A few years ago I organized a conference on detection procedures in consciousness science, in Paris. That’s when I met Axel Cleeremans. The title of his talk was something like: “Unconscious Perception: The details matter”. That’s a good title. It’s also a true title: the details do matter.
‘The details matter’ could have been the subtitle of Pournaghdali & Schwartz’s recent review on Continuous Flash Suppression (CFS). They demonstrate that whether a CFS study is worth anything depends on details that are often overlooked. In my humble opinion, anyone remotely interested in CFS should read this article.
So, what is CFS? In a nutshell, experimentalists present a participant with a salient moving pattern – a colored Mondrian or random noise – in one eye, while presenting a target stimulus to the other eye. When doing that, subjects consciously represent the salient Mondrian, and unconsciously represent the target. Or at least that's how it's supposed to work.
Some studies using CFS have reported that targets can be unconsciously represented at a high level of processing (e.g. semantic level). If that’s right, CFS studies demonstrate that you can read unconsciously. That’s amazing. Or perhaps that’s wrong. Incomplete suppression is always a possibility: it could be that participants see the Mondrian pattern but also have some conscious visual cues about the identity of the stimulus.
So, how do we know whether these studies are amazing, or wrong? We look at the details. Pournaghdali & Schwartz have fortunately provided a list of recommendations for optimal suppression. I won’t discuss all of them here. For the full list you can read their article.
First, check how fast the masking pattern moves around – its temporal frequency. CFS studies generally use temporal frequencies between 5Hz and 30Hz, namely, the Mondrian pattern changes every 33 to 200 milliseconds. The current literature suggests that low temporal frequencies are better at masking targets presented for a long time, while quickly presented targets should be masked with high temporal frequency masks. So, while choosing the right temporal frequency might seem a bit arbitrary at first, it’s not.
Second, the spatial frequencies of targets and masks matter as well. In general, the idea seems to be that masks with low spatial frequency have better suppressing effects (e.g. here). So, studies using masks with high spatial frequencies might have incomplete suppression effects. Again, it’s not immediately obvious why this should be the case – I’m actually not sure why, so experimentalists might often overlook this detail.
Finally, the color of the mask relative to the target matters too. Complete suppression of a chromatic target can’t be achieved if the mask is achromatic. Similarly, chromatic masks are pretty bad at masking achromatic targets. The problem is that a lot of CFS studies combine non-colored targets with colored masks. That’s not good. And even if those recommendations were followed, colors seem remarkably difficult to suppress with CFS. Hong & Blake (2009) have shown that even when the shape of a target is successfully suppressed with CFS, its color can still be identified above chance.
This last example illustrates an important point about visual suppression methods: one should be clear about what stimulus features are supposed to be suppressed. If there’s one thing that we learned from decades of masking studies, it’s that surface features (e.g. brightness, color) are not masked in the same way as contour features (e.g. edges, shape). For instance, Stober et al. (1978) have shown that contour and brightness are optimally masked at different Stimulus Onset Asynchronies (SOAs) (see also Breitmeyer et al. 2006). At a given SOA, a mask can be bad at suppressing the brightness level of the target, or its color, and good at suppressing its contour features.
The take-home message is that in CFS as in other forms of masking, masking parameters should be tailored to the specific feature that one aims to suppress, while keeping in mind that those parameters might be quite bad at masking other features of the stimulus. That’s one of the reasons why achieving complete invisibility is tough. Just because you successfully suppressed the shape of a stimulus doesn’t mean that participants don’t see its brightness or its color. Fortunately, subjective invisibility of a given feature of a stimulus might often be good enough.
So, where does this leave us? I think there are two main points here.
First, if you’re reading a CFS study showing unconscious high-level visual processing, and one of those details has been overlooked, chances are good that the visual suppression of the target wasn’t complete. Pournaghdali & Schwartz hold that this is the case for most studies claiming to demonstrate unconscious high-level visual processing with CFS, like facial expression. And as much as I’d like to think there’s unconscious high-level visual processing, their review has convinced me that this hasn’t been shown with CFS yet.
But maybe we can succeed if we follow all those recommendations? To be honest, I’m not sure. At the very least, with CFS, it’s going to be hard. Stimuli that are supposed to trigger high-level visual processing are typically quite complex, like faces, visual scenes, or words. They’re not simple, boring stimuli like Gabor patches. As we saw, it’s difficult to make sure that all the features of a stimulus are correctly masked, even for simple stimuli. So, for complex stimuli like faces, words or visual scenes, it’s not clear that complete suppression with CFS can be achieved anytime soon. We’ll see how the science evolves on this topic in the next few years. It's hard to predict. But learning more about the mechanisms of CFS itself seems like a good start.
I thank Ali Pournaghdali for his comments on a previous version of this blogpost.
MM
No comments:
Post a Comment