Using eye movements to isolate information use for faces and ecologically valid information sampling in the wild

The eyes sample the visual world at a high speed. The purpose of such fast sampling is to counteract the low-resolution information extracted by the extremities of the retina. As such, eye movements constantly position the targets at the retina centre with high resolution. Determining where the eyes land gives access to the information sampled and perhaps to the information processed by the brain, opening doors to cognition. However, it has been shown that the position of the fixation does not necessarily reflect information use, as the eyes may extract information from the periphery. To overcome this problem, gaze-contingent paradigms allow one to precisely control the extent of information available around the fixation. In addition to this limitation of eye movement research, most eye-tracking studies are performed in confined laboratory settings that do not guarantee the generalisability of their findings to the real world.

This thesis aims first to evaluate information use in facial recognition utilising gaze-contingent paradigms. Second, it aims to ensure the ecological validity of eye movement studies in the laboratory. Despite the assessment, almost 50 years ago, of the amount of information that can be read in a single fixation and the growing number of studies evaluating facial information sampling, the perceptual span for faces—the facespan—has not been determined yet. The first contribution of this thesis is therefore an evaluation of the facespan with a gaze-contingent moving window paradigm. Several window sizes showing the target facial information were used in a face recognition task. Our results indicate that an aperture of at least 17° of visual angle allows for normal performance and oculomotor behaviour. Within this aperture, 7° were preserved from the Gaussian aperture, hence, evaluating the facespan at 7°. However, as in reading, this quantity might be modulated by different parameters such as expertise, disorders, or age. For this reason, the second contribution is an evaluation of the modulation of the quantity of facial information extracted in the case of non-hearing individuals. We presented faces upside down or upright to a group of hearing-impaired persons and a control group in a face identification task. This time, a window extending according to the duration of the fixation was used.

Based on our results, the hearing-impaired people were less affected by the inversion effect than the control group. In addition, the hearing impaired extracted more information from a face at each fixation than the control group. Those studies performed in a confined laboratory raise the more general question of ecological validity for which previous studies did not find a consensus. Thus, the last contribution is an answer to this question. To this end, we compared the eye movements of participants in the real world and in the laboratory. Conditions were kept as similar as possible between the two environments. Our results revealed that, unlike in the laboratory, participants in the real world were regularly watching their next steps. Regarding similarities, participants in both groups looked at actionable objects, signs and faces in the same way. In conclusion, these contributions clarify the minimal quantity of facial information needed at each fixation and its modulation by deafness. In addition, the results of studies on eye movements obtained in the laboratory seem to be valid, but only under specific conditions.