Abstract
There is evidence that humans implicitly learn an average or prototype of previously studied faces, as the unseen face prototype is falsely recognized as having been learned (Solso & McCarthy, 1981, Br. J. Psychology). Here we investigated the extent and nature of face prototype formation in a multidimensional face space with a face learning and old/new discrimination experiment. Observers first studied eight synthetic faces defined by geometric information in a 37-dimensional face space (Wilson, Loffler, & Wilkinson, 2002, Vision Research). These faces were equidistant from the unseen prototype and comprised face, anti-face pairs along four axes with the unseen prototype as origin. After studying the faces for 30 s each, memory was tested using a series of test faces, each flashed for 240 ms. Test faces included previously studied faces, the unseen prototype, and eight novel distractor faces defined by four new axes in face space with a face and anti-face on each. Results showed that the unseen prototype was falsely identified as learned at a rate of 86%, whereas studied faces were identified correctly 66% of the time and the distractors 32%. This lasts at least one week. Additional studies demonstrated flexibility of prototype learning: the learned prototype could be either the face-space population mean or a highly distinctive non-mean face. Contrary to previous results (Cabeza et al., 1999, Memory & Cognition), this prototype effect for geometric face information also generalized across viewpoints, as the unseen prototype of faces rotated 20° was also falsely recognized after studying frontal views of the same faces (and vice versa). Further experiments suggest that head shape and internal features separately contribute to prototype formation. Thus, implicit face prototype extraction in a multidimensional space may be a very general aspect of geometric face learning.
This work was supported by CIHR Grant #172103 and NSERC Grant #OP227224 to H.R.W.