Object Function Guides Attention During Visual Search in Scenes PDF

629130 research-article2016 PSSXXX10.1177/0956797616629130Castelhano, WitherspoonObject Function and Attentional Guidance Research Article...

629130 research-article2016 PSSXXX10.1177/0956797616629130Castelhano, WitherspoonObject Function and Attentional Guidance Research Article Psychological Science How You Use It Matters: Object 2016, Vol. 27(5) 606–621 © The Author(s) 2016 Reprints and permissions: Function Guides Attention During sagepub.com/journalsPermissions.nav DOI: 10.1177/0956797616629130 Visual Search in Scenes pss.sagepub.com Monica S. Castelhano and Richelle L. Witherspoon Department of Psychology, Queen’s University Abstract How does one know where to look for objects in scenes? Objects are seen in context daily, but also used for specific purposes. Here, we examined whether an object’s function can guide attention during visual search in scenes. In Experiment 1, participants studied either the function (function group) or features (feature group) of a set of invented objects. In a subsequent search, the function group located studied objects faster than novel (unstudied) objects, whereas the feature group did not. In Experiment 2, invented objects were positioned in locations that were either congruent or incongruent with the objects’ functions. Search for studied objects was faster for function-congruent locations and hampered for function-incongruent locations, relative to search for novel objects. These findings demonstrate that knowledge of object function can guide attention in scenes, and they have important implications for theories of visual cognition, cognitive neuroscience, and developmental and ecological psychology. Keywords scene processing, visual attention, visual search, object recognition, eye movements Received 10/24/13; Revision accepted 1/6/16 “A place for everything, and everything in its place.” This showed that arbitrary targets were associated with specific old adage not only provides a prescription for keeping scene locations over multiple scene exposures. Additionally, everything tidy, but also summarizes a basic fact about Bayesian computational models (Eckstein, Drescher, & real-world scenes. Objects tend to appear in certain loca- Shimozaki, 2006; Ehinger, Hidalgo-Sotelo, Torralba, & Oliva, tions within scenes; this orderly structure of scenes is 2009; Najemnik & Geisler, 2005; Torralba, Oliva, Castelhano, referred to as scene context. One can take advantage of & Henderson, 2006) have demonstrated through machine these regularities when processing scenes. For instance, learning that associations between scene context and target scene context aids visual search (e.g., Neider & Zelinsky, placement can accurately predict human eye movements. 2006) because of observers’ expectations about object However, objects are acted on, not just viewed within con- placement. Disrupting these expectations impairs perfor- texts. In the present study, we investigated whether an mance such that it takes longer to find and recognize object’s function is related to its placement in a scene and a target (Castelhano & Heaven, 2011; Malcolm & whether knowledge of the object’s function can guide Henderson, 2010; Neider & Zelinsky, 2006). What is not search. clear, however, is how objects are associated with par- Gibson’s (1979) seminal work demonstrated that per- ticular locations in the scene. ception is not an end in itself, but rather occurs in the Most researchers posit that knowledge of scene context service of a larger task goal, usually in the performance arises from extensive experience with scenes and objects (see Henderson, 2003, for a review). In fact, many studies Corresponding Author: and computational models have shown that when a scene is Monica S. Castelhano, Queen’s University, Department of Psychology, viewed multiple times, one begins to learn object placement. 62 Arch St., Kingston, Ontario, K7L 3N6, Canada For example, Brockmole and Henderson (2006a, 2006b) E-mail: [email protected] Object Function and Attentional Guidance 607 of an action. Many studies have demonstrated an intimate the flash-preview moving-window paradigm. Without link between actions and perception. Recent studies have providing participants with firsthand experience viewing shown that actions are directly tied to object representa- these invented objects within a scene context, we were tions, even when unrelated to the task at hand (Beilock able to test whether the knowledge of object function & Holt, 2007; Grezes, Tucker, Armony, Ellis, & Passing- influenced attentional guidance. ham, 2003; Helbig, Steinwender, Graf, & Kiefer, 2010). Additionally, studies have shown that actions performed Pilot Study with objects have certain spatial associations (Downing- Doucet & Guérard, 2014; Tucker & Ellis, 2004). These To establish whether there was a connection between an results suggest that actions performed with objects not object’s function and its placement within a scene, we only may have an effect on object recognition and mem- conducted a pilot study to measure the amount of agree- ory, but also may influence how they fit within a larger ment among participants who saw only the description context. For instance, actions are associated with specific of the invented object and those who saw only a picture spatial constraints (e.g., reaching up, pressing a button), of the invented object. which could impose limitations on where objects are placed within a scene. Here, we investigated the connec- Method tion between object function and scene context and its impact on subsequent behavior. Participants. Twenty Queen’s University undergradu- We investigated the connection between object func- ate students received either $10 per hour or course credit tion and scene context in multiple ways. First, we con- for their participation. ducted a pilot study to directly test whether an object’s placement and its function were related. This was done Stimuli and apparatus. The stimuli were 36 invented for invented objects that were nonexistent but plausible. objects created in HomeDesign 5.0 (DataBecker, We found that knowledge of an object’s function did Düsseldorf, Germany). Each object description included affect its placement in a scene. Second, we examined its intended function and the action required to use it, whether knowledge of object function and scene context but excluded information about its placement in a scene. can guide search in scenes. Objects were designed to minimize participants’ ability to Guidance in scenes has been shown to arise from a guess their function from their appearance. We con- number of factors, including scene gist, spatial layout, and ducted two norming studies to measure participant’s abil- target features (Becker & Rasmussen, 2008; Castelhano ity to guess each object’s function. & Heaven, 2010, 2011; Castelhano & Henderson, 2007; In Norming Study 1, a separate group of participants Eckstein et al., 2006; Ehinger et al., 2009; Hillstrom, (N = 12) matched each functional description to an object Scholey, Liversedge, & Benson, 2012; Hollingworth, 2009; picture. Participants were presented with a matrix of all 36 Hwang, Wang, & Pomplun, 2011; Neider & Zelinsky, 2006; object images, while a description of one of the objects’ Torralba et al., 2006; Zelinsky, 2008). To examine the functions appeared on the left side of the screen. For each effect of the combined knowledge of object function and description, participants indicated with a mouse click scene context, we used the flash-preview moving-window which object picture matched. The matrix was the same paradigm. In this paradigm, participants are shown a brief across the study for each participant but was randomly preview of the entire search scene, after which they search determined across participants. Participants chose one through a gaze-contingent moving window that occludes object image per description and were not prevented visual information in all but the small area of the window from choosing the same object twice across descriptions. itself. The paradigm forces participants to rely on the scene Average accuracy in Norming Study 1 was 21% (SD = representation acquired during the initial preview, in addi- 0.2%) and was significantly above chance (1/36 or.0278) tion to knowledge about the target and the visual features in both a subject analysis, t(11) = 6.76, p <.001, d = 4.07, within the window. Within these limitations, this paradigm and an item analysis, t(35) = 5.41, p <.001, d = 1.83. This allowed us to examine whether an object’s function in means that, on average, participants matched the func- combination with knowledge of scene context could ben- tion to the correct object image for approximately 8 of efit search above and beyond the benefits provided by the 36 objects. This effect was largely driven by 4 objects peripheral visual features. that were correctly selected more than 50% of the time.1 In each of two experiments, there were two phases: a Although some objects were readily matched when both study phase and a search phase. During the study phase, the description and the image were available, we ran a participants studied 18 of a larger set of invented objects second test that more closely reflected the experience of for a memory test. Participants then searched for those the participants in each of the main experiments. objects and an additional 18 novel objects in a scene In Norming Study 2, participants (N = 10) were shown while their eye movements were tracked, according to a picture of each object separately and were asked to type 608 Castelhano, Witherspoon out a description in a text box beside it; this provided a the region or regions in the scene in which the object stringent test of the ability to guess function on the basis would most likely be found. They drew polygons with of appearance. Two independent raters judged the the mouse on the scene image and were not limited in response accuracy and were told to be as liberal in their the number, size, or shape of these polygons. When they scoring as possible. We found a high interrater reliability were ready to move on to the next scene, they pressed a (98% agreement; Cohen’s κ =.73, p <.001). Average accu- “Next” button on the lower left-hand side of the screen. racy in Norming Study 2 was 0.0263% correct (SD = The study took approximately 20 min to complete. 0.022%). It is difficult to assess chance guessing with such a test, but if a similar chance level as for Norming Study 1 Results and discussion is considered as the best-case scenario, this performance can be directly compared with the matching task. Analy- We calculated three measures to capture the spread and ses showed that participants were not able to correctly overlap of regions selected for each scene. Table 1 shows guess the function above this chance level (.0278) in both means for these three measures, and Figure 1 presents a subject analysis, t(9) = −.193, p >.1, d = −0.13, and an heat maps of the regions selected by each group for item analysis, t(35) = 0.41, p >.1, d = 0.14. On the basis of example objects. We determined the agreement among these results, we do not believe that object functions were participants by calculating the percentage of participants readily attainable from their appearance. who contributed to the highest peak (i.e., the most- selected region). We found that the percentage for the Procedure. For the pilot study itself, participants were description group (collapsed across participants for each split into two groups. One group was shown a picture of scene) was higher than for the picture group, t(70) = an object, and the other was shown a description of the 4.12, p <.001, d = 0.98. The overall average area of the object’s function (see Fig. 1). In both groups, this object scene selected was also higher for the description group information was presented on the left-hand side of the than for the picture group, t(70) = 2.51, p <.05, d = 0.60. screen, and a corresponding scene image was presented However, the description group selected fewer unique on the right-hand side. Participants were asked to select regions than the picture group, t(70) = −2.691, p <.01, Scene Object Picture Object Description Description Group Picture Group Pull downward on this device to dispense shampoo or 1 conditioner directly onto your head. 0 Hold the handle and pull out and down. This device will lower items from out of reach to a more comfortable level. The cartridge emits a glow that causes bacteria to fluoresce to help you maintain a clean, safe kitchen. Insert a new cartridge every fourteen days. Invert and rub onto paper and other surfaces to remove ink smudges and other damaging stains. Fig. 1. Stimuli and results for example scenes used in the pilot study. For each scene, participants were shown either a picture of an invented object or a description of that object. Participants’ task was to indicate the region or regions within the scene in which the object would most likely be found. The heat maps depict the frequency with which regions were selected for each scene, separately across all participants in the description group and all participants in the picture group; colors with values closer to 1 indicate regions that were selected more often. Object Function and Attentional Guidance 609 Table 1. Means for the Three Measures Assessed in the Pilot Study Description Picture Difference Measure group group between groups Percentage of participants who 66 (15) 52 (13) 14** selected the peak region Average area of the scene 9,459 (6,238) 6,191 (4,539) 3,268* selected (pixels2) Number of unique regions 5.4 (2.1) 6.8 (2.2) −1.4** selected Note: Standard deviations are in parentheses. *p <.05. **p <.01. d = 0.64. Taken together, these results show that com- credit for their participation. Sample size was determined pared with the picture group, the description group on the basis of previous scene-search studies (e.g., selected bigger but fewer regions overall, and there was Castelhano & Heaven, 2011). greater overlap between participants’ selections within the description group. Stimuli and apparatus. The stimuli were the same 36 invented objects used in the pilot study. For the study phase, each object picture was displayed on a gray back- Experiment 1 ground. For the search phase, each object was incorpo- rated into 1 of 36 scenes (e.g., a kitchen, a living room; Fig. Method 2). Targets and scenes were counterbalanced across partici- Participants. Thirty-two Queen’s University under- pants and across conditions. Targets were placed equally graduate students received either $10 per hour or course among the upper, middle, and lower regions of the scene. Study Phase Search Phase Previously Studied Novel Fig. 2. Objects and scenes from Experiment 1. The top row shows examples from the set of 36 invented objects presented in the study phase. The bottom rows show example scenes into which both studied and novel objects were placed during the search phase. Target objects in the scenes are highlighted in red. 610 Castelhano, Witherspoon Function Group Feature Group Insert your toothbrush into this device to clean and dry the bristles Study before putting it away. 1) Insert your toothbrush into this device to clean and dry the bristles before putting it away. 2) Insert your make-up brushes into this device to clean and dry the 1 2 bristles before putting it away. Test 3) Insert your make-up brushes into this device to untangle the bristles to increase their life-span. 4) Insert your toothbrush into this device to untangle the bristles to 3 4 increase their life-span. Fig. 3. Sample trial and memory test from the study phase of Experiment 1. Participants first studied each object individually; the function group was also given an on-screen description of the object. After viewing all objects, participants completed a memory test. In these examples, the correct answer for both groups is 1. For the test for the feature group, modifications were made to the base of the object that would attach to the wall (in this example, Distractors 2 and 3 have a darker blue color compared with the target object, and Distractors 2 and 4 have fewer layers). The correct item placement was randomized during test. Stimuli were presented on a 21-in. CRT monitor at a remained on the screen until they pressed a button. After resolution of 800 × 600 pixels (38.1° × 28.6°); the monitor viewing all objects, each participant completed a four- had a refresh rate of 100 Hz. Participants’ head move- alternative forced-choice memory test (Fig. 3, bottom ments were constrained by a chin rest, and their eye rows). Participants were required to have a minimum movements were tracked using an EyeLink 1000 eye 85% correct before beginning the search phase. tracker (SR Research, Kanata, Ontario, Canada) sampling For the search phase, the eye tracker was calibrated at 1000 Hz. using a 9-point calibration procedure, which resulted in an average spatial error no greater than 0.4°. To maintain Procedure. Participants were randomly assigned to accurate tracking, we checked calibration before each trial either the function or feature group. During the study and recalibrated the eye tracker when needed. We used the phase, all participants were shown 18 invented objects: flash-preview moving-window paradigm to encourage par- The function group saw the object image and a descrip- ticipants to use knowledge of scene context and to limit tion of its function, while the feature group saw only the their ability to use peripheral information during search image (Fig. 3, top row). Participants were instructed to (Castelhano & Henderson, 2007). For each trial, a preview study each object for a memory test, and each object of the search scene with the target absent was presented Object Function and Attentional Guidance 611 Tim e Fig. 4. Example trial sequence from the search phase of Experiment 1. Trials began with a blank screen, after which a preview of the search screen appeared with the target absent. A mask followed, and then participants were shown only the target. Finally, the search display with the target appeared, but partici- pants could view the image only through a gaze-contingent moving window of 2° radius. Their task was to locate the target visually and press a button. for 250 ms, followed by a visual mask for 50 ms (Fig. 4). feature) and target type (studied vs. novel) as factors. The target object was then presented on a gray background Each ANOVA was followed-up with two planned com- for 2,000 ms, followed by the search scene with a gaze- parisons that examined the difference between studied contingent moving window of 2° radius. Once they loca- and novel objects for each group; to achieve a family-wise ted the target, participants pressed a button. The trial ended error of.05, we used an alpha level of.025 for individual with the button press or after 20,000 ms had elapsed. tests. Attentional guidance was indexed from the onset of Data analysis. To investigate the effect of object func- the scene until the first fixation on the target object. tion and scene-context information on visual search, we Three eye-movement measures from this viewing examined both behavioral (accuracy and search time) period were calculated: latency to the target, number and eye-movement measures. To more closely examine of fixations prior to locating the target, and scan-path effects on gaze, we examined the eye-movement record ratio. Latency to the target was the total amount of using measures reflecting search (attentional guidance) time spent from the onset of the search scene up to and measures reflecting the processing and identifica- (but not including) the first fixation on the target. tion of the target (target processing). These measures Number of fixations prior to locating the target during are thought to reflect distinct processes (Castelhano & this period was also calculated (i.e., excluding the first Heaven, 2010; Castelhano, Pollatsek, & Cave, 2008; fixation on the target). Although related to latency to Malcolm & Henderson, 2009; Nuthmann, 2014). We were the target, this latter measure revealed whether partici- most interested in the effects of object function on guid- pants were effectively selecting likely target candidates ance during search, so we expected the experimental for fixation. Finally, the scan-path ratio was calculated manipulations to have the greatest effects on the atten- as the total distance covered by all fixations prior to tional-guidance measures. It was unclear whether learn- target fixation divided by the linear distance from the ing different object properties would also have an effect first fixation position to the center of the target. This on target processing. measure reflected the efficiency of the search, with a For each measure, we conducted a mixed-design higher value indicating a more indirect path to the analysis of variance (ANOVA) with group (function vs. target. 612 Castelhano, Witherspoon Experiment 1 Experiment 2 a Studied Novel c Studied Novel 1.00 1.00.95.95.90.90 Proportion Correct Proportion Correct.85.85.80.80.75.75.70.70.65.65.60.60.55.55.50.50 Function Feature Congruent Incongruent Group Congruency Condition b d Studied Novel Studied Novel 10,000 10,000 9,000 9,000 8,000 8,000 Search Time (ms) Search Time (ms) 7,000 7,000 6,000 6,000 5,000 5,000 4,000 4,000 3,000 3,000 2,000 2,000 1,000 1,000 0 0 Function Feature Congruent Incongruent Group Congruency Condition Fig. 5. Behavioral results for Experiments 1 (left column) and 2 (right column). For Experiment 1, the graphs show (a) mean accuracy and (b) mean search time as a function of group and target type. For Experiment 2, the graphs show (c) mean accuracy and (d) mean search time as a function of the congruency between the target’s function and its location in the scene, separately for each target type. Error bars show ±1 SEM. Target-processing measures indexed the decision pro- type revealed that the function group was significantly cesses assessing the object’s match to the target. For all more accurate than the feature group (mean difference = eye-movement analyses, each target was defined by a rect- 8%), F(1, 30) = 9.11, p =.005, η2 =.23, but no other effects angle surrounding it approximately 1° from each edge. were significant. For search time, there was a main effect Three measures reflecting early and later processing were of target type, F(1, 30) = 4.8, p =.036, η2 =.12, and a sig- calculated: first-fixation duration, first-gaze duration, and nificant interaction between group and target type, F(1, total time. First-fixation duration (the duration of the initial 30) = 4.95, p =.034, η2 =.12 (see Fig. 5b). While a main fixation on the target) is often seen as an indicator of the effect of target type suggests that there was an effect of initial processing of the object (Henderson, 1992; Rayner & having studied an object (manipulation check), our theo- Pollatsek, 1992). First-gaze duration was defined as the retical interest was the interaction between group and tar- sum of all fixations on the target from first entry to first exit get type. We expected the type of information viewed in the target region and reflects the time spent first exam- during the study phase to have differing effects on search ining the object. Total time on the target was calculated by performance; if knowledge of object function matters, summing all fixation durations on the target before the then there should be greater performance differences response button was pressed. between measures for the studied and novel objects in the function group than in the feature group. Further analyses revealed that the function group found studied targets sig- Results nificantly faster than novel targets, t(31) = 3.21, p =.009, Behavioral measures. Average search accuracy was d = −1.02, but there was no such difference for the feature high (82%; see Fig. 5a). The ANOVA with group and target group, t(31) = 0.43, n.s. Object Function and Attentional Guidance 613 Function Group Feature Group Studied Novel Fig. 6. Example eye-movement patterns (green lines) in Experiment 1. For each target type (studied and novel), eye movements of a single participant in the function group and the feature group are shown across a full trial. Eye-movement measures. Attentional-guidance mea- latencies to the target, t(15) = −2.99, p <.01, d = −1.14, sures showed considerable effects of function knowledge fewer number of fixations prior to locating the target, (Figs. 6 and 7). Analyses yielded a main effect of target t(15) = −3.28, p <.01, d = −1.07, and shorter scan paths, type for all measures of attentional guidance—latency to t(15) = −2.95, p <.01, d = −1.14, than for novel objects. the target: F(1, 30) = 4.85, p =.035, η2 =.12; number of Taken together, these measures indicate that the function fixations prior to locating the target: F(1, 30) = 6.17, p = group demonstrated better search performance than the.019, η2 =.14; and scan-path ratio: F(1, 30) = 6.03, p =.02, feature group, but only for studied objects. η2 =.15. Analyses also revealed significant interactions Target-processing measures are shown in Figure 8a between group and target type on each measure of atten- through 8c. Although one might expect differences in tional guidance—latency to the target: F(1, 30) = 5.26, p = how quickly studied and novel targets were recognized,.029, η2 =.13; number of fixations prior to locating the or differences between the two groups who studied the target: F(1, 30) = 6.59, p =.015, η2 =.15; and scan-path objects differently, we found no significant effects (Fs < 1, ratio: F(1, 30) = 6.03, p =.02, η2 =.15. Additionally, ps >.4). latency to the target showed a main effect of group, F(1, 30) = 6.23, p =.018, η2 =.17. No other main effects of Discussion group were significant. Further analyses investigating the interaction between Knowing an object’s function affected search above and group and target type revealed that there were no signifi- beyond what could be inferred about object placement on cant differences across measures between studied and the basis of its physical form. We found that targets were novel objects for the feature group (ps >.8). However, for located sooner and with fewer fixations, and that partici- studied objects in the function group, there were shorter pants’ scan paths led more directly to targets, when the 614 Castelhano, Witherspoon Experiment 1 Experiment 2 a d Studied Novel Studied Novel 9,000 9,000 8,000 8,000 Latency to Target (ms) Latency to Target (ms) 7,000 7,000 6,000 6,000 5,000 5,000 4,000 4,000 3,000 3,000 2,000 2,000 1,000 1,000 0 0 Function Feature Congruent Incongruent Group Congruency Condition b Studied Novel e Studied Novel 35 35 30 30 Number of Fixations Number of Fixations 25 25 20 20 15 15 10 10 5 5 0 0 Function Feature Congruent Incongruent Group Congruency Condition c Studied Novel f Studied Novel 8 8 7 7 6 6 Scan-Path Ratio Scan-Path Ratio 5 5 4 4 3 3 2 2 1 1 0 0 Function Feature Congruent Incongruent Group Congruency Condition Fig. 7. Eye-movement measures from Experiments 1 (left column) and 2 (right column). For Experiment 1, the graphs show (a) mean latency to the target, (b) mean number of fixations prior to locating the target, and (c) mean scan-path ratio as a function of group and target type. For Experi- ment 2, the graphs show (d) mean latency to the target, (e) mean number of fixations prior to locating the target, and (f) mean scan-path ratio as a function of the congruency between the target’s function and its location in the scene, separately for each target type. Error bars show ±1 SEM. targets’ function was provided than when it was not pro- within the scene. To examine whether object function led vided. This was shown by performance differences both participants to search in particular scene locations, we between studied and novel objects and between groups. conducted a second experiment in which target objects We hypothesize that the benefit arose because knowl- were placed in locations either congruent or incongruent edge about function constrained likely object placement with the objects’ functions. Object Function and Attentional Guidance 615 Experiment 1 Experiment 2 a d Studied Novel Studied Novel 450 450 First-Fixation Duration (ms) First-Fixation Duration (ms) 400 400 350 350 300 300 250 250 200 200 150 150 100 100 50 50 0 0 Function Feature Congruent Incongruent Group Congruency Condition b e Studied Novel Studied Novel 800 800 First-Gaze Duration (ms) First-Gaze Duration (ms) 700 700 600 600 500 500 400 400 300 300 200 200 100 100 0 0 Function Feature Congruent Incongruent Group Congruency Condition c Studied Novel f Studied Novel 900 900 800 800 Total Looking Time (ms) Total Looking Time (ms) 700 700 600 600 500 500 400 400 300 300 200 200 100 100 0 0 Function Feature Congruent Incongruent Group Congruency Condition Fig. 8. Target-processing results for Experiments 1 (left column) and 2 (right column). For Experiment 1, the graphs show (a) first-fixation duration, (b) first-gaze duration, and (c) total time spent looking at the target as a function of group and target type. For Experiment 2, the graphs show (d) first-fixation duration, (e) first-gaze duration, and (f) total time spent looking at the target as a function of the congruency between the target’s function and its location in the scene, separately for each target type. Error bars show ±1 SEM. Experiment 2 their participation. Sample size was determined on the basis of previous scene-search studies (e.g., Castelhano Method & Heaven, 2011). Participants. Twenty-four Queen’s University under- graduate students (none of whom took part in Experi- Stimuli, apparatus, and procedure. The stimuli, ment 1) received either $10 per hour or course credit for apparatus, and procedure were identical to those in 616 Castelhano, Witherspoon Fig. 9. Example scenes from Experiment 2. In each scene, the location of a target object (highlighted for purposes of illustration by a white box) was either (a) congruent or (b) incongruent with the target’s function. In these examples, the target was a device that had to be flicked to scan the room and turn off all the electronic devices. Experiment 1, with the exception that all participants We first calculated the proportion of times there was studied 18 invented objects’ function prior to the search overlap (across scenes) between participants’ preferred phase. In addition, studied and novel target objects were location and each target location (functionally congru- placed in locations that were either congruent or incon- ent and functionally incongruent). We found a signifi- gruent with their function (Fig. 9). cant main effect of location, F(1, 35) = 8.75, p <.05, η2 = The validity of object placements in the function-con-.25, as well as a significant interaction between location gruent and function-incongruent conditions was evalu- and group, F(1, 35) = 13.09, p <.01, η2 =.09, but a mar- ated in two norming studies. In the first, Norming Study ginal main effect of group, F(1, 35) = 3.18, p =.08, η2 = 3, a separate group of participants (N = 20) were shown.37. Further analyses showed that for the description a series of scenes, and they gave a “yes” or “no” response group, the preferred target location overlapped signifi- to rate whether a specific area in the scene (indicated by cantly more often with the congruent location than with a dot) was an appropriate place for a particular object, the incongruent location, t(35) = −4.17, p <.01, d = given a description of its function. The dots appeared in −1.41, but there was no difference in proportion of locations that were either congruent or incongruent with overlap between each location and the preferred loca- the functions of the target object. We found that the per- tion of the picture group, ts > 2.7, ps <.01. This is con- centage of “yes” responses was significantly higher for sistent with the notion that knowledge of an object’s the congruent placements (78.5%) than for the incongru- function was associated with the object placement most ent placements (19.5%), t(17) = 14.46, p <.001, d = 3.8. consistent with that function, while the object picture These results indicate that on the basis of the functional did not convey similarly useful information. description, there were clear expectations about the In addition to looking at direct overlap of the pre- appropriate placement of invented objects. ferred region with the target location, we also examined In Norming Study 4, we evaluated the congruent and the distance of the preferred region to the congruent and incongruent placement of invented objects using data incongruent locations when the regions did not overlap. from the pilot study. We examined the correspondence Average distance from the preferred region to the selected between participants’ preferred location (using the high- target locations is depicted in Figure 10b. We found a est peak from the pilot data) and the functionally congru- significant main effect of location, F(1, 35) = 14.07, p < ent and functionally incongruent target locations. For.01, η2 =.40, a marginal effect of group, F(1, 35) = 3.85, each group (description vs. picture), three measures were p =.058, η2 =.10, and a significant interaction between calculated: (a) the proportion of times the preferred loca- location and group, F(1, 35) = 5.93, p <.05, η2 =.17. Fur- tion overlapped with the target location, (b) the average ther analyses showed that the preferred region was sig- distance from the preferred location to the target loca- nificantly closer to the congruent location (8.76°) in the tion, and (c) the proportion of times the preferred loca- description group than in the picture group (> 14°), tion overlapped with the target scene region. Results are t(35) = −5.15, p <.01, d = −0.62. There was no such dif- presented in Figure 10. ference between the description and picture groups for Object Function and Attentional Guidance 617 Description Picture scene region of each target location. For this, we divided a the scene into three distinct regions: the upper walls and ceilings (upper); countertops, desktops, and tabletops.50 Proportion of Times Overlap.45 (middle); and floor (lower). We measured the proportion.40 of times the selected preferred region overlapped with the.35 scene region of the target location (target scene region) in Occurred.30.25 the congruent and incongruent conditions. If there was.20 no systematicity in the selection of the preferred target.15 region, then the proportion of times it overlapped with.10 the target scene region would be at chance (.33)..05.00 The average proportion of times the preferred region Congruent Incongruent and the target scene region overlapped is depicted in Figure 10c. We found significant main effects of location, b F(1, 35) = 26.51, p <.01, η2 =.76, and group, F(1, 35) = 25 7.57, p =.01, η2 =.22, as well as a significant interaction Distance From Preferred between location and group, F(1, 35) = 14.45, p <.01, to Target Location (°) 20 η2 =.41. Further analyses showed that for the congruent 15 scene region, the proportion of times the preferred regions overlapped was significantly greater when the 10 description of the function was seen (.83) than when the picture was seen (.42), t(35) = 5.0, p <.01, d = 1.69, and 5 greater than for the incongruent scene regions in each group, ts > 6.61, ps <.01. Also, the preferred region over- 0 Congruent Incongruent lapped with the incongruent scene region significantly c less often in the description group than in the picture group, t(35) = −2.02, p =.05, d = 0.68. Further, we found 1.0 that for the description group, the overlap between the.8 Proportion of Times Overlap Occurred preferred and congruent regions was significantly above.6 chance (.33), t(35) = 7.99, p <.001, d = −1.78, but the overlap between the preferred and incongruent regions.4 was significantly below chance, t(35) = −5.28, p <.001, d = −1.79. No other conditions differed significantly from.2 chance..0 Overall, congruent and incongruent locations were Congruent Incongruent validated across a number of measures. The results from Target Position these two norming studies showed that the placements chosen for the function-congruent condition were consis- Fig. 10. Results of Norming Study 4 from Experiment 2. The (a) mean proportion of times participants’ preferred location overlapped with the tent with expectations based on the functional descrip- target location, (b) mean distance from the preferred location to the tion. Likewise, the function-incongruent placements were target location, and (c) mean proportion of times the preferred location not consistent with such expectations. Further, the results overlapped with the target scene region are shown as a function of the were consistent with the notion that the description of congruency between the target’s function and its location in the scene, separately for each participant group. Chance (.33) is depicted by the the object’s function led to highly consistent selections of dotted line. Error bars show ±1 SEM. placement, while the object picture did not convey simi- larly useful information. the incongruent condition (p >.2), and the distance to the incongruent location from each of the preferred loca- Data analysis. As in Experiment 1, we examined both tions in the description and picture groups was farther behavioral (accuracy and search time) and eye-movement than the distance to the congruent location in the descrip- measures and divided the eye-movement record into tion group, ts > 4.1, ps <.001. measures of attentional guidance and target processing. Because the selection of the target location need not For each measure, we conducted a within-subjects ANOVA be an exact x-y position, but rather a general area of the with congruency (congruent vs. incongruent) and target scene (e.g., somewhere on the floor or a desk; see heat type (studied vs. novel) as factors. This was followed-up maps of participants’ selected regions in Fig. 1), we also by two planned comparisons that examined the differ- looked at each overlap of the preferred region with the ence between studied and novel objects for each level of 618 Castelhano, Witherspoon congruency; to achieve a family-wise error of.05, we p =.073, d = 0.50. Although previous research has shown used an alpha level of.025 for individual tests. that objects placed in incongruent locations are more dif- ficult to process than objects placed in congruent loca- tions (Malcolm & Henderson, 2010), target-processing Results measures (Figs. 8d–8f) showed no significant effects, Behavioral measures. Means for the behavioral mea- Fs < 1.5, ps >.6. sures are presented in Figures 5c and 5d. Average search accuracy was high (81% correct), and the ANOVA Discussion revealed a marginal interaction of congruency and target type, F(1, 23) = 3.36, p =.08, η2 =.04, but no other effects Results support the notion that an object’s function were significant, Fs < 1.8. Further analyses revealed mar- affected where participants searched for it in scenes. As ginally higher accuracy for studied objects than for novel in Experiment 1, participants located studied objects in objects in congruent locations, t(23) = 2.33, p =.029, d = function-congruent locations more efficiently than novel 0.97, but no difference between target types for incon- objects in function-incongruent locations. Interestingly, gruent locations. For search times, analyses revealed a when objects were in function-incongruent locations, main effect of congruency, F(1, 23) = 33.09, p <.001, η2 = search performance was worse for studied objects than.28, and a significant interaction of congruency and target for novel objects. These results indicate that participants type, F(1, 23) = 23.99, p <.001, η2 =.18. Further analyses had expectations about where studied objects were revealed that studied objects, compared with novel located, and when those expectations were violated, per- objects, were found significantly faster in congruent loca- formance worsened. tions, t(23) = −4.66, p <.001, d = −1.4, and significantly slower in incongruent locations, t(23) = 2.8, p =.01, General Discussion d = 0.76. In the present study, we investigated whether knowledge Eye-movement measures. Attentional-guidance mea- of an object’s function could affect the guidance of atten- sures for Experiment 2 are shown in Figures 7d through tion when searching for that object in scenes. We demon- 7f. Analyses revealed main effects of congruency across strated that search performance was significantly faster all measures—latency to the target: F(1, 23) = 30.66, p < when an object’s function was known than when it was.001, η2 =.27; number of fixations prior to locating the unknown. In Experiment 1, we found that participants target: F(1, 23) = 32.1, p <.001, η2 =.29; and scan-path trained on an object’s function had shorter search laten- ratio: F(1, 23) = 10.33, p =.004, η2 =.13. We also found cies, fewer fixations, and a more direct scan path toward significant interactions of congruency and target type target objects than participants who were familiar with across all measures—latency to the target: F(1, 23) = that object’s appearance only. 24.25, p <.001, η2 =.18; number of fixations prior to In Experiment 2, we examined whether knowing an locating the target: F(1, 23) = 31.07, p <.001, η2 =.21; and object’s function led participants to search in specific scan-path ratio: F(1, 23) = 19.37, p <.001, η2 =.16. There scene areas; invented objects were placed in locations were no main effects of target type, Fs < 1.4, ps >.4, either congruent or incongruent with those objects’ func- which was most likely due to opposing effects of congru- tions. Search performance benefitted from placement in ency and was not surprising. Of main theoretical interest congruent locations but was impaired for incongruent was the interaction between congruency and target type. locations. This impairment suggests that knowing an The interaction reflected the effect of positioning on object’s function informed search strategies initially and knowing the object function. Further analyses of objects that knowledge of function led to spatial expectations in congruent locations revealed that, compared with that harmed search performance when they were searches for novel objects, searches for studied objects violated. had shorter latencies to the target, t(23) = −4.62, p <.001, One outstanding question is how knowledge of object d = −1.37, fewer fixations prior to locating the target, function influences attention in relation to other sources t(23) = −5.62, p <.001, d = −1.5, and shorter scan paths, of guidance. Previous research has shown that when t(23) = −5.13, p <.001, d = −1.51. Additionally, analyses the whole scene is available during search, guidance is of objects in incongruent locations showed that searches heavily influenced by visual features in the periphery for studied targets, compared with searches for novel tar- (Hillstrom et al., 2012; Hollingworth, 2009; Pereira & gets, produced longer latencies to the target, t(23) = 2.87, Castelhano, 2014). In the present experiments, the use of p =.009, d = 0.79, and a greater number of fixations prior the flash-preview moving-window paradigm forced par- to locating the target, t(23) = 3.16, p =.004, d = 0.76, but ticipants to rely on the initial representation of the scene scan paths were not significantly longer, t(23) = 1.88, along with knowledge about the target. While the use of Object Function and Attentional Guidance 619 the moving window allowed us to examine whether these and other possible links, it remains unclear how knowledge of the object’s function combined with scene object function leads to improved search. context could benefit search, it is not clear whether object Finally, the present findings raise some interesting function would have the same benefit when peripheral questions about how scenes and objects are processed in information is available. Hillstrom et al. (2012) found that the brain. Many recent studies have revealed that differ- when the full scene was shown, the effect of a preview ent areas are recruited for different aspects of scene and was limited to the first few fixations. Thus, it could be object processing; such areas include the parahippocam- that the knowing the object’s function would last only as pal place area, retrosplenial complex, and transverse long as the preview information is useful. However, occipital sulcus (Baldassano, Beck, & Fei-Fei, 2013; Harel, because object function may convey vital information Kravitz, & Baker, 2013; Troiani, Stigliani, Smith, & Epstein, about object location, the influence of object function 2014). Studies have also found that object-recognition may persist as it narrows search to relevant locations. tasks using isolated objects lead to activation of motor Further research is needed to unpack how the character- responses or body parts used to perform actions with istics of surrounding scene information and object func- the objects (Helbig et al., 2010; Kellenbach, Brett, & tion combine to affect performance (e.g., Malcolm, Patterson, 2003). However, whether there are similar Nuthmann, & Schyns, 2014). links between entire environments and motor and body The influence of object function has many implica- representations remains unclear. It would be interesting tions for theories across a number of research areas. One to investigate whether similar activation of actions occurs interesting query is how scene context is initially acquired. when retrieving likely spatial locations of objects. It is While previous experience of seeing objects within a clear that further investigation into object functions will particular context surely has an effect on learning not only provide insights into how one processes visual (Brockmole & Henderson, 2006a, 2006b), the present information, but also into the development and influence study suggests that knowledge of object functions may of context on cognitive processing more generally. also benefit knowledge of object placement. This would apply to both development of scene-context knowledge Action Editor in infants and children as well as in adults, for whom Philippe G. Schyns served as action editor for this article. learning a new environment quickly is crucial. Future research could look at how using objects in context (or Author Contributions alternatively, watching someone use objects) may estab- lish contextual links. M. S. Castelhano developed the study concept. Both authors contributed to the study design. Stimulus construction, testing, On a related note, the present study has interesting and data collection were performed by R. L. Witherspoon. Both implications for environmental and ecological theories, authors analyzed and interpreted the data. M. S. Castelhano according to which real-world experience is central to drafted the manuscript, and R. L. Witherspoon provided critical cognitive function. In this study, participants did not have revisions. Both authors approved the final version of the manu- firsthand experience using the invented objects, but the script for submission. results raise some interesting questions about the role of action itself. For instance, the strong beneficial effect of Acknowledgments object function implies that passive learning of an object’s The authors thank Sian Beilock, Ingrid Johnsrude, Kevin Mun- intended purpose is quite effective for search guidance. hall, and Mareike Wieth for thoughtful comments. However, it also raises questions of how firsthand inter- actions with objects influence its link to the larger con- Declaration of Conflicting Interests text. Further research is needed to examine how different methods of learning object functions varies in its effec- The authors declared that they had no conflicts of interest with tiveness to guide attention. respect to their authorship or the publication of this article. The current results also have interesting implications for how object function and action is related to spatial Funding knowledge. For instance, if a device requires reaching up This work was supported by grants from the Natural Sciences high or down low, or requires the use of the hand or the and Engineering Research Council of Canada, Canada Founda- foot, participants may be able to use the action to index tion for Innovation, and Ontario Ministry of Research and Inno- spatial constraints within the context of a scene (e.g., vation to M. S. Castelhano. upper, middle, or lower regions of the scene; Pereira, Liu, & Castelhano, 2015). Alternatively, associations with Note other objects in the scene may lead to spatial constraints 1. We checked whether the inclusion of the objects affected (Bar, 2004; Fenske, Aminoff, Gronau, & Bar, 2006). With the results reported in the two main experiments. When we 620 Castelhano, Witherspoon removed these four objects from analysis in Experiments 1 and Grezes, J., Tucker, M., Armony, J., Ellis, R., & Passingham, R. E. 2, the pattern of results remained unchanged. (2003). Objects automatically potentiate action: An fMRI study of implicit processing. European Journal of Neurosci- ence, 17, 2735–2740. doi:10.1046/j.1460-9568.2003.02695.x References Harel, A., Kravitz, D. J., & Baker, C. I. (2013). Deconstructing Baldassano, C., Beck, D. M., & Fei-Fei, L. (2013). Differential visual scenes in cortex: Gradients of object and spatial lay- connectivity within the parahippocampal place area. Neuro- out information. Cerebral Cortex, 23, 947–957. doi:10.1093/ Image, 75, 228–237. doi:10.1016/j.neuroimage.2013.02.073 cercor/bhs091 Bar, M. (2004). Visual objects in context. Nature Reviews Helbig, H. B., Steinwender, J., Graf, M., & Kiefer, M. (2010). Neuroscience, 5, 617–629. doi:10.1038/nrn1476 Action observation can prime visual object recognition. Becker, M. W., & Rasmussen, I. P. (2008). Guidance of attention Experimental Brain Research, 200, 251–258. doi:10.1007/ to objects and locations by long-term memory of natural s00221-009-1953-8 scenes. Journal of Experimental Psychology: Learning, Mem- Henderson, J. M. (1992). Object identification in context: The ory, and Cognition, 34, 1325–1338. doi:10.1037/a0013650 visual processing of natural scenes. Canadian Journal of Beilock, S. L., & Holt, L. E. (2007). Embodied preference Psychology, 46, 319–341. judgments: Can likeability be driven by the motor sys- Henderson, J. M. (2003). Human gaze control during real-world tem? Psychological Science, 18, 51–57. doi:10.1111/j.1467- scene perception. Trends in Cognitive Sciences, 7, 498–504. 9280.2007.01848.x doi:10.1016/j.tics.2003.09.006 Brockmole, J. R., & Henderson, J. M. (2006a). Recognition Hillstrom, A. P., Scholey, H., Liversedge, S. P., & Benson, V. and attention guidance during contextual cueing in real- (2012). The effect of the first glimpse at a scene on eye world scenes: Evidence from eye movements. Quarterly movements during search. Psychonomic Bulletin & Review, Journal of Experimental Psychology, 59, 1177–1187. 19, 204–210. doi:10.3758/s13423-011-0205-7 doi:10.1080/17470210600665996 Hollingworth, A. (2009). Two forms of scene memory guide Brockmole, J. R., & Henderson, J. M. (2006b). Using real-world visual search: Memory for scene context and memory for scenes as contextual cues for search. Visual Cognition, 13, the binding of target object to scene location. Visual Cogni- 99–108. doi:10.1080/13506280500165188 tion, 17, 273–291. Castelhano, M. S., & Heaven, C. (2010). The relative contribu- Hwang, A. D., Wang, H.-C., & Pomplun, M. (2011). Semantic tion of scene context and target features to visual search in guidance of eye movements in real-world scenes. Vision scenes. Attention, Perception, & Psychophysics, 72, 1283– Research, 51, 1192–1205. doi:10.1016/j.visres.2011.03.010 1297. doi:10.3758/APP.72.5.1283 Kellenbach, M. L., Brett, M., & Patterson, K. (2003). Actions Castelhano, M. S., & Heaven, C. (2011). Scene context influ- speak louder than functions: The importance of manipula- ences without scene gist: Eye movements guided by spa- bility and action in tool representation. Journal of Cognitive tial associations in visual search. Psychonomic Bulletin & Neuroscience, 15, 30–46. doi:10.1162/089892903321107800 Review, 18, 890–896. doi:10.3758/s13423-011-0107-8 Malcolm, G. L., & Henderson, J. M. (2009). The effects of target Castelhano, M. S., & Henderson, J. M. (2007). Initial scene repre- template specificity on visual search in real-world scenes: sentations facilitate eye movement guidance in visual search. Evidence from eye movements. Journal of Vision, 9(11), Journal of Experimental Psychology: Human Perception and Article 8. doi:10.1167/9.11.8 Performance, 33, 753–763. doi:10.1037/0096-1523.33.4.753 Malcolm, G. L., & Henderson, J. M. (2010). Combining top- Castelhano, M. S., Pollatsek, A., & Cave, K. R. (2008). Typicality down processes to guide eye movements during real-world aids search for an unspecified target, but only in identifica- scene search. Journal of Vision, 10(2), Article 4. doi:10 tion and not in attentional guidance. Psychonomic Bulletin.1167/10.2.4 & Review, 15, 795–801. Malcolm, G. L., Nuthmann, A., & Schyns, P. G. (2014). Beyond Downing-Doucet, F., & Guérard, K. (2014). A motor similarity gist: Strategic and incremental information accumulation for effect in object memory. Psychonomic Bulletin & Review, scene categorization. Psychological Science, 25, 1087–1097. 21, 1033–1040. doi:10.3758/s13423-013-0570-5 doi:10.1177/0956797614522816 Eckstein, M. P., Drescher, B. A., & Shimozaki, S. S. (2006). Najemnik, J., & Geisler, W. S. (2005). Optimal eye move- Attentional cues in real scenes, saccadic targeting, and ment strategies in visual search. Nature, 434, 387–391. Bayesian priors. Psychological Science, 17, 973–980. doi:10.1038/nature03390 Ehinger, K. A., Hidalgo-Sotelo, B., Torralba, A., & Oliva, A. Neider, M. B., & Zelinsky, G. J. (2006). Scene context guides (2009). Modeling search for people in 900 scenes: A com- eye movements during visual search. Vision Research, 46, bined source model of eye guidance. Visual Cognition, 17, 614–621. doi:10.1016/j.visres.2005.08.025 945–978. doi:10.1080/13506280902834720 Nuthmann, A. (2014). How do the regions of the visual field Fenske, M. J., Aminoff, E., Gronau, N., & Bar, M. (2006). Top- contribute to object search in real-world scenes? Evidence down facilitation of visual object recognition: Object-based from eye movements. Journal of Experimental Psychology: and context-based contributions. Progress in Brain Research, Human Perception & Performance, 40, 342–360. doi:10.1037/ 155, 3–21. a0033854 Gibson, J. J. (1979). The ecological approach to visual percep- Pereira, E. J., & Castelhano, M. S. (2014). Peripheral guidance in tion. Boston, MA: Houghton Mifflin. scenes: The interaction of scene context and object content. Object Function and Attentional Guidance 621 Journal of Experimental Psychology: Human Perception in object search. Psychological Review, 113, 766–786. and Performance, 40, 2056–2072. doi:10.1037/a0037524 doi:10.1037/0033-295X.113.4.766 Pereira, E. J., Liu, Y. Q., & Castelhano, M. S. (2015). Inhibi- Troiani, V., Stigliani, A., Smith, M. E., & Epstein, R. A. (2014). tion of attention to irrelevant scene regions: Investigating Multiple object properties drive scene-selective regions. mechanisms of attention during visual search in scenes. Cerebral Cortex, 24, 883–897. doi:10.1093/cercor/bhs364 Manuscript submitted for publication. Tucker, M., & Ellis, R. (2004). Action priming by briefly presented Rayner, K., & Pollatsek, A. (1992). Eye movements and scene objects. Acta Psychologica, 116, 185–203. doi:10.1016/ perception. Canadian Journal of Psychology, 46, 342–376. j.actpsy.2004.01.004 Torralba, A., Oliva, A., Castelhano, M. S., & Henderson, J. M. Zelinsky, G. J. (2008). A theory of eye movements during tar- (2006). Contextual guidance of eye movements and atten- get acquisition. Psychological Review, 115, 787–835. doi:10 tion in real-world scenes: The role of global features.1037/a0013118