Raphaël Fournier^1,2, Paul E. Downing³, Richard Ramsey^1,4; ¹Neural Control of Movement Laboratory, Department of Health Sciences and Technology, ETH Zurich, Zurich, Switzerland, ²Neuroscience Center Zurich, University and ETH Zurich, Zurich, Switzerland, ³Department of Psychology, Bangor University, Bangor, United Kingdom, ⁴Social Brain Sciences Laboratory, Department of Humanities, Social and Political Sciences, ETH Zurich, Zurich, Switzerland

The ability to quickly and accurately recognise a familiar face is an important aspect of everyday life, as it guides social behaviour. The way that one might interact with a family member or friend is different to a stranger. Moreover, for familiar individuals, we not only know what they look like, but we also have access to person knowledge, such as their traits, attitudes, name and professional status. To date, most prior face recognition research has used manifest (observed) measures, such as response times and accuracy, to assess one's ability to recognise faces. As such, there is currently a rather limited understanding of the computational processes that support face recognition processes. Using evidence accumulation models (EAM), we investigated how different types of face familiarity (visual familiarity and stored person knowledge) impacted the signal-to-noise ratio (drift rate parameter) and response caution (threshold parameter) during a face memory task. We demonstrated that different computational processes support the recognition of different types of faces. In Experiment 1, visually familiar faces produced a lower signal-to-noise ratio (i.e., drift rate) than novel faces. In contrast, in Experiment 2, the combination of visual familiarity and semantic knowledge about a face induced a higher signal-to-noise ratio compared to novel faces. Given that visual familiarity is associated with the “core” face perception network in the ventral visual stream, whereas semantic knowledge is associated with the “extended” face perception network in more anterior brain regions, these results give some tentative insight into the computational processes that take place in different parts of the face processing network.