Perceptual ambiguity of bistable visual stimuli causes no or little increase in perceptual latency

Abstract

Cognitive ambiguity, such as found in categorical judgments, increases behavioral response latency. Here we examined whether perceptual ambiguity for bistable stimuli, stimuli in which two perceptual interpretations were mutually competitive, also increased perceptual latency. We presented a bistable stimulus and measured the observer's reaction time to judge which of two possible percepts was seen. Perceptual ambiguity was systematically manipulated and how it affected the response latency was examined. The first experiment used a motion-defined rotating cylinder. The observers judged the rotation direction, and the perceptual ambiguity was controlled by binocular disparity. The second experiment used Rubin's vase. The observers judged whether the figure was a vase or faces, and the perceptual ambiguity was controlled by luminance of the surround. In both experiments, we found that the perceptual ambiguity caused only a small or no increase in reaction time and, presumably, in the perceptual latency included in the reaction time. These findings suggest that perceptual competition does not have a strong effect on the latency of the initial perception of bistable stimuli. Given that many perceptual problems are under-constrained as in the cases of bistable stimuli, it is presumably ecologically functional for the brain to establish perception as quickly as possible regardless of the presence of potential alternatives.

Introduction

Visual latency

Due to processing delay, there is always a time lag between the projection of visual input to the retina and the establishment of the corresponding perceptual representation in the brain. We call this time lag “visual latency”. A number of important findings have been made about visual latency in the field of psychophysics, physiology, and brain imaging (see, e.g., Amano et al., 2006; Cook & Maunsell, 2002; Heekeren, Marrett, Ruff, Bandettini, & Ungerleider, 2006; Jaskowski, 1996; Luce, 1986; Roitman & Shadlen, 2002; Sternberg & Knoll, 1973), but our current understanding is far from complete. 

It is hard to estimate the absolute perceptual latency, since how to specify the timing of perception is a matter of controversy. Without specifying the timing of perception, however, one can consider relative changes in perceptual latency and investigate what factors do and do not change perceptual latency. 

Perceptual ambiguity

It is known that perceptual latency can be affected by simple physical parameters of a visual stimulus, such as intensity and spatial frequency (Breitmeyer, 1975; Ejima & Ohtani, 1987; Roufs, 1974). These effects can be ascribed to the ways visual signals activate early visual mechanisms. 

The present study examined whether perceptual latency is also affected by a stimulus parameter with slightly different characteristics, i.e., perceptual ambiguity of bistable stimuli. Unlike the previously examined parameters, perceptual ambiguity is a complex property that depends on many stimulus factors and process factors. 

We expected that the effect of perceptual ambiguity on visual latency would provide insight into the strategy taken by the visual system to cope with a basic problem of perception. The task of the visual system is to estimate objects and events in the environment based on the retinal images. Yet in many cases, retinal images are not sufficient for drawing to a unique interpretation of the environment. Many visual problems are under-constrained and ill-posed (Marr, 1982) and thus ambiguous. In this sense, ambiguous perception for bistable stimuli can be considered as a model case of perception in general, and whether or not extra delay is required to solve perceptual ambiguity of bistable stimuli would suggest how the visual system solves many other ill-posed problems. 

Estimation of perceptual latency from reaction time

To measure the perceptual latency, we used manual response time, a behavioral measure of the response latency of subjective judgments. The response time of a perceptual judgment inevitably includes nonperceptual components, such as the time required for preparation and execution of motor response (Donders, 1868–1869; Luce, 1986; Pins & Bonnet, 1996; Schweickert, Dahn, & McGuigan, 1988). However, by equating those residual components among compared conditions, it is possible to evaluate how a given factor affects perceptual latency. 

One might argue that what can be measured by reaction time is not the latency of perception but the latency of a perceptual decision. This criticism is based on the assumption that perception of an event is distinct from a perceptual decision of a given aspect of the event. An alternative view, one that we are taking as a working hypothesis, is that perception of an event may be a collection of many perceptual decisions about various aspects of the event, with each potentially having latency of its own. According to this view, measuring the latency of a perceptual decision is a proper step toward the estimation of perceptual latency. We do not exclude the possible existence of perceptual components beyond perceptual decisions, but currently we have no idea about how to scientifically analyze those components. 

Besides behavioral reaction time measurement, there are two popular methods for estimating perceptual latency. One is to measure the time course of cortical response evoked by the stimulus by means of invasive or noninvasive methodologies (Amano et al., 2006; Cook & Maunsell, 2002; Roitman & Shadlen, 2002). This approach is indispensable for revealing the neural dynamics underlying perceptual latency. However, to ascertain what the obtained cortical activities means, one should correlate them with the accompanied change in the observers' subjective state, which can be best inferred by behavioral reaction time. Therefore, we consider the measurement of behavioral reaction time to be a critical step toward a general understanding of perceptual latency, including the underlying neural dynamics. 

The other method is to ask observers to judge temporal relationships of two events, such as temporal order, simultaneity, and synchrony-based binding (e.g., Moutoussis & Zeki, 1997; Roufs, 1974; Tappe, Niepel, & Neumann, 1994). However, a subjective temporal judgment is the brain's interpretation of the time course of the events, which may not faithfully reflect the neural latency difference between the compared events (Dennett & Kinsbourne, 1992; Jaskowski, 1996; Libet, Wright, Feinstein, & Pearl, 1979; Nishida & Johnston, 2002; Sternberg & Knoll, 1973). On the other hand, reaction time is a measure of the objective time of perceptual processing, which is unlikely to be postdictively edited by the brain. This difference can account for a number of dissociations between reaction times and subjective temporal judgments (Jaskowski, 1996; Nishida & Johnston, 2002). We find no strong reason to take an alternative radical view that reaction times reflect the dynamics of unconscious perceptual processing and subjective temporal judgments reflect the dynamics of conscious perceptual processing (Neumann, 1990; Tappe et al., 1994). 

Arguments suggesting perceptual ambiguity may or may not increase perceptual latency

The main question addressed in this study is whether perceptual ambiguity increases perceptual latency as measured by reaction time. Several arguments suggest that this may be the case and others suggest it may not. 

The arguments that suggest that perceptual ambiguity may increase perceptual latency can be summarized as follows: First, it is known that the ambiguity in categorical judgment increases response latency (Grinband, Hirsch, & Ferrera, 2006; Ratcliff & Rouder, 1998). Similar delay may be observed for perceptual decisions. Second, the current popular model of reaction time, known as the diffusion model (Bogacz, Brown, Moehlis, Holmes, & Cohen, 2006; Ratcliff & Rouder, 1998; Smith & Ratcliff, 2004), predicts that the reaction time will be elongated when two alternatives of a binary choice are equally likely. The basic assumption of the model is that a decision is made when the accumulated sensory evidence exceeds a response criterion. In a binary choice, the evidence for one alternative and that for the other alternative are fed with opposite signs to an accumulation unit, and one alternative is chosen when the accumulated difference exceeds a threshold (to highlight this aspect, we will call this model “competitive” diffusion model). For bistable stimuli, since the sensory evidence is equally strong for two possible percepts, the accumulation of the evidence difference may be delayed. Third, it has been suggested that reciprocal inhibition among potential perceptual interpretations may cause perceptual alternation in bistable stimuli (for review, see Blake, 1989; Blake & Logothetis, 2002; Leopold & Logothetis, 1999; Pearson & Brascamp, 2008). If the competitive inhibition operates even at the initial perception of a bistable stimulus, mutual suppression may elongate the perceptual processing time. 

On the other hand, the following arguments suggest that perceptual ambiguity may not increase perceptual latency. 

First, as noted above, perceptual ambiguity is a universal property of perception. It would be functionally disadvantageous for the sensory system to have to take extra time to resolve perceptual ambiguity. Second, it is known that stimulus strength is a critical determinant of perceptual latency (Pins & Bonnet, 1996; Roufs, 1974). In many bistable stimuli, both possible percepts are clear and strong and thus likely to have short latencies. Third, it is not known whether the diffusion model (Ratcliff, 1978) is applicable to perceptual decisions. Even if one accepts that evidence accumulation is the basis of perceptual decision, one could imagine an alternative model in which the evidence is accumulated independently for each alternative percept, and a percept that first reaches the threshold wins the race (LaBerge, 1962; Pike, 1966; Vickers, 1970). This independent race model predicts little increase in perceptual latency by perceptual ambiguity. Finally, even if one accepts that competitive interaction between perceptual alternatives causes perceptual alternations, whether it works during the initial period of perception is controversial. Previous studies have shown that a duration of more than a few hundred milliseconds is necessary for the establishment of perceptual rivalry (Wolfe, 1983). It is also suggested that the onset period of binocular rivalry has a different characteristic from those of the perceptual alternation period (Carter & Cavanagh, 2007), and current models of “perceptual memory” consider the onset choice of perception as a process distinct from the subsequent alternation process (Noest, van Ee, Nijs, & van Wezel, 2007; Wilson, 2007). 

Overview of the present study

In this study, we presented a bistable stimulus and asked the observer to make a speeded response to indicate which of the two possible precepts they had seen. Perceptual ambiguity was systematically manipulated and how it affected the response latency was examined. 

We used two bistable stimuli: a motion-defined rotating cylinder ( Experiment 1) and Rubin's vase ( Experiment 2; Figure 1). 

Figure 1

 

Stimuli. (A) Motion-defined rotating cylinder used in Experiment 1. (Upper) Schematic illustration of a stimulus structure. The pattern consisted of 35 white dots placed on the surface of an otherwise invisible rotating cylinder. The velocity of each dot was a sinusoidal function of horizontal spatial position. The dots were given binocular disparity consistent with 3D rotation of a variable depth scale. The depth structure seen from the top is shown in the lower panel. Under the zero disparity condition, all the dots were imaged on the fixation plane. In this case, the subjects perceived a 3D cylinder, but the depth direction was ambiguous. Under nonzero disparity conditions, the perceived direction became less ambiguous. (B) Rubin's vase used in Experiment 2. Rubin's vase (two faces and a vase) was presented with a frame. The frame luminance was changed for the purpose of controlling the ambiguity of the figure. Vase perception dominated when the frame luminance was close to the luminance of the face (left), while face perception dominated when the frame luminance was close to the luminance of the vase (right). When the frame luminance was close to the mean, ambiguous perception was obtained.

The motion-defined rotating cylinder is a structure from motion stimulus, made by the orthographic projection of dots placed at random locations on the surface of a transparent cylinder rotating around the vertical axis (Treue, Husain, & Andersen, 1991; Ullman, 1979; Wallach & O'Connell, 1953). This stimulus produces a compelling sensation of a three-dimensional rotating cylinder, but its motion direction is ambiguous. The ambiguity of the perceived rotation direction is controlled by addition of binocular disparity cues (Dodd, Krug, Cumming, & Parker, 2001). This ambiguous stimulus has been used to examine perception-correlated neural activity in monkey electrophysiology (Dodd et al., 2001; Tsuchiya, Maier, Logothetis, & Leopold, 2008). 

Rubin's vase is a famous bistable picture perceived either as a vase or faces (Rubin, 1958). To control the perceptual ambiguity, we changed the surrounding background luminance. If the background luminance is closer to the vase luminance, the faces are more likely to be a figure. If it is closer to the face luminance, the vase is more likely to be a figure. 

For both bistable stimuli, we found that perceptual ambiguity had little effect on perceptual latency measured by reaction time. 

Experiment 1: Motion-defined rotating cylinder

Methods

Observers

Five adults with normal vision participated in the experiment (four males and one female, 21–45 years old). Two of them (ST, SN) were the authors. The other observers were naive and had previous experience with visual psychophysics. The experiment was approved by NTT Communication Science Laboratories Research Ethics Committee. 

Apparatus

Stimuli were generated on Apple PowerBook G4 1 GHz with Psychophysics Toolbox (Brainard, 1997; Pelli, 1997) and displayed on a gamma-corrected CRT monitor (Sony GDM-F520, monitor resolution: 1024 × 768 pixels, refresh rate: 85 Hz). In a dark room, an observer viewed a stereo pair at the viewing distance of 72 cm through the haploscope with the head fixated with a chinrest. 

Stimuli

The stimulus presented to each eye was a random-dot pattern subtending 3° in width and 3.2° in height. The pattern consisted of 35 white dots presented on a dark background (0.02 cd/m ²). Each dot was a white Gaussian blob ( SD = 0.05°, maximum luminance: 102 cd/m ²), which allowed us to control dot position and binocular disparity with sub-pixel precision. The 35 dots occupied about 25% of the cylinder drawing field. For the purpose of simulating orthogonal projection of a rotating transparent cylinder, each dot moved horizontally at sinusoidally changing velocity. The velocity peaked at the midline of the cylinder, and the direction reversed when a blob reached the edge. The cylinder rotation speed was 0.2 revolutions/s. 

The rotating cylinder was ambiguous with regard to the 3D rotation direction. Either the dots moving rightward or those moving leftward could be seen on the front surface. To control the magnitude of this perceptual ambiguity, we gave binocular disparities to the dots and changed the depth sign and depth scale ( Figure 1A). Disparity gradually changed over space to simulate a cylinder shape. At each location, the disparities of equal magnitude but opposite signs were given to the dots moving rightward and those moving leftward, respectively. The rotation center of the cylinder was always on the fixation plane. Each observer was shown four disparity conditions, including a zero disparity condition, for each disparity sign. At the beginning of the experiment of each observer, we carried out a pilot experiment and manually adjusted the disparity values to achieve the expected control of perceptual ambiguity. 

Procedure

A trial started after an inter-trial interval of about 5 s. The observer was first shown a gray nonius cross. After disappearance of the cross, a cylinder stimulus was presented for 480 ms. The observer's task was to report as quickly as possible the rotational direction by a key press. One key was to be pressed for front-leftward rotation and another for front-rightward rotation. To identify responses that did not reflect the observer's perception, we asked the observer to make a second correct key press immediately after he/she noticed an incorrect key press. These trials (4.0%) were excluded from the subsequent data analysis. Each observer participated in 10 sessions, and each session consisted of 80 trials (2 disparity sign × 4 disparity scale × 10 repeats). Exceptionally, one observer, BS, ran 5 sessions. Within a session, the disparity conditions were randomly ordered, which helped reduce the lock-in effect, which causes only one direction of motion to be continuously observed over trials. 

Reaction time from the onset of the cylinder until the button press was measured at the sampling rate of 80 Hz. We computed the distribution of reaction time for each luminance condition by collapsing responses for both rotation directions. Considering the nonsymmetric distribution of reaction time, we used the median as the measure of central tendency. The trials in which reaction times were over 1.5 s were excluded from the analysis, since they were likely to reflect nonperceptual delays. However, the excluded responses were not many (about 1%) and had little effect on the median RT even if they were included. 

Results

Figure 2 shows the choice probability of front-rightward rotation (upper panels) and the median reaction time (lower panels). The horizontal axis is the peak depth magnitude given to the rightward moving dots by binocular disparity. Positive values imply that the simulated depth was in front of the fixation plane, while negative values imply that the simulated depth was behind the fixation plane. The results are shown separately for each observer. Since the range of disparity was not common across observers, the group average was not computed. 

Figure 2

 

Results of Experiment 1. Individual data of choice probability for right rotation and median response time for each disparity condition. (Upper row) The horizontal axis indicates the value of disparity, and the vertical axis indicates the choice probability of right rotation. A negative disparity value means that the right moving surface is behind the fixation plane. (Lower row) The horizontal axis indicates the value of disparity, and the vertical axis indicates median response time for the rightward and leftward responses in each disparity condition. The error bar indicates the 95% confidence interval.

As we expected, the change in the probability of seeing front-rightward motion was consistent with the given binocular disparity. This suggests that our observers made responses based on their perception. It was theoretically possible that they had made choices irrelevant to the perceived rotation, but if this were the case, the choice probability would be uncorrelated with the disparity. The data also show that the point of perceptual ambiguity (50% point) was at around zero disparity. 

The median of response time was about 400 ms and only slightly affected by the disparity condition. There was no large difference between ambiguous conditions, where the choice probability was close to 50%, and unambiguous conditions, where it was close to 0 or 100%. 

To see the relationship between the perceptual ambiguity and response time in more detail, we computed the magnitude of ambiguity by the following equation:  

$\mathrm{A}\mathrm{m}\mathrm{b}\mathrm{i}\mathrm{g}\mathrm{u}\mathrm{i}\mathrm{t}\mathrm{y}=-p\log 2p-(1-p)\log 2(1-p),$

(1)

where p is the choice probability. Equation 1 tells us how much information about the choice is gained by observing the choice. We consider that a given choice is most unambiguous when the observation of the choice adds no additional information about the choice (since everything is predetermined), while a given choice is most ambiguous when the observation of the choice adds the maximum information about the choice (since nothing is predetermined). The ambiguity index is zero when the choice probability is 0 or 100%, while it is 1 when the choice probability is 50%. 

Figure 3 shows a scatter plot of the reaction times (35 points for seven disparity conditions of the five observers) against perceptual ambiguity. The median reaction times were normalized within each observer based on the overall mean for each observer. The plotted dots show one large group and the rate of increase of RT was almost flat. The slope of linear regression ( R ² = 0.36) was 0.043. Although the slope was significantly different from zero ( F(1,33) = 19.1, p < 0.001), the 4.3% increase in reaction time from the minimum to maximum degree of perceptual ambiguity was considerably small. 

Figure 3

 

The normalized reaction time plotted against the response ambiguity obtained with a motion-defined rotating cylinder ( Experiment 1).

In sum, the results of Experiment 1 indicate that the perceptual latency for the motion-defined rotating cylinder stimulus increases only slightly with increasing perceptual ambiguity. 

Experiment 2: Rubin's vase

Methods

Observers

Five adults with normal vision participated in the experiment (4 males and 1 female, 25–45 years old). Two of them (ST, SN) were the authors. The other observers were naive, had previous experience with visual psychophysics, and one of them also participated in Experiment 1. 

Apparatus

The apparatus was the same as that used in Experiment 1. 

Stimuli

A 2° × 2° picture of Rubin's vase surrounded by a 3° × 3° square frame ( Figure 1B) was presented on a uniform field of 50 cd/m ². The luminance was 81.6 cd/m ² for the face part and 20.4 cd/m ² for the vase part in half of the trials, and vice versa in the other half. For the purpose of controlling perceptual ambiguity, the frame luminance was randomly chosen from 20.4, 46.0, 51.0, 56.1, or 81.6 cd/m ². When the frame had the same luminance as either the face part or the vase part, the other part was highly likely to be seen as a figure (unambiguous conditions). When the frame luminance was close to the mean luminance of the two parts (51 cd/m ²), both the vase and face parts are equally likely to be seen as a figure (ambiguous conditions). 

Procedure

A trial started after an inter-trial interval of about 2 s. A stimulus was presented for 240 ms. No fixation point was presented. The observers were asked to report the perceived figure (face or vase) as quickly as possible by pressing one of two assigned keys. To identify responses that did not reflect the observer's perception, we asked the observer to make a second correct key press immediately after he/she noticed an incorrect key press. These trials (5.8%) were excluded from the subsequent data analysis. Each observer participated in 6 sessions of 200 trials (2 picture polarities × 5 frame luminance conditions × 20 repeats). This yielded 240 trials per condition per observer. 

Time from the onset of the stimulus until the button press was measured as reaction time. The trials in which reaction times were over 1.5 s were excluded from the analysis. We computed the distribution of reaction time for each luminance condition by collapsing vase and face responses obtained under the two contrast-polarity conditions (face dark and vase dark). 

Results

Figure 4 shows the choice probability for vase perception and median of reaction time for each frame luminance condition. As we expected, for all observers, the choice probability shows a monotonic decrease with frame luminance, passing the 50% point at the frame luminance of about 51 cd/m ². This indicates that the observers made responses on the basis of their perception. 

Figure 4

 

Results of Experiment 2. Individual data of choice probability for vase and median response time for each frame luminance condition. (A) Upper: The horizontal axis indicates the value of the frame luminance, and the vertical axis indicates the choice probability for vase. Lower: The horizontal axis indicates the value of the frame luminance, and the vertical axis indicates median response time for vase and face responses in each frame luminance condition. The error bar indicates a 95% confidence interval. (B) Average data of choice probability for vase and normalized response time for each frame luminance condition.

The median of RT was 400–600 ms. The longer RT than that obtained in the first experiment presumably indicates longer processing time for making a perceptual decision. The effect of the frame luminance on the reaction time was small and not consistent across observers. As a further statistical test, we averaged the reaction times over observers after normalizing the median reaction times by the overall mean for each observer. One-way ANOVA indicates no significant main effect of frame luminance ( F(4,20) = 1.246, p = 0.32). 

Figure 5 shows the normalized median RT as a function of choice ambiguity calculated by Equation 1. Linear regression analysis showed that the slope (−0.011) was not significantly different from zero ( F(1,23) = 0.35, p = 0.56). 

Figure 5

 

The normalized reaction time plotted against the response ambiguity obtained with Rubin's vase ( Experiment 2).

The results of Experiment 2 indicate that the perceptual latency for judging a figure for Rubin's face pictures does not change with the degree of perceptual ambiguity. 

Additional analyses

In the analyses described above, we computed the median RT for a given stimulus parameter with mixing the responses for two different percepts, one is more likely and the other is less likely. By considering RTs for the two percepts as separate data, we could examine the effect of perceptual likelihood on RT. Figure 6 shows the median normalized RT for each individual as a function of the choice probability of each percept. For a given stimulus value, there are two points for the two percepts. The figure indicates that RT tends to be elongated as the choice probability decreases. Linear regression analysis indicates that the slopes were significantly negative for both experiments ( p < 0.01): −0.071 and −0.078 when using median RTs for regression (without compensation for the difference in the number of samples), and −0.095 and −0.051 when using raw RT data. Shorter RT for more likely percept is consistent with a prediction of RT models that accumulate evidence over time, such as the diffusion model (Ratcliff, 1978; Ratcliff, Cherian, & Segraves, 2003) and the independent race model (LaBerge, 1962; Pike, 1966; Vickers, 1970). 

Figure 6

 

The normalized RT plotted against the choice probability of each of two possible percepts for the same stimuli. The results of all the observers are shown together. The data for more likely percepts appear on the right, while those for less likely percepts appear on the left. (A) Data obtained with a motion-defined rotating cylinder in Experiment 1. Filled circles and open squares indicate the data obtained when the observers reported leftward and rightward motions, respectively. (B) Data obtained with Rubin's vase in Experiment 2. Filled circles and open squares indicate the data obtained when the observers reported face perception and vase perception, respectively.

It is suggested that the perception of bistable stimuli is affected by the previous stimuli and/or the previous percepts (Kanai & Verstraten, 2005; Pearson & Brascamp, 2008). If such temporal context effects were very strong in our experiments, it could happen that the previous condition made the percept for ambiguous stimuli effectively unambiguous, thereby making its RT as short as that for unambiguous stimuli. To investigate this possibility, we analyzed the effects of the stimulus condition of the previous trial on the choice probability and RT of the current trial. Figure 7A shows the results of this analysis for Experiment 1. In this analysis, we classified disparity conditions into three categories according to the response of each observer. Leftward inducer includes negative disparity values at which rightward response was less than 25% of trials. Ambiguous inducer includes disparity values around zero at which rightward response was between 25% and 75% of trials. Rightward inducer includes positive disparity values at which rightward response was more than 75% of trials. We introduced this classification to average the data across different observers for whom we used different disparity values and to highlight the basic context effects, although the same conclusion could be drawn from the analysis using raw disparity values. Figure 7A shows that an ambiguous inducer following a leftward (rightward) inducer was seen as rightward in 32.3% (56.1%) of trials (see red line in the upper panel). This implies that the response of the current trial for ambiguous stimuli was affected in a positive (assimilative) way by the stimulus of the previous trial, but that the temporal context effect (at least by one trial before) was not strong enough to exclude perceptual ambiguity. In addition, RT for ambiguous stimuli was minimally affected by the stimulus condition in the previous trial, as indicated by red lines in the lower panel (F(2,8) = 0.1823, p > 0.1, one-way ANOVA), although RT would be longer for repeated presentation of ambiguous inducers (central red point) if perceptual ambiguity elongate RT under conditions of small stimulus context effects. Figure 7B shows the results of the similar analysis for Experiment 2. We again used three stimulus categories according to the observer's response, but the mapping rule became very simple in this case. The darkest background condition was the only vase inducer, the brightest background condition was the only face inducer, and the middle three background luminance conditions were ambiguous inducers. Response probability for ambiguous stimuli indicates again a positive context effect, but the magnitude was even smaller than that observed in Experiment 1 (49.0% and 62.1% vase choice after face and vase inducers, respectively). RTs for ambiguous stimuli were not significantly affected by the stimulus in the previous trial (F(2,8) = 0.6557, p > 0.1). Aside from this, Figure 7B indicates an RT reduction for successive presentation of the same unambiguous stimulus. Overall, there was a positive effect of previous stimuli on the perception of ambiguous stimuli, but it does not affect our conclusion that perceptual ambiguity has little effects on RT. 

Figure 7

 

Temporal context effects of the stimulus in the previous trial on the choice probability (upper panel) and normalized RT (lower panel) in the current trial. The stimulus conditions are classified into three types according to the choice probability of the observers. The horizontal axis indicates the stimulus type in the previous trial, while the plotting symbol and color indicate the stimulus type in the current trial. Error bar indicates ±1 SEM across observers. (A) The results of Experiment 1. (B) The results of Experiment 2.

We also carried out a similar analysis on the effect of the perception in the previous trial. In Experiment 1 ( Figure 8A), the percept for ambiguous stimuli was identical to that in the previous trial in 69.2% of trials. Despite the occurrence of this positive perceptual context effect, RT was similar for the same and different responses ( p > 0.1, paired t-test). In the second experiment ( Figure 8B), although RT was slightly shorter for the same responses than for different responses ( p = 0.012, paired t-test), the positive perceptual context effect was small (the percept for ambiguous stimuli was identical to that in the previous trial only in 57.5% of trials). In addition, since the stimulus context effect observed with unambiguous stimuli ( Figure 7B) indicate a reduction in RT for repetition of the same stimulus/response, the perceptual context effect on RT might also be related to the cost of changing response. In sum, the effects of previous percepts on the perception of ambiguous stimuli and on RT were not strong enough to affect our conclusion that perceptual ambiguity has little effects on RT. 

Figure 8

 

Temporal context effects of the perception in the previous trial on the choice probability (upper panel) and normalized RT (lower panel) in the current trial. The stimulus conditions are classified into two types, with the two unambiguous stimulus types in Figure 7 being merged into one. The horizontal axis indicates the response relation between the previous trial and the current trial, while the plotting symbol and color indicate the stimulus type in the current trial. Error bar indicates ±1 SEM across observers. (A) The results of Experiment 1. (B) The results of Experiment 2.

Discussion

Using two bistable stimuli, a rotating cylinder, and Rubin's vase, we found that the reaction time to report perception was not or only slightly affected by the degree of ambiguity. That is, there was little difference in latency to report percept “A” regardless of whether the stimulus was unambiguously seen as “A”, or ambiguously seen as “A” or “B”. 

This is not a conclusion specifically supported by the stimuli used in this study. We obtained a similar result by using an ambiguous rotational apparent motion (Takei & Nishida, 2009). The experiment is described in Figure 9. In a two-frame apparent motion display, four Gaussian blobs located at the corners of a virtual square were rotated around the center. The rotation angle was varied between −45° and +45°. Observers had to report the perceived direction of rotation as quickly as possible by pressing one of two keys. In general, the preferred rotation direction was the direction with the smaller jump angle. The reported direction was most ambiguous (50%) when the rotation angle was 0° (=90°) or 45°. The response time for perceptually ambiguous rotation angles (∼45°) was only slightly longer than that of unambiguous angles (∼20°). In contrast, response times obtained for very small angles, where motion signal weakness made the percept ambiguous, were significantly elongated. 

Figure 9

 

A related experiment using ambiguous apparent motion (Takei & Nishida, 2009). (A) Schematic illustration of a stimulus presentation sequence. After a 480-ms fixation period, the first frame consisting of four Gaussian blobs, arranged at vertices of a virtual 5° × 5° square, was presented for 480 ms. Following a 12-ms ISI, the second frame was presented, in which the four blobs were rotated, relative to the first frame, around the central fixation point. The rotation angle for each trial was randomly selected from the designated values in the range between ±45°. Note that X° clockwise rotation was physically equivalent to (90 − X)° counterclockwise rotation. The second frame was presented until the observer made a binary response about the perceived rotation (clockwise or counterclockwise). The reaction time from the onset of the second frame to the observer's response was measured. (B) Individual data. Error bars indicate the 95% confidential interval calculated by the Bootstrap method. (C) Group average data. Before averaging, we normalized reaction times by the minimum value for each observer. Error bars indicate ±1 SE across observers. In either (B) or (C), the upper panels indicate the choice probability of the rotation direction, which was “correct” in the sense that it was consistent with the smallest angle regardless of whether it was clockwise or counterclockwise. The correct direction was arbitrarily selected for 0° and 45°. The lower panels indicate the response time. In either case, the horizontal axis is the rotation angle (absolute value), with the data being collapsed across different signs of rotation angles. When the rotation angle was between 10° and 30°, the perceived direction was unambiguously determined by the stimulus. As the rotation angle was decreased below 10°, the choice probability was decreased and reached 50% (ambiguous response) at the 0° rotation. This change in choice probability was accompanied by a significant increase in reaction time. Similarly, as the rotation angle was increased beyond 30°, the choice probability was decreased and reached 50% at the 45° rotation. However, the increase in RT accompanying this change was much smaller than that found at small angles, except for one of five observers (TS). (D) Individual normalized RT plotted against response ambiguity index. As the ambiguity increases, RT increases rapidly for small rotation angles but very slowly for large rotation angles (except for the data of TS). For small angles, a reduction in directional signal intensity, not perceptual bistability, produces response ambiguity and causes an increase in RT. On the other hand, for large angles around 45°, perceptual bistability produces response ambiguity and causes little increase in RT, in agreement with the results of the main experiments. Adapted from Takei and Nishida (2009).

To be exact, our data do not exclude a possibility that perceptual ambiguity increases perceptual latency. In the first experiment, we found on average a 4.3% (∼15 ms) increase in RT as the perceptual ambiguity increased from zero to the maximum. While 4.3% is the proportion to the total RT including nonperceptual components, when those components are excluded by subtraction of the simple RT, which is ∼230 ms for the stimulus used in Experiment 1, the increase against the remaining components is 13.8%. In addition, the apparent motion experiment ( Figure 9) also shows a small RT increase as the rotation angle approaches 90°, and the perceptual ambiguity rises to the maximum. These aspects of the results suggest that perceptual ambiguity may have a small but significant effect on perceptual latency. On the other hand, we can also suggest several arguments against this interpretation. First, we did not find the effect of perceptual ambiguity on RT in Experiment 2. Even in Experiment 1 (as well as in the apparent motion experiment), the effect of perceptual ambiguity was not observed for some observers. Second, 15 ms is a small increase in comparison with RT changes introduced by other perceptual parameters: ∼100 ms for flash intensity (Roufs, 1974); ∼150 ms for luminance contrast (Ejima & Ohtani, 1987); ∼40 ms for hue (Bowen, 1981); ∼40 ms for spatial frequency (Ejima & Ohtani, 1987; Tappe et al., 1994); ∼120 ms for motion speed (Hohnsbein & Mateeff, 1992); ∼70 ms for rotation angle (Figure 9, small angles); ∼150 ms for motion coherency (Amano et al., 2006); and ∼200 ms for binocular disparity (Arnold & Wilcock, 2007). Finally, the observed effect of perceptual ambiguity might reflect the effect of stimulus strength. As shown in Figure 6, RT gradually increases as the likelihood of reporting a specific percept decreases, and the stimulus strength supporting that percept decreases. Given that the responses for unambiguous stimuli are mostly the responses for more likely percepts, the increase in RT with response ambiguity is hard to discriminate from an effect by a decrease in stimulus strength for ambiguous stimuli. Although further experimental investigation would be required to draw a definite conclusion about the presence or absence of RT modulation by perceptual ambiguity, we interpret our data as indicating that perceptual ambiguity has either no or very little effect on RT. The following argument is based on this interpretation. 

Given that the measured reaction time includes perceptual latency and that our manipulation of perceptual ambiguity is unlikely to affect reaction time components other than perceptual latency, the present findings suggest that perceptual ambiguity does not elongate perceptual latency, or even if it does, only slightly, in clear contrast to the effects of cognitive or decisional ambiguity (Grinband et al., 2006; Ratcliff & Rouder, 1998). Given that many perceptual problems are under-constrained and ill-posed, it is ecologically functional to establish perception as quickly as possible regardless of the presence of potential alternatives. The present findings with bistable stimuli could be considered as a line of evidence supporting this hypothesis. 

Our findings also support the view that the main factor of perceptual latency is stimulus strength (Pins & Bonnet, 1996; Roufs, 1974). As long as the stimulus is strong, even if the perception is multistable, the perceptual latency is short. With this regard, our findings are similar to the recent one by White, Stritzke, and Gegenfurtner (2008) that saccadic latency is increased when the target stimulus visibility is reduced by stimulus contrast, while it remains short when the target visibility is reduced not by reducing the target contrast but by adding a complex context including potential pseudo targets. 

Using a random-dot motion stimulus containing three coherent signal components, Niwa and Ditterich (2008) examined the response time to detect the dominant signal direction and found that even when the chosen signal had the same signal intensity, the response time was shorter when the three signals had the same strength than when one was stronger than the others. However, their signals were weak, their stimulus was not multistable, and their reaction times were long. We therefore suspect that they observed an effect of stimulus ambiguity on cognitive latency, not a perceptual one. 

Assuming that the perceptual latency is determined by the rate of accumulation of sensory evidence, the present findings are more consistent with the independent race model (LaBerge, 1962; Pike, 1966; Vickers, 1970) than with the competitive diffusion model (Ratcliff, 1978). The difference between the two models is whether the perceptual evidence for two alternative percepts is independently accumulated in separate units (independent race model) or integrated in a reciprocal manner in a single unit (competitive diffusion model). For the independent race model, decision latency is dependent on the stimulus strength but only slightly affected by the presence of alternative percepts. For the competitive diffusion model, on the other hand, when the input stimulus is bistable, mutual cancellation of the sensory evidence for the alternative percepts will elongate the decision latency. It should be added that both models predict a reduction of RT with the stimulus strength (Figure 6). 

Strictly speaking, we may not be able to exclude the competitive diffusion model from possible accounts of the present findings, since it is a complex model whose behavior depends on a number of parameters. For instance, when the stimulus strength is increased, sensory evidence signals become stronger and more variable while the decision criterion remains the same. Under this condition, even a competitive diffusion model may be able to account for a significant reduction of the latency difference between ambiguous and unambiguous conditions, since the variance increase could raise the chance for the evidence accumulation process to accidentally reach one of the boundaries in the short period of time (Niwa & Ditterich, 2008). 

Nevertheless, from a more general point of view, we still do not think that the competitive diffusion model serves as a general model of perceptual latency. This is because it seems unlikely that the brain always knows alternative interpretations of novel sensory inputs. There are some stimuli that have alternative interpretations that would never be taken by some observers. Although the competitive diffusion model is a powerful model for accounting for the reaction time of a binary cognitive decision in which alternative choices are well recognized by the observer, a different model, like the independent race model, may be more suitable for accounting for perceptual latency. 

Perceptual alternation in bistable stimuli may be caused by reciprocal inhibition among potential perceptual interpretations (Blake & Logothetis, 2002). The present findings, however, suggest that the competition mechanism does not delay the initial perception of one of two interpretations. This is consistent with the idea that the onset period of binocular rivalry is different in character from the perceptual alternation period (Carter & Cavanagh, 2007). 

Finally, by comparing our psychophysical findings with physiological responses to bistable stimuli, one could gain insights into neural correlates of perceptual latency, although it seems hard to draw a definite conclusion only from currently available data. Using a rotating cylinder similar to ours, Dodd et al. (2001) found that a correlation of monkey MT activity with the perceived direction of an ambiguous cylinder appeared rather quickly, i.e., within ∼100 ms from the stimulus onset. On the other hand, Tsuchiya et al. (2008) have shown in a preliminary report that the MT signals useful for decoding of perceived direction have substantially longer latency under an ambiguous zero disparity condition than under unambiguous conditions. Using a simple ambiguous motion display, Williams, Elfar, Eskandar, Toth, and Assad (2003) found in LIP, but not in MT, a neural activity correlated with the perceived direction. Since the correlated activity includes predictive components that appeared before stimulus onset, it is difficult to relate their intriguing data with perceptual latency. 

How long perception is delayed from input is a long-standing mystery of sensory science. The present study demonstrates that simple reaction time measurements are still useful in the quest to unravel this mystery. 

Acknowledgments

This work was partly supported by Global COE Program, “Photonics Integration-Core Electronics”, MEXT, Japan. 

Commercial relationships: none. 

Corresponding author: Shin'ya Nishida. 

Email: [email protected]. 

Address: 3–1 Morinosato Wakamiya, Atsugi–Shi, Kanagawa, Japan.