Skip to main content
NCBI home page
NIHPA Author Manuscripts logoLink to NIHPA Author Manuscripts
. Author manuscript; available in PMC: 2016 Jun 27.
Published in final edited form as: J Exp Psychol Gen. 2015 Mar 16;144(4):723–729. doi: 10.1037/xge0000063

Holistic Processing from Learned Attention to Parts

Kao-Wei Chua 1, Jennifer J Richler 1, Isabel Gauthier 1
PMCID: PMC4922746  NIHMSID: NIHMS764080  PMID: 25775049

Abstract

Attention helps us focus on what is most relevant to our goals, and prior work shows that aspects of attention can be learned. Learned inattention to parts can abolish holistic processing of faces, but it is unknown whether learned attention to parts is sufficient to cause a change from part-based to holistic processing with objects. Here, we trained subjects to individuate non-face objects (Greebles) from two categories, Ploks and Glips. Diagnostic information was in complementary halves for the two categories. Holistic processing was then tested with Plok-Glip composites that combined the kind of part that was diagnostic or non-diagnostic during training. Exposure to Greeble parts resulted in general failures of selective attention for non-diagnostic composites, but face-like holistic processing was only observed for diagnostic composites. These results demonstrate a novel link between learned attentional control and the acquisition of holistic processing.

Keywords: holistic processing, learning, attention, perceptual expertise

Introduction

Attention allows us to react to salient or surprising events (Theeuwes, 1994, 1998) and find information relevant to current goals (Folk et al., 1992; Bacon & Egeth, 1994). Attention can facilitate learning (Shiu & Pashler, 1992, Ahissar & Hochstein, 1993), but some learning effects are best characterized as changes in how we attend, or learned attention. Attention can be guided by statistical learning (Zhao, Al-Aidroos, and Turke-Brown, 2013), or reward and past selection history (Awh, Belopolsky, & Theeuwes, 2012). When task parameters correlate with item properties, mappings between items and the relevant attentional set may be learned (Jacoby et al., 2003; Bugg & Crump, 2012). Learned attentional settings can transfer to novel members of a category (Bugg et al., 2011). Similarly, learned attention could account for phenomena related to perceptual learning (Nosofsky, 1986; Goldstone, 1994). For example, eye movements reveal that subjects shift from attending to all stimulus dimensions equally to dimensions most diagnostic for categorization (Blair, 2009).

While most studies on learned attention use simple stimuli, complex objects like faces can also trigger attentional sets. For instance, in a study of cognitive control, subjects learned associations between face sex and proportions of congruent responses (Cañadas et al., 2013). Similarly, learned attention to dimensions of complex objects such as faces may account for expert visual object processing phenomena such as holistic processing, the tendency to process objects as unified wholes rather than parts (Young et al., 1987). In Chua et al., 2014, subjects learned to individuate faces from two novel face categories, Lunaris and Taiyos. Diagnostic information for identifying each face was found in complementary halves of the two categories. For example, the top halves of Taiyos and the bottom halves of Lunaris provided diagnostic information for individuation. After training, subjects saw composites made of diagnostic and non-diagnostic face parts in the composite paradigm, a common measure of holistic processing (Farah, 1998; Richler & Gauthier, 2014). In this task, subjects judge whether the target half (e.g., top) of two sequentially presented composite faces (made of top and bottom halves from different faces) is the same or different while ignoring the other part (e.g., bottom). Holistic processing is inferred when subjects cannot ignore information in the task-irrelevant half, which is typically only obtained for aligned face halves. In Chua et al. (2014), holistic processing was only found for face parts that were diagnostic at training, suggesting that learned attention to face parts may be responsible for holistic processing.

This is inconsistent with the prevalent idea that holistic processing is strictly a perceptual phenomenon (Rossion, 2013), for instance due to face representations where parts are not differentiated (Tanaka & Farah, 1993). The Lunari-Taiyo study challenged this sort of explanation, pointing to a role for learned attention in holistic processing. This conclusion was also supported by the finding that subjects showed no holistic processing for face composites made of parts that were non-diagnostic during training. Face parts with a history of not being attended did not trigger obligatory attention. However, while processing diagnostic and non-diagnostic composites differed, novices who had never seen Taiyo or Lunari faces processed them holistically. Thus, because the stimuli were faces, this experiment could not track the acquisition of holistic processing for diagnostic composites with learned attention, although it did show how learned inattention can abolish holistic processing for faces made of parts with a non-diagnostic history.

Here we address the acquisition of holistic processing with novel objects. We trained participants to individuate Greebles, objects that novices do not process holistically. We used two kinds of Greebles that contained diagnostic information in different parts and then tested holistic processing for Greebles combining parts never presented together before. In the Tayio-Lunari study, congruency effects that did not vary as a function of alignment were observed in the non-diagnostic condition (see Richler et al., 2009, for evidence that this is not face-like holistic processing), whereas in the diagnostic condition, the effect was abolished in the misaligned condition. The misaligned baseline should be more easily interpreted with novel objects because no congruency effect should be found in novices. We also included an additional phase-scrambled baseline not expected to be sensitive to training effects.

Methods

Subjects

Eighty subjects were randomly assigned to either the GlipTop/PlokBottom (18 male, 22 female, mean age = 21.6) or the GlipBottom/PlokTop (16 male, 24 female, mean age = 21.9) training condition. Group assignment dictated which part was diagnostic for each Greeble category during individuation training. A control group (n = 40) received no training (16 male, 24 female, mean age = 20.5). Subjects received $15/hour for participation. The study was approved by the Vanderbilt University IRB. Sample size was predetermined based on the group x congruency x alignment interaction in Chua et al. (2014), ηp2 = .04, and on our expectation that using fewer parts in the composite task should improve its reliability (see Ross et al., 2014). With 80 subjects and alpha = .05, power for the critical interaction should reach .90.

Stimuli

Stimuli were asymmetrical Greebles (Gauthier & Tarr, 1997; Rossion et al., 2004). Ploks and Glips had distinct body shapes, textures (Figure 1), and parts that pointed in different directions (up vs. down). All Greebles were presented in grayscale and tilted 40° clockwise to facilitate making composites without cutting any parts.

Figure 1.

Figure 1

Open in a new tab

Example Glips and Ploks for a subject who saw Glips with diagnostic top halves and Ploks with diagnostic bottom halves. The part that was diagnostic for each family was counterbalanced between two groups. Note that the non-diagnostic half did vary (there were 3 parts for each category, only one is shown), but there was ten times more variation for the diagnostic half.

During individuation training, different stimulus sets were used depending on condition. For example, in the GlipTop/PlokBottom condition, three Glip bottom halves were combined with 10 unique top halves each, and three Plok tops halves were combined with 10 unique Plok bottoms each, to create 30 Glips that varied 10 times more in the top than the bottom and 30 Ploks that varied 10 times more in the bottom than the top. A separate set of 60 Greebles was created for the GlipBottom/PlokTop condition, with the reverse assignment of part variability.

For the composite task, two sets of five unique top and bottom halves from the two categories that were not seen during training were used. Composite Greebles varied on the top and bottom. Therefore, all subjects were tested on Greebles with the same amount of variation; any differences in composite task performance could only be attributed to training. The top and bottom Greeble halves were randomly combined to form Plok-Glip composites (400 × 400 pixels). A white line 6 pixels thick separated Greeble halves. Misaligned composites were made by shifting the top and bottom halves 35 pixels to the left and right, respectively. An advantage of the Greebles being tilted is that the misaligned object remains inscribed within a box approximately the same width as the aligned object, eliminating a potential confound with the standard misalignment manipulation in which misaligned trials are wider. Another baseline condition was added wherein task-irrelevant parts were phased-scrambled (83% - see Figure 2) while luminance, contrast, and spatial frequency were preserved (Sadr & Sinha, 2004). With faces, performance does not differ between phase-scrambled and misaligned baselines (Richler et al., 2014).

Figure 2.

Figure 2

Open in a new tab

Example of an incongruent phase-scrambled trial in the composite task, where the top half was cued and the correct answer is “same.” In this condition, the task-irrelevant half is phase-scrambled. The cued halves are the same and the phase-scrambled halves are different, so this is an example of an incongruent phase-scrambled trial.

Experimental Procedure

Subjects completed 3 sessions of individuation training (approx. 60 minutes each) over the course of a week, followed by the composite task at the end of session 3.

Individuation Training

Subjects learned unique names for 16 Greebles, 8 from each category. During training trials, subjects were shown Greebles with one syllable names (e.g. Awg, Dak). Names were randomly assigned to objects for each subject. In all trials, subjects pressed the first letter of the Greeble’s name. Training trials were followed by test trials where Greebles were presented without names. If a Greeble appeared that had no learned name association, subjects pressed “n” to indicate “no name.” Of the thirty Greebles for a given category, eight were assigned names for participants to learn, while the remaining 22 required a “no name” response. Incorrect responses were followed by feedback showing the correct name (including “no name”). There were three phases on each training day (see Table 1). All trained Greebles were introduced by the end of Day 1.

Table 1.

Training structure for individuation training for one Greeble category. Both Ploks and Glips were trained in the same way.

Phase Greebles Training trials/block Test trials/block Day 1 Day 2 Day 3
1 4 named
6 unnamed		 8 21 2 blocks --- 2 blocks
2 6 named
14 unnamed		 12 42 2 blocks --- 2 blocks
3 8 named
22 unnamed		 24 63 6 blocks 10 blocks 6 blocks
Total Trials 1008 1260 1008
Open in a new tab

Composite Task

Stimuli in the composite task were Plok-Glip or Glip-Plok composites made of parts not seen during training (Figure 3). Five tops and five bottoms from each category, and their phase-scrambled versions, were used across all trials. For each subject, half of the composites were made of parts similar to parts that were diagnostic during training, and the other half made of parts similar to parts that were nondiagnostic during training. Each trial started with a fixation cross (200 ms), followed by a study Greeble (200 ms), a blank screen (500 ms), and a test Greeble (200 ms). Subjects were instructed to judge if the cued halves of the study and test composites were the same or different, while ignoring the other, irrelevant half. On congruent trials, the cued and irrelevant halves were associated with the same response (e.g., both parts same, or both parts different); on incongruent trials, the irrelevant and cued halves were associated with different responses (e.g., one part same, the other part different). A congruency effect (better performance on congruent vs. incongruent trials) indicates an inability to selectively attend: the irrelevant object half influenced performance, despite instructions to ignore it. On misaligned trials, only the test Greeble was misaligned to prevent pseudo-holistic effects that are not sensitive to configuration (see Richler et al., 2009; Richler et al., 2011). The signature of holistic processing is a congruency x alignment interaction, with larger congruency effects on aligned than misaligned trials (Richler & Gauthier, 2014). Likewise, congruency can be defined on phase-scrambled trials (Figure 2), depending on whether the task-irrelevant phase-scrambled half is the same or different between study and test. Therefore, holistic processing using the phase-scrambled baseline is also defined as an interaction, with larger congruency effects on aligned than phase-scrambled trials.

Figure 3.

Figure 3

Open in a new tab

Example stimuli for the Glip Top/Plok Bottom group. For this group, the tops of Glips and the bottom of Ploks were diagnostic during individuation training (top panel). During the composite task, stimuli were created from diagnostic and nondiagnostic Greeble parts (bottom panel). Composite task stimuli were either aligned, misaligned, or the task-irrelevant part was phase-scrambled.

There were 15 trials for each combination of composite condition (diagnostic/nondiagnostic), congruency (congruent/incongruent), alignment (aligned/misaligned/phase-scrambled), cued part (top/bottom) and correct response (same/different), for a total of 720 trials. Cued part was blocked (order counterbalanced across subjects). All other factors were randomized.

Results

Individuation Training

From Day 1 to Day 3, subjects become more accurate (Day 1: M = .92; Day 3: M = .96; F(2,79) = 110.6, p < .0001, ηp2 = .58) and faster (Day 1: M = 946.8 ms; Day 3: M= 709.0 ms; F(2,79) = 246.2, p < .0001, ηp2 = .75).

Composite Task

Data from two trained subjects and two control subjects were removed for below chance performance, leaving 78 trained subjects, and 38 untrained control subjects. Trials with reaction times less than 100 ms or greater than 2000 ms were discarded (1.62% of trials).

Sensitivity (d′) for control and trained subjects (separated into diagnostic and non-diagnostic composites) for aligned, misaligned, and phase-scrambled trials is presented in Figure 4. We found no evidence of holistic processing in control subjects, that is, there was no significant interaction between congruency (congruent/incongruent) and Trial Type (aligned/misaligned/phase-scrambled) (F(2,74) = .32, p = .73, ηp2 = .009).

Figure 4.

Figure 4

Open in a new tab

Sensitivity (d′) as a function of trial type (aligned/misaligned/phase-scrambled) and congruency for the control group, and non-diagnostic and diagnostic composites for the trained group. The aligned and misaligned conditions are connected to highlight the typical congruency x alignment interaction that is only observed in the diagnostic condition.

To assess training effects, all analyses were conducted twice, once comparing aligned trials to misaligned trials as the baseline, and once comparing aligned trials to phase-scrambled trials as the baseline. In all cases holistic processing is indexed by a congruency (congruent/incongruent) x Trial Type (aligned/misaligned or aligned/phase-scrambled) interaction. With each baseline, we first conducted a Condition (control/diagnostic/non-diagnostic) x Congruency (congruent/incongruent) x Trial Type (aligned/misaligned, or aligned/phase-scrambled) ANOVA, treating Condition as a between subjects variable even though diagnostic and non-diagnostic conditions are from the same individuals. The correlation between holistic processing across conditions was expected to be small (Ross et al., 2014), so there should be a negligible cost in power for this strategy1. To unpack any interaction with Condition, we ran within-subject ANOVAs comparing the diagnostic and non-diagnostic conditions in the trained subjects, and between-subject ANOVAs comparing each trained condition to the control group.

Misaligned baseline

The three-way interaction between Condition (control/diagnostic/non-diagnostic), Trial Type (aligned/misaligned), and Congruency was significant, F(2,191) = 3.16, p = .04, ηp2 = .03. With trained subjects only, the Condition x Congruency x Trial Type interaction was only marginally significant, F(1,77) = 3.06, p = .084, ηp2 = .04, with significant holistic processing (Congruency x Trial Type interaction) for diagnostic composites, F(1,77) = 6.67, p = .01, ηp2 = .08, but not for non-diagnostic composites, F(1,77) = 0.12, p = .73, ηp2 = .002. As expected, the interaction between Condition, Congruency, and Trial Type was not significant when the control group was compared to trained-non-diagnostic, F(1,114) = 0.59, p = .44, ηp2 = .0005, indicating that exposure to non-diagnostic parts during training did not result in holistic processing. In contrast, the same three way interaction was significant when the control group was compared to trained-diagnostic, F(1,114) = 4.50, p = .036, ηp2 = .038. This reveals more holistic processing for trained subjects judging diagnostic composites than for the untrained control group. In summary, with the misaligned baseline, the within-subject training effect was marginal, but only diagnostic composites were processed holistically by trained subjects and not untrained control subjects.

Phase-scrambled baseline

The three-way interaction between Condition, Trial Type, and Congruency was significant, F(2,191) = 6.24, p < .002, ηp2 = .06. With trained subjects only, the Condition x Congruency x Trial Type interaction was also significant, F(1,77) = 9.85, p < .002, ηp2 = .11, with significant holistic processing (Congruency x Trial Type interaction) for diagnostic composites, F(1,77) = 8.62, p < .0001, ηp2 = 0.20, but not for non-diagnostic composites, F(1,77) = 1.38, p = .24, ηp2 = .02. As expected, the interaction between Condition, Congruency, and Trial Type was not significant when the control group was compared to trained-non-diagnostic, F(1,114) = 1.31, p = 0.25, ηp2 = .001. However, this same interaction was significant when the control group was compared to trained-diagnostic, F(1,114) = 9.97, p = 0.002, ηp2 = .08, with greater holistic processing for diagnostic composites than the control group. In summary, there was more holistic processing in the diagnostic condition than in the non-diagnostic condition for trained subjects, and untrained control subjects did not process these objects holistically.

Discussion

Novel objects like the Greebles used in our study are not processed holistically by novices. We demonstrate that a history of learned attention to diagnostic Greeble parts was sufficient for the acquisition of holistic processing. This reveals a link between learned attentional strategies and visual object processing. Accounts of item-specific attentional strategies suggest that stimulus features and responses are jointly encoded in memory so that attentional filters are later cued by the perception of features with which they were previously associated (Crump et al., 2008). Here, novices do not process Greebles holistically at all, and trained subjects only exhibit holistic processing for diagnostic Greeble parts. We propose that when trained subjects were shown Greebles made of diagnostic parts in the composite task, they retrieved an attentional filter related to their past history of devoting attention to those parts. When both Greeble parts in the composite task are diagnostic, subjects cannot selectively attend to one Greeble part, producing face-like holistic processing.

In contrast, no holistic processing was acquired for non-diagnostic Greeble parts. However, training did have an impact since there was a main effect of congruency in the trained group not found in Greeble novices (F(1,114) = 4.10, p < .05, ηp2 = .044). Non-diagnostic parts may have become associated with an attentional routine to look for more useful information in a different part of the object, consistent with accounts of learned inattention and blocking during learning (Kruschke & Blair, 2000; Kruschke, 2003), whereby people learn to ignore irrelevant cues. In contrast, the attentional routines that our subjects acquired for diagnostic parts were specific to the aligned configuration, as typically observed for faces (Richler & Gauthier, 2014). The configural specificity of these congruency effects may be akin to learned attention that is specific to the encoding context (e.g., Chun & Jiang, 1998). However, the composites our subjects were tested with represent a new context in many ways: new exemplars were shown in a new task, parts were combined with parts from a different category with which they were never paired during training, and Greebles varied equally in both parts during the composite task, unlike during training. Despite these changes, the task-irrelevant diagnostic parts were difficult to ignore when they were aligned with a task-relevant diagnostic part. This points to spatial configuration as particularly critical for holistic processing, but not because of a special representational status for aligned objects. Rather, configuration may influence the allocation of attention during composite task judgments, perhaps through object-based attention for parts that are grouped perceptually (Vecera & Farah, 1994; Baker et al., 2004). Other work suggests a role of grouping in the composite task, based on evidence that misaligned colored backgrounds behind aligned face parts reduce holistic processing (Curby et al., 2013).

Our effects are similar to those obtained in a number of other paradigms where attention is biased toward information with a history of being attended. Because our effects are measured as reduced selective attention in a congruency task, they are similar to item-specific control during Stroop tasks (Cañadas et al., 2013; Bugg et al., 2011). Because our effects lead to spatial bias attached to a specific visual context, they evoke contextual cueing in which subjects implicitly learn spatial invariants in visual scenes (Chun & Jiang, 1998). However, in contrast to contextual cueing and item-specific control paradigms, our subjects carried attentional settings from a naming task to the composite task. This renders explanations based on conflict (which did not exist during naming) or learned response associations (because naming responses do not predict congruency) improbable. In category learning, attention can shift to a dimension that was useful in a previous categorization task, and, like our effects, this can occur with transfer objects and a new task (Goldstone, 1994), and sometimes last for days (Folstein et al., 2012). As in item-specific control, the effects can be stimulus-specific (Aha & Goldstone, 1992; Van Gulick & Gauthier, 2014) and may become category specific after multiple exemplars are encoded (Nosofsky 1986). Category-specific (or context-specific) control has also been described within Stroop paradigms (e.g,. Bugg et al., 2008). Finally, our results are also similar to effects obtained in the task-switching literature, where task-stimulus associations have long-term effects, primarily on trials when there is conflict, as in the composite task (Waszak et al., 2003). Which of these effects is most similar to holistic processing remains to be determined. Critically, these attentional accounts do not require encoding of unitary representations, a common account of holistic processing (Tanaka & Farah, 1993).

Together with our previous finding that holistic processing was only evident when both parts of a test face benefitted from a history of attention (Chua et al., 2014), the present results suggest that failures of selective attention to diagnostic parts are qualitatively different from those to non-diagnostic parts. Face-like holistic effects appear to require that both the task-relevant and task-irrelevant parts have a history of being attended and that the parts be perceptually grouped, allowing this attentional effect to apply to the entire object. More work is required to account for this configurally-specific learned attention in computational models of attention, but our results provide clear answers to the questions that motivated this work: holistic processing can be acquired for non-face objects, and all that is required is a history of attention to parts.

Acknowledgments

This work was supported by the NSF (Grant SBE-0542013), VVRC (Grant P30-EY008126) and NEI (Grant R01 EY013441-06A2). Riaun Floyd, Amit Khandhadia, and Nicole Goren aided with data collection. We would like to thank Matt Crump for his comments on the manuscript. We would also like to thank Dr. Chu Chang Chua for continued guidance.

Footnotes

1

As expected, there was no significant correlation between holistic processing in the diagnostic and non-diagnostic conditions using the misaligned baseline (r = −.10, p = .37) and that using the phase-scrambled baseline was significant but modest in size (r = .36, p = .001). This could represent an advantage of the phase-scrambled baseline for future work.

References

  1. Aha DW, Goldstone RL. Concept learning and flexible weighting. Proceedings of the fourteenth annual conference of the Cognitive Science Society; 1992. pp. 534–539. [Google Scholar]
  2. Ahissar M, Hochstein S. Attentional control of early perceptual learning. Proceedings of the National Academy of Sciences. 1993;90(12):5718–5722. doi: 10.1073/pnas.90.12.5718. [DOI] [PMC free article] [PubMed] [Google Scholar]
  3. Awh E, Belopolsky AV, Theeuwes J. Top-down versus bottom-up attentional control: a failed theoretical dichotomy. Trends in cognitive sciences. 2012;16(8):437–443. doi: 10.1016/j.tics.2012.06.010. [DOI] [PMC free article] [PubMed] [Google Scholar]
  4. Blair MR, Watson MR, Walshe RC, Maj F. Extremely selective attention: eye-tracking studies of the dynamic allocation of attention to stimulus features in categorization. Journal of Experimental Psychology: Learning, Memory, and Cognition. 2009;35(5):1196. doi: 10.1037/a0016272. [DOI] [PubMed] [Google Scholar]
  5. Bugg JM, Jacoby LL, Toth JP. Multiple levels of control in the Stroop task. Memory & cognition. 2008;36(8):1484–1494. doi: 10.3758/MC.36.8.1484. [DOI] [PMC free article] [PubMed] [Google Scholar]
  6. Bugg JM, Jacoby LL, Chanani S. Why it is too early to lose control in accounts of item-specific proportion congruency effects. Journal of Experimental Psychology: Human Perception and Performance. 2011;37(3):844. doi: 10.1037/a0019957. [DOI] [PubMed] [Google Scholar]
  7. Bugg JM, Crump MJ. In support of a distinction between voluntary and stimulus-driven control: A review of the literature on proportion congruent effects. Frontiers in psychology. 2012;3:367. doi: 10.3389/fpsyg.2012.00367. [DOI] [PMC free article] [PubMed] [Google Scholar]
  8. Cañadas E, Rodríguez-Bailón R, Milliken B, Lupiáñez J. Social categories as a context for the allocation of attentional control. Journal of Experimental Psychology: General. 2012;142:934–943. doi: 10.1037/a0029794. [DOI] [PubMed] [Google Scholar]
  9. Chua KW, Richler JJ, Gauthier I. Becoming a Lunari or Taiyo Expert: Learned Attention to Parts Drives Holistic Processing of Faces. Journal of Experimental Psychology: Human Perception and Performance. 2014;40(3):1174–1182. doi: 10.1037/a0035895. [DOI] [PMC free article] [PubMed] [Google Scholar]
  10. Crump MJC, Vaquero JMM, Milliken B. Context-specific learning and control: The roles of awareness, task relevance, and relative salience. Consciousness and Cognition. 2008;17:22–36. doi: 10.1016/j.concog.2007.01.004. [DOI] [PubMed] [Google Scholar]
  11. Curby KM, Goldstein RR, Blacker K. Disrupting perceptual grouping of face parts impairs holistic face processing. Attention, Perception, & Psychophysics. 2013;75(1):83–91. doi: 10.3758/s13414-012-0386-9. [DOI] [PMC free article] [PubMed] [Google Scholar]
  12. Farah MJ, Wilson KD, Drain M, Tanaka JN. What is “special” about face perception? Psychological review. 1998;105(3):482. doi: 10.1037/0033-295x.105.3.482. [DOI] [PubMed] [Google Scholar]
  13. Folstein J, Newton A, Van Gulick AB, Palmeri T, Gauthier I. Category learning causes long-term changes to similarity gradients in the ventral stream: A multivoxel pattern analysis at 7T. Journal of Vision. 2012;12(9):1106–1106. [Google Scholar]
  14. Gauthier I, Tarr MJ. Becoming a “Greeble” expert: Exploring mechanisms for face recognition. Vision research. 1997;37(12):1673–1682. doi: 10.1016/s0042-6989(96)00286-6. [DOI] [PubMed] [Google Scholar]
  15. Goldstone RL. Influences of categorization on perceptual discrimination. Journal of Experimental Psychology: General. 1994;123(2):178. doi: 10.1037//0096-3445.123.2.178. [DOI] [PubMed] [Google Scholar]
  16. Jacoby LL, Lindsay DS, Hessels S. Item-specific control of automatic processes: Stroop process dissociations. Psychonomic Bulletin & Review. 2003;10(3):638–644. doi: 10.3758/bf03196526. [DOI] [PubMed] [Google Scholar]
  17. Kruschke JK, Blair NJ. Blocking and backward blocking involve learned inattention. Psychonomic Bulletin & Review. 2000;7(4):636–645. doi: 10.3758/bf03213001. [DOI] [PubMed] [Google Scholar]
  18. Kruschke JK. Attention in learning. Current Directions in Psychological Science. 2003;12(5):171–175. [Google Scholar]
  19. Maurer D, Grand RL, Mondloch CJ. The many faces of configural processing. Trends in cognitive sciences. 2002;6(6):255–260. doi: 10.1016/s1364-6613(02)01903-4. [DOI] [PubMed] [Google Scholar]
  20. Nosofsky RM. Attention, similarity, and the identification–categorization relationship. Journal of experimental psychology: General. 1986;115(1):39. doi: 10.1037//0096-3445.115.1.39. [DOI] [PubMed] [Google Scholar]
  21. Richler JJ, Bukach CM, Gauthier I. Context influences holistic processing of nonface objects in the composite task. Attention, Perception, & Psychophysics. 2009;71(3):530–540. doi: 10.3758/APP.71.3.530. [DOI] [PMC free article] [PubMed] [Google Scholar]
  22. Richler JJ, Wong YK, Gauthier I. Perceptual expertise as a shift from strategic interference to automatic holistic processing. Current directions in psychological science. 2011;20(2):129–134. doi: 10.1177/0963721411402472. [DOI] [PMC free article] [PubMed] [Google Scholar]
  23. Richler JJ, Floyd RJ, Gauthier I. The Vanderbilt Holistic Face Processing Test: A short and reliable measure of holistic face processing. Journal of Vision. 2014;14(11):10. doi: 10.1167/14.11.10. [DOI] [PMC free article] [PubMed] [Google Scholar]
  24. Richler JJ, Gauthier I. A review and meta-analysis of holistic face processing. Psychological Bulletin. doi: 10.1037/a0037004. in press. [DOI] [PMC free article] [PubMed] [Google Scholar]
  25. Ross DA, Richler JJ, Gauthier I. Reliability of composite-task measurements of holistic face processing. Behavior Research Methods. 2014:1–8. doi: 10.3758/s13428-014-0497-4. [DOI] [PMC free article] [PubMed] [Google Scholar]
  26. Rossion B, Kung CC, Tarr MJ. Visual expertise with nonface objects leads to competition with the early perceptual processing of faces in the human occipitotemporal cortex. Proceedings of the National Academy of Sciences of the United States of America. 2004;101(40):14521–14526. doi: 10.1073/pnas.0405613101. [DOI] [PMC free article] [PubMed] [Google Scholar]
  27. Rossion B. The composite face illusion: A whole window into our understanding of holistic face perception. Visual Cognition. 2013;21(2):139–253. [Google Scholar]
  28. Sadr J, Sinha P. Object recognition and random image structure evolution. Cognitive Science. 2004;28(2):259–287. [Google Scholar]
  29. Shiu LP, Pashler H. Improvement in line orientation discrimination is retinally local but dependent on cognitive set. Perception & psychophysics. 1992;52(5):582–588. doi: 10.3758/bf03206720. [DOI] [PubMed] [Google Scholar]
  30. Tanaka JW, Farah MJ. Parts and wholes in face recognition. The Quarterly Journal of Experimental Psychology. 1993;46(2):225–245. doi: 10.1080/14640749308401045. [DOI] [PubMed] [Google Scholar]
  31. Theeuwes J. Exogenous and endogenous control of attention: the effect of visual onsets and offsets. Perception & psychophysics. 1991;49(1):83–90. doi: 10.3758/bf03211619. [DOI] [PubMed] [Google Scholar]
  32. Theeuwes J. Stimulus-driven capture and attentional set: Selective search for color and abrupt visual onsets. Journal of experimental psychology. Human perception and performance. 1994;20(4):799–806. doi: 10.1037//0096-1523.20.4.799. [DOI] [PubMed] [Google Scholar]
  33. Van Gulick A, Gauthier I. The perceptual effects of learning object categories that predict perceptual goals. Journal of Experimental Psychology: Learning, Memory, and Cognition. 2014 Sep;40(5):1307–1320. doi: 10.1037/a0036822. 2014. [DOI] [PMC free article] [PubMed] [Google Scholar]
  34. Vecera SP, Farah MJ. Does visual attention select objects or locations? Journal of Experimental Psychology: General. 1994;123(2):146. doi: 10.1037//0096-3445.123.2.146. [DOI] [PubMed] [Google Scholar]
  35. Waszak F, Hommel B, Allport A. Task-switching and long-term priming: Role of episodic stimulus–task bindings in task-shift costs. Cognitive psychology. 2003;46(4):361–413. doi: 10.1016/s0010-0285(02)00520-0. [DOI] [PubMed] [Google Scholar]
  36. Young AW, Hellawell D, Hay DC. Configurational information in face perception. Perception. 1987;16(6):747–759. doi: 10.1068/p160747. [DOI] [PubMed] [Google Scholar]
  37. Zhao J, Al-Aidroos N, Turk-Browne NB. Attention is spontaneously biased toward regularities. Psychological science. 2013;24(5):667–677. doi: 10.1177/0956797612460407. [DOI] [PMC free article] [PubMed] [Google Scholar]

RESOURCES