Free
Article  |   May 2013
The relation between gaze behavior and categorization: Does where we look determine what we see?
Author Affiliations
  • Mijke O. Hartendorp
    Experimental Psychology, Helmholtz Institute, Utrecht University, Utrecht, The Netherlands
    INCAS3, Assen, The Netherlands
    MijkeHartendorp@incas3.eu
  • Stefan Van der Stigchel
    Experimental Psychology, Helmholtz Institute, Utrecht University, Utrecht, The Netherlands
    s.vanderstigchel@uu.nl
  • Ignace Hooge
    Experimental Psychology, Helmholtz Institute, Utrecht University, Utrecht, The Netherlands
    i.hooge@uu.nl
  • Jeanette Mostert
    Experimental Psychology, Helmholtz Institute, Utrecht University, Utrecht, The Netherlands
    j.mostert@donders.ru.nl
  • Tanja de Boer
    Experimental Psychology, Helmholtz Institute, Utrecht University, Utrecht, The Netherlands
    mail@tanjadeboer.nl
  • Albert Postma
    Experimental Psychology, Helmholtz Institute, Utrecht University, Utrecht, The Netherlands
    Department of Neurology, University Medical Center Utrecht, Utrecht, The Netherlands
    a.postma@uu.nl
Journal of Vision May 2013, Vol.13, 6. doi:10.1167/13.6.6
  • Views
  • PDF
  • Share
  • Tools
    • Alerts
      ×
      This feature is available to Subscribers Only
      Sign In or Create an Account ×
    • Get Citation

      Mijke O. Hartendorp, Stefan Van der Stigchel, Ignace Hooge, Jeanette Mostert, Tanja de Boer, Albert Postma; The relation between gaze behavior and categorization: Does where we look determine what we see?. Journal of Vision 2013;13(6):6. doi: 10.1167/13.6.6.

      Download citation file:


      © ARVO (1962-2015); The Authors (2016-present)

      ×
  • Supplements
Abstract
Abstract
Abstract:

Abstract  When categorizing an object, we fixate our eyes on informative parts of that object. In the current study, morphed figures were used to investigate whether the interpretation of such unclear objects is reflected in the eye movement pattern. The morphed figures were created by interpolating two concrete objects. The intermediate steps represent figures that contain properties of both objects but in different proportions. Three experiments were conducted to investigate the relation between categorization and gaze behavior. In all three experiments, free-naming responses and eye movements were recorded simultaneously. In the first experiment, the relation between the interpretation and the fixated part of the figure was investigated: A strong relation was observed. Subsequently, it was investigated whether gaze patterns drive categorization or vice versa. In the second experiment, morphed figures were preceded by a prime word. A priming effect on categorization was found but not on gaze behavior. In the third experiment, a cue directed the observer's gaze to a particular location on the morphed figure. Interestingly, the cueing experiment showed a cueing effect not only on gaze behavior but also on categorization. Taken together, these findings suggest that where we look affects how we interpret a perceptually uncertain stimulus.

Introduction
The eyes are the entrance to the external visual world. In particular the foveal area of our retina provides us with sharp and detailed information. Therefore, if we examine someone's scan path we can register which information is focused on and is important for the interpretation of the visual input. Decades ago, the relation between categorization and gaze behavior was observed by Ellis and Stark (1978). They detected a correspondence between interpreting the Necker cube as facing leftwards or rightwards and the specific line of the cube which was fixated. Subsequently, others attempted to influence the relation between categorization and gaze behavior by either manipulating categorization or gaze behavior. Kovic, Plunkett, and Westermann (2009), for instance, investigated the relation between categorization and gaze behavior by examining the effect of mental representations on the gaze behavior of animate objects. The object (e.g., a picture of a cat) that was to be scanned was preceded by a sentence that could either be “look at the picture,” “what's this?” or “look at the cat.” The latter reinforces the activation of a mental representation of a cat before the picture of a cat is presented. This mental representation could guide participants towards focusing on cat-specific features whereas the other two questions left them without expectations. No differences in scan paths were observed between the different conditions. This suggests that the gaze behavior was not directly affected by the expectation activated by the previously presented word. Moreover, in a study by Georgiades and Harris (1997), the dominance of an ambiguous figure (the so-called “wife/mother-in-law figure”) was influenced by placing a cue at a critical feature. If the cue was placed near the position of the wife's eye, a tendency was reported to interpret the figure more often as the wife than as the mother-in-law. Although they did not record eye movements, their findings suggest that the feature that was looked at can affect the interpretation of the visual input. 
The response patterns observed in these two studies (Georgiades & Harris, 1997; Kovic et al., 2009) seem counterintuitive because expectations of the upcoming event play an important role in our gaze behavior (Hunnius & Bekkering, 2010; Land, Mennie, & Rusted, 1999). For instance, in studying gaze when performing procedural tasks, Land et al. (1999) noted that when making tea, the eyes are directed to the location important for the next step in executing the action. One possible explanation for the absence of a priming effect and the presence of a cueing effect might be the difference in stimulus materials. In the study by Kovic et al. (2009), clear and unambiguous pictures (e.g., picture of a cat) were used, while in the study by Georgiades and Harris (1997) an ambiguous stimulus was used (i.e., wife/mother-in-law figure). The difference in ambiguity might have different effects on categorization and, therefore, also yields a different impact on gaze behavior. 
Various studies have shown that gaze behavior is very stable even under difficult circumstances. For instance, gaze patterns were examined for faces viewed from different viewpoints (Chelnokova & Laeng, 2011; Sæther, Van Belle, Laeng, Brennen, & Øvervoll, 2009). Interestingly, the center part of the human face that was most frequently fixated when presented in frontal view was also most frequently fixated when the viewpoint was changed. Moreover, in a study by Judd, Durand, and Torralba (2011), the stability of the gaze behavior was further explored by varying the resolution of photographs of scenes. They showed that, despite the fact that some images were presented at a very low resolution, the pattern of fixation locations was still comparable to the ones found for the high-resolution images. While these are all examples of visual uncertainty, this begs the question of how stable gaze patterns are under circumstances of perceptual uncertainty. 
In the current study, we investigated the directionality of the relation between categorization and gaze behavior using perceptually uncertain stimuli: Does categorization direct gaze behavior or does gaze behavior direct categorization? The stimulus set consisted of so-called morphed figures: stimuli created by changing one object into another object with small linear steps. The intermediate steps represent figures that contain properties of both objects, but all in different proportions. Accordingly, different levels of morphing are accompanied by variations in level of uncertainty. Using morphed figures makes it possible to examine whether the uncertainty in categorization and the difference in uncertainty across morphing levels is also reflected in the gaze behavior (Experiment 1). In addition, by manipulating either the categorization or the gaze behavior of the morphed figures, new insights are provided on how they influence one another (Experiments 2 and 3, respectively). 
Experiment 1
The goal of the first experiment was to examine whether a relation exists between the interpretation of an individually presented morphed figure and the eye-fixation pattern. A free-naming experiment was conducted in which response and eye fixations were recorded simultaneously. The scan paths obtained for the extreme figures of a morph series (i.e., nonmorphed, 100% figures of a morph series) were used to identify the regions of interest (ROIs), such as the head of an animal. On the basis of these ROIs, we computed how often participants gave a free-naming response for the morphed figures which agreed with the area on which they fixated. 
Methods
Subjects
Twenty-one students of Utrecht University and Hogeschool Utrecht participated in this experiment (five of whom were male and 16 of whom were female; M = 23.8 years old, SD = 3.8 years). The experiment lasted approximately 30 min and participants received 6 euros or one course credit for their contribution. 
Materials
Morphed figures were created from black silhouette objects that were presented on a white background. Pairs of silhouette objects were interpolated in steps of 5% change using Sqirlz-Morph software (Xiberpix, Version 2.0), resulting in morph series consisting of 19 interpolations and two extreme figures (i.e., end extremes of morph continuum; in other words, the 100%0% and 0%100% figures; cf. Hartendorp et al., 2010, for a description of the morphing procedure). To reduce exposure to the same series as much as possible, only 7 out of the 19 interpolations from each morph series were included aside from the extreme figures. These were the 80%20%, 70%30%, 60%40%, 50%50%, 40%60%, 30%70%, and 20%80% figures (approximately 4.29° × 3.34°, 7 cm × 9 cm). 
Thirty morph series were used in this experiment. Fifteen morph series were similar to the ones used in Hartendorp et al. (2010), hereafter referred to as the original dataset. The original dataset was selected from a validated set of contour drawings of a wide range of living and nonliving objects (De Winter & Wagemans, 2004; Wagemans et al., 2008), which in turn were derived from a set of line drawings validated by Snodgrass and Vanderwart (1980). Additionally, one figure (i.e., the man figure) was selected from a set of contour drawings by Downing, Bray, Rogers, and Childs (2004). The remaining fifteen morph series, hereafter referred to as the new dataset, consisted of extreme figures that had the same object name as the extreme figures in the original dataset. These figures were selected from Flickr (www.flickr.com). The photos were restricted to be covered for the most part by one object (e.g., car, turtle, butterfly) and to have a high resolution. The selected photos were transformed into black silhouette objects presented on a white background using Adobe Photoshop CS (8.0, Adobe Systems, 2003). 
To validate the identification of the new black silhouettes, a verification task was conducted. In this task, 20 participants (different from those tested in the current experiment) were asked to verify whether an object name and a subsequently presented picture (black silhouette) were of the same object. Pictures were excluded for interpolation when they were rejected by more than four participants as belonging to the same object as the word. The included pictures were verified within a range of 450 to 700 ms with an average verification time of 582 ms for pictures of the original dataset and 587 ms for pictures of the new dataset. 
The reason for using equivalent extreme figures in both datasets was to generalize eye-movement patterns for different examples of the same category. As can be seen in Figure 1, great overlap in gaze behavior was observed between the equivalent extreme figures of the two datasets. A complete overview of all thirty morph series can be found in Appendix A and Appendix B
Figure 1
 
Heat map examples of extreme figures of the morph series. The upper row includes stimuli from the so-called original dataset and the lower row from the so-called new dataset. The blob in the left top corner reflects the location of the fixation cross where participants started their scan path.
Figure 1
 
Heat map examples of extreme figures of the morph series. The upper row includes stimuli from the so-called original dataset and the lower row from the so-called new dataset. The blob in the left top corner reflects the location of the fixation cross where participants started their scan path.
Apparatus
The experiment was designed using E-Prime (Psychology Software Tools Inc., Version 1.2) and the additional TET package (TET: Tobii Eye Tracking, Version 1.0.3.0). The eye tracker used was a Tobii 1750. For calibration of the Tobii eye tracker the Tobii software ClearView (Version 2.7.1) was applied. Calibration of the participant's eyes was conducted prior to the experiment and was repeated following each set of 90 experimental trials. Eye movements were measured at a frequency of 50 Hz. With an accuracy of 0.5° and a viewing distance of about 60 cm, the error of the eye-tracking data generated by the Tobii Eye Tracker is small enough to be of no impact and was therefore not taken into account. A voice key was used to end stimulus presentation when triggered by a sound cue (e.g., a participant's response) and a stimulus response box (SR box) was used to register the type of response. The stimuli were simultaneously presented on two separate displays. One display was used for presentation to the participant, the other to the experimenter. 
Procedure
Participants were asked to interpret a single figure by means of a free-naming response; any response was possible, as long as it was their interpretation of the figure. Although no specific time restrictions were included, participants were instructed to respond as quickly as possible. Participants were told to speak loudly and clearly and to avoid sighing and smacking their lips, since the voice key was sensitive to any kind of sound. In addition, their response could be any interpretation provided that it was limited to one word. Participants were informed that there were no correct or incorrect responses. The figures were presented randomly with the restriction that a subsequent trial should contain a figure of another morph series. 
The procedure of a trial was as follows. Participants fixated on a fixation cross in the top left corner of the screen for at least 1000 ms. To ensure participants' gaze behavior would start at the position of the fixation cross, the cross had to be fixated at least the last 200 ms of its presentation before continuation of the trial. The location of the fixation cross was chosen such that every scan path started at a neutral position outside the given figure. Subsequently, a figure of one of the morph series appeared at the center of the screen. The figure stayed on the screen until the participant's response triggered the voice key, after which the figure disappeared immediately. At the same time the eye movements were recorded. The recording of the eye movements started after the fixation cross disappeared and ended when the voice key was triggered. After the figure disappeared, the experimenter coded the response (see Responses section for an explanation of the coding procedure of the responses). An experimental run consisted of 270 trials (30 Morph Series × 9 Figures Per Series). Prior to the experimental run, four practice trials were included that followed the same sequence as the experimental trial, though the stimulus was an extreme figure not included in the experiment. 
Data analysis
Eye fixations
To test whether there is a relation between how we categorize an object and what we look at, we first had to decide which regions of an object are of our interest, so-called ROIs. As we know from the literature on gaze behavior, many objects have a preferred feature at which we look while categorizing the object. For instance, we have a strong tendency to fixate on the head of an animal (Kovic et al., 2009). In addition, we prefer to look at the center part of a human face independent of the orientation of the face (Sæther et al., 2009). In the current study, the gaze pattern was visualized by creating a heat map of the eye fixations during the categorization process of an object. A heat map is a graphical representation of the most frequently fixated areas of an image across all viewers represented by a color gradient overlay. A color gradient ranging from red to blue indicates the average viewing pattern. Red tones indicate the most frequently scanned spots of the image, whereas blue tones indicate lower fixation rates. The heat maps of the extreme figures of the tested morph series (100%0% and 0%100% figures) were used to determine the ROIs corresponding to a particular interpretation of the series. 
The determination of an ROI was done by placing a square on top of the part of the heat map colored red (an indication that this part of the figure was proportionally most frequently fixated). The size of the ROI was approximately the size of the red area on the heat map with the requirement that the two ROIs of the same morph series had the exact same size (squares were about 11° × 11°). The ROIs had the shape of a square to facilitate determining the x and y coordinates of the ROIs and inspecting overlap between ROIs, since the ROIs should not have any overlap. 
We were interested in whether the heat map of a morphed figure matched the heat map of one of its extreme figures. Each ROI covers a particular area of the visual display in which each pixel can be expressed in an xy coordinate. The eye fixations are expressed in an xy coordinate as well. Thus, if the xy coordinate of the eye fixation falls within the range of xy coordinates of the specified ROI, this eye fixation is interpreted as falling within a particular ROI. The area of the visual display that was not occupied by the ROIs of the extreme figures of a morph series is referred to as outside ROIs. The morph series in which the ROIs showed overlap (e.g., morph series heart-apple) were deleted from further analyzes, because eye fixations falling within this ROI could not be defined as falling exclusively either in one or another ROI. Eventually, 10 out of 30 morph series appeared to show overlap in their ROIs and were, therefore, discarded. In follow-up experiments, the discarded series no longer formed part of the experimental stimulus materials. As an illustration, the heat maps and their ROIs of the arm-banana series are presented in Figure 2
Figure 2
 
Heat maps of the arm-banana series. The first row presents the heat maps and their filter for each figure of the arm-banana series separately; the filters were based on all fixations and all response recorded for the extreme figures. The heat maps of the figures are based on all responses across all participants. The second row presents the heat maps of the first fixations for the “arm” responses across all participants. The third row presents the heat maps of the first fixations for the “banana” responses across all participants.
Figure 2
 
Heat maps of the arm-banana series. The first row presents the heat maps and their filter for each figure of the arm-banana series separately; the filters were based on all fixations and all response recorded for the extreme figures. The heat maps of the figures are based on all responses across all participants. The second row presents the heat maps of the first fixations for the “arm” responses across all participants. The third row presents the heat maps of the first fixations for the “banana” responses across all participants.
The next step was to match the xy coordinates of the eye fixations of a morphed figure to the xy coordinates of one of the ROIs. Here, we were particularly interested in the xy coordinates of the first fixation and whether they were corresponding to the xy coordinates of one of the ROIs. The recording of the eye movements started when the initial fixation cross was still presented on the screen. Participants were instructed to fixate on this fixation cross. We refer to the fixation following the first saccade (to the actual stimulus) as the first fixation. The fixation on the initial fixation cross is referred to as the zeroth fixation. In general, about three to four fixations were made before a figure was categorized (including the zeroth fixation). 
Responses
The verbal responses were coded by the experimenter as the object name of one or the other extreme of the morph series, as an alternative object name (not the object name of one of the extreme figures of the morph series), or as a voice key error. The latter was registered when the voice key was triggered by a sound other than the participant's response. For example for a figure of the arm-banana series, the response “banana” was coded as one, the response “arm” as two, the response “branch” as three, and a sigh as four. Due to the free-naming method, it is likely that participants would respond with a word other than the object names provided with the line drawings (Snodgrass & Vanderwart, 1980). Notice that responses were accepted as correct interpretations if the object name was an instance of the same category as the original line drawing (e.g., seal and sea lion are both instances of the same category). However, if the response was at the superordinate level (e.g., the response “bird” to the line drawing of a duck) the response was labeled as an alternative response. 
Fixations and responses
An inspection of the heat maps recorded for a morph series (see Figure 2) suggests a shift in the center of gravity of the eye fixations halfway along the morph continuum from one ROI (e.g., top right corner: ROI banana) to the other ROI (e.g., bottom left corner: ROI arm). If these heat maps are separated for type of response (e.g., heat maps for banana responses and for arm responses), a distinct pattern is observable with a preference for the ROI that fits the type of response. The heat maps of the morphed figures that were interpreted as a banana show a center of gravity of the eye fixations at the top right corner (at the location of the banana stem) and the heat maps of the figures interpreted as an arm show a center of gravity of the eye fixations at the bottom left corner (at the location of the hand). The strength of the relation between gaze behavior and the interpretation of the morphed figures was tested here by calculating the degree of correspondence between the eye fixations and the responses. First, we tested whether looking at a particular ROI resulted in a corresponding response (i.e., response in relation to region of interest). Next, subsequent testing was dedicated to whether giving a particular response led to a corresponding ROI (i.e., ROI in relation to response). If the response and the ROI of the first fixation corresponded, the trial was labeled as equal. If they did not correspond, but instead referred to the other extreme of the morph series, the trial was labeled as opposite. If they did not correspond at all (if the response was an alternative interpretation or if the fixation fell outside one of the ROIs), the trial was labeled as other. A relation between the categorization and the gaze behavior of morphed figures was ascertained when more equal trials (responses and ROIs) than opposite and other trials were observed. 
Results
In total, verbal responses and eye movement patterns were collected for 5670 trials (270 Trials Per Experimental Run × 21 Subjects). As mentioned earlier, 10 out of 30 morph series were excluded (33.3%), leaving us with 20 morph series. Subsequently, the trials containing a 100%0% and 0%100% figure as target were excluded (14.8%) because they were used to determine the ROIs. All trials with a reaction time (RT) below 200 ms (anticipatory trials) and above 10,000 ms (too slow) were discarded as were all trials registered as voice key errors or as missing gaze data (5.1%). These criteria left us with 2650 trials of the originally recorded number of trials (46.7%). 
The only trials that were included were those in which the zeroth fixation fell within the region of the fixation cross (65.8%) (top left corner, 12° < x < 36° and 9° < y < 27°). This region was defined as the region that covered the top left corner where the fixation cross appeared, though the size of the ROI was restricted so that there was no overlap with the surface of any of the morphed figures. From these trials, the responses were either one of the two interpretations (85.5%) or were an alternative interpretation (14.5%). Furthermore, the first fixation could fall in one of the two ROIs (61.5%), outside these ROIs (32.3%), or could not be recorded when only the zeroth fixation was made during a trial (6.2%). The mean RT of the included trials was 1414 ms (SD = 1124 ms) and the average number of fixations that were made per trial was 3.9 (SD = 3.3). 
The relation between fixated ROI and response was examined. The number of trials in which the fixated location corresponded to the response (equal trials: e.g., fixation of ROI arm and arm response) was compared to the number of trials in which the response was the opposite response (opposite trials: e.g., fixation of ROI arm and banana response) and to the number of trials in which the response was an alternative response (other trials: e.g., fixation of ROI arm and sock response) or in which the fixated location was outside one of the ROIs (other trials: outside ROIs and arm response). A higher frequency of equal trials in comparison to opposite and other trials confirms a relation between categorization and gaze behavior. First, we tested response in relation to ROI by conducting a chi-square test on the distribution of equal, opposite, and other trials. This chi-square test was highly significant, χ2(2) = 4.14E2 and p < 0.001. Paired samples t tests showed that the frequency of equal trials was higher than the frequency of opposite trials, which in turn was higher than the frequency of other trials, tequal-opposite (20) = 11.30 and p < 0.001, tequal-other (20) = 18.59 and p < 0.001, topposite-other (20) = 5.68 and p < 0.001, see also Figure 3. This means that there are more equal trials than opposite and other trials and more opposite than other trials. 
Figure 3
 
The proportions of trial type (equal, opposite, and other). Error bars express standard error of the mean (SEM). The solid black bars refer to the proportions of the first analysis, response in relation to ROI. For example, if the first fixation of an arm-banana figure was in the ROI arm, the response “arm” was labeled as equal, the response banana as opposite, and the response sock as other. The black-striped bars refer to the proportions of the second analysis, ROI in relation to response. For example, if the response to an arm-banana figure was “arm,” the first fixation falling within the ROI arm was labeled as equal, the first fixation falling within the ROI banana was labeled as opposite, and the first fixation falling outside the arm and banana ROI was labeled as other.
Figure 3
 
The proportions of trial type (equal, opposite, and other). Error bars express standard error of the mean (SEM). The solid black bars refer to the proportions of the first analysis, response in relation to ROI. For example, if the first fixation of an arm-banana figure was in the ROI arm, the response “arm” was labeled as equal, the response banana as opposite, and the response sock as other. The black-striped bars refer to the proportions of the second analysis, ROI in relation to response. For example, if the response to an arm-banana figure was “arm,” the first fixation falling within the ROI arm was labeled as equal, the first fixation falling within the ROI banana was labeled as opposite, and the first fixation falling outside the arm and banana ROI was labeled as other.
Second, we tested region of interest in relation to response by conducting a chi-square test on the distribution of equal, opposite, and other trials, χ2(2) = 2.93E2 and p < 0.001. Next, paired samples t tests were conducted to further investigate the bias in the distribution, tequal-opposite (20) = 13.39 and p < 0.001, tequal-other (20) = 4.57 and p < 0.001, topposite-other (20) = −7.49 and p < 0.001. These results show that there are more equal trials than opposite and other trials and that there are more other than opposite trials. The difference between this analysis and the first one is caused by the difference in other trials. From all responses, only a small amount was an alternative response (the other responses corresponded to the object name of the two extreme figures of a morph series), whereas the number of trials in which an area was fixated outside one of the ROIs was larger, as can be seen in Figure 3
Subsequently, the mean reaction time of the responses were analyzed to examine whether participants categorized a morphed figure faster when the ROI of the first fixation corresponded to the response than when they did not correspond. A repeated measures analysis of variance (ANOVA) was conducted with relation between response and ROI fixated (equal, opposite, other) as the within-subject variable, F(2, 40) = 35.34 and p < 0.001. A Bonferroni corrected post hoc comparison showed that all three levels differed significantly from one another at an alpha level of 0.05, with the shortest RTs for equal trials (M = 1171 ms and SE = 58 ms), followed by opposite trials (M = 1397 ms and SE = 106 ms), and the longest RTs were observed for other trials (M = 2544 ms and SE = 244 ms). These findings show that when a participant looked at a particular ROI and subsequently responded correspondingly to this ROI, participants were faster than when a participant's first fixation was in a ROI that did not correspond to the response of that trial. 
All previous results were analyzed across morphing level. An increase of morphing level (from 80%20% to 50%50% figures) also indicates an increase of uncertainty which may be reflected in the gaze behavior pattern. Chi-square tests were conducted for each morphing level separately to analyze whether the overall pattern of response in relation to ROI was also observed at all morphing levels. All chi-square tests were again highly significant, see Table 1
Table 1
 
The values of the chi-square tests and their paired-samples t tests conducted for each morphing level separately to analyze the distribution of equal, opposite, and other trials. The abbreviation eq-op stands for the paired sample equal-opposite, eq-ot for equal-other, and op-ot for opposite-other.
Table 1
 
The values of the chi-square tests and their paired-samples t tests conducted for each morphing level separately to analyze the distribution of equal, opposite, and other trials. The abbreviation eq-op stands for the paired sample equal-opposite, eq-ot for equal-other, and op-ot for opposite-other.
Experiment 1 Morphing
80%20% 70%30% 60%40% 50%50%
Chi-square test
 χ2 240.07 161.31 62.33 6.99
df 2 2 2 2
 p < 0.001 < 0.001 < 0.001 < 0.05
Paired-samples t test
eq-op eq-ot op-ot eq-op eq-ot op-ot eq-op eq-ot op-ot eq-op eq-ot op-ot
 t 8.9 18.1 6.2 9.8 21.0 4.5 5.3 6.7 1.4 1.3 2.3 0.8
df 20 20 20 20 20 20 20 20 20 20 20 20
 p < 0.001 < 0.001 < 0.001 < 0.001 < 0.001 < 0.001 < 0.001 < 0.001 > 0.15 > 0.20 < 0.05 > 0.40
The overall pattern of more equal than opposite and more opposite than other trials is recognizable for all morphing levels. However, this pattern appears to be weaker for the 60%40% and 50%50% levels than for the 80%20% and 70%30% levels. The proportions of equal, opposite and other trials for the different levels of morphing are presented in Table 2
Table 2
 
The proportions of equal, opposite, and other trials for the different levels of morphing (80%20%, 70%30%, 60%40%, and 50%50%).
Table 2
 
The proportions of equal, opposite, and other trials for the different levels of morphing (80%20%, 70%30%, 60%40%, and 50%50%).
Experiment 1 Morphing
80%20% 70%30% 60%40% 50%50%
Trial type
Equal 0.71 0.65 0.54 0.43
Opposite 0.24 0.25 0.26 0.29
Other 0.04 0.10 0.20 0.28
Discussion
In the current experiment, we examined the relation between categorization and gaze behavior of morphed figures using free naming. In particular, we looked at the frequency of trials in which the ROI of the first fixation corresponded to the response. For instance, we were interested in the number of trials in which the first fixation during target presentation fell in the ROI arm and was interpreted as arm. A strong relation was observed expressed by a higher frequency of so-called equal trials in comparison to the frequency of trials in which response and ROI did not correspond. Additionally, analyzing the reaction times showed that when the response corresponded to the ROI of the first fixation, categorization of the morphed figures was faster than when they did not correspond. Trials containing 80%20% and 70%30% figures showed a similar pattern to the overall pattern, meaning that more equal trials were observed than opposite trails and the other trials were the least observed. Trials containing a 60%40% figure showed a trend corresponding to the overall pattern, albeit much weaker than for the more defined figures (i.e., 80%20% and 70%30% figures). This suggests that an increase of ambiguity weakens the relation between categorization and gaze behavior. Overall, there were no effects for trials containing a 50%50% figure, save for the presence of more equal than other trials. This seems counterintuitive, since one might expect even stronger effects for the 50%50% figures in comparison to the other figures. This will be discussed further in the General discussion
Experiment 2
In this experiment, it was investigated whether having an expectation of the object directs the participant's eyes to a particular feature of the object using a priming paradigm. Each morphed figure was preceded by a prime word that corresponded to one of the two figures interpolated in the morphed figure or corresponded to one of the extreme figures of another morph series. 
Methods
Subjects
Twenty students of Utrecht University and Hogeschool Utrecht participated in this experiment (nine male and 11 female; M = 23.4 years old, SD = 2.2 years). None of these participants had participated in Experiment 1. Two of them were not included in the data analyses, since in about one-third of the trials no gaze data was recorded. The experiment lasted about 25 min and participants received three euros or a course credit for their contribution. 
Materials
Only 10 out of the 30 morph series used in Experiment 1 were selected for use in the current experiment. We limited our selection to only morph series of the original dataset to prevent an overlap in prime words. In Appendix A, the morph series used in this experiment are marked by an asterisk. The prime words were the Dutch object names of the extreme figures. For example, a figure of the arm-banana series was preceded by the prime word arm (in Dutch arm), by the prime word banana (in Dutch banaan), or by an unrelated prime word, which was an object name of one of the extreme figures of another morph series (e.g., church; in Dutch kerk). 
Procedure
The instructions were similar to those in Experiment 1, except for the fact that participants were informed that after the fixation cross a briefly presented word appeared that they were supposed to ignore. This word (i.e., prime word) was presented in black letters at the same position as the fixation cross, in the top left corner of the display. The prime word was presented for 100 ms after which a blank interval of 250 ms followed. In a pilot study, participants reported to have read the prime words when presented for 100 ms. In addition, the blank interval could be used to activate an expectation induced by the prime. Next, a figure was presented at the center of the screen that had to be named using free naming. The remainder of the procedure was similar to the procedure of Experiment 1. An experimental run consisted of 270 trials (10 Morph Series × 9 Figures × 3 Prime Words). 
Results
In this experiment, the verbal responses and eye movement patterns from 4,860 trials (270 Trials Per Experimental Run × 18 Subjects) were recorded. The trials with a RT below 200 ms and above 10,000 ms and trials that were registered as voice key errors or as missing gaze data were discarded from further analyses (3.1%). The remaining responses were either one of the two interpretations (92.3%) or were an alternative interpretation (7.7%). The first fixations could fall in one of the two ROIs (64.3%), outside these ROIs (22.2%), or could not be recorded when only one fixation was made during a trial (13.4%). The mean RT was 1016 ms (SD = 635 ms) and the average number of fixations that were made per trial was 3.0 (SD = 1.7). Notably, the zeroth fixations could fall in one of two ROIs (52.8%) or outside these ROIs (47.2%). The former percentage was much higher than expected, because the experiment was designed in such a way that participants were required to fixate the fixation cross for the last 200 ms of its presentation. The large amount of data loss is probably due to the presentation of the prime word and the blank interval that both did not require fixation. As a consequence, the prime word might not have been perceived by the participants on the trials in which fixation did not stay in the top left corner during presentation of the prime word. Therefore, only trials were selected in which the zeroth fixation fell in the region around the prime word (11.4%). 
The effects of priming on the verbal response and the fixation pattern were analyzed separately. First, it was analyzed whether priming had an effect on the verbal response. The verbal responses were divided into equal, opposite, and unrelated trials depending on their relation to the prime. For instance, if the response was “arm,” the trial was labeled equal if the prime word was arm, opposite if the prime word was banana, and unrelated if the prime word was church. Hence this relation does not depend on which extreme figure is more dominant in a morphed figure. The distribution of equal, opposite, and unrelated trials was tested by a chi-square test, χ2(2) = 11.07 and p < 0.01. In Figure 4, the distribution of the equal, opposite, and unrelated trials is presented graphically. To investigate the bias of the distribution further, paired samples t tests were conducted, tequal-opposite (17) = 4.87 and p < 0.001, tequal-other (17) = 4.38 and p < 0.001, topposite-other (17) = −4.16 and p < 0.01. These results show that there were more equal than unrelated trials and more unrelated than opposite trials. Thus, the prime word biased the verbal response in the direction of the prime word. 
Figure 4
 
The proportions a particular response was given in relation to the prime. Error bars express standard error of the mean (SEM). For instance, if the response to a target of the arm-banana series was arm, the trial was labeled as equal when preceded by the prime word arm, the trial was labeled as opposite when preceded by the prime word banana, and the trial was labeled as unrelated when preceded by the prime word church.
Figure 4
 
The proportions a particular response was given in relation to the prime. Error bars express standard error of the mean (SEM). For instance, if the response to a target of the arm-banana series was arm, the trial was labeled as equal when preceded by the prime word arm, the trial was labeled as opposite when preceded by the prime word banana, and the trial was labeled as unrelated when preceded by the prime word church.
In addition, it was tested whether priming also affected the time necessary to categorize a morphed figure by comparing the RTs found for the different types of trials. A repeated measures ANOVA was conducted with relation between prime and response (equal, opposite, and unrelated) as the within-subject variable and RT as dependent variable, F(2, 34) = 16.38 and p < 0.001. Bonferroni corrected post hoc comparisons showed that the fastest RTs were observed for the equal trials (M = 891 ms and SE = 40 ms) compared to the opposite (M = 991 ms and SE = 49 ms) and unrelated trials (M = 994 ms and SE = 43 ms). The latter two did not differ significantly. This finding shows that participants were faster in categorizing a morphed figure when their verbal response corresponded to the prime word. 
Next, it was investigated whether priming had an effect on the fixation pattern. The trials were divided into equal, opposite, and unrelated trials, depending on the relation between prime word and ROI of the first fixation. The distribution of equal, opposite, and unrelated trials was tested by a chi-square test, χ2(2) = 0.16 and p > 0.90. No significant difference was reported. This indicates that priming had no effect on where the first eye movement was made to (see Figure 5 for the participant means and their standard errors). 
Figure 5
 
The proportions of the first fixations that fell into a particular ROI in relation to the prime. Error bars express standard error of the mean (SEM). For example, if the first fixation for a target of the arm-banana series was in the ROI arm, the trial was labeled as equal when the prime word was arm, the trial was labeled as opposite when the prime word was banana, and the trial was labeled as unrelated when the prime word was church.
Figure 5
 
The proportions of the first fixations that fell into a particular ROI in relation to the prime. Error bars express standard error of the mean (SEM). For example, if the first fixation for a target of the arm-banana series was in the ROI arm, the trial was labeled as equal when the prime word was arm, the trial was labeled as opposite when the prime word was banana, and the trial was labeled as unrelated when the prime word was church.
In addition, it was tested whether the relation between prime word and ROI of the first fixation influenced the time necessary to categorize a morphed figure. A repeated measures ANOVA was conducted with relation between prime word and ROI (equal, opposite, and unrelated) as the within-subject variable and RT as dependent variable, F(2, 34) = 7.81 and p < 0.01. Bonferroni corrected post hoc comparisons showed that equal trials (M = 980 ms and SE = 47 ms) and opposite trials (M = 1026 ms and SE = 47 ms) did not differ in RT and opposite trials and other trials (M = 1095 ms and SE = 53 ms) did also not differ in RT. Only equal trials and other trials differed in RT. 
To investigate priming as a function of morphing level, we conducted two chi-square tests. First, the relation between prime and response (equal, opposite, unrelated) was examined as a function of morphing level, χ28020(2) = 0.91 and p > 0.60, χ27030(2) = 0.93 and p > 0.60, χ26040(2) = 7.53 and p < 0.05, and χ25050(2) = 3.97 and p > 0.10 (see Table 3, upper portion, for the proportions). A significant difference was only observed for the 60%40% level, tequal-opposite (15) = −1.14 and p > 0.25, tequal-unrelated (15) = 1.01 and p > 0.30, topposite-unrelated (15) = 2.36 and p < 0.05. The significant difference is due to more unrelated than opposite trials. The previously observed priming effect almost disappeared when analyzing the morphing levels separately. Second, the relation between prime and ROI (equal, opposite, unrelated) was also examined as a function of morphing level, χ28020(2) = 2.80 and p > 0.20, χ27030(2) = 0.58 and p > 0.70, χ26040(2) = 3.27 and p > 0.15, and χ25050(2) = 0.47 and p > 0.75 (see Table 3, lower portion, for the proportions). Overall, no priming effect was observed for the individual morphing levels. 
Table 3
 
The proportions of equal, opposite, and unrelated trials for the different levels of morphing (80%20%, 70%30%, 60%40%, and 50%50%), with the upper portion of the table representing the analysis of relation between prime and response and the lower portion of the table representing the analysis of relation between prime and ROI.
Table 3
 
The proportions of equal, opposite, and unrelated trials for the different levels of morphing (80%20%, 70%30%, 60%40%, and 50%50%), with the upper portion of the table representing the analysis of relation between prime and response and the lower portion of the table representing the analysis of relation between prime and ROI.
Morphing
80%20% 70%30% 60%40% 50%50%
Response
Equal 0.30 0.31 0.28 0.25
Ppposite 0.36 0.38 0.46 0.43
Unrelated 0.35 0.31 0.26 0.32
ROI
Equal 0.26 0.31 0.41 0.35
Opposite 0.37 0.31 0.33 0.29
Unrelated 0.38 0.37 0.26 0.35
Discussion
In this experiment, the effect of priming on the categorization and gaze behavior of morphed figures was investigated. The prime word was either similar to the name of one of the two extremes of the morph series to which the morphed figure belonged or was the name of an extreme of a different morph series. First, a priming effect on categorization was shown for both the response itself and the reaction time necessary to categorize the target. More equal trials and less opposite trials were observed with respect to the frequency of unrelated trials. This implies that the prime word was indeed processed to some extent. Nevertheless, an effect of priming on the gaze behavior (i.e., first fixation) was not found. Notably, the absence of any priming effect on the first fixation might be due to the small paucity of trials analyzed. Therefore, inferences should be considered with caution. However, even if more trials were analyzed, one may wonder whether a priming effect would have been observed given the large p value (p > 0.90). 
Experiment 3
In this cueing experiment, it was investigated whether the feature that the eyes fixate on directs the interpretation of the object. Each morphed figure was preceded by a cue at the location of a critical feature of one of the extreme figures interpolated in the morphed figure (such as the head of an animal). 
Methods
Subjects
Twenty students of Utrecht University and Hogeschool Utrecht participated in this experiment (five male and 15 female; M = 22.6 years old, SD = 2.8 years). These participants did not take part in Experiments 1 and 2. The experiment lasted about 15 min and participants received either three euros or a course credit for their contribution. 
Materials
In order to allow a comparison between the results of Experiment 2 and this experiment the same stimulus set as the one used in Experiment 2 was applied here. 
Procedure
The procedure of the current experiment was similar to the procedure of Experiment 2, though in this experiment, a prime word did not serve as a cue. Instead, the cue was a plus sign that was presented for 250 ms, right after the presentation of the fixation cross. Immediately after the presentation of the cue a figure was presented at the center of the screen. The cue appeared in the corresponding ROI of Extreme Figure A or Extreme Figure B of the morph series in question. For instance, if the ROI arm was positioned where the hand of the arm is, a plus sign was placed in the middle of this ROI. Participants were instructed explicitly to look at the cue. Each morphed figure was presented twice, once preceded by the cue that corresponded to the ROI of Extreme Figure A and once of Extreme Figure B, to examine whether a cue corresponding to ROI A caused more A responses and a cue corresponding to ROI B more B responses. An experimental run consisted of 180 experimental trials (10 Morph Series × 9 Figures × 2 Cues). 
Results
In this experiment, the verbal responses and eye movement patterns from 3600 trials (180 Trials Per Experimental Run × 20 Subjects) were recorded. The trials with a RT below 200 ms and above 10,000 ms and the trials that were registered as voice key errors or as missing gaze data were discarded from further analyses (8.6%). In addition, as the instructions were to look at the cue, we only analyzed the trials in which the zeroth fixation fell within the ROI of the cue (82.3% of remaining trials) and was longer than 250 ms (42.6% of remaining trials). In the end, this left us with 1155 trials (32.1% of total number of recorded trials). From these trials, the responses could either be one of the two interpretations (89.1%) or could be an alternative interpretation (10.9%). The mean RT of these trials was 1207 ms (SD = 1117 ms) and the average number of fixations per trial was 3.2 (SD = 1.7). 
Subsequently, it was analyzed whether starting the viewing pattern in the ROI where the cue was presented affected the response. Therefore, the relation between cue and response was investigated. This relation could either be equal, opposite, or other. Thus, consider the case in which the cue was presented in the ROI arm. When participants responded to the subsequent figure of the arm-banana series as “arm,” this response was labeled as an equal trial. When they responded with “banana,” it was labeled as an opposite trial. Finally, when participants offered some alternative interpretation (e.g., “sock”), the response was labeled as an other trial. The distribution of equal, opposite, and other trials was analyzed by a chi-square test, χ2(2) = 3.78E2 and p < 0.001. To investigate the bias of the distribution further, paired samples t tests were conducted, tequal-opposite (19) = 9.04 and p < 0.001, tequal-other (19) = 17.91 and p < 0.001, topposite-other (19) = 11.30 and p < 0.001. These results show that there are more equal trials than opposite trials and more opposite trials than other trials. In Figure 6, the distribution of the equal, opposite, and other trials is presented graphically. Taken together, it can be said that when one fixates on the cue position as the target appears, the visual information present at the ROI of the cue influences the interpretation of the figure. This shows that the interpretation of the visual input is affected by which visual information is perceived first. 
Figure 6
 
The proportions a particular response was given in relation to the cue preceding the target. Error bars express standard error of the mean (SEM). On the x axis the different types of trials are presented with equal referring to trials in which the response corresponds to the ROI of the cue, opposite to trials in which the response is opposite to the ROI of the cue, and other to trials in which the response is an alternative interpretation independent of the cue.
Figure 6
 
The proportions a particular response was given in relation to the cue preceding the target. Error bars express standard error of the mean (SEM). On the x axis the different types of trials are presented with equal referring to trials in which the response corresponds to the ROI of the cue, opposite to trials in which the response is opposite to the ROI of the cue, and other to trials in which the response is an alternative interpretation independent of the cue.
Since each morphed figure was presented twice, the first presentation might have influenced the second interpretation. Therefore, the same analyses were conducted again, but this time only trials were included containing a morphed figure that was presented for the first time. About half of the trials (44.2%) fulfilled this condition. First, a chi-square test was performed, χ2(2) = 128.13 and p < 0.001. To investigate the bias of the distribution further, paired samples t tests were conducted, tequal-opposite (19) = 5.63 and p < 0.001, tequal-other (19) = 12.00 and p < 0.001, topposite-other (19) = 5.58 and p < 0.001. These results showed that there are more equal trials than opposite trials and more opposite trials than other trials. These analyses suggest exactly the same pattern as the analyses containing all trials. These results provide evidence that our findings are not influenced by strong carryover effects. 
Subsequently, it was examined whether the different cues had an effect on the time necessary to categorize a figure. Again, only the trials in which the zeroth fixation was longer than 250 ms in the ROI of the cue were taken into account. A repeated measures ANOVA was conducted with relation between cue and response (equal, opposite, and other) as the within-subject variable and RT as dependent variable, F(2, 38) = 14.21 and p < 0.001. Bonferroni corrected post hoc comparisons showed that equal trials (M = 1223 ms and SE = 71 ms) and opposite trials (M = 1113 ms and SE = 50 ms) did not differ in RTs, but equal and opposite trials showed shorter RTs than the RTs observed for the other trials (M = 1710 ms and SE = 153 ms). This analysis demonstrates that the starting location of the gaze behavior had no effect on the reaction time when the response was a related response (i.e., equal or opposite). However, the reaction time increased markedly when an alternative interpretation was given. 
Finally, the effect of cueing as a function of morphing level was examined by conducting a chi-square test. The distribution of equal, opposite, and unrelated trials was analyzed for each morphing level separately (see Table 4). The results of the chi-square tests and their paired samples t tests are presented in Table 5. The results indicate that the 80%20%, 70%30%, and 60%40% morphing levels showed a similar pattern to the overall pattern: there were more equal trials than opposite trials and more opposite trials than other trials. However the 50%50% morphing level diverged from the other levels of morphing in the sense that the distribution of equal, opposite, and other trials did not differ significantly. This result shows that when the target was a 50%50% figure cueing had no effect on the response. 
Table 4
 
The proportions of equal, opposite, and other trials for the different levels of morphing (80%20%, 70%30%, 60%40%, and 50%50%).
Table 4
 
The proportions of equal, opposite, and other trials for the different levels of morphing (80%20%, 70%30%, 60%40%, and 50%50%).
Experiment 3 Morphing
80%20% 70%30% 60%40% 50%50%
Trial type
Equal 0.61 0.57 0.49 0.39
Opposite 0.34 0.31 0.32 0.37
Other 0.05 0.13 0.19 0.24
Table 5
 
The values of the chi-square tests and their paired-samples t tests conducted for each morphing level separately to analyze the distribution of equal, opposite, and other trials. The abbreviation eq-op stands for the paired sample equal-opposite, eq-ot for equal-other, and op-ot for opposite-other.
Table 5
 
The values of the chi-square tests and their paired-samples t tests conducted for each morphing level separately to analyze the distribution of equal, opposite, and other trials. The abbreviation eq-op stands for the paired sample equal-opposite, eq-ot for equal-other, and op-ot for opposite-other.
Experiment 3 Morphing
80%20% 70%30% 60%40% 50%50%
Chi-square test
 χ2 125.71 78.17 31.68 4.83
df 2 2 2 2
 p < 0.001 < 0.001 < 0.001 = 0.083
Paired-samples t test
eq-op eq-ot op-ot eq-op eq-ot op-ot eq-op eq-ot op-ot eq-op eq-ot op-ot
 t 6.4 16.5 9.1 6.0 10.4 5.0 4.2 5.3 2.7 x x x
df 19 19 19 19 19 19 19 19 19 x x x
 p < 0.001 < 0.001 < 0.001 < 0.001 < 0.001 < 0.001 < 0.001 < 0.001 < 0.05 x x x
Discussion
In this experiment, the effect of cueing on the categorization of morphed figures was investigated. Each target was preceded by a cue (a plus sign) that was located at the center part of one of the two ROIs of the target. In about 80% of the trials, the zeroth fixation fell in the ROI of the cue. The remaining trials were discarded from further analyses. Subsequently, we investigated whether the cue affected the response. Only the trials in which the eyes fixated on the cue location when the target appeared were analyzed. An effect of cueing was found on the response: More equal trials than opposite and other trials were reported. This indicates that the feature of the target to which the eyes were directed by the cue biased the response towards the interpretation corresponding to the ROI indicated by the cue. Moreover, an evaluation of the effect of cueing on morphing level showed that trials containing 80%20%, 70%30%, and 60%40% figures resulted in a pattern similar to the overall one. However, the 50%50% figures diverged from the overall pattern. The effect of cueing on categorization disappeared when the target was a 50%50% figure. Thus, cueing had an effect on the response as long as the target contained more information on one of the two objects. Notably, an explanation for a lack of an effect of the 50%50% morphing level may be the small number of trials analyzed: The analysis for the 50%50% figures is based on half the number of trials than for the other morphing levels, because the other levels were collapsed into two levels of morphing (i.e., 60%40% is based on 60%40% and on 40%60% figures, whereas 50%50% is only based on 50%50% figures). Furthermore, the high proportion of alternative interpretations (i.e., other trials) also reduced the number of related trials (i.e., equal and opposite trials). Hence statistical power might be low for this morphing level. 
General discussion
In this study, we have investigated whether there is a relation between where observers look and how they interpret a morphed figure. In Experiment 1, it was shown that when the first fixation was in a particular ROI (e.g., ROI arm), in a majority of the trials the free-naming response corresponded to this ROI (e.g., response “arm”) and vice versa. In addition, this finding was reinforced by shorter reaction times for the trials in which the ROI of the first fixation corresponded to the response. The strong relation between categorization and gaze behavior observed previously (Drewes, Trommershäuser, & Gegenfurtner, 2011; Kovic et al., 2009; Schütz, Braun, & Gegenfurtner, 2011) was confirmed in the current study in which morphed figures were used as visual stimuli. There are two explanations for this relation: (a) observers already form an expectation based on global peripheral information and subsequently direct their eyes towards a region of interest for verification of this expectation and (b) the eyes fixate on a particular feature. When this region contains valuable information about the visual input, categorization is easier and faster. 
To distinguish between these two explanations of what is affecting the relation between categorization and gaze behavior, we have manipulated these two aspects. In Experiment 2, it was tested whether having an expectation of the object's interpretation would influence which ROI was fixated on first. Each figure was preceded by a prime word that corresponded to one of the two objects or was unrelated to the target. The categorization was indeed influenced by the prime word, with a bias towards the prime word. This effect was reinforced by the shorter reaction times observed for the trials in which the response corresponded to the prime word. Despite the fact that priming had an effect on categorization time and the response itself, this priming effect was not observed for the gaze behavior. These findings indicate that having an expectation about the subsequently presented visual object affects categorization but does not affect gaze behavior. An alternative explanation for the missing correlation between priming and gaze behavior could be that gaze behavior is primarily used to exclude other possibilities, and therefore, indirectly confirms the expectation. If this is true, one may even expect a negative correlation. The most probable other possibility considered is the opposite interpretation (e.g., prime word banana leads to exclusion of the possibility arm). Therefore, exclusion of the opposite interpretation should have led to more opposite trials. Given the fact that this negative correlation was not observed, we consider this alternative explanation less likely. 
Subsequently, in Experiment 3, it was investigated whether fixating on a particular ROI influenced the interpretation of a morphed figure. Each figure was preceded by a cue presented at the location of one of the two ROIs of the figure (e.g., at the center of ROI arm or at the center of ROI banana). In this way, the visual input seen first is the information present in the ROI at the location of the cue. We have found that the cue biased the interpretation of the figure towards a response corresponding to the ROI of the cue. In other words, those visual features fixated on first influenced the interpretation of the morphed figure. 
Our findings resemble those by Kovic et al. (2009) and Georgiades and Harris (1997): Contrary to expectations, cueing has a clear effect on categorization. Whereas these previous studies used unambiguous figures and ambiguous faces, we now show for the first time that the strong correspondence in how a visual object is interpreted and which feature is fixated on first is also observed for morphed figures. Therefore, the strong relation between categorization and gaze behavior can be generalized to a great variety of visual stimuli. Based on the perceptual uncertainty of morphed figures expressed in the longer reaction times and greater variability in verbal responses one might not expect a strong relation between the final interpretation of a morphed figure and the feature which is first fixated on. To the contrary, we have shown that where an observer fixates first seems to be a strong indicator of how the morphed figure will be interpreted. 
Moreover, the categorization of a morphed figure can be influenced by the gaze behavior, while the gaze behavior cannot directly be influenced by the categorization of a morphed figure. In other words, where the observer fixates influences how the visual object is interpreted, not the other way around. In many studies on object recognition and even on scene perception, it was observed that particular features of objects (such as faces of humans and animals) were preferably fixated on (Cerf, Frady, & Koch, 2009; Crouzet, Kirchner, & Thorpe, 2010; Kovic et al., 2009; Thorpe, Fize, & Marlot, 1996; see also Schütz et al., 2011, for a review). However, it is still unknown what makes the eyes move towards these specific features. Our experiments have shown that the visual input itself plays an important role in the process of categorization. 
The contradiction of finding a priming effect on the response, but not on the first fixation, could be explained by the fact that priming took place at a basic level. A recent study by Poncet, Reddy, and Fabre-Thorpe (2012) showed that categorization of peripherally presented animals and artifacts is categorized at the superordinate level (e.g., animal and vehicle) twice as fast than at the basic level (e.g., dog and car). They explained this difference by the level of information processing. Categorization at a superordinate level takes place at a more global information processing level, whereas categorization at a basic level needs more detailed local information processing. Many studies support the idea that global information is processed more quickly than local information (Bar, 2003; Fei-Fei, Iyer, Koch, & Perona, 2007; Greene & Oliva, 2009; Schwarzer, Huber, & Dümmler, 2005). The prime words in our experiment were at the basic level and therefore might have been processed to slow to function as a strong prime word. Perhaps a prime at the superordinate level would have affected the first fixation. 
If we take morphing level into account, differences were found between morphing levels. In particular, the 50%50% figures diverged from the other levels of morphing. The lack of a clear relation between gaze behavior and categorization for 50%50% figures could be a consequence of the size of the dataset analyzed, as is mentioned in the Discussion section of Experiment 3. However, more fundamental reasons might have caused the differences between the levels of morphing. The literature on the categorization of morphed objects (Harnad, 1987; Hartendorp et al., 2010; Newell & Bülthoff, 2002; Verstijnen & Wagemans, 2004) describes that morphed stimuli are perceived categorically (i.e., categorical perception). This means that morphed figures at one half of the morph continuum are categorized as their nearest end extreme (i.e., dominant object). For instance, a 60%40% arm-banana figure is preferably categorized as arm, whereas a 40%60% arm-banana figure is preferably categorized as banana. However, the categorization of the 50%50% figure is more variable, as is also reflected in the current study by the high proportion of alternative interpretations for the 50%50% figures in comparison to the other levels of morphing. One might argue that the influence of gaze behavior on categorization should be most strongly observed for the 50%50% morphing level due to their perceptual uncertainty. However, this perceptual uncertainty might be too large to provide any bottom-up information and should therefore fully rely on top-down influences, such as wild guessing and memory of previous trials. Hence, results suggest that gaze behavior affects categorization as long as the visual input contains some clues for identification. 
Conclusions
This study has shown that the strong relation between categorization and gaze behavior was also found for morphed figures. The origin of this relation was investigated by manipulating the expectation of the upcoming target using priming and by manipulating the visual feature first fixated upon using cueing. Manipulation of the expectation affected categorization, but not gaze behavior. Moreover, manipulating the fixation location appeared to affect both gaze behavior and categorization. The relation between categorization and gaze behavior seems to be mostly determined by the feature of the visual object first fixated upon. We have demonstrated that the strong relation between categorization and gaze behavior is maintained when the location of the first fixation was manipulated. Taken together, these findings suggest that where we look affects how we interpret a perceptually uncertain stimulus. 
Acknowledgments
This work was supported by the project Unconscious Boundaries of Mind within the Consciousness in a Natural and Cultural Context (CNCC) program of the European Science Foundation (ESF) and by the Netherlands Organization for Scientific Research (NWO). We thank Hollie Burnett and Tjeerd Jellema from Hull University for providing the stimulus set and Matt Coler from INCAS3 for editing. 
Commercial relationships: none. 
Corresponding author: Mijke Olga Hartendorp. 
Email: MijkeHartendorp@incas3.eu. 
Address: INCAS3, Dr. Nassaulaan 9, Assen, The Netherlands. 
References
Bar M. (2003). A cortical mechanism for triggering top-down facilitation in visual object recognition. Journal of Cognitive Neuroscience, 15 (4), 600–609. [CrossRef] [PubMed]
Cerf M. Frady E. P. Koch C. (2009). Faces and text attract gaze independent of the task: Experimental data and computer model. Journal of Vision, 9 (12): 10, 1–15, http://www.journalofvision.org/content/9/12/10, doi:10.1167/9.12.10. [PubMed] [Article] [CrossRef] [PubMed]
Chelnokova O. Laeng B. (2011). Three-dimensional information in face recognition: An eye-tracking study. Journal of Vision, 11 (13): 27, 1–15, http://www.journalofvision.org/content/11/13/27, doi:10.1167/11.13.27. [PubMed] [Article] [CrossRef] [PubMed]
Crouzet S. M. Kirchner H. Thorpe S. J. (2010). Fast saccades toward faces: Face detection in just 100 ms. Journal of Vision, 10 (4): 16, 1–17, http://www.journalofvision.org/content/10/4/16, doi:10.1167/10.4.16. [PubMed] [Article] [CrossRef] [PubMed]
De Winter J. Wagemans J. (2004). Contour-based object identification and segmentation: Stimuli, norms and data, and software tools. Behavior Research Methods, Instruments & Computers, 36 (4), 604–624. [CrossRef]
Downing P. E. Bray D. Rogers J. Childs C. (2004). Bodies capture attention when nothing is expected. Cognition, 93, B27–B38. [CrossRef] [PubMed]
Drewes J. Trommershäuser J. Gegenfurtner K. R. (2011). Parallel visual search and rapid animal detection in natural scenes. Journal of Vision, 11 (2): 20, 1–21, http://www.journalofvision.org/content/11/2/20, doi:10.1167/11.2.20. [PubMed] [Article] [CrossRef] [PubMed]
Ellis S. R. Stark L. (1978). Eye movements during the viewing of Necker cubes. Perception, 7, 575–581. [CrossRef] [PubMed]
Fei-Fei L. Iyer A. Koch C. Perona P. (2007). What do we perceive in a glance of a real-world scene? Journal of Vision, 7 (1): 10, 1–29, http://www.journalofvision.org/content/7/1/10, doi:10.1167/7.1.10. [PubMed] [Article] [CrossRef] [PubMed]
Georgiades M. S. Harris J. P. (1997). Biasing effects in ambiguous figures: Removal or fixation of critical features can affect perception. Visual Cognition, 4 (4), 383–408. [CrossRef]
Greene M. R. Oliva A. (2009). The briefest of glances: The time course of natural scene understanding. Psychological Science, 20, 464–472. [CrossRef] [PubMed]
Harnad S. (1987). Introduction: Psychophysical and cognitive aspects of categorical perception: A critical overview. In Harnad S. (Ed.), Categorical perception: The groundwork of cognition (pp. 1–25). Cambridge: Cambridge University Press.
Hartendorp M. O. Van der Stigchel S. Burnett H. G. Jellema T. Eilers P. H. C. Postma A. (2010). Categorical perception of morphed objects using a free-naming experiment. Visual Cognition, 18 (9), 1320–1347. [CrossRef]
Hunnius S. Bekkering H. (2010). The early development of object knowledge: A study of infants' visual anticipations during action observation. Developmental Psychology, 46 (2), 446–454. [CrossRef] [PubMed]
Judd T. Durand F. Torralba A. (2011). Fixations on low-resolution images. Journal of Vision, 11 (4): 14, 1–20, http://www.journalofvision.org/content/11/4/14, doi:10.1167/11.4.14. [PubMed] [Article] [CrossRef] [PubMed]
Kovic V. Plunkett K. Westermann G. (2009). Eye-tracking study of animate objects. Psihologija, 42 (3), 307–327. [CrossRef]
Land M. Mennie N. Rusted J. (1999). The roles of vision and eye movements in the control of activities of daily living. Perception, 28, 1311–1328. [CrossRef] [PubMed]
Newell F. N. Bülthoff H. H. (2002). Categorical perception of familiar objects. Cognition, 85, 113–143. [CrossRef] [PubMed]
Poncet M. Reddy L. Fabre-Thorpe M. (2012). A need for more information uptake but not focused attention to access basic-level representations. Journal of Vision, 12 (1): 15, 1–16, http://www.journalofvision.org/content/12/1/15, doi:10.1167/12.1.15. [PubMed] [Article] [CrossRef] [PubMed]
Sæther L. Van Belle W. Laeng B. Brennen T. Øvervoll M. (2009). Anchoring gaze when categorizing faces' sex: Evidence from eye-tracking data. Vision Research, 49 (23), 2870–2880. [CrossRef] [PubMed]
Schütz A. C. Braun D. I. Gegenfurtner K. R. (2011). Eye movements and perception: A selective review. Journal of Vision, 11 (5): 9, 1–30, http://www.journalofvision.org/content/11/5/9, doi:10.1167/11.5.9. [PubMed] [Article] [CrossRef] [PubMed]
Schwarzer G. Huber S. Dümmler T. (2005). Gaze behavior in analytical and holistic face processing. Memory & Cognition, 33 (2), 344–354. [CrossRef] [PubMed]
Snodgrass J. G. Vanderwart M. (1980). A standardized set of 260 pictures: Norms for name agreement, familiarity, and visual complexity. Journal of Experimental Psychology: Human Learning and Memory, 6 (2), 174–215. [CrossRef] [PubMed]
Thorpe S. J. Fize D. Marlot C. (1996). Speed of processing in the human visual system. Nature, 381, 520–522. [CrossRef] [PubMed]
Verstijnen I. M. Wagemans J. (2004). Ambiguous figures: Living versus nonliving objects. Perception, 33, 531–546. [CrossRef] [PubMed]
Wagemans J. De Winter J. Op de Beeck H. Ploeger A. Beckers T. Vanroose P. (2008). Identification of everyday objects on the basis of silhouette and outline versions. Perception, 37, 207–244. [CrossRef] [PubMed]
Appendix A
Original stimulus set (the 100%0% and 0%100% figures were reproduced with permission from Pion Ltd, London (http://www.pion.co.uk and http://www.envplan.com, published in Wagemans, De Winter, Op de Beech, Ploeger, Beckers, & Vanroose, Perception, 2008, 37, 207–244). All fifteen morph series were used in Experiment 1,1534 and the series marked with an asterisk were used in Experiments 2 and 3. The first column represents the names of the morph series, with the first word referring to the leftmost object and the second word to the rightmost object. The first row represents the different morphing levels, with the first percentage reflecting the morphing percentage of the first object name of the morph series and the second percentage reflecting the morphing percentage of the second object name of the morph series. 
Appendix B
New stimulus set, only used in Experiment 1.1534 The first column represents the names of the morph series, with the first word referring to the leftmost object and the second word to the rightmost object. The first row represents the different morphing levels, with the first percentage reflecting the morphing percentage of the first object name of the morph series and the second percentage reflecting the morphing percentage of the second object name of the morph series. 
Figure 1
 
Heat map examples of extreme figures of the morph series. The upper row includes stimuli from the so-called original dataset and the lower row from the so-called new dataset. The blob in the left top corner reflects the location of the fixation cross where participants started their scan path.
Figure 1
 
Heat map examples of extreme figures of the morph series. The upper row includes stimuli from the so-called original dataset and the lower row from the so-called new dataset. The blob in the left top corner reflects the location of the fixation cross where participants started their scan path.
Figure 2
 
Heat maps of the arm-banana series. The first row presents the heat maps and their filter for each figure of the arm-banana series separately; the filters were based on all fixations and all response recorded for the extreme figures. The heat maps of the figures are based on all responses across all participants. The second row presents the heat maps of the first fixations for the “arm” responses across all participants. The third row presents the heat maps of the first fixations for the “banana” responses across all participants.
Figure 2
 
Heat maps of the arm-banana series. The first row presents the heat maps and their filter for each figure of the arm-banana series separately; the filters were based on all fixations and all response recorded for the extreme figures. The heat maps of the figures are based on all responses across all participants. The second row presents the heat maps of the first fixations for the “arm” responses across all participants. The third row presents the heat maps of the first fixations for the “banana” responses across all participants.
Figure 3
 
The proportions of trial type (equal, opposite, and other). Error bars express standard error of the mean (SEM). The solid black bars refer to the proportions of the first analysis, response in relation to ROI. For example, if the first fixation of an arm-banana figure was in the ROI arm, the response “arm” was labeled as equal, the response banana as opposite, and the response sock as other. The black-striped bars refer to the proportions of the second analysis, ROI in relation to response. For example, if the response to an arm-banana figure was “arm,” the first fixation falling within the ROI arm was labeled as equal, the first fixation falling within the ROI banana was labeled as opposite, and the first fixation falling outside the arm and banana ROI was labeled as other.
Figure 3
 
The proportions of trial type (equal, opposite, and other). Error bars express standard error of the mean (SEM). The solid black bars refer to the proportions of the first analysis, response in relation to ROI. For example, if the first fixation of an arm-banana figure was in the ROI arm, the response “arm” was labeled as equal, the response banana as opposite, and the response sock as other. The black-striped bars refer to the proportions of the second analysis, ROI in relation to response. For example, if the response to an arm-banana figure was “arm,” the first fixation falling within the ROI arm was labeled as equal, the first fixation falling within the ROI banana was labeled as opposite, and the first fixation falling outside the arm and banana ROI was labeled as other.
Figure 4
 
The proportions a particular response was given in relation to the prime. Error bars express standard error of the mean (SEM). For instance, if the response to a target of the arm-banana series was arm, the trial was labeled as equal when preceded by the prime word arm, the trial was labeled as opposite when preceded by the prime word banana, and the trial was labeled as unrelated when preceded by the prime word church.
Figure 4
 
The proportions a particular response was given in relation to the prime. Error bars express standard error of the mean (SEM). For instance, if the response to a target of the arm-banana series was arm, the trial was labeled as equal when preceded by the prime word arm, the trial was labeled as opposite when preceded by the prime word banana, and the trial was labeled as unrelated when preceded by the prime word church.
Figure 5
 
The proportions of the first fixations that fell into a particular ROI in relation to the prime. Error bars express standard error of the mean (SEM). For example, if the first fixation for a target of the arm-banana series was in the ROI arm, the trial was labeled as equal when the prime word was arm, the trial was labeled as opposite when the prime word was banana, and the trial was labeled as unrelated when the prime word was church.
Figure 5
 
The proportions of the first fixations that fell into a particular ROI in relation to the prime. Error bars express standard error of the mean (SEM). For example, if the first fixation for a target of the arm-banana series was in the ROI arm, the trial was labeled as equal when the prime word was arm, the trial was labeled as opposite when the prime word was banana, and the trial was labeled as unrelated when the prime word was church.
Figure 6
 
The proportions a particular response was given in relation to the cue preceding the target. Error bars express standard error of the mean (SEM). On the x axis the different types of trials are presented with equal referring to trials in which the response corresponds to the ROI of the cue, opposite to trials in which the response is opposite to the ROI of the cue, and other to trials in which the response is an alternative interpretation independent of the cue.
Figure 6
 
The proportions a particular response was given in relation to the cue preceding the target. Error bars express standard error of the mean (SEM). On the x axis the different types of trials are presented with equal referring to trials in which the response corresponds to the ROI of the cue, opposite to trials in which the response is opposite to the ROI of the cue, and other to trials in which the response is an alternative interpretation independent of the cue.
Table 1
 
The values of the chi-square tests and their paired-samples t tests conducted for each morphing level separately to analyze the distribution of equal, opposite, and other trials. The abbreviation eq-op stands for the paired sample equal-opposite, eq-ot for equal-other, and op-ot for opposite-other.
Table 1
 
The values of the chi-square tests and their paired-samples t tests conducted for each morphing level separately to analyze the distribution of equal, opposite, and other trials. The abbreviation eq-op stands for the paired sample equal-opposite, eq-ot for equal-other, and op-ot for opposite-other.
Experiment 1 Morphing
80%20% 70%30% 60%40% 50%50%
Chi-square test
 χ2 240.07 161.31 62.33 6.99
df 2 2 2 2
 p < 0.001 < 0.001 < 0.001 < 0.05
Paired-samples t test
eq-op eq-ot op-ot eq-op eq-ot op-ot eq-op eq-ot op-ot eq-op eq-ot op-ot
 t 8.9 18.1 6.2 9.8 21.0 4.5 5.3 6.7 1.4 1.3 2.3 0.8
df 20 20 20 20 20 20 20 20 20 20 20 20
 p < 0.001 < 0.001 < 0.001 < 0.001 < 0.001 < 0.001 < 0.001 < 0.001 > 0.15 > 0.20 < 0.05 > 0.40
Table 2
 
The proportions of equal, opposite, and other trials for the different levels of morphing (80%20%, 70%30%, 60%40%, and 50%50%).
Table 2
 
The proportions of equal, opposite, and other trials for the different levels of morphing (80%20%, 70%30%, 60%40%, and 50%50%).
Experiment 1 Morphing
80%20% 70%30% 60%40% 50%50%
Trial type
Equal 0.71 0.65 0.54 0.43
Opposite 0.24 0.25 0.26 0.29
Other 0.04 0.10 0.20 0.28
Table 3
 
The proportions of equal, opposite, and unrelated trials for the different levels of morphing (80%20%, 70%30%, 60%40%, and 50%50%), with the upper portion of the table representing the analysis of relation between prime and response and the lower portion of the table representing the analysis of relation between prime and ROI.
Table 3
 
The proportions of equal, opposite, and unrelated trials for the different levels of morphing (80%20%, 70%30%, 60%40%, and 50%50%), with the upper portion of the table representing the analysis of relation between prime and response and the lower portion of the table representing the analysis of relation between prime and ROI.
Morphing
80%20% 70%30% 60%40% 50%50%
Response
Equal 0.30 0.31 0.28 0.25
Ppposite 0.36 0.38 0.46 0.43
Unrelated 0.35 0.31 0.26 0.32
ROI
Equal 0.26 0.31 0.41 0.35
Opposite 0.37 0.31 0.33 0.29
Unrelated 0.38 0.37 0.26 0.35
Table 4
 
The proportions of equal, opposite, and other trials for the different levels of morphing (80%20%, 70%30%, 60%40%, and 50%50%).
Table 4
 
The proportions of equal, opposite, and other trials for the different levels of morphing (80%20%, 70%30%, 60%40%, and 50%50%).
Experiment 3 Morphing
80%20% 70%30% 60%40% 50%50%
Trial type
Equal 0.61 0.57 0.49 0.39
Opposite 0.34 0.31 0.32 0.37
Other 0.05 0.13 0.19 0.24
Table 5
 
The values of the chi-square tests and their paired-samples t tests conducted for each morphing level separately to analyze the distribution of equal, opposite, and other trials. The abbreviation eq-op stands for the paired sample equal-opposite, eq-ot for equal-other, and op-ot for opposite-other.
Table 5
 
The values of the chi-square tests and their paired-samples t tests conducted for each morphing level separately to analyze the distribution of equal, opposite, and other trials. The abbreviation eq-op stands for the paired sample equal-opposite, eq-ot for equal-other, and op-ot for opposite-other.
Experiment 3 Morphing
80%20% 70%30% 60%40% 50%50%
Chi-square test
 χ2 125.71 78.17 31.68 4.83
df 2 2 2 2
 p < 0.001 < 0.001 < 0.001 = 0.083
Paired-samples t test
eq-op eq-ot op-ot eq-op eq-ot op-ot eq-op eq-ot op-ot eq-op eq-ot op-ot
 t 6.4 16.5 9.1 6.0 10.4 5.0 4.2 5.3 2.7 x x x
df 19 19 19 19 19 19 19 19 19 x x x
 p < 0.001 < 0.001 < 0.001 < 0.001 < 0.001 < 0.001 < 0.001 < 0.001 < 0.05 x x x
×
×

This PDF is available to Subscribers Only

Sign in or purchase a subscription to access this content. ×

You must be signed into an individual account to use this feature.

×