.2 Results and Discussion -7.7 4.3 EXPERIMENT 8 81. 4.3.1 Methods 32. 4.3.2 Results and Discussion 83 5. THE ACTIVITY DISTRIBUTION MODEL &7 5.1 INPUT L E V E L 93. 5.2 FILTER L E V E L 9-7. 5.3 ATTENTION L E V E L 99. 5.4 A N EXPLANATION OF C U E EFFECTS 9.9 6. TESTING PREDICTIONS 106 6.1 EXPERIMENT 9 1Q7 6.1.1 Methods 108 6.1.2 Results and Discussion 108 6.2 EXPERIMENT 10 I l l 6.2.1 Methods 115 6.2.2 Results and Discussion 116 6.3 EXPERIMENT 11A 1.19 6.3.1 Methods 121 6.3.2 Results and Discussion 121 6.4 EXPERIMENT 1 I B 12.4 6.4.1 Methods 125 6.4.2 Results and Discussion 126 6.5 EXPERIMENT 12 128 6.5.1 Methods 130 6.5.2 Results and Discussion 132 7. GENERAL DISCUSSION 136 7.1 EMPIRICAL CONTRIBUTIONS OF THIS RESEARCH 137 7.2 CONNECTION OF UNIQUE-CUE P A R A D I G M TO THE VISUAL ATTENTION LITERATURE 141 7.3 A C O M M E N T ON PHYSIOLOGICAL CONNECTIONS L42 7.4 ACTIVITY DISTRIBUTION M O D E L A N D OTHER DIRECT-CUE FINDINGS 143 7.5 CONNECTION TO OTHER NOTIONS OF T H E CONTROL OF ATTENTION 145 7.5.1 Symbolic-Cue effects: 149 7.5.2 Object-based Attention: 150 7.5.3 Visual Search: LSI 8. REFERENCES 155 APPENDIX A : M E A N RESPONSE T I M E TABLES 162 V LIST OF TABLES Table P a § e 1 Differences in cue-effect patterns for Direct Cues and Symbolic Cues 10 2 Mean Response Times and Cue Effects as a function of Target Location 36 in Experiment 1 (p-values for planned comparisons in brackets). 3 Mean Response Times and Cue Effects as a function of Target Location 39 in Experiment 2 (p-values for planned comparisons in brackets). 4 Mean Cue Effects as a function of Number of Standard Cues in 44 Experiment 3 (p-values for planned comparisons in brackets). 5 Mean Response Times (in ms) on Cued and Uncued trials as a function 45 of the type of stimulus adjacent to the Uncued location (E = empty/no cue; U = Unique Cue; S = Standard Cue) for each subject set. 6 Mean Cue Effects as a function of Cue Duration in Experiment 4 (p- 51 values for planned comparisons in brackets). 7 Mean Cue Effects as a function of Cue Luminance in Experiment 5 (p- 57 values for planned comparisons (Standard-Cue vs. Uncued trials) in brackets). 8 Mean Cue Effects and p-values for planned comparisons (Standard-Cue 62 vs. Uncued trials) as a function of the Number of Standard Cues in Experiment 6. 9 Mean Cue Effects as a function of Unique-Cue Validity in Experiment 7a 71 (p-values for planned comparisons in brackets). 10 Mean Cue Effects as a function of Unique-Cue Validity in Experiment 78 7b (p-values for planned comparisons in brackets). 11 Mean Cue Effects as a function of Cue-Target-Onset-Asynchrony 84 (CTOA) in Experiment 8 (p-values for planned comparisons in brackets). 12 Mean Cue Effects as a function of Cue-Target-Onset-Asynchrony 109 (CTOA) in Experiment 9 (p-values for planned comparisons in brackets). 13 Mean Cue Effects as a function of Cue-Target-Onset-Asynchrony 118 (CTOA) and Cue Luminance in Experiment 10 (p-values for planned comparisons in brackets). vi 14 Mean Cue Effects as function of Cue-Target-Onset-Asynchrony (CTOA) 122 in Experiment 11a (p-values for planned comparisons in brackets). 15 Mean Cue Effects as function of Cue-Target-Onset-Asynchrony (CTOA) 126 in Experiment 1 lb (p-values for planned comparisons in brackets). 16 Mean Cue Effects as a function of Cue-Cue-Onset-Asynchrony (CCOA) 132 in Experiment 12 (p-values for planned comparisons in brackets). 17 Analogous results found both in the Unique-Cue paradigm and the Visual Attention Literature. vi i LIST OF FIGURES Figure Page Typical stimulus display sequence used in Symbolic-Cue (A) and 4 Direct-Cue (B) experiments. Typical time course of Cue Effects for Symbolic Cues (dashed line) and 7 Direct Cues (solid line) as a function of Cue-Target-Onset-Asynchrony (CTOA). Example of basic stimulus display used in the Unique-Cue paradigm. 30 The fixation cross is visible for the Inter-Trial-Interval (ITI), then the cues precede the target by the Cue-Target-Onset-Asynchrony (CTOA). Finally the target appears at either the Unique-Cue, a Standard-Cue, or an Uncued location. 4 Mean Response Times and mean Cue Effects as a function of Number 43 of Standard Cues in Experiment 3. 5 Mean Response Times and mean Cue Effects as a function of Cue 50 Duration in Experiment 4. 6 Mean Response Times and mean Cue Effects as a function of Cue- 58 Luminance level in Experiment 5. 7 Mean Response Times and mean Cue Effects as a function of Number 63 of Standard Cues in Experiment 6. 8 Mean Response Times and mean Cue Effects as a function of Unique- 72 Cue Validity in Experiment 7a. 9 Mean Response Times and mean Cue Effects as a function of Unique- 79 Cue Validity in Experiment 7b. 10 Mean Response Times and mean Cue Effects as a function of Cue- 86 Target-Onset-Asynchrony (CTOA) in Experiment 8. 11 A depiction of the problem associated with information transmission 90 between parallel and serial levels of processing. Serial-level processing is least efficient (error-free) i f all sources of information at the parallel level transmit their information at the same rate (A). Events that increase the rate of information transmission from a source wil l increase the efficiency of serial-level processing of that source, however, these streams wil l still be "non-viable" for correct processing (B). Optimal efficiency is only achieved by filtering out information from all vii i unwanted sources, resulting in a "viable" information stream. 12 An example of the different components of the Activity Distribution 95 model and how they would operate in the Unique-Cue paradigm. The onset of the cues triggers the formation of stimulus-driven activity distributions at all cued locations in the Luminance Map and a goal-driven activity distribution in Goal-Driven Input. Activity Distributions are only passed on to the Interaction Map from the Luminance Map i f there is an index associated with a cued location. At the Interaction Map, activation from the Luminance Map and Goal-Driven Input combine to form the Activation Topography which drives the Filter to open an Attention Channel at the most active location. 13 Operation of the Lateral Inhibition Network over time. Each location 98 inhibits all other locations by an amount proportional to the size of the corresponding activity distributions. 14 Mean Response Times and mean Cue Effects as a function of Cue- 110 Target-Onset-Asynchrony (CTOA) in Experiment 9. 15 Mean Response Times and mean Cue Effects for Dim (circles) and 117 Bright (triangles) cues as a function of Cue-Target-Onset-Asynchrony (CTOA) in Experiment 10. 16 Mean Response Times and mean Cue Effects as a function of Cue- 123 Target-Onset-Asynchrony (CTOA) in Experiment 1 la. 17 Mean Response Times and mean Cue Effects as a function of Cue- 127 Target-Onset-Asynchrony (CTOA) in Experiment 1 lb. 18 Stimulus display used in Experiment 12. The fixation cross is visible 131 for the Inter-Trial-Interval (ITI), then the Unique-Cue appears alone in the display for the duration of Cue-Cue-Onset-Asynchrony (CCOA). After this, the Standard-Cues appear for the duration of the Cue-Target-Onset-Asynchrony (CTOA). Finally the target appears at either the Unique-Cue, a Standard-Cue, or an Uncued location. Note that all stimuli remain visible until subjects make a response. 19 Mean Response Times and mean Cue Effects as a function of Cue-Cue- 133 Onset-Asynchrony (CCOA) in Experiment 12. ix ACKNOWLEDGEMENTS The work in this dissertation was conducted in the Psychophysics Laboratory in the Department of Psychology at the University of British Columbia, Vancouver Canada, with some help from the Vision Laboratory of Simon Fraser University, Vancouver Canada. I would like to thank my supervisor Dr. Lawrence Ward for his invaluable contributions to this dissertation. I would also like to thank Drs. Jim Enns, Geoff Hall, and Richard Tees for their discussions of earlier versions of this dissertation. Additional thanks to Drs. John McDonald, and Richard Wright along with D. Prime, Godwin Chan, and Nellie for the help and distraction that made the completion of this work possible. And special thanks to Karen and Nicholas for their support and patience. This work was financially supported by post-graduate research scholarships from the Natural Sciences and Engineering Research Council (NSERC) of Canada and from NSERC grants to Drs. Ward and Wright. 1 1. INTRODUCTION Most people are familiar with the phenomenon of "looking out of the corner of your eyes." A common way to experience this is while driving. For example, while maintaining your eyes on the car in front of you, it is still possible to keep track of a pesky bike courier off to your right that looks like he is about to dart out in front of you. In this case, by tracking the courier in your visual periphery, you are selectively processing visual information from a specific location in your visual field to the exclusion of other locations that you could have processed instead. This selective visual processing is also known as visual attention. Thus, this is an example of paying attention to something that you are not directly looking at. Helmholtz performed one of the first controlled demonstrations of this phenomenon almost 150 years ago (see Warren & Warren, 1968). He set up a screen containing large printed letters in a darkened field. The only item visible in this display was an illuminated pinhole on which he fixated his eyes. A brief electrical spark then illuminated the letters in the display, and his objective was to identify these letters (note that spark duration was not long enough for Helmholtz to move his eyes to any of the letters while they were visible). Helmholtz found that he was initially only able to identify the letters in the central region. He did, however, find it possible to perceive groups of letters farther out in the periphery by deciding in advance to focus on that region of the field. Helmholtz also found that one consequence of concentrating on an area was that he could not perceive letters outside of this region, including those positioned near the fixation point. Helmholtz concluded that, without moving his eyes, 2 he could voluntarily attend to sensations from a particular part of the visual field to the exclusion of sensations from other parts. In Helmholtz's demonstration, two factors were important for determining which letters he perceived. The first was the structure of his retinae, which reduced the acuity of his vision for letters presented away from fixation. Holding this factor constant, it is also apparent that the act of "concentrating" on or attending to a region of the visual display also affected his perception of the scene. In particular, the letters within the region of focus were easier to perceive and remember than letters outside this area. Thus, in this instance, visual attention enhanced or improved the extraction of information from specific locations in the retinal image. In contemporary investigations of this phenomenon, researchers have developed more precise procedures for measuring the effects of attending to a location. One such procedure is the spatial-cueing paradigm. This technique is based on the notion that attention facilitates or enhances processing at a location (e.g., Posner, 1980). Consequently, i f an item requiring processing (i.e. a target stimulus in a detection task) appears at an attended location, it should be processed faster than if it appears at an unattended location. In the latter case, attention would have to be aligned with the previously unattended location before the target could be adequately processed. In the spatial-cueing paradigm, the locus of attention is controlled with visual cues that indicate potential target-onset locations. The information provided by these cues can be used to direct attention to an appropriate location in advance of the target onset. Different combinations of cued and target locations are possible in a presentation sequence. Trials in which the target appears at the same location as that indicated by the 3 cue are called valid-cue trials. This is because the cue provides the correct target-location information. On the other hand, trials in which the target appears at a location other than the one indicated by the cue are called invalid-cue or uncued trials. This is because the cue provides incorrect target-location information. Thus, under the appropriate conditions, it is typical to observe faster or more accurate mean responses on valid-cue trials than on invalid-cue trials because attention is in the correct location to process the target on valid-cue trials. I wil l refer to the difference in responses between these two types of trials as the cue effect. That is, the cue effect reflects the net effect that a valid cue has (usually facilitative) on target responses. In spatial-cueing tasks, two different types of cues can convey information about potential target locations. The first is called a symbolic cue and it indirectly indicates the target position with information presented at a non-target location (see Figure la). One example is an arrow-shaped stimulus presented at fixation that points to a location in the peripheral visual field (e.g., Posner, 1980). In this case, subjects must voluntarily interpret the cue and compute the indicated location before initiating the appropriate preparatory processes. The second is called a direct cue, and it involves the presentation of information directly at a potential target location (see Figure lb): One example of this type of cue is a bar marker that brightens at a location near to which a target can appear (e.g., Jonides, 1981). In this case, precise location information is directly available from the location of the cue onset. 4 Figure 1: Typical stimulus display sequence used in Symbolic-Cue (A) and Direct-Cue (B) experiments. 5 Several models have been developed to explain the operation of visual attention in spatial-cueing paradigms. Although there are many differences in specific elements of these models, there is also a lot of consensus on how this process can be described. Generally, visual attention is described as a mechanism that facilitates or enhances information processing at a location in visual space (e.g., Eriksen & Yeh, 1985; Jonides, 1981; Posner, 1980; Sperling & Weichselgartner, 1995). This has led to metaphors describing attention as a "spotlight" or a "zoom lens" that improves the quality or extraction of information at certain locations. Another common property is that attention is often regarded as a unitary, indivisible processing focus (e.g., Keefer & Siple, 1987; McCormick & Klein, 1989; Posner, Snyder, & Davidson, 1980). I wil l refer to this description as the typical view of attention in the spatial-cueing literature. 1.1 BASIC SYMBOLIC- AND DIRECT- C U E DIFFERENCES Positing that attention operates as a processing focus has allowed researchers to uncover properties about the orienting of attention. One recurrent finding is that orienting attention seems to occur differently depending on the type of cue used to shift attention. More specifically, although both symbolic and direct cue are effective in orienting attention, the particular operations involved in each case seem to differ. One difference is that the cue-types have different computational requirements before attention shifts can be initiated. More specifically, because location information conveyed with a symbolic cue is abstract, this type of cue must first be interpreted, and then the appropriate target position must be generated. These computations may not be necessary with direct cues because these cues appear directly at potential target locations. 6 Another difference between the cues is that observers are not obliged to interpret target information from symbolic cues (they can just ignore these cues), but they may have no choice with direct cues because it may not be possible to suppress neural activity generated by a direct-cue onset. These differences are manifested empirically as differences in cue effects, and they reveal functional differences in the processing initiated by each type of cue. One of the ways in which these cues differ is in their time course of cue effectiveness. The time course measures changes in cue effects over time and it generally reflects the amount of time that subjects have to prepare for a target onset, presumably by initiating an attention shift to the cued location (e.g., Shulman, Remington, & McLean, 1979). It is measured by varying the Cue-Target-Onset-Asynchrony (CTOA), which is the delay between the onset of a cue and the onset of the target. The results of several studies indicate that cue effects generated by symbolic cues build up gradually and level off after around 300 ms (e.g., Cheal & Lyon, 1991; Muller & Findlay, 1988; Muller & Rabbitt, 1989; see Figure 2). In contrast, direct-cue effects are transient, peaking quickly around 100 ms and dropping thereafter (Muller & Rabbitt, 1989; Nakayama & Mackeben, 1989; Weichselgartner & Sperling, 1987; see Figure 2). The slower build-up of symbolic-cue effects is consistent with the idea that additional time is required to interpret the cue and to compute the corresponding destination before attention can be shifted. 7 CO E O LU LU LU Z> O T 1 1 " I 100 200 300 400 CTOA (ms) Figure 2: Typical time course of Cue Effects for Symbolic Cues (dashed line) and Direct Cues (solid line) as a function of Cue-Target-Onset-Asynchrony (CTOA). 8 Symbolic- and direct- cue effects are also differentially affected by cue validity. Cue validity is a measure of the probability that a target will appear at a cued location (the cue-target contingency). A cue's validity can range from low, if it does not indicate the target's actual location better than chance, to high if the cue indicates the correct target location on most of the trials. The results of some studies indicate that direct cues produce reliable cue effects when cue validity is both high and low, whereas symbolic cues are only effective under conditions of high cue validity (e.g., Jonides, 1981; Riggio & Kirsner, 1997; Weichselgartner & Sperling, 1987). The absence of symbolic-cue effects in low-validity conditions is attributable to the fact that these cues are not very useful, and that subjects can probably perform these tasks with less effort by simply ignoring the cues. This suggests that subjects have to voluntarily use information conveyed by symbolic cues i f these are to produce cue effects. In fact, some studies report significant cue effects with low-validity cues i f subjects are given specific instructions to use them (Jonides 1981). This is consistent with the notion that symbolic-cue effects depend more on voluntary, "top-down " operations (mental processing directed by high-level computational goals), while direct-cue effects may be more a product of involuntary processing triggered by the cue onset. A similar pattern is found when the interruptability of the cues is considered. Interruptability relates to how much cue effectiveness is disrupted by factors unrelated to a cueing task (i.e., secondary tasks). Investigations of interruptability consistently report that symbolic-cue effects, but not direct-cue effects are attenuated under these conditions (e.g., Jonides, 1981; however, see Yantis & Jonides, 1990). This pattern occurs most often in situations in which the task-irrelevant factor imposes some type of cognitive 9 load, such as a secondary counting task (Jonides, 1981), competing instructions to ignore cues (Jonides, 1981), or increased task difficulty (Weichselgartner & Sperling, 1987). These data lead to conclusions similar to the previous ones - that direct-cue effects are more involuntary or automatic than symbolic-cue effects. And finally, practice is another factor that influences symbolic-cue effects more than direct-cue effects. In particular, comparisons of responses from before and after extended practice in spatial-cueing tasks indicate that while direct-cue effects are unchanged, symbolic-cue effects tend to occur at shorter CTOAs in practiced conditions (Cheal & Lyon, 1991; Weichselgartner & Sperling, 1987, Warner, Juola, & Koshino, 1990). These practice effects could be explained by asserting that subjects develop strategies that increased the efficiency of the processes mediating symbolic-cue effects. Another account is that these processes become more "automatic" (e.g., Smffrin & Schneider, 1977; Schneider & Sluffrin, 1977). This still leads to a similar conclusion because the automatization of function is said to occur for behaviours that initially require voluntary control on the part of the subject. Empirically, it is quite clear that different cognitive operations mediate the generation of cue effects for symbolic cues and direct cues (see top 4 lines of Table 1). Direct-cue effects occur rapidly, are involuntarily, and are unaffected by secondary tasks and practice. These effects can be described as stimulus-driven because processing activity generated by the onset of the cue in the display seems to be the primary determinant of the properties associated with these effects. In contrast, symbolic-cue effects occur more slowly, require voluntary participation on the part of the observer, and are affected by both distracter tasks and practice. These effects can be described as goal-10 driven because "top-down" factors seem to be the most important determinant of i f and how cue effects wil l occur. Table 1: Differences in cue-effect patterns for Direct Cues and Symbolic Cues Direct Cue Cue Type Symbolic Cue CTOA Fast, transient time course Cue Validity Cue effects occur with both high and low cue validity Interruptability Practice Divisibility No effect No effect Cue effects can occur at multiple locations Slow, sustained time course Cue effects only occur with high cue validity Secondary tasks attenuate cue effects Accelerates time course Cue effects can only occur in a single continuous region 1.2 MODELS OF ATTENTION ORIENTING 1.2.1 Single-Mechanism View Although researchers have proposed several different models to account for the specific properties associated with cue effects produced by symbolic and direct cues, many of these involve a similar relationship between attention and cue effects. More specifically, these models can generally be classified as involving a single attention mechanism or focus that is aligned with different locations in the display. Moreover, this attentional focus is said to be the exclusive source of cue effects in spatial-cueing tasks (e.g., Sperling & Weichselgartner, 1995). I wil l refer to this category of models as the single-mechanism view. This attention mechanism has been described as a "spotlight" (e.g., Posner, 1980) or a "zoom-lens" (e.g., Eriksen & Yeh, 1985) that enhances or facilitates the processing of visual information at the location with which it is aligned. Furthermore, a central property of this mechanism is that it is said to be a unitary 11 indivisible focus (e.g., Sperling & Weichselgartner, 1995). This is also supported by data from symbolic-cue experiments that show that attention cannot be divided between two different locations under these conditions (e.g., Keefer & Siple, 1987; McCormick & Klein, 1989). But more important perhaps, this assumption preserves the perception of a "... single stream of consciousness..." (James, 1890) that seems to dominate our introspective sensations of visual attention. According to the single-mechanism view, while attention is responsible for producing cue effects, the specific cue-effect patterns observed under different cue conditions are actually due to the methods used to control the attentional focus (e.g., Jonides, 1981). More specifically, the properties of goal-driven cue effects, such as the slow time course and dependence on cue usefulness, occur because shifting attention is said to require voluntary planning and execution on the observer's part. In contrast, the properties of stimulus-driven cue effects, such as a rapid time course and insensitivity to cue validity, occur because salient visual signals are said to automatically capture or pull attention to the relevant location (e.g., Yantis & Jonides, 1984). Thus, this view holds that while cue effects are exclusively the product of attention, different sets of cue effect properties can occur because this focus can be shifted in different ways. 1.2.2 Dual-Mechanism View: The assumption that cue effects are exclusively the product of attention may be an unnecessarily limiting constraint in the case of direct cues. These cues are associated with luminance changes in the peripheral visual field and the primate visual system is particularly sensitive to this type of stimulation (e.g., Breitmeyer & Ganz 1976), with neural representations of this input in several areas of the brain (e.g., Posner, Petersen, 12 Fox, & Raichle, 1988; Wurtz & Albano, 1980). Furthermore, abrupt luminance changes often signal important visual events (e.g., the sudden appearance of a predator), and they seem to be processed with higher priority than stimuli associated with smaller or no luminance changes (e.g., Theeuwes, 1995; Todd & Van Gelder, 1979; Yantis & Jonides, 1984). These properties lead to the possibility that luminance changes may also facilitate stimulus processing independently of focused visual attention. While the majority of attention models fall into the single-mechanism category, another group departs from this explanation of cue effects in important ways. This alternative view holds that cue effects, along with being produced by attention orienting as above, can also be the product of additional operations involved in processing direct-cue onsets (e.g., Tassinari, Aglioti, Chelazzi, Peru, & Berlucchi, 1994; Tepin & Dark, 1992; Wright, 1994). More specifically, this view holds that sensory activity generated by the appearance of a direct cue may also facilitate stimulus processing independently of focused visual attention. I will refer to this as the dual-mechanism view because cue effects are said to be due to both attention and sensory-related processing of direct-cue onsets. 1.3 C U E EFFECTS AT DIFFERENT STAGES OF PROCESSING The dual-mechanism view offers several advantages over the single-mechanism view. In particular, it provides a simple explanation of data suggesting that direct cues and symbolic cues may operate at different stages of visual processing. The results of one set of studies suggest that the processes invoked by direct cues seem to operate at a stage of visual processing involved in binding visual features into accurate perceptions, 13 while the processes invoked by symbolic cues seem to operate at a different stage (e.g., Briand, 1998; Briand & Klein, 1987; Klein, 1994). This notion is also consistent with other findings that suggest that, while direct cues seem to affect perceptual sensitivity, symbolic cues may have a greater affect on decision bias (e.g., Muller & Humphreys, 1991; see also Muller, 1994). Similarly, other researchers have found that symbolic cues and direct cues produce additive effects on response times when they occur simultaneously in the same display (Riggio & Kirsner, 1997). These authors argue that this result is consistent with the notion that different stages are involved because changing the magnitude of one type of effect has no impact on the magnitude of the other effect (c.f., Sternberg, 1969). Evidence for different stages is easily accommodated by the dual-mechanism view because this proposal explicitly attributes each type of effect to separate levels of processing. In contrast, explaining these data with the single-mechanism view requires the further assumption that attention can operate at different stages of processing and that access to these stages is limited by cue type. 1.4 C U E EFFECTS AT M U L T I P L E LOCATIONS Another advantage of acknowledging that direct cue effects may have a separate sensory component is that it provides a way to reconcile data from symbolic-cue experiments, which suggest that attention is indivisible, with data from direct-cue experiments, which suggest that attention is divisible. The initial basis of this contradiction stems from symbolic-cue studies, which show that focused visual attention cannot be simultaneously divided between different locations. More specifically, in one 14 set of experiments, two different locations were indicated with symbolic cues and response times at or near the cued locations were measured (Keifer & Siple, 1987; McCormick & Klein, 1990). The data from these studies indicated that target detection times were fastest in regions near the cued locations and became progressively slower for targets appearing farther out. Moreover, responses to targets appearing in between the two cued locations were just as fast as responses to targets at cued locations. This suggests that, i f faced with demands requiring attention at two different symbolically cued locations, subjects may adopt a strategy of attending to an intermediate position (cf. Klein & McCormick, 1989). This finding is supported by other data involving electrophysiological recordings. In one experiment, subjects determined whether two non-adjacent letters were the same or different, a task which forced them to voluntarily attend to both locations simultaneously (Heinze, Luck, Miinte, Mangun, & Hillyard, 1994; however see Kramer & Hahn, 1995). Electrophysiological responses to subsequent probe items presented at locations in between target letters were similar to responses to probes presented at the target locations. These responses, however, were different from those at other non-target locations not in between the two letter positions. That is, all locations within the contiguous region encompassing the target locations produced the same electrophysiological responses to the probe while all locations outside of this region produced a different response. Based on these findings, certain models of attention (e.g., Sperling & Weichselgartner, 1995) explicitly assert that focused attention is indivisible. A different pattern of results, however, seems to occur with direct cues. The data from experiments using these cues suggest that cue effects can occur simultaneously at multiple locations (e.g., Wright, 1994; see also Kramer & Hahn, 1995). This seems to 15 hold for up to four simultaneously presented direct cues (Richard, 1995; see also Krose & Julez, 1989), even if the cued locations are on opposite sides of fixation (Wright, Richard, & McDonald, 1996). In these studies, cue effects were unchanged as the number of simultaneously presented cues increased. This is inconsistent with predictions based on a single indivisible attentional focus because this view holds that cue effects should decrease inversely with the number of simultaneously presented cues1. Another finding is that direct-cue effects seem to be confined to cued locations, because no cue effects occur on trials in which the target appears directly between two cued locations (Richard, 1995). Thus, unlike cue effects produced by symbolic cues, cue effects produced by direct cues do not seem to be confined to a single contiguous region of visual space. 1.5 SIMULTANEOUS GOAL-DRIVEN AND STIMULUS-DRIVEN C U E EFFECTS Another set of findings that support the notion that different stages of processing may be involved in cue effects comes from data indicating that both stimulus-driven and goal-driven cue effects can occur simultaneously in the same display. One source of evidence comes from studies reporting significantly larger direct-cue effects with high-validity direct cues than with low-validity direct cues (e.g., Lambert, Spencer, & Mohindra, 1987; Muller & Rabbitt, 1989). In one study, this finding was interpreted as suggesting that, while low-validity direct cues produce stimulus-driven facilitation, high-validity direct cues produce both stimulus-driven and additional goal-driven facilitation 1 This is based on the assumption that if only a single cued location is attended, then on multiple-cue trials the target can appear at unattended cued locations. If this happens then response times in the cued condition become a mixture of attended (fast) and unattended (slower) cued response times, which brings the average response time closer to that found in purely uncued conditions (it reduces the cue effect). Since the probability of this occurring increases as the number of cued locations increases then cue effects should decrease systematically as a function of the number of cued locations. 16 arises from the usefulness of the cue (Muller & Rabbitt, 1989). Similar findings were obtained in another study that involved presenting both a high-validity symbolic cue and a low-validity direct cue in the same display. The results showed that overall cue effects were significantly larger in the condition in which both types of cues indicated the correct target location when compared to conditions in which only one of the cues indicated the correct target location (Riggio & Kirsner, 1997). In this case, symbolic and direct cues seemed to produce separate but additive effects that were attributed to separate stimulus-driven and goal-driven processes. And finally, similar conclusions were reached in another multiple-cue study. More specifically, in this study, cue effects were larger on trials that involved a single direct cue than on trials that involved multiple simultaneously presented direct cues in which there was no single unambiguous shift destination (Richard, 1995). This difference in magnitudes was interpreted as reflecting the possibility that attention was captured by the single cue (producing both sensory-related and attention-related cue effects), but was not captured by any cued locations on multiple-cue trials (producing only sensory-related cue effects). Thus, in all these cases, conditions involving factors conducive to attentional orienting (high cue validity & attentional capture) produced larger direct-cue effects than conditions that did not involve these factors. Another form of this simultaneous cue-effect argument comes from studies showing that separate stimulus-driven and goal-driven cue effects can occur simultaneously at different locations in the same visual display. In one study, a direct cue signalled a likely target onset at a location in the opposite hemifield (Tepin & Dark, 1992). Thus, the direct cue acted as a direct cue in its onset position and also as a 17 symbolic cue to a position in the opposite side of the display. The results showed that target processing was facilitated both at the direct-cue location and at the symbolically cued location. Similar findings were obtained in an experiment that used simultaneous direct cues and symbolic cues to indicate the target location on a trial (Riggio & Kirsner, 1997). In this case, cue effects occurred at all cued locations, even if the cues indicated different competing locations. Consistent with these findings are the results of another experiment that employed a multiple-cue procedure to measure the simultaneous occurrence of goal-driven and stimulus-driven cue effects in the same display (Richard, 1995). In particular, one red direct cue (Unique Cue) was presented with three grey direct cues (Standard Cues). The Unique Cue was highly valid, while the Standard Cues did not indicate the target location any better than chance. The results indicated that cue effects occurred at all cued locations, however, they were significantly greater at the Unique-Cue location. Furthermore, if the Unique-Cue was made as uninformative as the Standard-Cues (chance), the larger Unique-Cue effect on those trials disappeared. These data are consistent with the notion that, while all cued locations benefited from stimulus-driven processing facilitation, the Unique-Cue also benefited from additional attention-related facilitation (if it was highly valid). Based on this review it is possible to conclude that the single-mechanism proposal has difficulties mamtaining the integrity of its central assumptions. Specifically, with this view it is not possible to assert that the same attentional mechanism is responsible for both stimulus-driven and goal-driven cue effects while retaining the notion that the focus of attention is indivisible. These claims are directly contradicted by data indicating that 18 1) direct-cue effects can occur simultaneously at multiple locations, and 2) that stimulus-driven and goal-driven factors can independently influence cue effects. On the other hand, a dual-mechanism view of cue effects can account for these findings while retaining the notion of an indivisible focus of attention. The main reason for this is that the preattentive level of processing, where much of the stimulus-driven effects are said to occur, does not seem to have the same constraints as the attentional level of processing. More specifically, stimulus-driven processing at the preattentive level can occur at multiple locations in parallel (e.g., Neisser 1967). Thus, i f some direct-cue effects are the product of preattentive processing directly triggered by a cue onset, this processing could be activated at all cued locations simultaneously, without compromising the assumptions made about the operation of the attentional focus. Another problem with the view that a single facilitatory attention mechanism is solely responsible for cue effects is that this notion diminishes an important aspect of visual attention that may only be readily apparent under multiple-cue conditions in spatial cueing paradigms. In particular, multiple-cue procedures emphasize the importance of attention as a selection mechanism. This is because, with multiple cues, different potential attention-shift destinations must compete for access to a single attention mechanism. Consequently, additional operations are required, such as filtering processes, to determine which of the competing attention-shift destinations wil l receive attention. This is not a problem in single-cue procedures because the cue is typically the only attention-shift destination that occurs before the target onset. 19 1.6 SELECTIVE ATTENTION MODELS The notion of attention as a selection process was a central idea in several early models of attention (e.g., Broadbent, 1958; Treisman 1960). According to these models, attention acts as a filter that selectively reduces the flow of information from unwanted streams or channels. Another important aspect of these models is that the selection of a particular information stream for attention is based on the specific properties of that stream. This means that selection can be based on sensory properties such as sound pitch (e.g., Broadbent, 1958) or higher-level properties such as grammatical structure (e.g., Treisman, 1960). This notion is critical because it allows for control over what information is attended. The idea that attention involves the selection of information based on sensory or strategic factors is also evident in some recent models of attention. One of these models is the Attentional Control Settings theory of attention (Folk, Remington, & Johnston, 1992). According to this view, selection involves activating one out of a set of mutually exclusive attentional control settings that determine what type of visual stimulus wil l command attention. Adopting a particular setting renders stimuli associated with other settings unable to capture attention - effectively filtering them out. Moreover, determining which control setting is employed in a situation seems to be under strategic control depending on the specific task-response requirements. In particular, i f subjects perform a task requiring them to respond to a target belonging to one category (activating a particular setting) then only stimuli belonging to that category will capture attention. Thus, in this model, selection occurs because a strategically determined subset of visual stimuli have the preferential ability to command attention. 20 A similar type of model has been derived from visual search experiments in which observers search for a target stimuli embedded in a field of similar non-target stimuli. This is essentially a multiple-stimulus display (without a cue) in which observers must selectively attend to individual display items in order to determine i f a particular item is a target. One model used to explain the results of this type of experiment is the Guided Search model (Wolfe, 1994). According to this proposal, visual information is decomposed into constituent features (e.g. red, straight, horizontal) and these are stored in separate spatiotopically-organized representations based on the feature category (e.g., colour, curvature, orientation, etc.). In each map, a saliency value is computed for each feature location based on how unique or conspicuous a particular feature is relative to its neighbouring features. The saliency values for all locations in all feature maps are then combined to produce a global saliency map indicating the overall salience associated with each location in the visual field. From this point, the control of selection is based simply on the allocation of attention to the most salient locations in sequence until the desired target is found. In addition to this sensory-based selection, "top-down" processes are said to exert control over the orienting of attention by modifying the relative saliency of specific features (e.g., Wolfe, 1992). For example, i f the colour red is particularly important for performing a task, "top-down" processes can boost the saliency associated with the feature "red". Consequently, all items in the display containing the boosted feature will have a higher overall saliency and command attention with a higher priority. Thus, according to this model, the goal-driven selection of a target stream of information 21 involves voluntarily boosting the inherent sensory-based saliency associated with the location of this stream so that it surpasses the saliency of competing locations. The central ideas of the Guided Search model also form the core of a more general model of selective attention called the Activity Distribution model (LaBerge & Brown, 1989). More specifically, this model operates in a similar manner; in fact, the Guided Search model can actually be described as a specialized implementation of the Activity Distribution model. The main difference between the two is that in the Activity Distribution model saliency values are described more generally as activity distributions that play the same role but can be generated by several different processes. That is, along with being generated by feature saliency and "top-down" modification of this saliency, activity distributions can also be produced by other processes such as the designation of actual locations or objects as requiring attention (as opposed to the just the features associated with specific locations or objects). Thus, this model allows several different types of processes to control the allocation of attention. Another difference between the two models is that the selection mechanism is developed in greater detail in the Activity Distribution model. Rather than simply stating that the most salient location commands attention, the Activity Distribution model holds that selection is the product of interactions between activity distributions and a lateral-inhibition network. In particular, combined activity distributions from a global map drive the lateral-inhibition network so that activation from all but the strongest source is attenuated. In this case, selection occurs because the ability of competing locations to command attention is gradually diminished until only the strongest "attention-demanding" signal remains. In other words, once the strongest activity distribution 22 surpasses a threshold, a channel of attention (analogous to a focus of attention) opens at that location. This additional detail about the filter mechanism is useful because it provides testable hypothesis such as the prediction that the ability of competing signals to capture attention should dimmish over time (c.f., Theeuwes, 1991; Yantis & Jonides, 1990). It is also possible to connect the Activity Distribution model with the previously described dual-mechanism view that was argued to be a more complete description of cue effects in multiple-cue paradigms. In particular, along with sharing in common a single attentional channel or focus, both views also posit a significant role for parallel preattentive processing. While the dual-mechanism view holds that preattentive processing may facilitate target-related processing in a non-attentional manner, the Activity Distribution model holds that preattentive accumulation of activity distributions control the allocation of visual attention. If activity distributions can also be associated with a facilitatory effect on processing (perhaps by speeding the opening of an attention channel relative to locations without activity distributions), then it is possible to combine the two views. According to this combined proposal, cue effects can occur i f a target appears at an attended location, or at an unattended location with a large enough activity distribution. The latter condition would produce cue effects because opening an attention channel to process the target could occur faster at a cued location if there was a preexisting activity distribution at that location relative to uncued locations for which this would not be the case. Note that this explanation would be consistent with the dual-mechanism explanation of cue effects at multiple locations as long as the same factors 23 that were said to trigger preattentive processing facilitation (e.g., direct cues) could also generate activity distributions. Another advantage of using the Activity Distribution model to explain cue effects under multiple-cue conditions is that it also contains provisions to account for both stimulus-driven and goal-driven control of attention orienting. This is because activity distributions at a particular location can be generated and modified by both stimulus-driven (e.g., feature saliency) and goal-driven (e.g., "top-down" modification of feature saliency) processes. Thus, by modifying the topography of activity distributions, these processes could indirectly control the allocation of attention in visual space. 1.7 GOALS OF THIS RESEARCH The Activity Distribution model may provide a solution to the previously discussed problem about the inability of the single-mechanism view to adequately account for cue effects under multiple-cue conditions. The reason for this is that, through activity distributions and the processes that generate and modify them, this model is able to account for data indicating both stimulus-driven cue effects independent of attention, and simultaneous stimulus-driven and goal-driven cue effects. Before developing a comprehensive account of stimulus-driven and goal-driven control of visual attention under multiple-cue condition, however, it is necessary to uncover further information about the operation of these processes. Apart from the findings suggesting that stimulus-driven and goal-driven cue effects can occur simultaneously at the same or different locations, not much else is known about the control of attention under multiple-cue conditions. 24 One candidate for this type of investigation is the Unique-Cue paradigm. Although this procedure was initially designed as a simple control study to help explain an anomalous cue effect pattern in a multiple direct-cue study (Richard, 1995), the results it yielded suggest that it may be a useful method for studying the effects of stimulus-driven and goal-driven processing on the control of visual attention. In particular, the multiple direct-cue aspect associated with the multiple Standard Cues seems to be an effective method for isolating the specific stimulus-driven processing effects generated by direct cues. Unlike in single-direct-cue procedures, because multiple Standard Cues appear simultaneously there is no single unambiguous attention-shift destination to which attention can be involuntarily captured Thus, cue effects at those locations should be relatively independent of attention. While the multiple Standard Cues provide a measure of stimulus-driven cue effects, the Unique Cue also provides a way to simultaneously measure the involvement of attention and other "top-down" processing effects. Unlike the Standard Cues, the Unique Cue does provide information about a single unambiguous shift destination, which means that it may also have the ability to capture attention. Furthermore, because the increased facilitation at the Unique-Cue location is affected by "top-down" factors such as cue validity, it seems that this procedure also provides a good measure of additional goal-driven effects. Another benefit of the Unique-Cue paradigm is that, unlike more traditional methods for studying goal-driven and stimulus-driven factors in attention, this procedure also provides measures of each type of activity as they occur simultaneously in the same display. Other methods usually involve measuring each factor separately in different 25 blocks of trials or in different experiments (e.g., Jonides, 1981), which makes these investigations susceptible to strategic differences between presentations (Jonides & Mack, 1984). Because stimulus-driven and goal-driven factors are measured simultaneously in the Unique-Cue paradigm, each should be influenced by the same strategic factors. A final advantage is that because stimulus-driven and goal-driven effects are measured at separate locations, it allows these effects to be decoupled or dissociated. More specifically, because Standard-Cue trials provide an independent measure of stimulus-driven processing, cue effects on these trials can be used to parse out stimulus-driven effects from the overall cue effects on Unique-Cue trials. Since other techniques do not have these simultaneous measures of each effect (e.g., Riggio & Kirsner, 1997), any parsing is based on effects obtained in different blocks or sessions, which again makes them susceptible to strategic differences. Thus, the Unique-Cue paradigm has the potential to provide a more effective way to study the involvement of stimulus-driven and goal-driven processes in the control of visual attention. In this dissertation, I will use the Unique-Cue paradigm to investigate how stimulus-driven and goal-driven processes affect the control of attention in spatial cueing tasks. Because it is an unproven paradigm, this investigation wil l be separated into two different parts. The experiments in the next three chapters wil l establish some of the basic empirical characteristics of this phenomenon. This includes ruling out alternative explanations for the observed data pattern and examining the roles that stimulus-driven and goal-driven processes play in generating cue effects. Following this, a model of attention orienting based on the Activity Distribution model will be presented in Chapter 5 to explain the results of these experiments and Chapter 6 wil l provide empirical support 26 for some of the central assumptions of the model. Finally, the last chapter will summarize the results of the previous chapters and relate them to the larger attention literature. Note that although this dissertation wil l adopt the Activity Distribution model as a framework for explaining stimulus-driven and goal-driven control of attention in this paradigm, this account wil l not be applied to the results until later in the dissertation (starting in Chapter 5). The reason for this is that the purpose of the experiments presented in the earlier chapters is to uncover further details about the characteristics of stimulus-driven and goal-driven cue effects under multiple-cue conditions. Consequently, the results wil l instead be interpreted as properties of the stimulus-driven and goal-driven processes that mediate these effects. Once more information is known about these processes, the Activity Distribution model will be used to provide a comprehensive explanation of the data. 27 2. ESTABLISHING THE PHENOMENON The purpose of the experiments in this chapter is to establish the Unique-Cue paradigm as a valid method for investigating visual attention. Although this paradigm is derived from similar spatial-cueing procedures that are used to study visual attention, it contains some differences in the stimulus display and procedure. Thus, before applying the Unique-Cue paradigm to study the control of attention, it is necessary to confirm that the results from this procedure are relevant to visual attention and not just an artifact of its unique aspects. The basic paradigm involved presenting a single red direct cue {Unique Cue) together with three grey direct cues (Standard Cues) 100 ms before the target appeared in the display (Richard, 1995). The target was most likely to appear at the Unique-Cue location (66.7% of trials), otherwise it was equally likely to appear at either one of the Standard-Cue locations (16.7% of trials) or at any one of the four Uncued locations (16.7% of trials). Under these conditions, cue effects occurred on both Unique-Cue and Standard-Cue trials, however, they were significantly larger on Unique-Cue trials (Richard, 1995). In the context of an attention-related interpretation, this result suggests that, while all cues produced stimulus-driven effects, the Unique Cue might also have produced additional goal-driven effects because it could be singled out and subjects had an incentive to attend to it. Thus, cue effects on Unique-Cue trials seem to involve a combination of stimulus-driven and goal-driven processing facilitation. There are, however, a few unconventional aspects of this paradigm. One is that goal-driven cue effects are associated with a stimulus defined by a unique visual property (the red colour of the Unique Cue). Generating these cue effects therefore requires 28 isolating the Unique Cue from the Standard Cues based on its unique colour and this involves a set of operations not typically associated with spatial-cueing tasks. Thus, it is possible that the processes mediating cue effects in the Unique-Cue paradigm differ in important ways from those mediating cue effects found with more traditional spatial-cueing procedures. If this paradigm is to be validly applied to studying stimulus-driven and goal-driven control of attention, then it is necessary to confirm that these effects are indeed due to goal-driven and stimulus-driven processing (as opposed to other factors such as stimulus-display peculiarities or alternative response strategies). Thus, the purpose of this chapter is to confirm this important assumption. 2.1 GENERAL METHODS 2.1.1 Apparatus: A microcomputer (PC) controlled the experiment timing and stimulus presentation. Stimuli were displayed on a 14-inch colour monitor, and response times were recorded with a button box interfaced with a dedicated timing board in the computer. Subjects were tested in a dimly lit room (to minimize reflections), and an adjustable chin rest was used to maintain head position at a distance of approximately 60 cm from the computer monitor. 2.1.2 Stimuli: A l l stimuli were presented on a black (unlit) background. A light-grey fixation cross (0.4 x 0.4°) remained visible in the centre of the display throughout the experiment. The Unique Cue was a red bar (0.8 x 0.2°), and the Standard Cues were light grey bars with the same dimensions. The target was a white line (1.1 x 0.1°) tilted either to the left (50% of the trial) or to the right (50% of the trials) at a 45° angle 2 . The target was also very easy to detect under these conditions, and this was done to reduce the likelihood that subjects would execute an eye movement to the target before responding. There were eight possible cue and target positions arranged in a circle around the fixation cross (see Figure 3). The midpoints of all cue locations were 6.2° from the centre of the fixation cross and 5.5° from the midpoints of adjacent cue locations. The target appeared just above a cue location so that cues and targets did not overlap i f both occurred at the same position. 2.1.3 Procedure: Subjects were instructed to keep their eyes on the fixation cross at all times and to press the response button as quickly as possible as soon as the target appeared. They were also instructed to "focus" on the Unique Cue when it appeared on every trial, and told that they could earn extra money by consistently doing this. What it meant to "focus" on the Unique Cue was demonstrated by having them fixate the central cross and "notice" or attend to the Unique Cue in their visual periphery. The results from pilot studies suggested that this behavioural control minimized differences in strategic approaches that individual subjects employed during an experiment session. 2 The target was left/right tilted so that the same target stimuli could be used in identification-task versions of this paradigm (Experiments 1 & 7b). 30 Figure 3: Example of basic stimulus display used in the Unique-Cue paradigm. The fixation cross is visible for the Inter-Trial-Interval (ITT), then the cues precede the target by the Cue-Target-Onset-Asynchrony (CTOA). Finally the target appears at either the Unique-Cue, a Standard-Cue, or an Uncued location. 31 Each trial began with a 1000 ms inter-trial interval (ITI). Following this, the Unique Cue and three Standard Cues appeared at randomly-selected display locations (see Figure 3). After a second delay (CTOA), the target appeared at one of the possible target locations and remained visible until the subject pressed the response button. Response times were measured as the interval between the target onset and the button press. A l l cues and targets were extinguished following the observer's response, which marked the end of the trial. A n experiment session was divided into several blocks of approximately 50 trials. The first block was considered practice, and the data from this block were not analyzed. One quarter of the trials in a block were catch trials with 1500 ms CTOAs. These were randomly interspersed with the data trials. Catch-trial responses were collected but not analyzed because their sole purpose was to minimize response anticipation errors. 2.1.4 Design: A l l the experiments involved a Target-Location variable. This variable described where the target appeared in the display; either at the Unique-Cue location, one of the Standard-Cue locations, or at an Uncued location. In most of these experiments Unique-Cue trials occurred most often (66.7% of the trials), while the other two types of trials were equally likely to occur in the remaining trials (16.7% of the trials each). In all experiments, 25% of the trials were Catch trials with 1500 ms CTOAs. These were divided among the different levels of Target Location in the same 4:1:1 ratio as data trials. 32 2.1.5 Subjective Luminance Matching: Subjective Luminance Matching was done using a variation of the flicker photometry technique. Subjects viewed grey and green patches (20° x 20°) that alternated in the centre of the screen at 60 Hz. The luminance of the grey patch remained constant while subjects adjusted the luminance of the green patch. A Method of Limits procedure with ascending and descending stimulus series was used to find the luminance level for the green patch that produced the least flicker. The final luminance for each subject was calculated as the average of four runs (two alternating ascending and descending runs). 2.1.6 Eye-Movement Monitoring: Eye movements were monitored in Experiments 7b and 10. Eye-movement monitoring was done by recording E O G using tin electrodes placed 1 cm lateral to the left and right outer canthi. The E O G activity was amplified with a bandpass of 0.1 - 30 Hz and continuously digitized at a rate of 128 Hz. Electrode impedance was kept below 5 kO. Eye position was calibrated at the beginning of each test session by having subjects make saccades to the cue/target positions. Trials contaminated by eye movements (> 2°) were discarded prior to analysis. Because eye movements were not monitored in all of the experiments, there exists a possibility that eye movements to cued locations in advance of the target onset may have biased or contaminated the data in these experiments. There are several arguments against this claim. The first is that the stimuli were designed so that the tasks could be performed easily without the need to foveate/look at the items. 33 The second is that in the experiments in which eye movements were monitored, eye movements occurred on less than 1% of the trials. Furthermore, the low eye-movement rates occurred even though the experiments involved factors that should have maximized the incentive to execute eye movements (e.g., 80% Unique-Cue validity, target identification task, long CTOAs). The third argument against this claim is that the data in the eye-movement monitoring experiments were comparable to data involving similar conditions, thus it is unlikely that monitoring eye movements affected strategic factors determining how subjects performed a task. The final argument is that the data pattern from experiments that involved long CTOAs but did not monitor eye movements is inconsistent with the notion that eye movements occurred. Three factors are involved in this pattern: 1) eye movements to a location should affect response (speeding Unique-Cue responses and slowing the others), 2) eye movements require time before execution, & 3) eye movements are discrete movements, jumping almost immediately from one location to the next. If eye movements occurred during the course of the CTOA, then these factors should have produced a specific pattern in cue effects. Specifically, there should have been a large step-like increment in Unique-Cue effects at a CTOA commensurate with regular eye-movement execution latencies (200 - 300 ms). The data from Experiments 9 and 12, however, show no change in Unique-Cue effects over CTOA. This pattern suggests that eye movements did not occur during the interval between the cue onset and the target onset. 34 2.1.7 Data Analysis: Before any statistics were calculated, response times less than 100 ms and greater than 1000 ms were excluded from the analysis as errors. Following this, response times greater than three standard deviations from the corresponding trial-type means were also removed. Subjects with error rates of greater than 10% were excluded from further analysis (Mean error rates were below 4% in all experiments except Experiment 8 in which the error rate was 5.7%). Analysis of variances (ANOVAs) were run on error rates in each experiment to check for speed-accuracy trade-offs. These analyses are not reported because none of them showed a significant speed-accuracy trade-off. Only mean cue effects are presented in tables in the main text (mean response times are presented in figures). A more complete description of responses (mean response time, standard error, and error rates), is presented in Appendix A for each experiment. Statistical analyses were performed on the remaining data by using repeated measures A N O V A s . Huynh-Feldt-corrected degrees of freedom were used to determine probability values for all factors with more than two levels i f the assumption of sphericity was violated. Specific hypothesis were tested with planned comparisons using the Bonferroni inequality to control the familywise error rate (set at 0.15). 2.2 EXPERIMENT 1 The purpose of the first experiment was to confirm that the pattern of cue effects obtained with the Unique-Cue paradigm generalizes beyond simple-detection response 35 tasks. Although for the most part, little distinction is made between simple detection and identification tasks in the spatial-cueing literature, the findings of some reviews suggest that there may be some conditions under which these tasks activate potentially different sources of cue effects (Tassinari et al., 1994). If the claim is to be made that the Unique-Cue paradigm permits the investigation of fundamental operations involved in the control of attention, it is therefore necessary to establish that cue effects in this paradigm are not exclusive to simple-detection tasks. To this end, the goal of the present experiment was to replicate the basic Unique-Cue effect pattern with an identification task. 2.2.1 Methods Subjects: 12 University of British Columbia undergraduates were paid $5 for participating in one 1-hour session. A l l subjects had normal or corrected-to-normal vision. Procedure: The procedure was the same as described in the General Methods except for the following changes. Subjects were instructed to press a button with their left hand i f the top of the target was tilted to the left (\) and to press a button with their right hand if the top of the target was tilted to the right (/). The CTOA on data trials was 100 ms. Design: Only the Target-Location variable was measured in this experiment. In total, there were 720 data trials consisting of 480 Unique-Cue, 120 Standard-Cue, and 120 Uncued trials. Also, 240 catch trials with 1500 ms CTOAs were divided according to the same 4:1:1 trial ratio. 2.2.2 Results and Discussion The mean error rate in this experiment was 3.2%. 36 A one-way repeated measures A N O V A was run on pooled mean response times for all subjects in each condition of the Target Location factor (Unique-Cue, Standard-Cue, or Uncued). Table 2 shows mean response times averaged over all subjects. The main effect of Target Location was significant, F(2,22) = 24.60, MSE = 274.82,/) < 0.001. Mean cue effects are also presented in Table 2 (see Appendix A for further information on mean response times, standard errors and error rates, for this and all following experiments). Significant cue effects occurred for both types of cues, which accounts for the main effect of Target Location. Furthermore, cue effects were significantly larger on Unique-Cue trials than on Standard-Cue trials, as indicated by the Unique-Cue Advantage (Unique-Cue effects - Standard-Cue effects). These data provide a clear replication of the Unique-Cue effect with a target identification task. Table 2: Mean Response Times and Cue Effects as a Junction of Target Location in Target Location Unique Cue Standard Cue Uncued Mean Response Time (ms) 413 486 519 Mean Cue Effect (ms) 46 (< 0.001) 33 (0.003) — Unique-Cue Advantage(ms) 13 (0.007) — — 2.3 EXPERIMENT 2 The Unique-Cue paradigm is a potentially informative procedure if the data that it generates can be interpreted in the context of stimulus-driven and goal-driven mediation of attentional control. In order to do this, it is necessary to establish that it is these processes that are responsible for the observed pattern of results and not some other non-attentional processes. One factor that raises uncertainty about this possibility is that the Unique-Cue paradigm introduces new stimulus elements that are not typically used in 37 spatial-cueing experiments. Thus, the first step in establishing the relevance of stimulus-driven and goal-driven processes in the Unique-Cue paradigm is to rule out some of these stimulus-based factors as alternative explanations for the observed pattern of results. One of the main differences between the Unique-Cue procedure and other direct-cue procedures is that the Unique-Cue procedure uses a cue that differs in color from the other cues (red vs. grey). Furthermore, within the Unique-Cue paradigm, mean response times are faster i f target appears at the Unique-Cue location than i f it appears at a Standard-Cue location. Because one of the goals of this paradigm is to interpret this difference as an attention-related effect, it is necessary to establish that this difference is not purely due to the specific stimulus-based differences associated with the Unique Cue. To this end, the present experiment employed a Unique Cue that that was both a different colour (green) and had the same subjective luminance as the Standard Cues. Using a green Unique Cue controlled for the possibility that the response-time difference was simply a red colour-specific effect. Additionally, equating the subjective luminance of the cues also controlled for the possibility that the Unique Cue may have appeared brighter, which may in addition, have yielded larger facilitatory effect at that location (see Experiment 5). Another potential confound involves configural properties that emerge when the target appears at a cued location. Because of the proximity of the horizontal cue and the diagonal target, it is possible to perceive them as a single "angle" shape. If the visual system is somehow more sensitive to this configuration than to just the target alone, this could provide an alternative explanation for why responses are faster i f the target appears at a cued location (Klein, 1996, personal communication). To control for this possibility 38 in the present experiment, the target appeared as a small box above the cue. Upon inspection, the box target did not seem to form as strong a perceptual configuration with the cue as it did when it was a diagonal line. The purpose of this experiment was to eliminate these confounds as possible alternative explanations for the observed cue-effect pattern. It involved a simple detection task with the following controls: the Unique Cue was green instead of red, the subjective luminance of the Unique and Standard cues were matched, and the target was a box instead of a diagonal line. 2.3.1 Methods Subjects: 15 University of British Columbia undergraduates were paid $5 for participating in one 1-hour session. A l l subjects had normal or corrected-to-normal vision. Apparatus and Stimuli: These were the same as described in the General Methods except for the following changes: 1) the Unique-Cue was green instead of red, 2) the subjective luminance of the Unique Cue was matched with the subjective luminance of the Standard (grey) Cues, and 3) the target was a small box (0.2 x 0.2°) instead of a diagonal line. Procedure: The C T O A on data trials was 100 ms. Subjective Luminance Matching Procedure: The procedure was the same as described in the General Methods. Design: Only the Target-Location variable was measured in this experiment. In total, there were 720 data trials consisting of 480 Unique-Cue, 120 Standard-Cue, and 120 39 Uncued trials. Also, 240 catch trials with 1500 ms CTOAs were divided according to the same 4:1:1 trial ratio. 2.3.2 Results and Discussion The mean error rate in this experiment was 1.8%. A one-way repeated measures A N O V A was run on pooled mean response times for all subjects in each condition of the Target Location factor (Unique-Cue, Standard-Cue, or Uncued). Table 3 shows mean response times averaged over all subjects. The main effect of Target Location was significant, F(2,28) - 40.02, MSE = 153.89,/? < 0.001. Mean cue effects are also presented in Table 3. Significant cue effects occurred for both types of cues, which accounts for the main effect of Target Location. Furthermore, cue effects were significantly larger on Unique-Cue trials than on Standard-Cue trials, as indicated by the Unique-Cue Advantage. These data permit the rejection of the notion that the previously discussed stimulus properties were exclusively responsible for the observed response-time pattern. Table 3: Mean Response Times and Cue Effects as a junction of Target Location in Experiment 2 (p-values for planned comparisons in brackets). Target Location Unique Cue Standard Cue Uncued Mean Response Time (ms) 348 371 388 Mean Cue Effect (ms) 40 (< 0.001) 17 (< 0.001) Unique-Cue Advantage (ms) 23 (< 0.001) 2.4 EXPERIMENT 3 The previous experiment discounted some alternative explanations of the observed cue-effect pattern that were based on the stimulus-specific properties present in the display. Another alternative explanation is based on the particular response strategies 40 that subjects could have adopted to perform the task. One possibility is that the observed cue-effect pattern occurred simply because subjects used a serial search strategy to find the target (inspecting each display item separately in sequence until the target was found). Although the unique feature properties of the target suggest that this strategy would be an inefficient way to perform the task (e.g., Treisman & Gelade, 1980), a serial search strategy under the present conditions would exactly predict the observed response-time pattern. This strategy would involve first searching the Unique-Cue location, then all of the Standard-Cue locations, and finally all of the Uncued locations until the target was found. This particular sequence is consistent with the predictions of visual search models that take into account both goal-driven and stimulus-driven factors in visual search tasks (e.g., Wolfe, 1994). One way to test i f a serial search strategy is involved is to vary the number of Standard Cues that appear on a trial. If subjects search all of these locations before moving on to the Uncued locations, then mean cue effects of Standard Cues should decrease as the number of Standard Cues in a trial increased. This is because subjects would, on average, have to search more Standard-Cue locations and fewer Uncued locations before they found the target. This would lengthen mean response times on Standard-Cue trials resulting in a smaller difference between the two (a smaller Standard-Cue effect). In this experiment, the plausibility of the serial search explanation was tested by varying the number of standard cues presented in a trial from 1 to 3 cues. If the serial search hypothesis is correct, then Standard-Cue effects should decrease with increasing numbers of cues. 41 2.4.1 Methods Subjects: 14 University of British Columbia undergraduates were paid $10 for participating in two 1-hour sessions. A l l subjects had normal or corrected-to-normal vision. Procedure: The procedure was the same as described in the General Methods except that the Unique Cue appeared with either 1, 2, or 3 Standard Cues. The C T O A on data trials was 100 ms. Design: A Number of Cues (1, 2, or 3 Standard Cues) variable was crossed with the Target-Location variable. In total, there were 1296 data trials consisting of 288 Unique-Cue, 72 Standard-Cue, and 72 Uncued trials for each level of Number-of-Cues. Also, 432 catch trials with 1500 ms CTOAs were divided according to the same 4:1:1 trial ratio. 2.4.2 Results and Discussion The mean error rate in this experiment was 2.4%. A 3x3 repeated measures A N O V A was run on pooled mean response times for all subjects in each condition. The within-subjects factors were Number of Cues (1,2, or 3 Standard Cues), and Target Location (Unique-Cue, Standard-Cue, or Uncued). Figure 4 shows mean response times averaged over all subjects. While the main effect of Target Location, F(2,26) = 10.22, MSE = 2964.01,p < 0.01, was significant, neither the main effect of Number of Cues, F(2,26) = 1.35, MSE = 72.09,/> = 0.27, nor the Number of Cues x Target Location interaction, F(4i52) = 0.071, MSE = \%l.ll,p = 0.99, was significant. 42 Mean cue effects are presented in Table 4 and in Figure 4. Significant cue effects occurred on both Unique- and Standard- cue trials, which accounts for the main effect of Target Location. Further inspection reveals that cue effects were unchanged across the Number of Cues variable, which indicates that this factor did not affect responses on Standard-Cue trials. This is confirmed by the lack of a Number of Cues x Target Location interaction. Because cue effects did not decrease as a function of Number of Cues, it is possible to reject the notion that subjects used a serial search strategy to respond to the target. Table 4: Mean Cue Effects as a function of Number of Standard Cues in Experiment 3 (p-values for planned comparisons in brackets). Number of Standard Cues Cue Effects Unique Cue 54 (0.004) 51 (0.001) 54 (0.004) Standard Cue 21(0.001) 18 (0.016) 22(0.011) This result is consistent with the findings of another multiple direct-cue experiment that used a similar procedure (Richard, 1995). In particular, this study found that cue effects were unchanged as the number of direct cues presented in a trial varied from two to four. Furthermore, subsequent analysis of these data revealed that cue effects were confined to cued locations, and did not overlap with nearby uncued locations. This pattern was interpreted as consistent with the notion that cue effects were the product of independent processing at each cued location. 43 EXPERIMENT 3: MEAN RESPONSE TIMES LU LU CO z o CL CO LU 400' 390' 380' 370' 360' 350 340 330 320 310 300 UNIQUE CUE STANDARD CUE O - UNCUED O - @, © 1 2 3 NUMBER OF STANDARD CUES CO E, CO h-o LU LU LU O EXPERIMENT 3: CUE EFFECTS 80' 70' 60' 50 40 30 20 10 0 -10 -20 UNIQUE CUE STANDARD CUE © ^ — ® 1 2 3 NUMBER OF STANDARD CUES Figure 4: Mean Response Times and mean Cue Effects as a function of Number of Standard Cues in Experiment 3. 44 It is also possible to investigate the spatial distribution of cue effects in the present experiment. Because the positions of the cues and target were randomly assigned, many different cue-target configurations were possible in this experiment. For example, on Uncued trials, the target could appear adjacent to either type of cue or adjacent to empty locations. If the effects of the cues on target processing extended beyond the cued location, then response times on Uncued trials should differ according to what was adjacent to the target on a particular trial. The spatial distribution of cue effects was examined by reanalyzing the data from the current experiment. This analysis involved separating uncued trials into different categories depending on the target's proximity to other cues. Specifically, uncued trials were separated into five different categories. These were: 1) target adjacent to two empty locations ( E T E ) , 2) target adjacent to the Unique Cue and one empty location (U_T_E), 3) target adjacent to one Standard Cue and one empty location ( S T E ) , 4) target adjacent to two Standard Cues (S_T_S), and 5) target adjacent to the Unique Cue and one of the Standard Cues (U_T_S). The condition in which the target appeared between two empty locations ( E T E ) should provide the most valid baseline measure of performance on Uncued trials. This is because it is the Uncued condition in which the target appears farthest from the other cues and it should be relatively free of any effects that those cues have on target processing. This baseline was used as a comparison to determine if adjacent cues had any effect on target processing at empty uncued locations. Specifically, i f response times were faster on Uncued trials in which the target location was adjacent to one or more cues, then this would suggest that nearby cues influenced processing at uncued locations. Otherwise, i f no differences were found, then this would suggest that cue effects were more location specific. Because of how the experiment design was implemented, not all subjects produced data for each of the five types of Uncued trials. In particular, four subjects had no response times in the S_T_S conditions and four different subjects had no response times in the U_T_S trials. Instead of excluding all subjects with missing data from the analysis (leaving N = 6), subjects were separated into two different sets depending on which type of trials they had data for. The first set excluded the four subjects that did not have responses for the U T S trials. The composition of this set still allowed for meaningful comparisons between the trials involving only Standard Cues and the baseline condition. Similarly, the second set excluded the four subjects that did not have responses for the S_T_S trials (this set included the 6 subjects from the previous set with responses for all trial types). This set permitted comparisons between the trials involving the Unique Cue and the baseline condition. Response times are displayed in Table 5. Table 5: Mean Response Times (in ms) on Cued and Uncued trials as a Junction of the type of stimulus adjacent to the Uncued location (E = empty/no cue; U = Unique Cue; S = Standard Cue) for each subject set. Cued Trials Uncued Trials Unique Standard E T E U T E S T E S T S UTS Sell 313 338 369 363 361 362 — Set 2 308 349 375 368 369 343 The effects of nearby standard cues were small or negligible, including trials in which the uncued location was surrounded by two Standard Cues. If there was some spreading of cue effects beyond the cued location, the magnitude of this effect was small compared to cue effects on trials in which the target appeared directly at the Standard-Cue location (7-8 ms vs. 31 ms). These data therefore suggest that processing facilitation 46 was distributed as a relatively tight gradient around the cued location. This finding is consistent with other experiments which report cue effects that fall off within 1.5 degrees of a low-validity, direct-cue (e.g., Muller & Humphreys, 1991; however some studies do report a broad gradient under similar conditions, e.g., Henderson 1991). The data involving an adjacent Unique Cue were more complicated. The U_T_E condition showed a small advantage relative to baseline (7 ms), which was similar to the pattern observed with Standard Cues. This suggests that Unique Cues in isolation may have a location-specific gradient similar to the one found with Standard Cues. A different pattern occurred, however, i f the target appeared in between the Unique Cue and a Standard Cue. On these trials, target processing at the uncued location was facilitated to a similar extent as in the Standard-Cue valid condition (32 ms vs. 26 ms). This is either an anomalous result, or it suggests a strong spatial interaction between the Unique Cue and nearby Standard Cues. The possibility that nearby Unique Cue and Standard Cues may interact was further investigated by separating Standard-Cue valid trials into different groups depending on what occurred at adjacent locations. Three categories were used: 1) A Standard Cue next to two empty locations, 2) A Standard Cue next to the Unique Cue, and 3) a Standard cue next to one or two other Standard Cues. Mean response times for each condition were 342, 347, and 345 ms respectively. These data suggest that adjacent Unique or Standard cues did not interact with cue effects at a Standard-cue location. Thus, the previously observed Standard-cue/Unique-cue interaction either only occurs over a relatively long spatial range at uncued locations, or it is an anomalous finding. 47 In sum, an analysis of the effects of spatial proximity of cues on Uncued trials seems to provide evidence for location-specific cue effects. This notion is consistent with the idea that the cue effects in the Unique-Cue paradigm can involve spatially restricted processing at multiple cued locations, simultaneously in the visual scene. Moreover, this is inconsistent with the notion that the cue effect pattern observed in this paradigm is solely due to specific visual search strategies. 2.5 EXPERIMENT 4 The results from Experiment 3 suggest that Standard-Cue effects may involve location-specific preattentive processing. Although this view is consistent with the Activity Distribution model, it is also compatible with another non-attentional explanation of direct-cue effects. More specifically, other researchers have proposed that facilitative cue effects triggered by low-validity direct cues may be the product of spatial summation between the cue and the target (e.g., Tassinari et. al., 1994). In spatial summation, the physiological responses evoked by stimuli that appear closely together in space and time interact resulting in a brighter perceived target, which can be responded to faster. In this experiment, researchers found indirect evidence suggesting that facilitative effects on target processing only occurred i f the cue and target overlapped in time. More specifically, the facilitative effect of the cue was minimal i f the cue disappeared from the display before the target appeared (transient cue). In the Unique-Cue paradigm, all cues are sustained. That is, they remain visible until the subject responds to the target, which means that for the entire duration that the target is visible, the cues are also visible. Furthermore, because sensory summation is a spatially local phenomenon (e.g., Hallett, 48 1963), any effects would be confined to the cued location - similar to the spatial pattern observed in Experiment 3. Thus, it is possible that this type of sensory summation may be involved in generating stimulus-driven cue effects in the Unique-Cue paradigm. The purpose of the present experiment was to test the possibility that sensory summation may play a role in producing the direct-cue effects observed in the Unique-Cue paradigm. This was done by including transient-cue trials in which the cues disappeared from the display before the target appeared. If sensory summation is necessary for cue effects to occur, then cue effects should be eliminated or diminished with transient cues. Although the Unique Cue also appears as a direct cue, additional non-stimulus-driven (goal-driven) effects are said to be associated with this type of trial. If this is the case, then eliminating the stimulus-driven component should not necessarily eliminate the goal-driven component unless it is also an artifact of sensory summation. Thus, i f Unique-Cue effects are related to goal-driven processes rather than sensory processes, then these effects should occur with both sustained and transient cues. 2.5.1 Methods Subjects: 14 University of British Columbia undergraduates were paid $5 for participating in one 1-hour session. A l l subjects had normal or corrected-to-normal vision. Procedure: The procedure was the same as described in the General Methods except that on half of the trials (Transient trials) all of the cues disappeared from the display 33 ms after their onset. An equal number of Sustained and Transient trials were randomly intermixed in each block. The CTOA on data trials was 100 ms. 49 Design: A Cue Duration (Sustained or Transient) variable was crossed with the Target-Location variable. In total, there were 900 data trials consisting of 300 Unique-Cue, 75 Standard-Cue, and 75 Uncued trials for each level of Cue Duration. Also, 300 catch trials with 1500 ms CTOAs were divided according to the same 4:1:1 trial ratio. 2.5.2 Results and Discussion The results of one subject were excluded from analysis due to error rates that surpassed 10%. The mean error rate for the remaining subjects was 1.7%. A 2x3 repeated measures A N O V A was run on pooled mean response times for all subjects in each condition. The within-subjects factors were Cue Duration (Sustained & Transient), and Target Location (Unique-Cue, Standard-Cue, or Uncued). Figure 5 shows mean response times averaged over all subjects. Both the main effects of Cue Duration, F ( U 2 ) = 4.92, MSE = 569.26,/? = 0.047, and Target Location, F(2,24) = 42.02, MSE = 182.81,/? < 0.001 were significant. Furthermore, the C T O A x Target Location interaction, F(2,24) = 4.27, MSE = 87.19,/? = 0.026, was also significant. Mean cue effects are presented in Table 6 and in Figure 5. As indicated in Figure 5, the significant effect of Cue Duration seems to be due to generally slower response times on Transient-Cue trials. Additionally, the main effect of Target Location appears to be due to faster response times on Unique-Cue and Standard-Cue trials relative to Uncued trials (see Figure 5). This pattern is supported by significant cue effects in all but Transient Standard-Cue trials (see Table 6). 50 EXPERIMENT 4: MEAN RESPONSE TIMES 400-390-, , 380-to E, 370-ME 360-H 350-LLI 340-CO z o 330-a. CO 320-LU OH 310-300-- • — UNIQUE CUE - © — STANDARD CUE O - UNCUED O - - --O SUSTAINED TRANSIENT CUE DURATION CO E, to o LU U_ LL LU LU Z) o EXPERIMENT 4: CUE EFFECTS 80 70 60 50 40 30 20 10 0 + -10 -20 UNIQUE CUE STANDARD CUE SUSTAINED TRANSIENT CUE DURATION Figure 5: Mean Response Times and mean Cue Effects as a Junction of Cue Duration in Experiment 4. 51 Table 6: Mean Cue Effects as a function of Cue Duration in Experiment 4 fa-values for planned comparisons in brackets). Cue Duration (nis) Cue Effects Sustained Transient Unique Cue 41 (< 0.001) 27 (< 0.001) Standard Cue 18 (< 0.001) 7 (0.082) Unique-Cue Adv. 23 (<0.001) 20 (0.001) Several fmdings are of note in these results. The first is that the small (7 ms) cue effect observed on Transient Standard-Cue trials was not significant. This finding is consistent with the notion that some form of sensory-based, cue-target summation or interaction may be involved in cue effects that occur at low-validity direct-cue locations. In their study, Tassinari et al. (1994) refer to sensory summation in the context of spatial summation with the rods on the periphery of the retina (e.g., Hallett, 1963). It is unlikely, however, that this type of summation is responsible for the cue effects found in studies involving low-validity direct cues. One reason is that, as the authors point out in fheir review, summation effects seem to be a factor in spatial cueing studies that involve simple-target-detection responses but not those involving target-discrimination responses. It seems improbable that a relatively "high-level" factor such as task-type should influence activity at the receptor level. Another reason why sensory summation probably does not involve interactions at the retinal level is that this form of spatial summation yields diminished effects for light waves in the "red" part of the light spectrum (Hallett, 1963). In the context of the Unique-Cue paradigm, this should be associated with reduced summation effects at the red Unique-Cue location, especially when the contribution of other goal-driven response-facilitating factors is eliminated (c.f. Experiment 7b). There is no evidence of this pattern in the current investigation. Thus, although the data are consistent with the idea that some form of cue-target sensory 52 interaction may be involved in direct-cue effects, the underlying sensory-summation process is probably not based exclusively on spatial summation at the retinal level. Although the retinal-level explanation may not hold, the more general notion of a sensory interaction still retains explanatory appeal. Cue effects based on sensory interactions between cues and targets would be associated with properties, such as location specificity and an involuntary initiation, that are consistent with the properties of Standard-Cue effects observed in this investigation. The notion that sensory interactions are involved in stimulus-driven cue effects was investigated further in Experiment 5. The second notable finding in this experiment was that although Unique-Cue effects were diminished with Transient Cues, the Unique-Cue Advantage remained the same. This suggests that using a transient cue did not affect the contribution made by goal-driven processes in Unique-Cue effects. Thus, the notion that Unique-Cue effects are solely due to sensory interactions can be rejected. Similarly, the fact that the Unique-Cue Advantage did not change with transient cues also has implication for the interpretation of how stimulus-driven and goal-driven processes interact in this paradigm. If stimulus-driven cue effects are assumed to have decreased by the same amount at the Unique-Cue and the Standard-Cue locations, then the unchanging Unique-Cue Advantage would suggest an additive relationship between the two types of cue effects (cf. Riggio & Kirsner, 1997). That is, cue effects at the Unique-Cue location may be composed of separate and independent stimulus-driven and goal-driven components and the decrease in Unique-Cue effects was due to a corresponding decrease in the stimulus-driven component. 53 The Unique-Cue paradigm has the potential to provide an informative method for investigating stimulus-driven and goal-driven influences on the control of attention orienting. The purpose of this section was to establish the validity of the Unique-Cue paradigm for this task. The results from this section suggest that the Unique-Cue effect pattern is both replicable and not simply an artifact of non-attentional stimulus-display or search-strategy factors. 54 3. S T I M U L U S - D R I V E N E F F E C T S The experiments from the previous chapter revealed several properties of stimulus-driven cue effects. The purpose of the experiments in this chapter was to explore these properties in further detail. In particular, they focused on the notion that stimulus-driven cue effects may be mediated by preattentive processes that operate independently and simultaneously across the visual field. Several findings from the earlier experiments are consistent with this idea. These include data indicating that stimulus-driven cue effects occur in the absence of "top-down" goals to attend to the cues, that they seem to be location specific, that they may be affected by sensory interactions between cues and targets, and that they have an additive relationship with goal-driven processing facilitation. Because the emphasis in this chapter was on the stimulus-driven cue effects observed in this paradigm, the Unique Cue was not used in the experiments reported here. This was done in order to minimize the influence of goal-driven processes on cue effects at Standard-Cue locations. 3.1 EXPERIMENT 5 In Experiment 4, Standard-Cue effects were small and non-significant on Transient-Cue trials. Although spatial summation at the retinal level is probably not responsible for this result, some form of sensory interaction at higher levels of visual processing remains as a potential explanation. As the results of the Experiment 4 and Tassinari et. al.'s (1994) experiment suggest, cue-target overlap is one method that taps into this sensory interaction. Another 55 method that may also tap into this sensory interaction is the direct manipulation of stimulus luminance. In this case, different luminance magnitudes may represent intermediate points on the same continuum defined by the presence (sustained cue) or absence (transient cue) of sensory-related activity generated by the cue. The magnitude of the luminance change may affect the degree to which the cue-related sensory activity persists at the time of the target onset to interact with target-related sensory activity. Support for relevance of stimulus luminance comes from one study in which the luminance of a stimulus directly affected its ability to capture attention (e.g., Theeuwes, 1995; however see Yantis & Hillstrom 1994). That is, targets associated with greater luminance changes 'stood out' more effectively in a field of similar distracters and were found more easily. Moreover, the effectiveness of attentional capture was described by a monotonic function that increased directly as a function of stimulus luminance. The purpose of the present experiment was to provide further support for the notion that sensory interactions, specifically those associated with luminance changes, may be involved in direct-cue effects. To this end, cue luminance was varied randomly across trials3 (all cues had the same luminance within a trial). If luminance-based sensory interactions do play a direct role in cue effects, then cue-effect magnitudes should vary as a function of cue luminance. To ensure that goal-driven factors did not affect the results, the Unique Cue was replaced with another Standard cue. With multiple uninformative direct cues, subjects should have neither any single unambiguous attention-shift destination, nor any reason to ' This procedure was chosen over the alternative procedure of giving each cue a different luminance on each trial for two reasons. The first was to minimize the effects of the spatial configuration of the cues on brightness (e.g., the dimmest cue might appear to have a different brightness if it was adjacent to the brightest cue compared to if it was 56 engage goal-driven processing at any of the cued locations. This procedure was expected to provide a relatively unbiased indicator of stimulus-driven direct-cue effects. 3.1.1 Methods Subjects: 25 Simon Fraser University undergraduates were paid $5 for participating in one 1-hour session. A l l subjects had normal or corrected-to-normal vision. Stimuli: Each set of cues presented on a trial had one of four possible luminance levels (0.07, 1.52, 6.29, 14.4 cd/m2). The background luminance was 0.8 cd/m2 and the luminance of the target was 55.52 cd/m 2. Procedure: The procedure was the same as described in the General Methods except that four Standard Cues were presented on each trial, and no Unique Cue was used. Subjects were given 10 minutes to dark adapt before the experiment began. The CTOA on data trials was 100 ms. Design: A Luminance variable (1,2, 3, or 4) was crossed with a Target-Location variable (Standard-Cue or Uncued) that reflected where the target appeared in the display. In total, there were 720 data trials consisting of 90 Standard-Cue, and 90 Uncued trials for each level of Luminance. Also, 240 catch trials with 1500 ms CTOAs were divided according to the same 1:1 trial ratio. 3.1.2 Results and Discussion The results from 4 subjects were excluded from analysis due to error rates that surpassed 10%. The mean error rate for the remaining subjects was 2.0%. adjacent to empty locations). The second reason was to eliminate the possibility that the differentiation between cues could lead to the singling out of one of the cues (the brightest) as a stimulus that could capture attention. 57 A 4x2 repeated measures A N O V A was run on pooled mean response times for all subjects in each condition. The within-subjects factors were Cue Luminance (1,2, 3, & 4), and Target Location (Standard-Cue, or Uncued). Figure 6 shows mean response times averaged over all subjects. The main effect of Cue Luminance, FQ^Q) ~ 33.98, MSE = 96.78,/? < 0.001, was significant. This arises from a general decrease in response times as Cue Luminance increases, which probably reflects an alerting effect of brighter cue luminance. Additionally, the main effect of Target Location, F(i,20)= 49.95, MSE = 340.83,/? < 0.001 was also significant and this arises from generally faster response times on Standard-Cue trials. Finally, the Cue Luminance x Target Location interaction, FQ, SO) = 18.74, MSE - 72.88,/? < 0.001, was significant. This seems to be due to cue effects that increase as a function of Cue Luminance. An examination of the cue effects (Table 7) indicates that only the cue effects at the two higher luminance levels were significant. Table 7: Mean Cue Effects as a Junction of Cue Luminance in Experiment 5 (p-values for planned comparisons (Standard-Cue vs. Uncued trials) in brackets). Cue Luminance Cue Effects Standard Cue A (0.129) 6 (0.057) 12 (0.003) 20 (<0.001) Cue effects in this experiment increased directly as a function of cue luminance, which is consistent with the notion that sensory interactions between the cue and target may have been responsible for generating these effects. It is also possible to argue that goal-driven processes were not involved in producing this luminance effect. More specifically, other research involving cue luminance shows that, under conditions conducive for producing goal-driven cue effects (symbolic cues with 1-2 sec CTOAs), cue luminance affected overall reactions times in an additive manner (similar to an alerting effect), but did not directly affect cue-effect magnitudes (Hughes, 1984). In 58 EXPERIMENT 5: MEAN RESPONSE TIMES 400-390-[ms) 380-LU 370-360-1-LU 350-CO z 340-o D. 330-CO LU 320-01 310-300-STANDARD CUE O - UNCUED O - - . 1 2 3 CUE LUMINANCE CO CO o LU LU LU D O EXPERIMENT 5: CUE EFFECTS 80 70 60 50 40 30 20 10 1 -10-1 -20 © — - © i i i 1 2 3 CUE LUMINANCE Figure 6: Mean Response Times and mean Cue Effects as a Junction oj Cue-Luminance level in Experiment 5. 59 contrast, cue luminance directly affected cue-effect magnitudes in the present experiment,which suggests that the processes affected by luminance differ from those involved in goal-driven cue effects. The dependence of cue effects on luminance also provides a mechanism to explain the diminished Transient-Cue effects observed in Experiment 4. The brief sensory stimulation associated with Transient cues may have been represented in the visual system as having a lower overall "luminance energy" integrated over time. Consequently, a Transient cue might have the same effectiveness as a low-luminance direct cue. 3.2 EXPERIMENT 6 Based on findings so far in this investigation, it might be possible to posit a direct, perhaps involuntary link between the onset of the direct cues and the occurrence of cue effects. In particular, direct cue effects vary in magnitude as a function of cue luminance, they are confined in space to a region directly around the cue, and they occur even if there is no "top-down" incentive to use the cues. In other reports, similar findings have led to conclusions that direct cues automatically trigger cue effects or that they automatically capture attention (Jonides, 1981, Yantis & Jonides, 1984; however, see Yantis & Jonides, 1990). This purely involuntary account of stimulus-driven cue effects would predict that cue effects would occur at all locations associated with an abrupt luminance change. According to this view, it should be possible to observe cue effects at many locations 60 simultaneously, with the only limiting factor being the number of different visual-field positions that could be individually represented in the visual system. A n alternative possibility is that cue effects may only occur at a limited subset of the cued locations. This notion is based on the results of several studies involving stimuli associated with abrupt luminance changes, which find that the processing of these stimuli may be mediated by a limited-capacity tracking mechanism. More specifically, i f target items have abrupt onsets, observers can simultaneously track with accuracy, up to four or five of them as they move in a quasi-random manner in a field of other moving items (Pylyshyn & Storm, 1988; however see Yantis, 1992). Similarly, observers can also rapidly count up to four of five abrupt-onset items in a parallel manner (Trick & Pylyshyn, 1993), whereas counting seems to be serial i f the stimuli do not have an abrupt onset (Wright & Richard, 1995). Finally, up to four or five abrupt-onset items seem to be searched with higher priority than items that do not have an abrupt onset (Yantis & Johnson, 1990; Yantis & Jones, 1991). In all these situations, target sets that contain more than four or five abrupt-onset stimuli seem to be processed differently (they contain a serial-search component). Because the direct cues used in this paradigm also appear abruptly in the display, it is possible that the number of locations at which cue effects can occur is also mediated by this capacity limit. The purpose of this experiment was to determine if there is a limit to the number of cued locations at which direct-cue effects could occur independently. This was done by varying the number of direct cues presented in the display from one to eight. If cue effects occur at all cued locations regardless of number, then mean cue effects should remain constant across the number of cues presented because each cued location would 61 independently trigger stimulus-driven processing (c.f. Experiment 3). In contrast, i f there is a limit, then average cue-effect magnitudes should begin to drop once this limit is surpassed. For example, i f the limit extends to five locations, then cue-effect magnitudes should be the same for up to five cues. On trials with more than five cues, however, mean cue effects should begin to drop as a function of the number cues over the limit. The reason for this is that if there are more cued locations than locations at which cue effects can occur, then on some proportion of trials, the target will appear at "non-cue-effected", cued location. This would result in average response times for those trials that were composed of a mixture of "cue-effected" (same as cued) and "non-cue effected" (same as uncued) response times, yielding reduced mean cue effects. Thus, a drop in cue-effect magmtudes would be consistent with the notion of a capacity limit. 3.2.1. Methods Subjects: 28 Simon Fraser University undergraduates were paid $10 for participating in two 1-hour sessions. A l l subjects had normal or corrected-to-normal vision. Stimuli: The stimulus-display used in this experiment configuration differed from the circular array used in the other experiments. In particular, to accommodate the greater number of cues, twelve possible cue and target positions were used instead of eight. The positions were arranged in a circle array like the face of an analogue clock. Additionally, to maintain the same distance between adjacent positions as the other experiments (5.5°), all positions were moved farther from centre so that they were (8.1°) from the central fixation point. A l l other aspects of the stimulus display were as described in the General Methods. 62 Procedure: The procedure was the same as the previous experiment except that one to eight Standard Cues were presented on each trial and cue luminance did not change (it was the same as the Standard-Cue luminance in the other experiments). Design: A Number of Cues variable (1, 2, 3,4, 5, 6, 7, & 8) was crossed with a Target-Location variable (Standard Cue or Uncued) that reflected where the target appeared in the display. In total, there were 960 data trials consisting of 60 Standard-Cue, and 60 Uncued trials for each level of Number of Cues. Also, 320 catch trials with 1500 ms CTOAs were divided according to the same 1:1 trial ratio. 3.2.2 Results and Discussion The results from 3 subjects were excluded from analysis due to error rates that surpassed 10%. The mean error rate for the remaining subjects was 2.3%. A n 8x2 repeated measures A N O V A was run on pooled mean response times for all subjects in each condition. The within-subjects factors were Number of Cues (1, 2,3, 4, 5, 6, 7, & 8), and Target Location (Standard-Cue, or Uncued). Figure 7 shows mean response times averaged over all subjects. The main effect of Number of Cues, F(7y JSS) -4.254, MSE = 247.51,/? = 0.002, was significant. This seems to be due to a general decrease in overall response times on multiple-cue (2-8 cues) trials. Additionally, the main effect of Target Location, F(i,24) = 46.01, MSE = 842.56, p < 0.001 was also significant and this seems to be due to generally faster response times on Standard-Cue trials. Finally, the Number of Cues x Target Location interaction, F(7i m> = 6.21, MSE = 267.11,/? < 0.01, was significant. This seems to be due to the drop in cue effects after the one- and five- cue conditions. Mean cue effects are presented in Table 8 and in Figure 7. 63 EXPERIMENT 6: MEAN RESPONSE TIMES 450-1 440-430-CO E, 420-ME 410-1- 400-UJ 390-CO z 380-O D_ CO 370-LU a: 360-350-- © — STANDARD CUE O - UNCUED -O -O ,o ©- O — Q _ - ©- -© I I 1 — I 1 - 1 1 1 1 2 3 4 5 6 7 8 NUMBER OF STANDARD CUES CO E, CO H-O LU LU LU ZD o EXPERIMENT 6: CUE EFFECTS 80' 70' 60 50 40 30 20 10 0 + -10 -20 1 2 3 4 5 6 7 8 NUMBER OF STANDARD CUES Figure 7: Mean Response Times and mean Cue Effects as a function of Number of Standard Cues in Experiment 6. 64 Table 8: Mean Cue Effects and p-values for planned comparisons (Standard-Cue vs. Uncued trials) as a function of the Number of Standard Cues in Experiment 6. Number of Standard Cues 1 2 3 4 5 6 7 8 Cue Effects 43 26 23 20 19 8 10 9 p <0.001 <0.001 <0.001 0.001 0.001 0.184 0.069 0.048 Two findings were of interest in this experiment. The first was the decrease in cue effects on trials involving more than five cues. The second was the larger cue effect observed on single-cue trials. The results indicate that cue effects were approximately equal on multiple-cue (2-8 cues) trials for up to five cues. On trials with more than five cues, however, cue effects were diminished and not significant. This pattern is consistent with the idea that there is a limit to the number of cued locations at which direct-cue effects can occur. What is the source of the capacity limit? It is unlikely to occur at the earliest stages of visual processing where processing occurs in parallel across the visual scene. Otherwise, why should an area that is responsive when a single cue appears at that location no longer be responsive when one of eight cues appears at that location (while other locations remain responsive)? It makes more sense to posit that the same activity occurs at that level whenever a cue appears, but that the ability of that activity to affect processing at higher levels is nullified by a limited-capacity structure (c.f, Broadbent, 1958). That is, although the processes involved in generating cue effects may still occur at all cued locations, the links to higher-level processes that enable the expression of these effects may be limited to five cued locations. The advantage of this idea is that it permits an involuntary association between an abrupt stimulus onset and the initiation of stimulus-driven processing, but still allows a limit on the number of affected locations. 65 One mechanism that has been proposed to mediate abrupt-onset capacity limitations involves spatial indexes (e.g,. Pylyshyn, 1989; Pylyshyn, Burkell, Fisher, Sears, Schmidt, & Trick, 1994; Wright, 1994). Indexes are said to encode the location information of important or significant stimuli in the visual field (e.g., abrupt onsets). If other processes require access to the locations of these stimuli, then they can reference the indexes for immediate access. In other words, indexes are said to function by providing a link between higher level processes and the locations associated with important visual stimuli. Based on the findings of research involving abrupt-onset stimuli (e.g., Pylyshyn & Storm, 1988), some researchers have posited that the number of available indexes is limited to around four or five (e.g., Pylyshyn 1989; Pylyshyn et. al., 1994). This limit provides an explanation for the current results. More specifically, i f more than four or five abrupt onsets appear in a display, then only a subset of these stimuli can be indexed. Consequently, while processing may still occur at unindexed locations, without an index, the results of that processing can not be passed on to affect higher levels of processing. Thus, a target appearing at an unindexed cued location would not benefit from the stimulus-driven processing initiated by the cue onset. Another notable finding in this experiment is the larger cue effect on single-cue trials. This result replicates the cue effect pattern obtained in another multiple direct-cue experiment that used a similar procedure (Richard, 1995). It suggests that, on single-cue trials, some form of additional processing advantage was available that was not available on multiple-cue trials. One possibility is that the single cue may have captured attention because it represented a single potential shift destination and because it had a relatively 66 higher cue validity4. These two properties make the single cue analogous to the Unique-Cue (because it can also be singled out based on its colour and it has a higher cue validity), thus it is not surprising that cue effects were larger on these trials. The possibility that attentional or goal-driven operations were only involved on single-cue trials has consequences for the interpretation of the effects of these processes. More specifically, the larger cue effect magnitude on single-cue trials was due to relatively slower responses on single Uncued trials and not due to faster responses on single Standard-Cue trials (see Figure 7). This suggests that the effect of these goal-driven or attention-related processes was not to facilitate processing at the cued location, but rather to inhibit processing at uncued locations. This finding is consistent with the view of attention as a selective process. In this case, focusing attention at a location inhibits processing at unattended locations (c.f. Broadbent, 1958; Treisman 1960; Yantis & Jonides, 1990). A more focused investigation on the properties of stimulus-driven cue effects in this paradigm reveals that while luminance-dependent effects suggests an involuntary relationship between direct cues and stimulus-driven cue effects, this relationship seems to be mediated by a limited-capacity mechanism. 4 Because the target appeared at a cued location on 50% of the trials, the cue had a higher cue validity on single-cue trials than any particular cue on multiple-cue trials (50% vs. 25% or less cue validity). Thus, in the context of the experiment, this higher cue validity may have encouraged subjects to voluntarily attend to the location of the single cue. 67 4. GOAL-DRIVEN EFFECTS The working hypothesis in the Unique-Cue paradigm is that both stimulus-driven and goal-driven factors contribute to cue effects. The involvement of stimulus-driven factors is supported by data indicating that Standard-Cue effects occur even though subjects have no incentive to attend to those locations, and also that these effects are closely tied to stimulus properties (they are location specific and affected by cue luminance). Up to this point, however, the involvement of additional goal-driven processing associated with the Unique-Cue has largely been an unsubstantiated assumption, made based on task conditions. More specifically, because the Unique-Cue was highly valid (making it useful for carrying out the task) and because subjects were given explicit instructions "to focus on the red cue," it was assumed that they would voluntarily engage the processing required to focus attention at that location. Although these procedures seem sufficient to invoke goal-driven processing, it is still necessary to confirm this assumption before making any further claims about goal-driven involvement in Unique-Cue effects. The experiments in this chapter represent attempts to find evidence for the involvement of goal-driven processes in Unique-Cue effects by using alternative methods to indirectly induce subjects to initiate goal-driven processing at the Unique-Cue location. This means that the explicit instruction to "focus" on the Unique Cue given to subjects in previous experiments was not included in these experiments. Instead, the experiments in this chapter employed other manipulations to direct goal-driven processing at the Unique-Cue location. 68 4.1 EXPERIMENT 7A As discussed in the introduction, one factor that affects goal-driven cue effects is cue validity. Cue validity determines how useful a cue is for predicting the location of an impending target. If a cue is not highly valid, then subjects have no incentive to use the cue to voluntarily prepare their attention at the indicated location. Consequently, with symbolic cues, which depend on goal-driven preparation for their effectiveness (e.g. voluntarily interpreting the cue, computing a shift destination and making a shift), cue effects do not occur under low-validity conditions (e.g. Jonides, 1981). This contrasts with direct cues, whose effects occur regardless of cue validity (e.g. Jonides, 1981), presumably because these effects are involuntarily triggered by the cue onset. The purpose of the present experiment was to validate the assumption that goal-driven operations are involved in the larger Unique-Cue effects by varying the validity of the Unique Cue. With low cue validity, subjects should not find the Unique Cue particularly useful for performing the task, and should therefore not "waste" unnecessary mental processing invoking goal-driven preparations at that location (it would be more efficient just to wait for the target onset). Consequently, goal-driven cue effects should not occur on these trials. In contrast, with high Unique-Cue validity the cue would be useful and it would be worth expending the mental effort required to prepare for the target onset at the indicated location, which would result in additional goal-driven cue effects at that location. This experiment used three different levels of Unique-Cue validity (33, 66, or 80% valid). It was predicted that the Unique-Cue Advantage, which reflects goal-driven processing, would be negligible on 33% valid trials and would get progressively larger on 69 the higher validity trials as the cue became more useful. Another prediction is that Unique-Cue validity should have no effect on Standard-Cue effects because these are stimulus-driven and independent of top-down processing. 4.1.1 Methods Subjects: 21 University of British Columbia undergraduates were paid $15 for participating in three 1-hour sessions. A l l subjects had normal or corrected-to-normal vision. Apparatus and Stimuli: These were the same as described in the General Methods. Procedure: The procedure was the same as described in the General Methods except that cue validity varied between sessions. This separation was done to maintain a consistent incentive regime that would promote uniform goal-driven conditions within a session. The order in which subjects ran in each cue validity session was counterbalanced so that each permutation of sequence order occurred at least twice. The number of subjects starting in each cue validity condition was also equated. Another major difference was that subjects were only informed about the probability that the target would appear at each type of location, and they were not given any special instructions or incentive to focus on the Unique-Cue location. The CTOA on data trials was 100 ms. Design: The Target-Location variable was completely crossed with a Unique-Cue Validity variable (33, 66, or 80%) that represented the proportion of trials in which the target appeared at the Unique-Cue location. On 33%-valid trials, the target appeared at the Unique-Cue location on 33.3% of trials (120 data trials), at one of the Standard-Cue locations on 33.3% of trials (120 data trials), or at one of the Uncued locations on 33.3% 70 of trials (120 data trials). On 66%-valid trials, the target appeared at the Unique-Cue location on 66.7% of trials (480 data trials), at one of the Standard-Cue locations on 16.7% of trials (120 data trials), or at one of the Uncued locations on 16.7% of trials (120 data trials). And on 80%-valid trials, the target appeared at the Unique-Cue location on 80% of trials (792 data trials), at one of the Standard-Cue locations on 10% of trials (99 data trials), or at one of the Uncued locations on 10% of trials (99 data trials). In addition to data trials, 25% of the trials in each experiment were catch trials with 1500 ms CTOAs. Catch trials were presented in the same proportions as indicated by the value of the Unique-Cue Validity level. 4.1.2 Results and Discussion The results of three subjects were excluded from analysis due to error rates that surpassed 10% in one or more of the sessions. The mean error rate for the remaining subjects was 1.7%. A 3x3 repeated measures A N O V A was run on pooled mean response times for all subjects in each condition. The within-subjects factors were Unique-Cue Validity (33%, 66%, or 80%), and Target Location (Unique-Cue, Standard-Cue, or Uncued). Figure 8 shows mean response times averaged over all subjects. The main effect of Unique-Cue Validity was not significant, FQM) = 1.43, MSE - 2230.20,p = 0.25, however, both the main effect of Target Location, FQM) = 64.45, MSE = 164.26,/? < 0.01, and the Unique-Cue Validity x Target Location interaction, F(4I68) = 8.26, MSE = 104.36,/? < 0.01, were highly significant. 71 Mean cue effects are presented in Table 9 and in Figure 8. Significant cue effects occurred in all conditions, which accounts for the main effect of Target Location. In addition, the significant Target Location by Unique-Cue Validity interaction suggests that cue effects increased with increasing Unique-Cue Validity. Table 9: Mean Cue Effects as a Junction of Unique-Cue Validity in Experiment 7a (p-values for planned comparisons in brackets). UNIQUE-CUE VALIDITY Cue Effects 33 % 66% 80% Unique Cue 15 (< 0.001) 25 (< 0.001) 42 (< 0.001) Standard Cue 11 (< 0.001) 16 (< 0.001) 24 (< 0.001) Unique Cue Adv. 4 (0.53) 9 (< 0.001) 18 (< 0.001) Although the finding in this experiment that Unique-Cue Validity affected Unique-Cue effects is consistent with the hypothesis that goal-driven processes are associated with Unique-Cue effects, these results must be interpreted with caution. One reason for this is that Unique-Cue Validity also directly affected the magnitude of Standard-Cue effects. This finding was confirmed by an additional 2-way A N O V A run on the mean response times from the Standard-Cue and Uncued conditions of the Target Location factor and the Unique-Cue Validity factor. There was a significant Target Location by Unique-Cue Validity interaction, F(2,34) = 3.47, MSE = 96.62,/? = 0.039, indicating Standard-Cue effects changed over Unique-Cue Validity. If Standard-Cue effects were purely stimulus driven, then they should be unaffected by a "top-down" factor such as Unique-Cue Validity. This irregularity is joined by another unexpected result involving uncharacteristically small Standard-Cue effects on 33%-validity trials. In the previous experiments, Standard-Cue effects were nearly twice as large unless they were attenuated by sensory-based factors. These observations suggest that subjects may 72 EXPERIMENT 7a: MEAN RESPONSE TIMES CO E, LU 380 370H 360 350 340 H 330 LU CO 320 O 310 Q_ CO 300 LU , a: 290-I 280 - • — UNIQUE CUE - © — STANDARD CUE O - UNCUED O 33% 66% 80% UNIQUE-CUE VALIDITY CO E, CO I -o LU LU LU Z> o EXPERIMENT 7a: CUE EFFECTS 80 70 60 50 40 30 20 10 0 + -10 -20 UNIQUE CUE STANDARD CUE 33% 66% 80% UNIQUE-CUE VALIDITY Figure 8: Mean Response Times and mean Cue Effects as a function of Unique-Cue Validity in Experiment 7a. 73 not have performed the target detection task in this experiment in the same manner as they performed it in the previous experiments. A n explanation for this performance difference may be found in one of the major procedural differences between the current experiment and the previous experiments. More specifically, in this experiment subjects were not given an explicit response strategy to use (they were not instructed to "focus" on the Unique Cue). This, coupled with the fact that cue validity is said to affect "top-down" processing, suggests that higher-level strategic factors may have influenced the current results. One possibility is that Unique-Cue Validity affected the probability that subjects used a response strategy that did not produce cue effects. In other words, under some conditions, subjects may have ignored cue-related processing and used other visual information to direct their responses5. This alternative strategy is made possible by the fact that the target differed from all other stimuli in the display because it was defined by a unique set of features (it was the only item that was white or diagonal). Other researchers (e.g., Treisman & Gelade, 1980) have proposed that this type of feature information is collected rapidly and in parallel across the visual scene, and that it is also stored in separate representations based on particular feature dimensions (e.g., separate colour and line-orientation representations). Furthermore, this information is said to be accessible by other higher-level processes, which enables the strategy of detecting a unique target simply by "checking" these representations for the presence of a specific feature (Treisman & Gelade, 1980). Using a strategy that involves "feature checking" 5 This notion is supported by several pilot experiments of similar design that did not involve giving subjects special instructions to "focus" on the Unique Cue. In these experiments, cue effects were absent in all cue conditions, suggesting that subjects employed a different response strategy than the one used in experiments involving the "focus" instructions. 74 should not produce cue effects because it bypasses the processing pathway responsible for generating these effects. If subjects used a "feature-checking" strategy on some proportion of trials in the lower Unique-Cue Validity conditions, then mean response times on these trials would be a composite of trials in which cue effects did and did not occur. The more often that subjects used this checking strategy, the smaller that cue effect would be. Thus, it is possible that Unique-Cue Validity may not have affected the probability that subjects attended to the Unique-Cue location, but rather the likelihood that subjects employed a feature-checking response strategy. 4.2 EXPERIMENT 7B The idea just discussed that different response strategies may have been involved in task performance is similar to another proposal that holds that some visual search tasks can be performed with different strategies, depending on task parameters (e.g., Bacon & Egeth, 1994). According to this view, visual search can be carried out by either searching for a unique target feature, called feature-search mode, or by allowing attention to be captured by the most visually salient stimulus in a display, called singleton-detection mode. Although the singleton-detection mode seems to be the default strategy under most conditions (see Theeuwes, 1991a, 1992), one study showed that subjects could be forced into using a feature-search strategy with a stimulus display that rendered the singleton-detection strategy ineffective (Bacon & Egeth, 1994). If subjects in Experiment 7a probabilistically shifted between a feature-search and an attention-related strategy, it should be possible to use a similar approach to force 75 subjects to exclusively use an attention-related response mode (the mode that subjects seemed to consistently employ in the previous experiments). One way to do this is by requiring subjects to perform an identification task rather than a detection task. As shown in Experiment 1, Unique-Cue effects also occur with identification tasks. More importantly, because this task requires spatially-localized processing at the target location to identify the target orientation, the feature-search strategy of simply checking feature-based representations for the presence of the unique target feature should not work (e.g., Treisman & Gelade, 1980). Thus, the identification task should force subjects to use a response strategy that involves focusing attention at the target location on all trials. This procedure should not only confirm the dual-strategy hypothesis, but also permit a valid reexamination of the predictions from the previous experiment. More specifically, this involves the prediction that cue validity should only affect the additional goal-driven cue effects associated with the Unique-Cue location. Two levels of Unique-Cue Validity were used in this experiment. While the High Unique-Cue Validity condition was the same as in the previous experiment (80%), the Low Unique-Cue Validity condition had a validity that was lower than the previous experiment (12.5%). This was to maximize the "top-down" incentive to use the feature-search strategy. The greater incentive should not matter, however, i f the identification task makes this strategy ineffective. This is because subjects will have no choice but to use the attention-related strategy to respond to the target. Thus, the use of this strategy would be verified by the occurrence of Standard-Cue effects at their expected levels (approximately 20 ms) in all Unique-Cue Validity conditions. 76 In a departure from the procedure used in the previous experiments, eye movements were monitored in this experiment. This was to control for a potential confound in the data introduced by eye movements. More specifically, one explanation for the larger cue effects on Unique-Cue trials is that subjects may have moved their eyes or prepared to move their eyes to the Unique-Cue location in advance of the target onset. This is especially problematic in the present experiment for two reasons. The first is that the unusually high Unique-Cue Validity gives subjects even more incentive to move their eyes to the Unique-Cue location, and the second is that because the identification task requires more detailed analysis, foveating the target would facilitate performance. To rule out this explanation, eye movements were monitored and trials in which they occurred were excluded from analysis. 4.2.1 Methods Subjects: 15 University of British Columbia undergraduates were paid $10 for participating in two 1-hour sessions. A l l subjects had normal or corrected-to-normal vision. Apparatus and Stimuli: These were the same as described in the General Methods. Eye Movement Monitoring: See General Methods. Procedure: The procedure was the same as in the previous experiment except that subjects identified the orientation of the target line by pressing the appropriate button. If subjects pressed the wrong button on a trial, they received a feedback tone. The only other differences were that eye movements were monitored, and that subjects ran in only two sessions. The CTOA on data trials was 100 ms. 77 Design: The Target-Location variable was completely crossed with a Unique-Cue Validity variable (12.5%, or 80%) that represented the proportion of trials on which the target appeared at the Unique-Cue location. On 12.5%-valid trials, the target appeared at the Unique-Cue location on 12.5% of the trials (60 trials), at one of the Standard-Cue locations on 37.5% of the trials (180 trials), or at one of the Uncued locations on 50% of the trials. This means that the target was equally likely to appear at any of the display locations on a given trial in this condition. On 80%-valid trials, the target appeared at the Unique-Cue location on 80% of trials (480 data trials), at one of the Standard-Cue locations on 10% of trials (60 data trials), or at one of the Uncued locations on 10% of trials (60 data trials). In addition to data trials, 25% of the trials in each experiment were catch trials with 1500 ms CTOAs. Catch trials were divided up into the same proportions as indicated by the value of the Unique-Cue Validity level. 4.2.2 Results and Discussion The results of one subject were excluded from analysis due to error rates that surpassed 10% in one or more of the sessions. The mean error rate for the remaining subjects was 2.0%. Eye movement occurred on less than 1% of the trials for all subjects. If subjects failed to make a significant number of eye movements in this experiment, in which the incentive to do so was maximized, it is probably safe to assume that they did not move their eyes in the other experiments that involved similar conditions (e.g., short CTOAs). A 2x3 repeated measures A N O V A was run on pooled mean response times for all subjects in each condition. The within-subjects factors were Unique-Cue Validity 78 (12.5% or 80%), and Target Location (Unique-Cue, Standard-Cue, or Uncued). Figure 9 shows mean response times averaged over all subjects. While the main effect of Unique-Cue Validity was not significant, F(U3) = 0.90, MSE = 3191.47,/? = 0.360, the main effect of Target Location was highly significant, F(2,26)= 25.69, MSE = 265.70,/? < 0.001. This arose from faster response times on cued trials. Furthermore, the Unique-Cue Validity x Target Location interaction, F(2, 26j = 5.92, MSE = 180.24,/? = 0.008, was also significant. This was due to the larger Unique-Cue effect on 80%-valid trials. Mean cue effects are presented in Table 10 and in Figure 9. Table 10: Mean Cue Effects as a Junction oj Unique-Cue Validity in Experiment 7b (p-values for planned comparisons in brackets). Unique-Cue Validity Cue Effects 12.5% 80% Unique Cue \9(<0.001) 42 (< 0.001) Standard Cue 17 (< 0.001) 20 (0.022) Unique Cue Adv. 2 (0.453) 22 (0.004) In contrast to the previous experiment, Standard-Cue effects did not change with Unique-Cue Validity. This finding is consistent with the notion that subjects did not . alternate probabilistically between a feature-detection and an attention-related strategy. The reason for this is that switching between strategies would have been associated with smaller mean Standard-Cue effects on Low Unique-Cue validity trials as these would be "watered down" by the inclusion of trials in which no cue effects occurred because the feature-detection strategy was used. Thus, the finding that Unique-Cue validity did not affect Standard-Cue effects supports the notion that the identification task forced subjects to consistently use a response strategy that involved attending to the target location. 79 EXPERIMENT 7b: MEAN RESPONSE TIMES 550' 540' 530' 520' 510' £: I - 500 LU CO 490 CO E, LU O D_ CO LU OH 480 470 460 450 - • — UNIQUE CUE - © — STANDARD CUE O - UNCUED . O 12.5% 80% UNIQUE-CUE VALIDITY CO E, CO H o LU LU LU O EXPERIMENT 7b: CUE EFFECTS 80 70 60 50 40 30 •{ 20 10 0 + -10 -20 UNIQUE CUE STANDARD CUE 12.5% 80% UNIQUE-CUE VALIDITY Figure 9: Mean Response Times and mean Cue Effects as a function of Unique-Cue Validity in Experiment 7b. 80 The current results are also consistent with the assumption that goal-driven processes are involved in the larger Unique-Cue effects. On Low-Validity trials, subjects had no incentive to initiate additional top-down processing at the Unique-Cue location. This was reflected by the complete absence of a Unique-Cue Advantage on these trials. In contrast, on High-Validity trials, the strong incentive to use the cue yielded a large Unique-Cue Advantage. Thus, the effects of varying Unique-Cue Validity support the notion that addition goal-driven processing was involved at the Unique-Cue location. Also of interest in this experiment is the relationship between overall mean response times and Unique-Cue Validity. Assuming that attention was involved on High-Validity trials, then mean response times can provide information about the absolute effects of attention on visual processing. In particular, i f attention has a facilitatory effect on processing at the attended location, then valid-cue response times should be faster when attention is involved (High-Validity Unique Cue trials) than when attention is not involved (Low-Validity Unique Cue trials). On the other hand, i f attention has an inhibitory effect on processing at unattended locations, then response times should be slower at unattended locations (Standard-Cue and Uncued trials). The data are consistent with the latter pattern. Moreover, a similar inhibitory pattern was also observed in Experiments 6 and 7a, which suggests that visual attention in the Unique-Cue paradigm operates not as a facilitatory processing focus but as a location-based filter that inhibits processing at unattended locations. 81 4.3 EXPERIMENT 8 The previous experiment implicated goal-driven processing in Unique-Cue effects by using Unique-Cue Validity to control the likelihood that subjects would engage this processing at the Unique-Cue location. The present experiment had the same goal, but used a different technique to induce goal-driven processing. Specifically, this experiment involved performing a secondary task at the Unique-Cue location that required attention during the interval before the target appeared. This was done to force subjects to attend to that location in a goal-driven manner, in absence of other factors such as high Unique-Cue validity or specific instructions to "focus" on the cue. The secondary task involved having subjects identify the location of a briefly presented (50 ms) gap that appeared on the left or the right side of the Unique Cue and disappeared before the target onset. The gap was not easily visible and seemed to require attentional processing for successful localization (in a pilot study in which the gap was present on the Unique Cue but subjects were not required to perform the gap-localization task, subjects reported being completely unaware of the gap). In addition to the secondary task, other changes were implemented in the Unique-Cue procedure to control for possible confounds. First, to control for the chance that the onset or removal of the gap might somehow affect attention orienting, additional gaps were also embedded in the Standard Cues. Second, to provide a more sensitive measure of attention orienting, a second (200 ms) C T O A was added in case subjects required more time to attend to the Unique-Cue location with this procedure. In all other respects, 82 the design of this experiment was the same as the low-validity condition of Experiment 7b, which was shown to be devoid of goal-driven cue effects. If goal-driven processes are involved in the larger cue effects at the Unique-Cue location, then forcing subject's attention to this location with an attention-demanding task should produce a Unique-Cue advantage. 4.3.1 Methods Subjects: 15 University of British Columbia undergraduates were paid $10 for participating in two 1-hour sessions. A l l subjects had normal or corrected-to-normal vision. Stimuli: A l l cues were identical to those described in the General Methods except that each cue appeared with a gap (0.2° x 0.2°) located randomly either 0.2° inside the right or left edge of the cue. The gap involved a complete break in the continuity of the cue's form and was only visible for 50 ms, after which all gaps were "filled in." Procedure: C U E I N G T A S K : The cueing task was identical to the cueing task described in the General Methods except that subjects were not given a financial incentive to focus on the Unique-Cue location. Instead subjects were given a financial incentive to perform the gap task as accurately as possible and to respond to the target as quickly as possible (pilot studies indicated that it was necessary to give incentive for performance on both the target-detection and gap-localization tasks). The CTOA on data trials was 100 ms on one half of the trials and 200 ms on the other half of the trials in a block. 83 G A P T A S K : On 2 5 % of the trials, after responding to the target, subjects were presented a display screen that queried them about the position of the gap on the "red" cue (Querying only occurred on 2 5 % of the trials because pilot studies indicated that subjects tended to sacrifice performance in the target detection task if querying occurred more frequently). They indicated the position of the gap by pressing a button. Responses were not timed and subjects were instructed to guess if they were unsure about the gap position. The next trial resumed after subjects made a gap response. Subjects were given feedback about their accuracy at the end of a block. P R A C T I C E SESSION: Before beginning the experiment, subjects were given practice in the secondary gap task. The practice session was identical to a test session except that no targets were presented, and the gap position was queried on each trial. Subjects completed blocks of 5 0 trials until, 1) their accuracy in the gap task for a block exceeded 80%, and 2) they responded that they felt comfortable with the task. Two subjects were not able to perform to these standards and were excused from further participation. Design: A CTOA variable (100, 2 0 0 ms) was crossed with the Target-Location variable. Unique-Cue validity was 12.5%, Standard-Cue validity was 37 .5%, and Uncued validity was 50%, which means that the target was equally likely to appear at any of the display locations on a given trial. In total, there were 576 data trials consisting of 7 2 Unique-Cue, 2 1 6 Standard-Cue, and 288 Uncued trials for each level of CTOA. Also, 3 0 0 catch trials with 1500 ms CTOAs were divided according to the same 1:3:4 trial ratio. 4.3.2 Results and Discussion The mean error rate in this experiment was 5.7%. A 2x3 repeated measures A N O V A was run on pooled mean response times for all subjects in each condition. The within-subjects factors were C T O A (100, 200 ms), and Target Location (Unique-Cue, Standard-Cue, or Uncued). Figure 10 shows mean response times averaged over all subjects. Both the main effect of CTOA, F(U4) = 83.95, MSE = 653.91,/? < 0.001, and Target Location, F(2,28) = 12.47, MSE = 415.46,/? < 0.001 were highly significant. The CTOA x Target Location interaction, however, F(2t 28) = 2.99, MSE = 216.80,/? = 0.067, was not significant. Mean response times are presented in Figure 10, and mean cue effects are presented in Table 11. As indicated in Figure 10, the significant effect of CTOA seems to arise from generally faster response times on 200-ms CTOA trials, which is probably due to an alerting effect. Additionally, the main effect of Target Location appears to arise from faster response times on Unique-Cue and Standard-Cue trials relative to Uncued trials (see Figure 10). Table 11: Mean Cue Effects as a function of Cue-Target-Onset-Asynchrony (CTOA) in Experiment 8 (p-values for planned comparisons in brackets). CTOA (ms) Cue Effects 100 200 Unique Cue 36 (< 0.001) 18 (0.012) Standard Cue 17 ( 0.025) 11 (0.044) Unique-Cue Adv. 19 ( 0.035) 7 (0.350) On the most relevant C T O A trials (100 ms), the cue effect pattern closely resembled the Unique-Cue effect pattern found in the other experiments. More importantly for the current hypothesis, the Unique-Cue advantage was significant even though subjects were given neither specific instructions nor a cue-validity incentive to attend to that location. The Unique-Cue advantage must therefore be attributed to the secondary attention demanding task that subjects performed at the Unique-Cue location. 85 EXPERIMENT 8: MEAN RESPONSE TIMES 540-530-520-CO 510-IME 500-H 490-LU 480-CO Z o 470-D_ 460-CO LU or 450-440-UNIQUE CUE STANDARD CUE - O - UNCUED 100 200 CTOA (ms) 0 0 E CO I -O LU U_ U. LU LU o EXPERIMENT 8: CUE EFFECTS 80 70 60 50 40 30 20 10 0 + -10 -20 UNIQUE CUE STANDARD CUE 100 200 CTOA (ms) Figure 10: Mean Response Times and mean Cue Effects as a function of Cue-Target-Onset-Asynchrony (CTOA) in Experiment 8. 86 Also of note is the fact that Standard-Cue effects also occurred in this experiment. This is consistent with the idea that stimulus-driven processing is directly triggered by the onset of the cues, and not due to the "focus" instructions or to high cue validity. Although not central to the purpose of this experiment, the results from the 200 ms C T O A trials indicate that cue effects were smaller on both Unique-Cue and Standard-Cue trials. The decrease in Unique-Cue effects probably occurred because subjects had completed the gap-task and no longer required their attention at the Unique-Cue location. Additionally, the decrease in Standard-Cue effects is probably a reflection of the transient nature of direct-cue effects (e.g., Nakayama & Mackeben, 1989). The experiments in this chapter provide converging support for the notion that goal-driven processes are involved in larger cue effects at the Unique-Cue location. In particular, alternative methods of inducing subjects to initiate goal-driven processing at the Unique-Cue location were effective in generating the expected pattern of cue effects in this paradigm, in absence of specific instructions to do so. 87 5. THE ACTIVITY DISTRIBUTION MODEL Enough data have accumulated in this investigation to attempt to put together a detailed account of cue effects observed in the Unique-Cue paradigm. The relevant findings that require an explanation can be categorized as stimulus-driven cue effects (mediated by Standard Cues) or as goal-driven cue effects (mediated by the Unique Cue). These are as follows: Stimulus-driven cue effects: • do not require the involvement of "top-down" processes • are location specific and can occur at multiple locations simultaneously • vary as a function of cue luminance and may involve a sensory-based cue-target interaction • are mediated by a limited capacity system Unique-Cue effects: • are initiated by "top-down" processes • may interact additively with stimulus-driven cue effects • involve inhibition of unattended locations In the introduction, it was stated that the Activity Distribution model would be used as a conceptual framework to guide the interpretation of the results from this investigation. While this seems to work for interpreting the results at a general level, in its present form, the Activity Distribution model falls short of being able to account for all of the data presented above. Most notable in this regard is that this model has no provisions for dealing with stimulus-driven cue effects such as the direct link between luminance change and cue effect magnitudes or the capacity limits also associated with 88 Standard-Cue effects. The purpose of this chapter is to describe modifications of this model to account for these findings, and to describe in detail how it can explain the control of attention under multiple-cue conditions. Before getting into the specifics of the Activity Distribution model, it would be useful to elaborate on the motivation for this model. The essential purpose of this model is to solve a fundamental problem associated with information processing in the brain. More specifically, at the lowest levels of neural representation, information processing occurs in parallel simultaneously throughout the visual scene. At some higher levels, however, information processing occurs in a less parallel manner (e.g., Neisser, 1967; to simplify this discussion, I will take the strong form of this view that processing at this level is essentially serial or specific to a single source of information). The basic problem is that passing information from a parallel processing level to a serial processing level results in a "bottleneck" as the large amount of incoming information surpasses the capacity of the serial level to process all of that information. This can lead to computational errors or reduced processing efficiency at the serial level (e.g., Neisser, 1967). One way to characterize this problem is to discuss it in terms of the flow of information between levels. Information could flow from multiple sources in the parallel level, however, the serial level would only be optimized to process information from a single source. If information was flowing from all possible sources in the parallel level at the same rate, the serial level would probably process any particular stream of information inefficiently and with errors (c.fi, illusory conjunctions, Treisman & Schmidt, 1982). For example, i f the goal of serial processing was to identify the 89 orientation of a target line and information about the orientation of irrelevant lines was also passed to the serial process, then there is a good chance that this irrelevant information could "misinform" the serial process and cause it to produce an erroneous computation. This is particularly problematic i f information from all sources is flowing to the serial process at the same rate because irrelevant information would be at least equally i f not more likely to inform the serial process than would the relevant information (see Figure 11a). In contrast, this problem is alleviated somewhat i f information is flowing from the relevant source at a greater rate than from irrelevant sources. This is because the greater rate from the relevant source decreases the relative proportion of total incoming information that is erroneous, which means that there is a greater likelihood that the serial process wil l be informed by the correct information (see Figure l i b ) . In this context, information processing can be can be controlled to some degree by processes that increase the rate of transmission of the correct information to the serial level (e.g., abrupt luminance changes or salient local features). Although increasing the rate of transmission of correct information would reduce the occurrence of errors, it would not entirely eliminate this possibility. This goal could only be achieved by completely stopping the flow of information from all irrelevant sources. One way to do this is to introduce a filter mechanism that suppresses the flow of information from all but a single dominant source6 (see Figure 1 lc). When competing 6 Note that it is not necessary to posit that filtering is absolute. A "leaky" filter can also be used if the requirements for error-free input are relaxed. More specifically, if there are conditions in which there exists an acceptable error rate, the degree of filtering can be adjusted so that the output error matches the acceptable error rate. Thus, the absolute filtering position simply represents the endpoint of this continuum. 90 CD ^ CO c CO (U tt1" c o c O ' _ CO " E r a to Q£ c CO E c Serial Level A Parallel Level "Non-Viable" processing streams I I B Salient Feature Luminance Change "Top-Down" Intervention "Viable" Processing Stream A A A A A A A A A A • Filter Figure 11: A depiction of the problem associated with information transmission between parallel and serial levels ofprocessing. Serial-level processing is least efficient (error-free) if all sources of information at the parallel level transmit their information at the same rate (A). Events that increase the rate of information transmission from a source will increase the efficiency ofserial-level processing of that source, however, these streams will still be "non-viable "for correct processing (B). Optimal efficiency is only achieved by filtering out information from all unwanted sources, resulting in a "viable" information stream. 91 information from irrelevant sources is eliminated, the serial level can process the relevant information at maximum efficiency. The adequate filtering out of irrelevant information also marks a transition in the flow of information. More specifically, the relevant source changes from being a "non-viable" source of correct information (because it still can still be associated with erroneous computations) to being a "viable" source of correct information. Note that although it is possible for a relevant source to dominate the informing of serial processing with a higher transmission rate, until it leads to error-free processing, it is just as "non-viable" as other, irrelevant, sources with lower transmission rates. Thus, a qualitative transition occurs in the viability of a relevant source of information once irrelevant information is sufficiently filtered out. Based on this qualitative difference, it is possible to separate the same stream of information into two different constructs. The first involves the "viable" stream of information. This represents the selected stream of information that can be processed however required by the serial level. I will call a stream of information in this state the Attention Stream or Attention Channel. The second construct is apparent before the transition occurs. In this case, although a stream of information is "non- viable" for processing, it still plays a crucial role in determining i f and when (via its interaction with the filter mechanism) a particular stream of information becomes an attention channel. I wil l call streams of information in this state Activity Distributions. Thus, interactions between activity distributions and the filter mechanism control the opening of an attention channel. 92 One notion that follows from these constructs is that stimulus-driven or goal-driven processes that affect the rate of information transmission from a source (which corresponds to changes in the relative size of an activity distribution) can also indirectly control the opening of an attention channel. For example, i f an event such as a luminance change or the appearance of a highly salient feature was able to produce a relatively large activity distribution, then it would increase the chance that the corresponding source of information would win the filter-based competition to become an attention channel. This idea forms the basis of the explanation of attentional control provided by the Activity Distribution model. After describing the motivation underlying the Activity Distribution model, it is now possible to describe the model in more formal terms. The core of the model involves the three basic components described in the previous section. These include the topography of activity distributions generated by low-level sources of information, a filter mechanism that eliminates activation from all but one of these sources, and the attention channel that results from this filtering process (see Figure 12). Thus, the Activity Distribution model provides a description of the processes involved in opening an attention channel, which ultimately allows serial higher-level operations to perform elaborative processing on one out of several possible sources of lower-level information. The nature of those higher-level operations will not be specified in detail in the present model; however, it would include all of the processes involved in elaborating visual information, decision making, response planning and execution, etc., - everything involved in performing the expected response behaviour. 93 As alluded to in the previous discussion it may be possible for various factors (such as luminance changes or feature saliency) to modify the magnitude of activity distributions. The consequence of this is that these factors could control the likelihood that a specific source of information would win the filter-based competition to become the attention channel. Thus, this model accounts for the control of attention by stimulus-driven and goal-driven factors by positing that this control occurs indirectly through modifications in the relative size of the activity distributions associated with different sources of information. In the Activity Distribution model, these factors are implemented as different Input Maps based on various stimulus-driven representations and their interactions with higher-level processes (e.g., feature saliency boosting). These input maps generate the activity distributions that drive the control of attention. These ideas lead to the more formal description of the version of the Activity Distribution model that wil l be used to explain the cue effects obtained in this dissertation. This version departs from LaBerge & Brown's (1989) original version in that it includes a new input component based on luminance changes, and it excludes details about the operations from other components not specifically needed to explain cue effects in the Unique-Cue paradigm. The current model can be divided into separate Input, Filtering, and Attentional levels. The following section provides a detailed description of the operations involved in each level. 9 4 Activity Distribution Model Figure 12: An example of the different components of the Activity Distribution model and how they would operate in the Unique-Cue paradigm. The onset of the cues triggers the formation of stimulus-driven activity distributions at all cued locations in the Luminance Map and a goal-driven activity distribution in Goal-Driven Input. Activity Distributions are only passed on to the Interaction Map from the Luminance Map if there is an index associated with a cued location. At the Interaction Map, activation from the Luminance Map and Goal-Driven Input combine to form the Activation Topography which drives the Filter to open an Attention Channel at the most active location. 95 5.1 INPUT L E V E L Luminance Map: The Luminance Map is a spatiotopic representation that codes luminance changes in the visual scene. Visual processing in this map is involuntary and parallel. The Luminance Map responds to visual input at a location by generating a location-specific activity distribution with a magnitude proportional to the magnitude of the luminance change at the corresponding location. Furthermore, activity distributions generated by separate events can combine if they overlap sufficiently in space and time. Indexing stage: This stage is closely connected to the Luminance Map. It involves a limited number of indexes that provide a link between activity in the Luminance Map and higher-level representations such as the filter in the present case. If the location of an activity distribution is not linked to the filter with an index, the effects of that activity distribution cannot be communicated upward. In this case, an activity distribution will have no effect on subsequent processing in the Interaction Map. Goal-Driven Input: This processing allows top-down operations to influence the control of attention. In the Unique-Cue paradigm, goal-driven processes "find" the red-cue location and generate an activity distribution at that location because it is a particularly useful location. The details of how this occurs are not implemented in the current model, however, there are several pre-existing models that provide examples of how this can be done (Koch &Ullman, 1985; LaBerge & Brown, 1989; Treisman & Sato, 1990; Wolfe, 1994). Most notable is the Guided Search model (Wolfe, 1994) in which a saliency value (similar to an activity distribution) is computed for each spatial location based on the uniqueness of stimulus features (e.g., colour, line orientation, etc.) present at a location relative to other locations. Furthermore, the salience contributed by a particular feature can be modulated by "top-down" demands. Thus, a higher saliency (larger activity distribution) could be produced at the location of a red item if the feature "red" is particularly significant to the observer. The goal-driven control of attention can occur in two different forms. The first can be described as a persistent strategic involvement, similar to adopting an experiment-wise goal or setting that chronically boosts activation associated with a source (e.g., the a priori boosting of an important feature). The second can be described as more "on-the-fly," and it involves the generation of activity distributions based on the immediate processing requirements (e.g., boosting target-related activation in response to a target onset). 5.2 FILTER L E V E L Interaction Map (Activation Topography): Activity distributions from all sources, including the Luminance Map and Goal-Driven Input, are passed to this map. These activity distributions are combined to produce an "Activation Topography" that represents the total amount of activation associated with each visual field location. Filter: The Activation Topography then feeds into the Filter. It operates with a lateral inhibition network, whereby each location inhibits each other location by an amount proportional to the magnitude of its associated activity distribution. The net effect of this is that activation is eventually suppressed at all locations but the most active one (c.f., 97 Koch & Ullman, 1985; LaBerge & Brown, 1989; Wolfe, 1994). See Figure 13 for details. Note that the filter only plays a passive role in the control of attention. It "fine-tunes" preexisting activation from the Interaction Map so that only a single source dominates. The actual control of attention occurs external to the filter at the Input Level. 5.3 ATTENTION L E V E L Attention Map: After the filter designates a single location (relative activation surpasses a threshold), a channel of attention is opened at this location. Information associated with this location passes from the lower-level representations to the higher-level processes, which use it to carry out the required behaviours. According to the Activity Distribution model, all these components must work together to control the orienting of attention. Given the complexity of this proposal, it is important to be explicit about just what constitutes attention in this model. For the purpose of this dissertation, attention is defined as the process of elaborating the information that is passed to higher-level processes by way of the attention channel (the "viable" information stream). Thus, information is actually processed in an attentional manner at levels above those described in this model. In functional terms, however, attention can be equated to the opening of an attention channel at a location because this provides the access to information that permits attentional processing to begin. The remaining components of the model deal with the control of where the attention channel wil l open. Figure 13: Operation of the Lateral Inhibition Network over time. Each location inhibits all other locations by an amount proportional to the size of the corresponding activity distributions. 99 5.4 AN EXPLANATION OF CUE EFFECTS With the formal description of the control of attention provided by the Activity Distribution model, it is now possible to explain the cue effects obtained in the Unique-Cue paradigm. Before this is done, however, it is important to clarify the role in determining cue effects played by the performance requirements associated with spatial-cueing tasks. In these tasks, the performance requirements involve a speeded response to the onset of the target. An underlying assumption is that, before subjects can make this response, they must first confirm the presence of the target in the display. This confirmation is assumed to require focused attention at the target location (unless the specific design of the display allows an alternative response strategy, such as a feature-detection strategy). Thus, this model hold that, regardless of where the target appears relative to the cues, an attention channel must always be opened at the target location before a response to the target can be made. A consequence of this is that because targets appearing at cued or uncued locations ultimately receive the same attentional processing, cue effects, by definition, solely reflect the time required to open an attention channel at a target location. This contrasts with another possibility which holds that a target appearing at an uncued location can still be processed without attention, however, because information sampling/processing would occur more slowly outside the attentional focus, responses to these targets would be slower (e.g., Johnson & Yantis, 1995). The assumption that cue effects are exclusively related to the time required to open an attention channel at the target location results in two subtly different explanations of cue effects. The first involves cue effects that occur at short CTOAs, 100 before an attention channel is open. In this case, cue effects reflect the ongoing processing involved in opening an attention channel. More specifically, the Activity Distribution model holds that opening an attention channel involves boosting the relative activation at a designated location past a threshold value by filtering out activation at all but the designated location. Consequently, the time required to surpass threshold wil l be affected by preexisting activation in the system. For example, preexisting activation at the target location would give the opening of a channel at the target location a "head start" because total activation at that location would be closer to threshold. In contrast, preexisting activation at other locations would slow the opening of a channel because this activation would first have to be filtered out (the other locations have a "head start" that the target location must overcome). Thus, at this early stage cue effects are a product of the influence that the preexisting Activation Topography has on the opening of an attention channel. The second explanation involves cue effects that occur at longer CTOAs. In this case, cue effects reflect the fact that an attention channel is already open at the target location before the target appears. This represents the end-point of the channel-opening continuum. That is, a designated location does not only have a "head start" but has already crossed the finish line by the time the target appears. More specifically, this saves all of the time required to open a channel once the target appears. Consequently, this leads to faster responses relative to other locations in which an attention channel has to first be opened before the target is processed. Note that this account of cue effects is similar to how many attention-orienting models explain cue effects - that a cue leads to attention being focused at a potential target location in advance of the target onset (e.g., 101 Posner, 1980; Sperling & Weichselgartner, 1995). Thus, at this later stage, cue effects occur because attention is already focused at the target location, and ready to process the information necessary to make a response. Although these two explanations of cue effects are a part of the same continuum, there are subtle differences in the impact they have on cue effects. At short CTOAs, cue effects are determined by operations that have a graded or continuous influence on cue effect magnitudes. For example, in this situation, changing activation levels in the system wil l have a corresponding effect on opening an attention channel - it wil l speed or slow the passing of threshold by a related amount, depending on the location and the magnitude of the change. In contrast, at longer CTOAs, cue effects are determined by operations that have a binary or discrete influence on cue effect magnitudes. In this situation, changing activation levels would have no effect on the speed with which the attention channel is opened because the channel is already open. This makes cue effects at the attended location insensitive to changes in activation levels. With these points in mind, the Activity Distribution model can be used to explain cue effects in the Unique-Cue paradigm. Because all of the experiments described so far use short CTOAs, the most relevant description of cue effects involves the initial channel-opening process. In this case, cue effects are essentially determined by the interaction between target-related activation and the Activation Topography that exists at the time of the target onset. In the Unique-Cue paradigm, two significant sources of activity distributions define the layout of the Activation Topography. The first is the stimulus-driven activation generated at all cued locations by the luminance change associated with the onset of the cues. The second is the goal-driven activation generated 102 exclusively at the location of the Unique-Cue because the conditions present in this paradigm dictate that that location is "useful" for performing the task. Still undefined in this explanation is the source of target-related activation. Target-related activation is responsible for opening an attention channel at the target location before it can be processed. In the Activity Distribution model, target-related activation will be implemented in the following manner. Some activation will be generated in the Luminance Map because the target appears with an abrupt luminance increment. It may combine with preexisting cue-related activation at the same location in a manner similar to how sensory summation may occur (c.f. Experiment 4). The remaining activation wil l come from "top-down" sources directed through Goal-Driven Input, in much the same way as it is generated for the Unique-Cue (the onset of target-related activation may coincide with the termination of Unique-Cue-related activation from Goal-Driven Input). This target-related activation wil l then be sustained until an attention channel is opened. Given these conditions, the following operations would be involved in each trial type of the Unique-Cue paradigm. Unique-Cue trial: The target appears at a location with a large pre-existing activity distribution. The relative size of this distribution may be near or already past threshold, and very little additional target-related activation would be required to bring this location past threshold. Uncued trial: The target appears at a location at which there are no preexisting activity distributions. Target-related activation must overcome activity distributions already 103 established at non-target locations before an attention channel opens, which takes longer than on Unique-Cue trials. Standard-Cue trial: These trials are similar to Uncued trials, except that there is a small activity distribution remaining from the cue onset. Target-related activation at this location still has to overcome the larger activity distribution at the Unique-Cue location, however, the pre-existing activity distribution gives channel-opening a "head start" relative to Uncued trials. With this description of how the Activity Distribution model can account for cue effects, it is possible to return to the previously mentioned set of findings that need to be explained. Stimulus-Driven Cue effects: • do not require the involvement of "top-down" processes • are location specific and can occur at multiple locations simultaneously • vary as a function of cue luminance and may involve a sensory-based cue-target interaction • are mediated by a limited capacity system The specific properties attributed to the Luminance map can explain the first three points. This is because this representation is said to respond involuntarily and in parallel to luminance changes by generating a location-specific activity distribution of a corresponding magnitude. Furthermore, because this map is said to transmit this activation to higher levels via links provided by the indexing stage, capacity limits are introduced by the number of available indexes. 104 Goal-Driven Cue effects: • are initiated by "top-down" processes • may interact additively with stimulus-driven cue effects • involve inhibition of unattended locations According to the Activity Distribution model, the first point is explained by the fact that Goal-Driven Input contributes activation specific to the Unique-Cue location. In addition, the second point is explained by the fact that opening an attention channel involves an interaction between stimulus-driven and goal-driven processes at the Activation-Topography level. Finally, because targets appearing at locations other than the Unique-Cue location have to overcome larger activation, opening a channel at these would take longer, producing a relative inhibition of response times at these locations. Thus, the Activity Distribution model seems to provide a reasonable account of the data obtained so far with the Unique-Cue paradigm. One point of note is that this model differs from other attention models as to how cue effects are interpreted. In particular, with other models (e.g., Eriksen & Yeh, 1985; Posner, 1980; Sperling & Weichselgartner, 1995), cue effects largely reflect the properties of the attention mechanism. For example, location-specific cue effects would be interpreted as suggesting that the attentional focus has a small spatial extent. In contrast, with the activity distribution model, cue effects instead reflect the properties of the processes that control the opening of an attention channel. In this case, i f a direct cue generates location-specific cue effects, it is only because the corresponding stimulus-driven processing is location specific. If a less spatially precise method is used to open an attention channel (e.g., goal-driven activation of a symbolically-cued location), then 105 the attentional focus would change accordingly. Thus, interpretation of cue effects with the Activity Distribution model provides details about the processes involved in controlling attention, not the properties of attention itself. 106 6. TESTING PREDICTIONS The Activity Distribution model provides the framework that this investigation wil l use to explain the results from the Unique-Cue paradigm. Not only does this model provide a reasonable account of the findings thus far, but it also explains cue effects in ways not previously done in other models. Several of the ideas associated with the Activity Distribution model, such as the explanation of cue effects at short CTOAs, constitute a significant departure from conventional explanations. The purpose of the experiments reported in this chapter was to validate some of these ideas. If these assumptions can be confirmed along with their connection to similar attentional phenomena in the spatial-cueing literature, then the Activity Distribution model has the potential to provide a new and more comprehensive description of the control of attention. The central focus of this chapter was on examining three fundamental assumptions. The first was the notion that separate sources of inputs (stimulus-driven and goal-driven) can contribute to cue effects. This idea differs from many other models of attention that hold that all cue effects are due to a single, central attentional process that is controlled in different ways (e.g., Jonides, 1981; Sperling & Weichselgartner, 1995). The second assumption was that at short CTOAs, the interaction between stimulus-driven and goal-driven inputs influences the time required to open an attention channel at a location and that it is these interactions and not facilitated processing by focused attention that form the basis of cue effects. The final assumption to be tested was that filtering operations are involved in the opening of an attention channel. Confirming 107 these assumptions wil l help establish the Activity Distribution model as a viable explanation of the control of attention. 6.1 EXPERIMENT 9 This experiment addressed the assumption that stimulus-driven and goal-driven cue effects are primarily caused by processes that function independently at separate non-attentional stages of the Activity Distribution model. More precisely, in the Unique-Cue paradigm, stimulus-driven cue effects would be the product of Luminance Map activation while goal-driven cue effects would be the product of Goal-Driven Input. If these processes can affect cue effects independently, then it should be possible to find dissociations in their effects across certain factors. One factor over which goal-driven and stimulus-driven effects seem to differ is CTOA. As previously discussed, goal-driven cue effects are sustained over time, whereas without goal-driven input, stimulus-driven cue effects are more transient, (e.g., Nakayama & Mackeben, 1989). If these separate stimulus-driven and goal-driven processes are involved in generating these cue-effects, then a similar pattern should occur with cue effects in the Unique-Cue paradigm. Specifically, cue effects at the Unique-Cue location should be sustained while cue effects at the Standard-Cue locations should be transient. Varying the time course of cue and target presentation may also provide information about how stimulus-driven and goal-driven processes interact over time. According to the Activity Distribution model, the nature of the interaction should depend on the CTOA. At short CTOAs, before an attention channel is open, stimulus-driven and goal-driven effects should be additive (cf. Experiment 4). In contrast, at longer CTOAs, after the attention channel is open, the Activity Distribution model states that the input-108 related activity distribution should no longer affect the magnitude of the cue effects. Thus, one consequence of opening an attention channel at later CTOAs is the introduction of a non-linearity in cue effects over time. This should produce a significant interaction across C T O A between cue effects on Unique-Cue and Standard-Cue trials. The present experiment measured changes in cue effects over time by varying the CTOA. CTOAs ranged from 100 to 400 ms and included the optimal CTOAs for both stimulus-driven (100 ms) and goal-driven cue effects (300-400 ms). 6.1.1 Methods Subjects: 15 University of British Columbia undergraduates were paid $10 for participating in two 1-hour sessions. A l l subjects had normal or corrected-to-normal vision. Procedure: The procedure was the same as described in the General Methods except that CTOAs on data trials were either 100, 200, 300, or 400 ms. Design: The Target-Location variable was completely crossed with a CTOA variable (100, 200, 300,400 ms). The target appeared at either the Unique-Cue location (66.7% of trials), one of the Standard-Cue locations (16.7% of trials), or at one of the Uncued locations (16.7% of trials). In total, there were 1800 data trials consisting of 300 Unique-Cue, 75 Standard-Cue, and 75 Uncued trials for each level of CTOA. Also, 600 catch trials with 1500 ms CTOAs were divided according to the same 4:1.1 trial ratio. 6.1.2 Results and Discussion The mean error rate in this experiment was 1.7%. 109 A 4x3 repeated measures A N O V A was run on pooled mean response times for all subjects in each condition. The within-subjects factors were C T O A (100, 200, 300, & 400 ms), and Target Location (Unique-Cue, Standard-Cue, or Uncued). Figure 14 shows mean response times averaged over all subjects. Each of the main effects of CTOA, F(3,42) = 30.88, MSE = 508.64,/? < 0.01, Target Location, F(2,2s) = 16.69, MSE = 1810.79,/? < 0.01, and the C T O A x Target Location interaction, F(6M) = 6.76, MSE = 86.75,/? < 0.01, were highly significant. Mean cue effects are presented in Table 12 and in Figure 14. Table 12: Mean Cue Effects as a Junction of Cue-Target-Onset-Asynchrony (CTOA) in Experiment 9 (p-values for planned comparisons in brackets). CTOA (ms) Cue Effects 100 200 300 400 Unique Cue 44 (0.001) 40(0.001) 38(0.001) 39 (< 0.001) Standard Cue 20 (<0.001) 2 (0.654) -1 (0.905) -8(0.096) The data indicate that cue-effect magnitudes did not vary over C T O A on Unique-Cue trials. This is consistent with other studies showing that goal-driven direct-cue effects can occur rapidly and remain sustained over time (e.g., Cheal & Lyon, 1991). In contrast, cue-effect magnitudes were largest at 100 ms on Standard-Cue trials and dropped thereafter. This is consistent with other studies showing that stimulus-driven cue effects are transient over time (e.g., Nakayama & Mackeben, 1989). The dissociation in time course provides support for the notion that stimulus-driven and goal-driven cue effects are mediated by separate processes that function differently over time. 110 EXPERIMENT 9: MEAN RESPONSE TIMES CO LU LU CO z o D_ CO LU 01 UNIQUE CUE STANDARD CUE O - UNCUED 100 200 300 400 CTOA(ms) CO E, CO I— o LU LU LU O EXPERIMENT 9: CUE EFFECTS 80' 70 60 50 40 30 20 10 0 -10 -20 UNIQUE CUE STANDARD CUE - - 0m -100 200 300 400 CTOA(ms) Figure 14: Mean Response Times and mean Cue Effects as a function of Cue-Target-Onset-Asynchrony (CTOA) in Experiment 9. I l l With regard to the nature of the interaction, it is not possible to determine from the data i f cue effects were additive on 100 ms C T O A trials. The problem is that Unique-Cue effects at this CTOA do not differ from Unique-Cue effects at later CTOAs, in which an attention channel is presumed to be open. Thus, it is unknown whether the magnitude of this effect is the same because the additive interaction coincidentally yields the same magnitude as attention effects or if cue effects hit ceiling because an attention channel was opened during that interval. On the other hand, it is clear that cue effects are not purely additive on longer C T O A trials. This is because Unique-Cue effects remain unchanged as Standard-Cue effects dropped after 100 ms. If the effects were purely additive, a corresponding drop should have occurred with Unique-Cue effects, although the overall magnitude should have remained high. This suggests that Unique-Cue effects are consistent with a threshold process. The data from this experiment are consistent with the Activity Distribution model predictions that cue effects can be generated by different inputs, and that once an attention channel is open, cue effects are not additive. 6.2 EXPERIMENT 10 The dissociation between Unique-Cue and Standard-Cue effects over CTOA found in the previous experiment suggests that separate and independent sources of processing may contribute to cue effects. This finding has specific implications for the interpretation of Unique-Cue effects because both stimulus-driven and goal-driven factors are involved at that location. The Activity Distribution model takes this possibility into account in its explanation of Unique-Cue effects. It states that activation 112 from separate stimulus-driven and goal-driven sources combine, producing a larger activity distribution at that location. This speeds the opening of an attention channel, yielding a large cue effect. A n important component of this explanation is that stimulus-driven and goal-driven cue effects combine at the Interaction Map. This idea is supported by the results of Experiment 4, in which transient cues decreased cue effects on Unique-Cue and Standard-Cue trials by the same amount. According to the Activity Distribution model, this finding is explained by positing that a reduced level of activation from the Luminance Map sums linearly with activation from Goal-Driven Input. The smaller combined activity distribution at the Unique-Cue location would take longer to surpass threshold, resulting in longer response times relative to Sustained Unique-Cue trials (which would have a larger luminance-based activity distribution). Thus, according to this explanation, changes in the magnitude of Unique-Cue effects are attributed to interactions between peripheral input levels of the Activity Distribution model. While the results of Experiment 4 were consistent with the notion of an additive peripheral interaction, the results from Experiment 9 were not able to confirm this at short CTOAs. The results were indeterminate because Unique-Cue effects did not change across C T O A (see discussion in Experiment 9 for explanation). This constant pattern of cue effects is also consistent with an alternative explanation for Unique-Cue effects. In particular, this account holds that cue effects on Unique-Cue trials are exclusively due to attentional processing at that location regardless of C T O A (as long as the C T O A is appropriate for cue effects to occur). Furthermore, according to this view, magnitude changes in cue effects (c.f., Experiment 4) are not due to interactions 113 involving peripherally-generated activation, but to a qualitative change in the processing efficiency of the attentional focus. In specific terms, this proposal holds that attention can be described as a pool of "attentional resources" that can be flexibly allocated to visual field locations as required (e.g., Shaw, 1983; Shaw & Shaw, 1977). Furthermore, the degree of attention-related processing facilitation is directly related to the amount of resources allocated to a location. Thus, according to this Attentional Resource proposal, smaller Unique-Cue effects in Experiment 4 would have occurred because the reduced luminance energy of the transient cue would have caused fewer attentional resources to be assigned to that location. These two views can be differentiated based on the time course of their effects. More specifically, because stimulus-driven cue effects are transient (e.g., Nakayama & Mackeben, 1989; see also Experiment 8), Unique-Cue effects based on peripheral interactions with this type of process should also be transient. In contrast, because changing the level of attentional resources is said to have a persistent effect on processing over time (lasting several seconds; e.g., Hughes, 1984), Unique-Cue effects based on this type of process should be sustained. The Unique-Cue effects observed in Experiment 9 are consistent with this notion. These cue-effect magnitudes remained constant over time, affording the interpretation that the amount of attentional resources initially allocated to the Unique-Cue location stayed the same. The purpose of this experiment was to test the assumption made by the Activity Distribution model that changes in cue effects at short CTOAs are primarily mediated by interactions between peripheral inputs, as opposed to a central change in the quality of attentional processing. In the present experiment, Unique-Cue effect magnitudes were 114 manipulated by varying cue-luminance levels (Dim or Bright). This factor was chosen because, as results from Experiment 5 indicate, luminance seems to affect stimulus-driven cue effects and this provides a method for inducing what should be peripheral changes in activation levels. The critical variable for evaluating these two possibilities was the CTOA. According to the Activity Distribution model, both Unique-Cue and Standard-Cue effects should be larger on Bright luminance trials relative to Dim luminance trials at short CTOAs (e.g., 100 ms). However, at longer CTOAs (e.g., 400 ms) after the stimulus-driven effects of luminance have attenuated, cue effects on Bright trials should return to "normal." That is, cue effects should be the same on Dim and Bright trials (both of these should also be similar to the long C T O A trials in Experiment 9). In contrast, according to the Attentional Resource view, any changes in Unique-Cue effects attributable to luminance should be present at both short and long CTOAs. This means that Unique-Cue magnitudes should remain elevated on Bright trials at long CTOAs. Eye movements were also monitored in this experiment. Because the long CTOA trials provided enough time to execute an eye movement to the highly-valid Unique-Cue location, it was important to discount eye movements as a factor in the Unique-Cue Advantage. Thus, eye movements were monitored and trials in which eye movements occurred were excluded from analysis. 115 6.2.1 Methods Subjects: 15 University of British Columbia undergraduates were paid $10 for participating in two 1-hour sessions. A l l subjects had normal or corrected-to-normal vision. Apparatus and Stimuli: These were the same as described in the General Methods except that the cues on Bright and Dim trials differed in luminance. Subjective luminance was equated within trials for the Unique Cue and the Standard Cues using the same procedure as described in the General Methods except that a red colour patch was used instead of a green one. Eye Movement Monitoring: See General Methods. Procedure: The procedure was the same as described in the General Methods except that CTOAs on data trials were either 100 or 400 ms. Design: The Target-Location variable was completely crossed with a CTOA variable (100 or 400 ms), and with a Cue Luminance variable (Bright or Dim). The target appeared at either the Unique-Cue location (66.7% of trials), one of the Standard-Cue locations (16.7% of trials), or at one of the Uncued locations (16.7% of trials). In total, there were 1800 data trials consisting of 300 Unique-Cue, 75 Standard-Cue, and 75 Uncued trials for each level of Cue Lurninance and CTOA. Also, 600 catch trials with 1500 ms CTOAs were divided according to the same 4:1:1 trial ratio. 116 6.2.2 Results and Discussion The results of two subjects were excluded from analysis due to error rates that surpassed 10% in one or more of the sessions. The mean error rate for the remaining subjects was 2.9%. Eye movement occurred on less than 1% of the trials for all subjects. If subjects failed to make a significant number of eye movements in this experiment, in which the incentive to do so was maximized, it is probably safe to assume that they did not move their eyes in the other experiments that involved similar conditions (e.g., long CTOAs; see also Appendix B). A 2x2x3 repeated measures A N O V A was run on pooled mean response times for all subjects in each condition. The within-subjects factors were Target Location (Unique-Cue, Standard-Cue, & Uncued), CTOA (100 & 400 ms), and Cue Luminance (Bright & Dim). Figure 15 shows mean response times averaged over all subjects. The main effects of Target Location, F(2,24) = 21.69, MSE = 1889.48, p < 0.001 and CTOA, F ( U 2 ) = 100.94, MSE- 1731.84, p < 0.001, were highly significant, whereas the main effect of Cue Luminance just approached significance, F ( U 2 ) = 4.52, MSE = 394.28, p = 0.055. Only two interactions reached significance. The first was the 2-way C T O A x Target Location interaction, F(2I24) = 9.62, MSE = 227.38,/? = 0.002. This was due to the disappearance of Standard-Cue effects on 400 ms C T O A trials. The second was the 3-way C T O A x Cue Luminance x Target Location interaction, F(2I24) = 9.27, MSE = 108.57, p = 0.005. This was due to relatively larger decrease in cue effects on Bright Unique-Cue trials at the 400 ms C T O A (a separate A N O V A showed a Target Location x CTOA 117 EXPERIMENT 10: MEAN RESPONSE TIMES 6 . CO E, LU LU to z o Q. CO LU OH 500-480-460-440-420-400' 380' 360' 340 320 300 a. *\, O DIM A BRIGHT UNIQUE CUE STANDARD CUE • - - UNCUED I 100 I 400 CTOA (ms) CO E, CO I -o LU LU LU O EXPERIMENT 10: CUE EFFECTS 80' 70 60 50 40 30 20 10 0 + -10 -20 o DIM A BRIGHT UNIQUE CUE — — STD. CUE 100 400 CTOA (ms) Figure 15: Mean Response Times and mean Cue Effects for Dim (circles) and Bright (triangles) cues as a function of Cue-Target-Onset-Asynchrony (CTOA) in Experiment 10. 118 interaction on Bright trials; see below). Neither the 2-way C T O A x Cue Luminance, F(i,i2) = 0.03, MSE = 179.34,/) = 0.862, nor the 2-way Cue Luminance x Target Location, F(2,24) = 1.04, MSE = 251.62,/? = 0.370, interaction was significant. Mean cue effects are presented in Table 13 and in Figure 15. Table 13: Mean Cue Effects as a function of Cue-Target-Onset-Asynchrony (CTOA) and Cue Luminance in Experiment 10 (p-values for planned comparisons in brackets). Cue Luminance Bright Dim CTOA (ms) CTOA (ms) Cue Effects 100 400 100 400 Unique Cue 1A (0.001) 44(0.001) 49 (0.002) 52 (0.002) Standard Cue 35 (0.001) 4(0.650) 21(0.001) 2 (0.837) At 100 ms, cue effects were larger on Bright trials. After 400 ms, however, the effects of luminance disappeared. This was confirmed by an additional A N O V A performed on the Cue Luminance and Target Location variables for 400 ms C T O A trials. Neither the main effect of Cue Luminance, F(jj2) = 1.85, MSE = 428.22,/? = 0.198, nor the Cue Luminance x Target Location interaction, F(2,24) = 1-04, MSE - 238.34,/? = 0.369, was significant, indicating that luminance had no effect at the later CTOA. The absence of a sustained luminance effect on Unique-Cue trials is inconsistent with the Resource Allocation proposal. If luminance affected the qualitative nature of attentional processing in the sustained manner found in previous experiments (e.g., Hughes, 1984), then cue effects should have remained elevated on Unique-Cue trials. Instead, cue effects on Bright trials drop over time and converge on the values observed on Dim trials (the cue-effect pattern also closely matches the one observed on 400 ms C T O A trials in Experiment 9). Thus, this transient effect of luminance on Unique-Cue 119 effects is consistent with the view put forth by the Activity Distribution model that luminance affects a peripheral input stage of the system, such as the Luminance Map. While these data are consistent with a stimulus-driven and goal-driven interaction at short CTOAs, the nature of the interaction differs from the one observed in Experiment 4. In that experiment, cue duration affected cue effects by the same amount on both Unique-Cue and the Standard-Cue trials, suggesting additive effects. In the present experiment, however, cue luminance had a larger effect on Unique-Cue trials than on Standard-Cue trials (25 vs. 14 ms, respectively). This differential effect was confirmed by an additional 2x3 A N O V A performed on the Cue Luminance and Target Location factors in the 100 ms C T O A condition. There was a significant Cue Luminance x Target Location interaction, = 8.37, M S E - 121.85, p = 0.003, indicating that the effect of luminance was not purely additive. Although this finding is still consistent with the Activity Distribution assumption that peripheral processes interact to produce cue effects, it tempers the claim that this interaction is purely additive in nature (cf. Experiment 4). 6.3 EXPERIMENT 11A The results of the two previous experiments are consistent with the notion that cue effects at short CTOAs arise from peripheral input from representations such as the Luminance Map and Goal-Driven Input. These inputs, through the activity distributions they generate, produce cue effects by influencing the time required to open an attention channel at the target location. Note that this influence is felt after the target has appeared in the display. That is, these effects occur indirectly as target-related activation interacts with the pre-existing Activation Topography (formed by cue-related activation). This 120 position marks a departure from the typical interpretation of cue effects. In particular, cue effects are usually said to occur because attention is aligned with a target location before the target appears (e.g., Shulman, Remington, & Pierce, 1979; Sperling & Weichselgartner, 1995). This is said to produce a "head start" for the attentional operations involved in responding to the target, resulting in a response-time savings. Thus, there is a fundamental difference about the point in time relative to the target onset that cue effects occur. The Activity Distribution model states that cue effects are due to post-target-onset operations, while other attentional models state that they are due to pre-target-onset operations. This difference can be addressed empirically. Because attention-related cue effects require attentional operations in advance of the target onset, they should not occur if the C T O A is too short to carry out these operations. That is, according to this view, cue effects would not occur with a 0 ms C T O A because there is insufficient time to prepare attention before the target appears. On the other hand, cue effects are possible at this C T O A with the Activity Distribution model. This is because cue effects are the product of the interaction between target-related activation and persisting cue-related activation. The only requirement for cue effects, therefore, is that these activity distributions overlap in time in the Interaction Map. If we assume that a similar amount of time is required to transmit cue-related and target-related activation, then it should be possible to have both occur at the same time (0 ms CTOA), and get the same cue-effect-producing overlap in activation. The purpose of this experiment was to determine i f cue effects could appear at very short CTOAs. The presence of such cue effects would contradict predictions made by an attention-related account of cue effects and support explanations based on post-target-onset operations, such as the Activity Distribution model. This experiment used the same procedure as Experiment 9 except that the CTOAs used were all less than or equal to 100 ms (0,33,67, & 100 ms). 6.3.1 Methods Subjects: 11 University of British Columbia undergraduates were paid $10 for participating in two 1-hour sessions. A l l subjects had normal or corrected-to-normal vision. Apparatus and Stimuli: These were the same as described in the General Methods. Procedure: The procedure was the same as described in the General Methods except that CTOAs on data trials were 0, 33, 67, or 100 ms. Design: The Target-Location variable was completely crossed with a CTOA variable (0, 33, 67, or 100 ms). The target appeared at either the Unique-Cue location (66.7% of trials), one of the Standard-Cue locations (16.7% of trials), or at one of the Uncued locations (16.7% of trials). In total, there were 1800 data trials consisting of 300 Unique-Cue, 75 Standard-Cue, and 75 Uncued trials for each level of CTOA. Also, 600 catch trials with 1500 ms CTOAs were divided according to the same 4:1:1 trial ratio. 6.3.2 Results and Discussion The mean error rate in this experiment was 1.9%. A 4x3 repeated measures A N O V A was run on pooled mean response times for all subjects in each condition. The within-subjects factors were Target Location (Unique-122 Cue, Standard-Cue, or Uncued) and CTOA (0,33,67, & 100 ms). Figure 16 shows mean response times averaged over all subjects. Both the main effects of Target Location, F(2,20) = 31.60, MSE = 1268.52,/? < 0.01 and CTOA, F(3,30) = 114.13, MSE = 204.30,/? < 0.01, were highly significant. The CTOA x Target Location interaction, FQM) = 1.58, MSE = 117.83,/? = 0.16, however, was not significant. Mean cue effects are presented in Table 14 and in Figure 16. Table 14: Mean Cue Effects as Junction of Cue-Target-Onset-Asynchrony (CTOA) in Experiment 11a (p-values for planned comparisons in brackets). CTOA (ms) Cue Effects 0 33 67 100 Unique Cue 51(0.001) 58(0.001) 43 (0.001) 49(0.001) Standard Cue 30(0.004) 37 (0.001) 20 (0.004) 22 (0.001) The data indicate that significant direct-cue effects occurred even when the cue appeared at the same time as the target. Cue effects on both Unique-Cue and Standard-Cue trials were significant at all CTOAs. Furthermore, there was no significant CTOA x Target Location interaction, which suggests that these effects did not change over this short time interval. This result is consistent with the notion that cue effects are the result of post-target-onset processes, and are inconsistent with predictions from standard attention models that hold that cue effects are due to preparatory attention shifts1. If the latter would have been true, then cue effects on both types of trials should have been 7 Because these data also show an alerting effect (overall response times were slower on shorter CTOA trials), it is not possible to unequivocally reject the notion that cue effects were due to preparatory processes. In particular, if the response strategy that subjects used was so dominant that it overrode other processing, then it is possible that it would have forced processing to follow the same course (i.e., the system might have waited until attention had shifted before processing the target onset). This would have delayed overall response times in the 0 ms CTOA condition by the time required to first shift attention to the Unique-Cue location. This idea could be tested by replicating this experiment with an additional temporal-cue factor. A temporal cue that was location non-specific (i.e., a change in background luminance) could provide a measure of the alerting effect associated with the temporal warning provided by the cue. If this effect was similar to as the drop in overall response times over CTOA, then it would suggest that this response time change in the current condition was due to generalized alerting effects and not to a processing delay. 123 EXPERIMENT 11a: MEAN RESPONSE TIMES LU 380' CO Z 370' o D_ 360 CO LU 350 01 340 — • — UNIQUE CUE — © — STANDARD CUE - O - UNCUED 33 66 CTOA (ms) 100 CO E, CO I— o LU LU LU O EXPERIMENT 11a: CUE EFFECTS 80-70' 60' 50' 40' 30 20 10 0 -10 -20 UNIQUE CUE STANDARD CUE I 33 66 CTOA(ms) 100 Figure 16: Mean Response Times and mean Cue Effects as a Junction oj Cue-Target-Onset-Asynchrony (CTOA) in Experiment 11a. 124 non-existent on 0 ms C T O A trials as they are in several other experiments (e.g., Nakayama & Mackeben, 1989). 6.4 EXPERIMENT 11B Although the results of the previous experiment are consistent with the post-target processing notion, the occurrence of cue effects on 0 ms CTOA trials permits another explanation of these data. This explanation is premised on the notion that observers can selectively "boost" the activation associated with certain stimulus features in a goal-driven manner, resulting in relatively higher levels of activation at locations containing these features (cf. Wolfe, 1994). In the current version of the Activity Distribution model, this boosting process was an implicit operation occurring in the Goal-Driven Input component. Because the Unique-Cue location is very useful for task performance and it is defined by a unique feature (the colour red), the model holds that activation associated with this feature would be voluntarily "boosted" or increased. This elevated activation from the Goal-Driven Input component would contribute to the larger cue effects at the Unique-Cue location. It is also possible, however, that feature boosting was not merely restricted to the "red" feature of the Unique Cue. In this case, because the Standard Cues also shared some features in common with the Unique Cue (e.g., same size, shape, lack of curvature, orientation, etc.,), Standard-Cue locations may have also benefited from elevated feature-based activation levels. Thus, it is possible that the cue effects observed in this paradigm were entirely due to feature boosting. Furthermore, because the Standard-Cues do not contain the complete set of boosted features, they would not be associated with as high a 125 level of activation. This would result in cue effects that, while significant, would not be as high as on Unique-Cue trials, which is precisely the observed results. The present experiment tested this possibility by presenting Unique and Standard Cues that did not share any of the same features. While the Standard Cues were the same grey bars as used in previous experiments, the Unique Cue was a red disk that appeared above the target location. If cue effects were simply due to shared features, then cue effects should not occur on Standard-Cue trials because these cues no longer shared any features with the Unique Cue. The CTOAs in this experiment were 0 and 100 ms, covering the same range of CTOAs as the previous experiment. 6.4.1 Methods Subjects: 12 University of British Columbia undergraduates were paid $5 for participating in a single 1-hour session. A l l subjects had normal or corrected-to-normal vision. Apparatus and Stimuli: These were the same as described in the General Methods except that the Unique Cue was a red disk (0.5 x 0.5°) that appeared above the target location instead of below. Procedure: The procedure was the same as described in the General Methods except that CTOAs on data trials were either 0 or 100 ms. Design: The Target-Location variable was completely crossed with a CTOA variable (0 or 100 ms). The target appeared at either the Unique-Cue location (66.7% of trials), one of the Standard-Cue locations (16.7% of trials), or at one of the Uncued locations (16.7% of trials). In total, there were 900 data trials consisting of 300 Unique-Cue, 75 Standard-126 Cue, and 75 Uncued trials for each level of CTOA. Also, 300 catch trials with 1500 ms CTOAs were divided according to the same 4:1:1 trial ratio. 6.4.2 Results and Discussion The mean error rate in this experiment was 1.3%. A 2x3 repeated measures A N O V A was run on pooled mean response times for all subjects in each condition. The within-subjects factors were Target Location (Unique-Cue, Standard-Cue, or Uncued) and C T O A (0 & 100 ms). Figure 17 shows mean response times averaged over all subjects. Both the main effects of Target Location, F(2,22) = 18.06, MSE = 500.51,/? < 0.01, and CTOA, F(U1) = 79.87, MSE = 570.30,/? < 0.001, were highly significant. The C T O A x Target Location interaction, F(2,22) = 0.38, MSE = 119.65,/? = 0.69, however, was not significant. Mean cue effects are presented in Table 15. Table 15: Mean Cue Effects as Junction ojCue-Target-Onset-Asynchrony (CTOA) in Experiment lib (p-values Jorplanned comparisons in brackets). CTOA (ms) Cue Effects 0 100 Unique Cue 37 (0.003) 41 (< 0.001) Standard Cue 19 (0.018) 18 (0.005) The data indicate that, although the Standard Cues did not share any features with the Unique Cue, significant cue effects still occurred on Standard-Cue trials. This finding is more consistent with the previous notion that Standard-Cue effects are linked to the onset of these cues in the display. Furthermore, these data also replicate the finding of cue effects on 0 ms C T O A trials, which also supports the idea of a post-target-onset source of cue effects. 127 EXPERIMENT 11b: MEAN RESPONSE TIMES 440' 430 ^ 4201 & 410-| 400H P 390' LU CO 380' O 370' CL „ „ CO 360 LU 0C 350 340 UNIQUE CUE STANDARD CUE - O - UNCUED 0 100 CTOA (ms) CO E, CO I— o LU LU LU O EXPERIMENT 11b: CUE EFFECTS 80' 70' 60 50 40 30 20 10 0 + -10 -20 UNIQUE CUE STANDARD CUE o - _ _ _ _ _ _ o 100 CTOA (ms) Figure 17: Mean Response Times and mean Cue Effects as a function of Cue-Target-Onset-Asynchrony (CTOA) in Experiment lib. 128 6.5 EXPERIMENT 12 The previous experiments have found evidence for the existence of various operations associated with the Activity Distribution model. These include the notion that cue effects are due to peripheral input-level operations and that cue effects seem to be due to post-target-onset interactions. One important aspect of the model that still requires verification, however, is the Filter stage, which is responsible for selecting a location for the attention channel. The Filter mechanism involves a lateral-inhibition network that has the effect of suppressing activity distributions at all but the most active location. On the surface, the suppressive function of the filter may seem to contradict the idea that cue effects can occur simultaneously at multiple locations. How can multiple activity distributions affect processing i f they are filtered out? The answer lies in the time course of the filtering operation. In particular, filtering based on a lateral-inhibition network is an iterative process and requires a certain amount of time to run to completion. At the shortest intervals, filtering is relatively ineffective and activation can briefly persist at multiple locations. At longer intervals, however, all but the most dominant activation is suppressed. The hypothesized change in filter effectiveness over time provides a way to test for the involvement of this type of operation in the Unique-Cue paradigm. More specifically, at short intervals it should be possible to observe cue effects at multiple locations, while at longer intervals, cue effects should only occur at a single attended location (e.g., the Unique-Cue location). This pattern of results over time was already measured in Experiment 9, however, those data on their own cannot provide adequate confirmation of the operation of a filter mechanism. The problem is that direct-cue 129 effects are said to be transient, thus it is impossible to determine if the absence of Standard-Cue effects at the later CTOAs was due to filtering or to a time-based attenuation of direct cue effectiveness. It is possible, however, to modify the Unique-Cue paradigm to avoid this confound. One option is to present the Unique-Cue at various intervals before the Standard-Cues appear. The delay between the Unique-Cue and the Standard-Cues permits control over the time allowed for filtering to occur. Moreover, with this technique, Standard-Cues can be presented with an optimal interval separating these cues from the target (100 ms), which should make the Standard-Cues equally effective on all trials i f their effects are not suppressed. The purpose of this experiment was to determine i f cue effects on Standard-Cue trials could be eliminated or diminished by giving the putative Filter mechanism enough time to suppress processing at all but the Unique-Cue location. To this end, the time interval between the onset of the Unique Cue and the Standard Cues (called the Cue-Cue-Onset-Asynchrony or CCOA) was varied from 0 to 300 ms in 100 ms increments. Also of note is the fact that the interval between the Standard-Cues and the target (CTOA) was always 100 ms, which means that these cues should have been optimally effective on all trials. If filtering is involved in this paradigm, then Standard-Cue effects should be diminished or suppressed on longer C C O A trials because enough time is given for the filtering operations to run to completion. 130 6.5.1 Methods Subjects: 12 University of British Columbia undergraduates were paid $10 for participating in two 1-hour sessions. A l l subjects had normal or corrected-to-normal vision. Apparatus and Stimuli: These were the same as described in the General Methods. Procedure: The procedure was the same as described in the General Methods except that the Unique Cue preceded the onset of the Standard cues by either 0, 100, 200, or 300 ms (and remained visible until the subject responded to the target onset; see Figure 18). On 0 ms C C O A trials the Unique Cue appeared at the same time as the Standard Cues. The CTOA between the Standard Cues and the target was always 100 ms on data trials. Design: The Target-Location variable was completely crossed with a CCOA variable (0, 100, 200, & 300ms). The target appeared at either the Unique-Cue location (66.7% of trials), one of the Standard-Cue locations (16.7% of trials), or at one of the Uncued locations (16.7% of trials). In total, there were 1800 data trials consisting of 300 Unique-Cue, 75 Standard-Cue, and 75 Uncued trials for each level of Number-of-Cues. Also, 900 catch trials with 1500 ms CTOAs were divided according to the same 4:1:1 trial ratio. 131 Figure 18: Stimulus display used in Experiment 12. The fixation cross is visible for the Inter-Trial-Interval (ITT), then the Unique-Cue appears alone in the display for the duration ofCue-Cue-Onset-Asynchrony (CCOA). After this, the Standard-Cues appear for the duration of the Cue-Target-Onset-Asynchrony (CTOA). Finally the target appears at either the Unique-Cue, a Standard-Cue, or an Uncued location. Note that all stimuli remain visible until subjects make a response. 132 6.5.2 Results and Discussion The mean error rate in this experiment was 1.5%. A 4x3 repeated measures A N O V A was run on pooled mean response times for all subjects in each condition. The within-subjects factors were Target Location (Unique-Cue, Standard-Cue, or Uncued) and C C O A (0, 100, 200, & 300 ms). Figure 19 shows mean response times averaged over all subjects. Both the main effects of Target Location, F(2M) = 9.69, MSE = 3535.73,p < 0.01, and CCOA, Fa33) = 17.80, MSE = 289.10,p< 0.01, were highly significant. Additionally the C T O A x Target Location interaction, F^M)= 2.40, MSE = 132.03,p = 0.03, was also significant. Mean cue effects are presented in Table 16. Table 16: Mean Cue Effects as a junction of Cue-Cue-Onset-Asynchrony (CCOA) in Experiment 12 (p-values for planned comparisons in brackets). CCOA (ms) Cue Effects 0 100 200 300 Unique Cue 55 (0.003) 53 (0.009) 50 (0.004) 52 (0.003) Standard Cue 28 (< 0.001) 24 (0.002) 8 (0.234) 12 (0.119) In previous experiments, presenting Standard Cues 100 ms before the target onset produced cue effects. In the present experiment, whether or not cue effects occurred on these trials (with the same 100 ms CTOA) depended on how much time subjects had to respond to the appearance of the Unique Cue. If the Unique Cue preceded the Standard Cues by 200 ms or more, then cue effects on the latter trials were significantly diminished. This observation is supported by the significant C C O A x Target Location interaction and the fact that Standard-Cue effects were only significant on 0 and 100 ms C C O A trials. Thus, the attenuation of the Standard-Cue effects at longer CCOAs 133 E X P E R I M E N T 12: M E A N R E S P O N S E T I M E S 420-410-'in E, 400-LU 390-380-H LU 370-CO z 360-o D_ 350-CO LU 340-or 330-320-UNIQUE CUE STANDARD CUE O - UNCUED i i i > 100 200 300 C C O A (ms) CO E, to I -o LU LU LU O E X P E R I M E N T 12: C U E E F F E C T S 80' 70' 60' 50 40 30 20 10 0 -10 -20 UNIQUE CUE STANDARD CUE i i 1 1 100 200 300 C C O A (ms) Figure 19: Mean Response Times and mean Cue Effects as a function of Cue-Cue-Onset-Asynchrony (CCOA) in Experiment 12. 134 supports the hypothesis derived from the Activity Distribution model that opening an attention channel at a location involves filtering-out processing at unattended locations. Another notable finding in this experiment is the disassociation between the time course of the Unique-Cue effects and the time course of suppression. More precisely, i f the time course of suppression is interpreted as reflecting the time required to open an attention channel at a location, it differs from the time required to obtain cue effects. Although opening a channel of attention required 200 ms, Unique-Cue effects were apparent at the 0 ms C C O A (100 ms before the target onset), which suggests that cue effects actually precede the opening of an attention channel. This dissociation is consistent with the explanation of cue effects provided by the Activity Distribution model, which states that cue effects are not due to attentional processing, but rather to interactions between target-related activation and the pre-existing activation topograpy. The suppression effects found in the present experiment are similar to a phenomenon called attentional engagement in the spatial attention literature. Attentional engagement involves a similar procedure; a potential target location is indicated by a highly valid cue (cf. Unique Cue), and the effects of competing stimuli appearing at other locations (cf. Standard Cues) are measured. The basic finding is that i f subjects are given enough time (200 ms) to engage their attention at the location indicated by the highly valid cue, the subsequent appearance of the competing abrupt-onset stimuli wil l have minimal effects on target processing (Theeuwes, 1991b; Yantis & Jonides, 1990). Given the similarity between these procedures, the similar time course of effectiveness (200 ms interval), and the similar suppression of unattended stimuli, it seems reasonable to 135 conclude that the Unique-Cue paradigm taps into the same processes studied in the attentional engagement paradigm. The experiments in this chapter validated some of the basic assumptions of the Activity Distribution model. The assumption that separate peripheral processes are involved in generating cue effects was confirmed by the dissociation between Unique-Cue and Standard-Cue effects over CTOA. Another assumption is that, at short CTOAs, cue effects are not due to focused attention but to the influence that interactions between input-level processes have on the opening of an attention channel. This notion was supported by data suggesting that 1) luminance changes had a transient impact on both Unique-Cue and Standard-Cue effects, 2) cue effects can be caused by pre-target-onset operations, and 3) Unique-Cue effects seem to precede the opening of an attention channel. The final assumption is that filtering operations are involved in the opening of an attention channel. Support for this assumption comes from the finding that i f enough time is given for the filtering operations to become effective, sensory events such as the onset of Standard-Cues at other locations will have attenuated effects on processing. Taken together, these findings suggest that the Activity Distribution model provides a consistent explanation for cue effects in the Unique-Cue paradigm. 136 7. GENERAL DISCUSSION This dissertation was initially motivated by a problem in the spatial attention literature. Most notions regarding the control of attention were based on a single-mechanism view which held that all cue effects were the product of a single indivisible focus of attention that could be controlled in a stimulus-driven or a goal-driven manner (e.g., Sperling & Weichselgartner, 1995). Given mounting evidence, however, that this single focus is not the exclusive source of all direct-cue effects (e.g., Richard, 1995), we are left with an incomplete description of the control of attention. The primary goal of this dissertation was to address this shortcoming by investigating the control of attention under the multiple-cue conditions present in the Unique-Cue paradigm. To this end, several steps leading to a more complete account of direct-cue effects were achieved in this dissertation. In particular, this study has established that the pattern of cue effects obtained with the Unique-Cue paradigm is not simply an artifact of the idiosyncratic procedure used. Furthermore, the experiments in this study have uncovered several properties about how stimulus-driven and goal-driven processes operate under the conditions present in the Unique-Cue paradigm. These properties were used to develop a new explanation of direct-cue effects, and finally, some of the major assumptions of this explanation were verified. On a general level, the explanation of cue effects that emerges from this investigation shares some general principles in common with the single-mechanism view, however, at a specific level there are substantial differences. The similarities include the fact that both views describe attention as a processing focus and that both allow stimulus-driven and goal-driven processes to exert control over the shifting of this focus in visual 137 space. The most significant difference involves what is described as the source of cue effects. In particular, the single-mechanism view holds that cue effects arise exclusively from processing facilitation afforded by the attentional focus. In contrast, the Activity Distribution model holds that cue effects do not arise from attentional processing, but rather from prioritizing attention-shift destinations with activity distributions and the resulting consequences on the time required to open an attention-channel at a specific location. Moreover, because, according to the latter view, the control of attention by stimulus-driven and goal-driven processes is based on parallel operations, this proposal can account for evidence of simultaneous stimulus-driven and goal-driven cue effects at multiple locations in a display. Thus, this new approach solves the problem of extending an explanation of cue effects to cover the data from multiple-cue conditions, which was the initial motivation for this investigation. 7.1 EMPIRICAL CONTRIBUTIONS OF THIS RESEARCH Along with providing a new explanation of direct-cue effects, this investigation has also uncovered several new empirical findings that are relevant to the spatial-cueing literature. The first of these involves the demonstration that stimulus-driven and goal-driven cue effects can occur independently at different locations. Because this was one of the major aims of this dissertation, several experiments provide data supporting this claim. Some of this evidence centres on dissociations between stimulus-driven and goal-driven cue effects across certain factors. In particular, goal-driven cue effects but not simultaneously occurring stimulus-driven cue effects were affected by Unique-Cue Validity (Experiment 7b), and goal-driven cue effects were sustained over time while 138 stimulus-driven cue effects were transient (Experiments, 9 & 10). Similarly, giving subjects sufficient time to open an attention channel at the Unique-Cue location diminished stimulus-driven Standard-Cue effects but did not change goal-driven Unique-Cue effects (Experiment 12). Another finding that supports the notion of independent stimulus-driven and goal-driven cue effects is that they occurred simultaneously at different locations and the respective effects were confined mostly to the corresponding cued locations (Experiment 3). Although other studies have reported simultaneous stimulus-driven and goal-driven cue effects in the same display (e.g., Riggio & Kirsner, 1997; Tepin & Dark, 1992), none have provided evidence from so many different converging sources. Additionally, the validity of the conclusions from these other studies was potentially diminished by methodological problems such as the possible biasing of critical baseline conditions due to blocking by trial type (see Jonides & Mack, 1984). Thus, this investigation provides unprecedented support for the notion that stimulus-driven and goal-driven cue effects can occur simultaneously in the same display. The second new finding is the dissociation between the occurrence of cue effects and the opening of an attention channel found in Experiment 12. Although the data showing the suppressive effects of engaged attention are not new (e.g., Theeuwes, 1991b; Yantis & Jonides, 1990), the possibility that this attention-related suppression is separate from the factors mediating cue effects is new. Previously, cue effects and attentional engagement were thought to reflect the same underlying process (focused attention). As the results of Experiment 12 indicate, however, Unique-Cue effects were evident at least 100 ms before the suppressive effects of focused attention. Thus, the differing time 139 courses for Unique-Cue effects and attentional-engagement effects that were found in this experiment suggest that different processes may be involved. There are a few reasons why this result may not have been accessible to earlier studies. The first is that they did not use a time course that measured as many points before the onset of the second set of cues (they only used CTOAs that measured cue effects on targets appearing after an attention channel was supposedly open). Another reason is that attention in these experiments was directed by a symbolic cue. Cue effects produced in this manner can be interrupted or briefly eliminated by abrupt onsets appearing at other locations before attention is engaged (e.g., Muller & Rabbitt, 1989). Thus, cue effects at the location of the symbolic cue could have been interrupted by the onset of the second abrupt-onset stimulus. In contrast, the highly-valid Unique Cue used to "engage" attention in the present investigation was a direct cue. Because direct cues do not seem to be as susceptible to interruptions (e.g., Muller & Rabbitt, 1989), cue effects at this location would have been preserved. The third set of new results involves the effect of cue luminance on stimulus-driven cue effects. Although the effects of luminance on symbolic cues (e.g., Hughes, 1984) and abrupt onsets in visual search (e.g., Theeuwes, 1995) have been examined, to my knowledge, no one has investigated these effects for stimulus-driven cue effects. The data from Experiment 5 show a monotonic relationship between luminance and direct-cue effects. The fourth new finding is that, although stimulus-driven cue effects can occur simultaneously at multiple locations, these effects may be mediated by a limited-capacity mechanism that only permits the expression of cue effects at a limited number of cued 140 locations (Experiment 6). While similar capacity limits have been reported for other tasks (e.g., Pylyshyn & Storm, 1988; Yantis & Johnson, 1990), they have not yet been shown with cue effects. The current study shares with the other studies both the fact that the onsets of the relevant stimuli are associated with abrupt luminance changes, and that a similar capacity limit is involved (4-5 items). This suggests that a common capacity-limiting structure based on luminance changes (e.g., an indexing mechanism) may mediate these effects. The final new result involves data suggesting that subjects may have probabilistically alternated between different response strategies under some conditions in the Unique-Cue paradigm (Experiment 7a). The two strategies identified in the present investigation, which involved a feature-detection strategy and an attention-related strategy, are similar to response strategies identified in some visual search tasks (e.g., Bacon & Egeth 1994). In addition, other direct-cue studies have reported the existence of "attentional-control settings" that seem to determine what type of stimulus change in a display wil l capture attention (Folk, Remington, & Johnston, 1992; Folk, Remington, & Wright, 1994; however see Yantis 1993; and Theeuwes, 1994). Together, these findings underscore the flexibility that subjects may have in performing spatial-cueing tasks as well as the need to consider the effect that experiment conditions may have in biasing subjects towards using particular strategies. In many of the experiments reported in this investigation, cue-validity and other incentives were used to induce subjects to adopt an attention-related strategy. An experiment design involving similar incentives may provide a way to reduce inter-subject variability i f it forces them to adopt the same strategy. 141 This dissertation contributed several new empirical findings to the visual attention literature. These new findings highlight the usefulness of the Unique-Cue paradigm as a research tool for investigating visual attention. 7.2 CONNECTION OF UNIQUE-CUE PARADIGM TO T H E VISUAL ATTENTION LITERATURE The relevance of the previously described findings depends on the assumption that the Unique-Cue paradigm taps into the same processes that are typically investigated in visual attention studies involving abrupt onsets. While the Unique-Cue paradigm shares several aspects in common with more traditional spatial cueing approaches, there are enough differences in task demands and procedures to require justification before any connection is made. In defence of the notion that both procedures investigate the same basic attention-related processes, many common spatial-cueing results, as well as some less intuitive findings (e.g., 4-5 item capacity limit) are found in both types of investigation. See Table 18 for a summary of the major similarities between these approaches. 142 Table 17: Analogous results found both in the Unique-Cue paradigm and the Visual Attention Literature. Result Unique-Cue Paradigm Visual Attention Literature Goal-driven cue effects require high cue validity Experiment 7a, 7b Muller & Rabbitt, 1989; Jonides, 1981; Nakayama & Mackeben, 1989; Weischelgartner & Sperling, 1997 Stimulus-driven cue effects occur regardless of cue validity Experiments 7a, 7b Muller & Rabbitt, 1989; Jonides, 1981; Nakayama & Mackeben, 1989; Weischelgartner & Sperling, 1997 Goal-driven cue effects are sustained Experiments 9,10 Cheal&Lyon, 1991; Muller & Rabbitt, 1989; Nakayama & Mackeben, 1989; Stimulus-driven cue effects are transient Experiments 9, 10 Muller & Rabbitt, 1989; Nakayama & Mackeben, 1989; Cue effects are location-specific Experiment 3 Muller & Humphreys, 1991; however see Henderson 1991 Processing of abrupt onset stimuli is capacity-limited Experiment 6 Yantis & Johnson, 1990; Yantis & Jones, 1991 Attentional engagement occurs after 200 ms CCOAs Experiment 12 Yantis & Jonides, 1990; Theeuwes, 1991b 7.3 A C O M M E N T ON PHYSIOLOGICAL CONNECTIONS Along with establishing connections between the Unique-Cue paradigm and other findings in the spatial-cueing literature, it is also tempting to search for connections between the Activity Distribution model and physiological activity in the brain. This approach, however, will not be taken in this dissertation, although other researchers have attempted to make this connection elsewhere (e.g., LaBerge, 1995; LaBerge & Brown, 1989). The main reason why this approach is avoided is because the elements of the Activity Distribution model as presented in this dissertation are purely cognitive constructs. I do not think that it is clear as to what would constitute the physiological correlates of abstract concepts such as a "source of information" or a "top-down process." 143 In this case, the basic elements of the model are just not yet approachable from a physiological perspective. While it is certainly possible to derive more general elements from physiological data (e.g., that the visual system contains areas particularly responsive to luminance transients; e.g., Wurtz & Albano, 1980), applying this approach to more complex and abstract elements simply yields ad-hoc associations. Thus, fortius reason, the dissertation wil l avoid attempts to link cognitive constructs to brain function. 7.4 ACTIVITY DISTRIBUTION M O D E L AND OTHER D I R E C T - C U E FINDINGS Although physiological connections will not be made, because the findings from the Unique-Cue paradigm seem to be relevant to spatial-cueing studies in general, it is still possible to apply the current explanation of cue effects in the Unique-Cue paradigm to data from other direct-cue studies. In particular, this explanation would have to account for previous data involving a single direct cue, as well as more recent data indicating that stimulus-driven direct-cue effects can occur at multiple locations independently from goal-driven cue effects. To this end, some of the major findings involving direct cues (see Table 1) are explained below using the Activity Distribution model. Fast time course of cue effectiveness: Following the onset of a single direct cue, a single involuntary activity distribution would be formed in the Luminance Map. Without competition from other activity distributions, the lateral-inhibition network in the filter mechanism would open an attention channel quickly. In addition, cue effects would still be present before an attention channel was open because the cued location would be closer to threshold and have a "head start" (yielding a cue effect) relative to other uncued locations. 144 Cue effects still occur with low cue validity: Because the generation of activity distributions is said to be an involuntary consequence of the onset of the direct cue, a corresponding activity distribution would be formed regardless of cue validity. If cue validity was high enough to induce additional activation from Goal-Driven input, then the combined activity distribution for that location at the Interaction Map, and the corresponding cue effects, would be even greater (cf. Muller & Rabbitt, 1989). Direct cue effects are not interrupted by secondary cognitive tasks: Again, because initial the generation of activity distributions is involuntary, they will occur regardless of what additional activity is occurring at higher levels of processing. Moreover, because activity distributions are generated independently at multiple locations, the onset of secondary direct cues would also not affect the generation of the activity distribution at an initial direct-cue location, unlike in symbolic-cue experiments in which secondary direct cues do interrupt cue effects (Muller & Rabbitt, 1989). Practice has no effect on the time course of direct-cue effects: Because the pathway from the Luminance Map to the filter mechanism does not involve any "controlled" or voluntary operations that can be automatized (e.g., SHffrin & Schneider, 1977), practice should have no effect on the rate at which cue effects occur at a direct-cue location. Direct-cue effects can occur at multiple locations: Because activity distribution are said to be involuntarily triggered independently at multiple locations in the Luminance Map, the locations of the multiple direct cues would have a "head-start" relative to uncued locations (as long as there was an available index). Moreover, because there would not be any single dominant activity distribution, it is likely that none of the 145 activity distributions would surpass the relative threshold on its own before all the activity distributions faded out. This means that on multiple-cue trials, none of the cues would capture attention (c.f. Richard, 1995). Not only does the Activity Distribution model account for the data reported in the present dissertation, but it can also provide a plausible explanation of some of the major findings in the direct-cue literature. 7.5 CONNECTION T O O T H E R NOTIONS O F T H E CONTROL O F ATTENTION The discussion in the previous section shows how the Activity Distribution model can produce a reasonable explanation of some of the most common findings in the direct-cue literature. This is not the only area, however, in which sMfting attention in visual space is relevant. Visual attention has also been implicated in symbolic-cueing tasks, object-based attention tasks, and visual search tasks. The Activity Distribution model as presented in this dissertation cannot deal with these cases because it is insufficiently developed. The critical question, however, is whether or not it can be further elaborated to explain data from these areas of the larger visual attention literature. The answer is yes. In fact, the original version of this model (LaBerge & Brown, 1989) actually contained the elements necessary to allow it to account for data from these areas. The reason behind this flexibility is that the core of the Activity Distribution model (the Interaction Map and the Filter) operates in a simple way to solve a very general problem. The problem that it solves is that of opening up an access channel (attention) between low-level information (e.g., luminance changes, stimulus features), and the higher-level processes that require this low-level information to control behaviour. For example, the 146 high-level decision processes involved in determining the orientation of a target line would require the feature information about the line orientation coded at the target location in the low-level feature maps. This information would be obtained by opening an attention channel at the target location (cf. Treisman & Gelade, 1980). The way the Activity Distribution model solves this problem is that it allows input sources (e.g., Luminance Map) to generate activity distributions that affect where the passive attention channel opens. Put another way, the Interaction Map combined with the activity distributions provides a "universal" interface that allows representations based on potentially incompatible information to interact using a common "currency" (activity distributions) to influence the allocation of attention. Thus, any input source that can generate or influence the appropriate activity distributions can exert some control over attention. External sources of activation must exhibit two basic properties. The first is that they must represent a type of information that is useful for directing attention. This information must be represented spatiotopically (compatible with the Interaction Map & the Filter), or if it is organized according to another dimension (e.g., object-based) then the representation must be able to convert its output into location-based units. The second property is that the representation must have a means of generating activity distributions. This process can be involuntary (e.g., luminance changes in the Luminance Map) or it can be based on "top-down" goals. The latter case requires access methods or routines so that higher-level processes can interact with a representation to actively generate appropriate activity distributions (e.g., boost the activation associated with the feature "red"). A collateral effect of these access routines is that is that they may provide 147 alternative non-attentional ways to perform certain tasks. For example, i f an access routine for a feature representation could determine the presence or absence of a specific feature, such as a diagonal line orientation, then this information could be used to direct a response in a target-orientation identification task (c.f. Experiment 7a). By adding additional representations with these properties to the Activity Distribution model, it is possible to expand the model so that it can also deal with data from other studies involving attention shifts in visual space. The purpose of this section is to show that it is possible to expand the Activity Distribution model to account for data from symbolic-cueing, object-based attention, and visual search tasks. This will involve a description of a way to implement the representations to deal with each type of task, along with a brief description of some of the predictions that this type of model would make. One caution that must be mentioned when modifying the Activity Distribution model in this way is the problem of making this model completely arbitrary or "ad-hoc" by adding new properties or components whenever new data arise. The ideal method for adding new components would be to derive them from first principles, such as those dictated by the physiological plausibility of specific additions. For example, the properties of the new components would have to be constrained by how the putative structures in the brain were observed to function. This approach is being preliminarily pursued for this model by some researchers (LaBerge, 1995). However, because most of these first principles are largely unspecified at this point, the new components in this dissertation will instead be based on the type of information used to control attention. In 148 particular, these additions wil l cover attentional control associated with feature-, location-and object- based information. Another strategy for adding some validity to additional components is to identify global constraints derived from the general operating principles of the model and to show how the new components conform to these constraints. More specifically, this involves finding evidence of the effects of the core aspects of the model (e.g., the attention-channel opening process), on empirical data from the areas covered by the new additions. One constraint is that the activation outputs from new components must interact with other activation in the Interaction Map. This has identifiable consequences on the expression of these effects. Most notably, there should be evidence of location-based effects, and the effects of a new component should be able to interact directly with effects generated from other components (e.g., similar to the interaction between Luminance Map and Goal-Driven Input activation in Experiments 4 & 10). Another constraint is that if a new component can be influenced by "top-down" elements, then an interface for this interaction is required. One consequence of this is that it may provide alternative non-attentional methods for performing a task (cf. Experiment 7a). Note that this effect is in direct opposition to the previously described consequence involving the presence of location-based effects. In this case, procedures must be used to systematically control the occurrence of "top-down" involvement (cf. Experiment 7b). Thus, the addition of new properties to the Activity Distribution model must comply with the pre-existing constraints of the model and it should also be possible to observe the effects of these constraints. 149 With this caveat in mind, it is possible to apply the basic principles of the Activity Distribution model to the larger attention literature. The following discussion will avoid an extensive review of the literature and focus mostly on reporting evidence of the previously described integration constraints. 7.5.1 Symbolic-Cue effects: As described in the Introduction, symbolic-cue effects reflect the voluntary allocation of visual attention to a specific location in visual space (e.g., Posner, 1980). In the Activity Distribution model, this process would involve a location-based representation that would be similar to the Luminance Map. One difference, however, is that activity distributions would not be an involuntary consequence of stimulus-driven events, but rather they would be generated by mtervening "top-down" processes. Thus, this representation would require access routines that deal with location information (c.f. LaBerge & Brown, 1989). "Top-down" processes would select a location and an activity distribution would be created at the selected location and persist as long as the "top-down" processes continue to designate that location. Once the activity distribution passed to the Interaction Map, it would influence the opening of an attention channel in the same way as input from other sources would. The effectiveness of this modification in accounting for symbolic cueing results is reviewed elsewhere (LaBerge & Brown, 1989). One consequence of this modification is that the constraints governing the operation of the "top-down" processes would determine the properties associated with the corresponding activity distributions. If it takes time to select a location, then cue effects 150 would have a slower-rising time course of effectiveness (e.g. Nakayama & Mackeben, 1989). Similarly, i f the "top-down" processes are disrupted by factors such as secondary tasks and low cue validity, then cue effects would also be disrupted (e.g. Jonides, 1981). Because this type of activation is inherently location-based, the constraint necessitating the presence of location-based effect is somewhat meaningless. Rather, the evidence that provides some validation of the type of component comes from a study that showed that observers have goal-driven control over the spatial extent of goal-driven cue effects produced by symbolic cues (LaBerge & Brown, 1989). These cue effects were distributed in space as a gradient (similar in shape to the putative activity distributions), and it was not possible to account for this pattern using a moving spotlight model of attention. 7.5.2 Object-based Attention: Object-based attention shifts are similar to attention shifts that occur in response to symbolic cues. The major difference, however, is that rather than selecting a specific location for attention, attention is directed to specific objects. Accordingly, the representations involved with this process would encode the perceptual objects present in the visual scene. Furthermore, "top-down" processes would access and select an object from this representation and this selection would result in the generation of activity distributions that corresponded to all of the surface area (contiguous or discrete) associated with the single selected object. The resulting activity distribution (sustained as long as the object was selected) would make it likely that an attention channel would eventually open at all of the locations associated with that object. If an object was 151 partially occluded and separated into non-contiguous regions, the sustained activity distributions would permit multiple attention channels at all the relevant locations allowing all the information associated with the same object to be passed on (multiple channels would be unstable without sustained activity distributions; LaBerge & Brown, 1989). Consistent with the integration constraint requiring location-based effects, some studies indicate that object-specific effects related to attention also seem to be accompanied by additional location-based effects (Egly, Driver, & Rafal, 1994; Tipper, Driver, & Weaver, 1991). Another constraint specific to the Activity Distribution model is that shifting an attention channel between two objects should not be affected by the distance between two objects. This is because selecting an object is said to generate a corresponding activity distribution regardless of its position in space. Thus, the cost of switching would only involve the time required to open an attention channel and not any cost associated with traversing attention over different distances in visual space. The results of some studies indicate a similar insensitivity to spatial separation (e.g., Duncan 1984; Vecera & Farah, 1994). 7.5.3 Visual Search: The Activity Distribution model also has the capacity to allow specific features to affect the search sequence based on the priority assigned to certain features present in the display. An example of this is was the opening of an attention channel at the Unique-Cue location based on its unique feature (the colour "red"). As previously mentioned, this type of charmel-opening process was implicitly assumed to be part of the Goal-Driving 152 Input module. It involves a feature-based representation that is organized into feature-maps (e.g., colour, line orientation, curvature maps) that code the value of a specific feature at a specific location. Activity would be generated at all the locations containing features, and the magnitude of the activity distributions would be based on how unique or conspicuous the collection of features at a location was. Moreover, the relative magnitude of these activity distributions could be modified by "top-down" processes through the appropriate access routines. Thus, i f a certain feature (e.g., "red") was particularly useful for directing attention, then the activation associated with that feature could be voluntarily boosted. This additional activation would increase the likelihood that that location would be attended with a higher priority because it would have the largest sustained activity distribution in the Interaction Map. The effectiveness of a similar type of feature-based guidance of attention in explaining visual search results is reviewed elsewhere (e.g., Wolfe, 1994). One prediction of this type of representation is that if the magnitude of feature-based activity distributions was affected by the "uniqueness" of a feature relative the surrounding features, then it should be possible to produce cue effects based on feature differences rather than just on luminance changes. In one target identification task, a target was preceded by a field of homogeneous vertical or horizontal lines that included one line oriented perpendicular to the other lines (Joseph & Optican, 1996). These researchers found significant cue effects at the location of the unique line, even in absence of goal-driven incentives to attend to this stimulus. This result is consistent with the notion that the relative uniqueness of the perpendicular line may have generated a 153 larger activity distribution at that location, thus speeding the opening of an attention channel. There is also some evidence for some of the integration constraints. In the case of visual search, location-based effects are manifest in a specific form. More specifically, the Activity Distribution model holds that attentional control based on feature information would not have the effect of improving the quality of the perception of a designated feature, but rather it would improve the probability that the locations at which that feature occurs would be attended to with a higher priority. The results of several experiments are consistent with this prediction (e.g., Moore & Egeth, 1998; Shih & Sperling, 1996; Tsal & Lavie, 1993). Another constraint is that because "top-down" access routines are said to be involved, it should be possible to perform some tasks using either a non-attention-related strategy involving these routines or an attention-related strategy. This notion is consistent with the findings from one study in which researchers induced subject into using one of these strategies based on experiment conditions (Bacon & Egeth, 1994; see also Experiment 7b). The Activity Distribution model can be modified to accommodate the findings from several research areas involving visual attention. The general properties of data involving symbolic-cue tasks, object-based attention tasks, and visual search tasks, can be explained by adding new components to the Activity Distribution model that allow activity distributions associated with the representations that code the information used in these tasks to interact with a common filter mechanism. This putative interaction provides a way to test the central notion that activity distributions from different types of representation interact to drive the filter mechanism. More specifically, according to this 154 common filter arrangement, it should be possible to find evidence that activation associated with one type of representation can be filtered out or suppressed i f another type of representation causes an' attention channel to open (cf. Experiment 12). This type of evidence would provide some validation for this particular approach to expanding the Activity Distribution model to explain other data involving shifting of visual attention. In summary, the Unique-Cue paradigm provides a new way to investigate stimulus-driven and goal-driven control in spatial-cueing paradigms. This technique has proven useful for understanding the role of stimulus-driven and goal-driven processes under multiple direct-cue conditions. It has lead to the discovery of several new findings, and to the development of a new explanation of cue effects. It has advantages over previous techniques because it can take into account and isolate direct-cue effects at multiple locations. This has made this paradigm useful in filling the explanatory void left by the inability of single-mechanism views of attention 155 8. REFERENCES Bacon, W. F., & Egeth, H.E. (1994). Overriding stimulus-driven attentional capture. Perception & Psychophysics, 55,485-496. Briand, K . A. , & Klein, R. M . (1987). Is Posner's "beam" the same as Treisman's "glue"?: On the relation between orienting and Feature Integration Theory. Journal of Experimental Psychology: Human Perception & Performance, 13, 228-241. Briand, K . A . (1998). Feature integration and spatial attention: More evidence of a dissociation between endogenous and exogenous orienting. Journal of Experimental Psychology: Human Perception & Performance, 24,1243-1256. Breitmeyer, B. G., & Ganz, L . (1976). Implications of sustained and transient channels for theories of visual pattern masking, saccadic suppression, and information processing. Psychological Review, 83, 1-36. Broadbent, D. (1958). Perception and communication. Oxford: Pergamon. Cheal, M . , & Lyon, D. R. (1991). Central and peripheral precueing of forced-choice discrimination. Quarterly Journal of Experimental Psychology,4 3A, 859-880. Duncan, J. (1984). Selective attention and the organization of visual information. Journal of Experimental Psychology: General, 113, 501-517. Egeth, H. E., & Yantis, S. (1997). Visual Attention: Control, representation, and time course. Annual Review of Psychology, 48,269-297. Egly, R., Driver, J., & Rafal, R. D. (1994). Shifting visual attention between objects and locations: Evidence from normal and parietal lesion subject. Journal of Experimental Psychology: General, 123,161-177. Eriksen, C. W., & Yeh, Y . (1985). Allocation of attention in the visual field. Journal of Experimental Psychology: Human Perception & Performance, 11, 583-597. Folk, C. L , Remington, R.W., & Johnston, J. C. (1992). Involuntary covert orienting is contingent on attentional control settings. Journal of Experimental Psychology: Human Perception & Performance, 18,1030-1044. Folk, C. L , Remington, R.W., & Wright, J. H . (1994). The structure of attentional control: Contingent attentional capture by apparent motion, abrupt onset, and color. Journal of Experimental Psychology: Human Perception & Performance, 20, 317-329. Hallett, P. E. (1963). Spatial Summation. Vision Research. 3, 9-24. 156 Henderson, J. (1991). Stimulus discrimination following covert attentional orienting to an exogenous cue. Journal of Experimental Psychology: Human Perception & Performance, 17, 91-106. Heinze, H . , Luck, S. J., Munte, T. F., Gos A. , Mangun, G. R., & Hillyard, S. A . (1994). Attention to adjacent and separate positions in space: An electrophysiological analysis. Perception & Psychophysics, 56,42-52. Hughs, H . C. (1984). Effects of flash luminance and positional expectancies on visual response latency. Perception & Psychophysics, 36, 177-184. James, W. (1890). Principles of psychology, Volume 1. New York: Holt. Johnson, D.N. , & Yantis, S. (1995). Allocating visual attention: Tests of a Two-Process model. Journal of Experimental Psychology: Human Perception & Performance, 21, 1376-1390. Jonides, J. (1981). Voluntary vs. automatic control over the mind's eye's movement. In J.B. Long & A.D. Baddeley (Eds.), Attention and performance IX, (pp. 187-203). Hillsdale, NJ: Erlbaum. Jonides, J., & Mack, R. (1984). On the cost and benefit of cost and benefit. Psychological Bulletin, 96, 29-44. Joseph, J. S., & Optican, L. M . (1996). Involuntary attentional shifts due to orientation differences. Perception & Psychophysics, 58, 651-665. Kiefer, R. J., & Siple, P. (1987). Spatial constraints on the voluntary control of attention across visual space. Canadian Journal of Psychology, 41,474-489. Klein, R. (1994). Perceptual-motor expectancies interact with covert visual orienting under conditions of endogenous but not exogenous control. Canadian Journal of Experimental Psychology, 48, 167-181. Klein, R., & McCormick, P. (1989). Covert visual orienting: Hemifield activation can be mimicked by zoom lens and midlocation placement strategies. Acta Psychologica, 70, 235-250. Koch, C , & Ullman, S. (1985). Shifts in selective visual attention: Toward the underlying neural circuitry. Human Neurobiology, 4, 219-227. Kramer, A . F., & Hahn, S. (1995). Splitting the beam: Distributions of attention over noncontinguous regions of the visual field. Psychological Science, 6,381-385. Krose, B . J. & Julez, B . (1989). The control and speed of shifts of attention. Vision Research, 28,1607-1619. 157 LaBerge, D. (1995). Attentional processing: The brains's art of mindfulness. Cambridge, M A : Harvard University Press. LaBerge, D., & Brown, V . (1989). Theory of attentional operations in shape identification. Psychological Review, 96, 101-124. Lambert, A. , Spencer, E, & Mohindra, N . (1987). Automaticity and the capture of attention by a peripheral display change. Current Psychological Research & Reviews, 6, 136-147. McCormick, P. A., & Klein, R. (1990). The spatial distribution of attention during covert visual orienting. Acta Psychologica, 75, 225-242. Moore, C. M . , & Egeth, H. (1998). How does feature-based attention affect visual processing? Journal of Experimental Psychology: Human Perception & Performance, 24, 1296-1310. Muller, H . J. (1994). Qualitative differences in response bias from spatial cueing. Canadian Journal of Experimental Psychology, 48, 218-241. Muller, H . J., & Findlay, J. M . (1988). The effect of visual attention on peripheral discrimination thresholds in single and multiple element displays. Acta Psychologica, 69, 129-155. Muller, H . J., & Humphreys, G. W. (1991). Luminance-increment detection: Capacity-limited or not? Journal of Experimental Psychology: Human Perception & Performance, 17, 107-124. Muller, H . J., & Rabbitt, P. M . A . (1989). Reflexive and voluntary orienting of visual attention: Time course of activation and resistance to interruption. Journal of Experimental Psychology: Human Perception & Performance, 75, 315-330. Nakayama, K. , & Mackeben (1989). Sustained and transient components of focal visual attention. Vision Research, 11,1631 -1647. Neisser, U . (1967). Cognitive Psychology. New York: Appleton-Century-Crofts. Posner, M . I. (1980). Orienting of attention. 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 & Performance X, (pp. 531-556). Hillsdale, New Jersey: Erlbaum. 158 Pylyshyn, Z. W. (1989). The role of location indexes in spatial perception: A sketch of the FINST spatial-index model. Cognition, 32, 65-97. Pylyshyn, Z. W., & Storm, R. W. (1988). Tracking multiple independent targets: Evidence for a parallel tracking mechanism. Spatial Vision, 3,179-197. Pylyshyn, Z. W., Burkell, J., Fisher, B. , Sears, C , Schmidt, W., & Trick, L. (1994). Multiple Parallel Access in Visual Attention. Canadian Journal of Experimental Psychology, 48, 167-181. Richard, C. M . (1995). Simultaneous response facilitation at multiple locations in visual space. Unpublished masters thesis, Simon Fraser University. Riggio, L. , & Kirsner, K . (1997). The relationship between central cues and peripheral cues in covert visual orientation. Perception & Psychophysics, 59, 885-899. Schneider, W., & SMffrin, R. M . (1977). Controlled and automatic information processing: I: Detection, search, and attention. Psychological Review, 84,1-66. SMffrin, R. M . , & Schneider, W. (1977). Controlled and automatic information processing: I: Perceptual learning, automatic attending, and a general theory. Psychological Review, 84, 127-190. Shih, S., & Sperling, G. (1996). Is there feature-based attentional selection in visual search? Journal of Experimental Psychology: Human Perception & Performance, 22, 758-779. Shulman, G. L. , Remington, R. W., & McLean, J. P. (1979). Moving attention through visual space. Journal of Experimental Psychology: Human Perception & Performance, 5, 522-526. Sperling, G., & Weichselgartner, E. (1995). Episodic theory of the dynamics of spatial attention. Psychological Review, 102, 503-532. Sternberg, S. (1975). The discovery of processing stages: Extensions of Donder's method. Acta Pschologica, 30,276-315. Tassinari, G., Aglioti, S., Chelazzi, L. , Peru, A. , & Berlucchi, G. (1994). Do Peripheral non-informative cues induce early facilitation of target detection? Vision Research, 34, 179-189. Tepin, M . B., & Dark, V . J. (1992). Do abrupt-onset peripheral cues attract attention automatically? Quarterly Journal of Experimental Psychology, 45A, 111-132. Theeuwes, J. (1991). Exogenous and endogenous control of attention: The effect of visual onsets and offsets. Perception & Psychophysics, 49, 83-90. 159 Theeuwes, J. (1992). Perceptual selectivity for color and form. Perception & Psychophysics, 51, 599-606. Theeuwes, J. (1994). Stimulus-driven capture and attentional set: selective search for color and visual abrupt onsets. Journal of Experimental Psychology: Human Perception & Performance, 20,799-806. Theeuews, J. (1995). Abrupt luminance change pops out; abrupt colour change does not. Perception & Psychophysics, 57, 637-644. Tipper, S., Driver, J., & Weaver, B. (1991). Object-centred inhibition of return of visual attnention. Quarterly Journal of Experimental Psychology, 43(A), 289-298. Todd, J. T., & Van Gelder, P. (1979). Implications of a transient-sustained dichotomy for the measurement of human performance. Journal of Experimental Psychology: Human Perception & Performance, 6, 625-638. Trick, L . M . , & Pylyshyn, Z.W. (1993). What enumeration studies can show us about spatial attention: Evidence for limited capacity preattentive processing. Journal of Experimental Psychology: Human Perception & Performance, 19,331-351. Treisman, A. (1960). Contextual cues in selective listening. Quarterly Journal of Experimental Psychology, 12, 242-248. Treisman, A. , & Gelade, G. (1980). A feature integration theory of attention. Cognitive Psychology, 12, 97-136. Treisman, A. , & Sato, S. (1990). Conjunction search revisited. Journal of Experimental Psychology: Human Perception & Performance, 15,459-478. Treisman, A. , & Schmidt, H. (1982). Illusory conjunctions in the perception of obj ects. Cognitive Psychology, 14, 107-141. Tsal, Y . , & Lavie, N . (1993). Location dominance in attending to color and shape. Journal of Experimental Psychology: Human Perception & Performance, 19, 131-139. Vecera, S. P, & Farah, M . J. (1994). Does visual attention select objects or locations? Journal of Experimental Psychology: General, 123, 146-160. Warner, C. B. , Juola, J. F., & Koshino, H . (1990). Voluntary allocation versus automatic capture of visual attention. Perception & Psychophysics, 48, 243-251. Warren, R. M . , & Warren, R. P. (1968). Helmholtz on perception: Its physiology and development. New York: Wiley. 160 Weichselgartner, E., & Sperling, G. (1987). Dynamics of automatic and controlled visual attention. Science, 238, 778-780. Wolfe, J. M . (1994). Guided Search 2.0: A revised model of visual search. Psychonomic Bulletin & Review, 1, 202-238. Wright, R. D. (1994). Shifts of visual attention to multiple simultaneous location cues. Canadian Journal of Experimental Psychology, 48,205-216. Wright, R.D., & Richard, C M . (1995). Subitizing abrupt-onset visual stimuli. Poster session presented at the annual meeting of the Canadian Society for Brain, Behaviour, and Cognitive Science, Halifax Nova Scotia. Wright, R.D., & Richard, C M . (In Preparation). Simultaneous cue effects at multiple visual locations. Wright, R. D., Richard, C. M . , & McDonald, J. J. (1996). Neutral location cues and cost/benefit analysis of visual attention shifts. Canadian Journal of Experimental Psychology, 49, 540-548. Wurtz, R. H. , & Albano, J. E. (1980). Visual-motor function of the primate superior colliculus. Annual Review ofNeuroscience, 3, 189-226. Yantis, S. (1992). Multielement visual tracking: Attention and perceptual organization. Cognitive Psychology, 24, 295-340. Yantis, S. (1993). Stimulus-driven attentional capture and attentional control settings. Journal of Experimental Psychology: Human Perception & Performance, 19, 676-681. Yantis, S., & Hillstrom, A. P. (1994). Stimulus-driven attentional capture: evidence from equiluminant visual objects. Journal of Experimental Psychology: Human Perception & Performance, 20, 95-107. Yantis, S., & Jones, E. (1991). Mechanisms of attentional selection: Temporally modulated priority tags. Perception & Psychophysics, 50,166-178. Yantis, S., & Johnson, D. N . (1990). Mechanisms of attentional priority. Journal of Experimental Psychology: Human Perception & Performance, 16, 812-825. Yantis, S., & Jonides, J. (1984). Abrupt visual onsets and selective attention: Evidence form visual search. Journal of Experimental Psychology: Human Perception & Performance, 10, 601-621. 161 Yantis, S., & Jonides, J. (1990). Abrupt visual onsets and selective attention: Voluntary versus automatic allocation. Journal of Experimental Psychology: Human Perception & Performance, 16, 121-134. 162 APPENDIX A : M E A N RESPONSE T I M E TABLES Mean response times (RT), standard error (SE), and error-rates in percent (%E) are presented for each experiment. Experiment 1 Target Location RT SE % E Unique Cue 473 19 2.6 Standard Cue 485 19 3.3 Uncued 519 24 3.7 Experiment 2 Target Location RT SE % E Unique Cue 341 16 2.3 Standard Cue 371 18 1.6 Uncued 388 19 1.6 Experiment 3 NUMBER OF STANDARD CUES Target 1 2 3 Location RT SE % E RT SE % E RT SE % E Unique Cue 313 11 2.5 311 11 2.5 313 12 2.6 Standard Cue 346 22 2.2 344 27 2.2 345 23 2.4 Uncued 367 25 2.3 362 22 2.4 367 25 2.5 Experiment 4 C U E DURATION Target Sustained Transient Location RT SE % E RT SE % E Unique Cue 319 15 1.3 336 17 1.5 Standard Cue 342 17 1.9 357 18 1.9 Uncued 360 18 1.8 363 16 1.9 163 Experiment 5 C U E LUMINANCE Target 1 2 3 4 Location RT SE %E RT SE %E RT SE %E RT SE %E Standard Cue 374 12 1.8 354 13 2.1 342 13 1.7 335 14 1.8 Uncued 378 13 2.2 360 13 2.3 354 14 2.2 355 14 1.8 Experiment 6 Target NUMBER OF STANDARD CUES Location 1 2 3 4 5 6 7 8 RT 373 373 371 372 373 377 378 383 Standard Cue SE 13 13 13 14 14 15 14 13 % E 2.3 2.3 2.3 2.0 2.3 2.0 2.1 2.8 RT 416 399 394 392 392 385 388 392 Uncued SE 14 15 14 15 15 14 14 14 % E 2.7 2.3 2.1 2.1 2.7 2.1 2.5 2.3 Experiment 7a UNIQUE- C U E VALIDITY Target 33% 66% 80% Location RT SE % E RT SE % E RT SE % E Unique Cue 305 9 1.5 301 11 1.7 305 9 2.0 Standard Cue 309 8 1.8 310 11 2.5 323 10 1.4 Uncued 320 10 1.6 326 13 1.5 347 13 1.6 Experiment 7b UNIQUE - C U E VALIDITY Target 12.5% 80% Location RT SE % E RT SE % E Unique Cue 476 11 2.3 474 11 1.7 Standard Cue 478 10 1.8 496 11 1.9 Uncued 495 11 1.8 516 12 2.3 Experiment 8 CTOA Target 100 ms 200 ms Location RT SE %] E RT SE % E Unique Cue 488 36 6. 3 449 37 5.1 Standard Cue 506 38 6. D 455 37 6.3 Uncued 524 38 4. 7 466 39 5.8 Experiment 9 CTOA Target 100 ms 200 ms 300 ms 400 ms Location RT SE %E RT SE %E RT SE %E RT SE %E Unique Cue 330 11 1.5 291 12 1.2 285 12 2.0 291 11 1.3 Standard Cue 354 13 1.8 329 14 1.6 324 16 1.6 338 16 1.2 Uncued 374 15 1.8 331 14 2.5 323 15 1.9 330 14 2.3 Experiment 10 CTOA 100 ms 400 ms Target Dim Bright Dim Bright Location RT SE %E RT SE %E RT SE %E RT SE %E Unique Cue 439 21 2.5 420 21 2.6 362 20 5.1 363 21 2.5 Standard Cue 467 25 1.8 458 24 2.5 413 22 4.8 403 23 1.8 Uncued 487 26 2.7 493 26 1.6 415 21 4.1 407 21 2.7 Experiment 11 a CTOA Target 0 ms 33 ms 66 ms 100 ms Location RT SE %E RT SE %E RT SE %E RT SE %E Unique Cue 385 24 2.2 360 25 1.8 342 24 1.8 330 24 1.7 Standard Cue 412 26 1.4 381 27 1.4 365 27 2.1 357 25 1.7 Uncued 442 31 2.4 418 31 1.9 385 30 2.0 379 26 2.0 165 Experiment 1 lb C T O A Target Oms 100 ms Location RT SE % E RT SE % E Unique Cue 397 14 1.7 344 13 1.0 Standard Cue 415 14 1.7 367 13 0.6 Uncued 434 17 1.7 385 14 0.8 Experiment 12 C C O A Target 0 ms 100 ms 200 ms 300 ms Location RT SE %E RT SE %E RT SE %E RT SE %E Unique Cue 355 17 1.9 332 20 2.4 326 20 2.1 329 21 1.7 Standard Cue 382 21 1.2 361 21 1.7 368 24 1.8 369 24 2.2 Uncued 410 24 1.9 385 24 2.9 376 25 2.3 381 24 2.5 

Cite

Citation Scheme:

        

Citations by CSL (citeproc-js)

Usage Statistics

Share

Embed

Customize your widget with the following options, then copy and paste the code below into the HTML of your page to embed this item in your website.
                        
                            <div id="ubcOpenCollectionsWidgetDisplay">
                            <script id="ubcOpenCollectionsWidget"
                            src="{[{embed.src}]}"
                            data-item="{[{embed.item}]}"
                            data-collection="{[{embed.collection}]}"
                            data-metadata="{[{embed.showMetadata}]}"
                            data-width="{[{embed.width}]}"
                            async >
                            </script>
                            </div>
                        
                    
IIIF logo Our image viewer uses the IIIF 2.0 standard. To load this item in other compatible viewers, use this url:
http://iiif.library.ubc.ca/presentation/dsp.831.1-0089346/manifest

Comment

Related Items