Effects of repetition learning on upright, inverted and contrast-reversed face processing using ERPs

Abstract

The effects of short-term learning on memory for inverted, contrast-reversed and upright faces were investigated using event-related potentials (ERPs) in a target/nontarget discrimination task following a learning phase of the target. Subjects were equally accurate for all three face types although responding more slowly to inverted and negative faces compared to upright faces. Face type affected both early ERP components P1 and N170, and long-latency components at frontal and parietal sites, reflecting the difficulty of processing inverted faces. Different effects of face type were found for P1 and N170 latencies and amplitudes, suggesting face processing could start around 100–120 ms and is sensitive to facial configuration. Repetition effects were also found on both early and long-latency components. Reduced N170 latency and amplitude for repeated targets are likely due to perceptual priming. Repetition effects on the N250 were delayed for inverted and negative faces, suggesting delayed access to stored facial representations for these formats. Increased frontopolar positivity at 250–300 ms and parietal positivity from 300 to 500 ms reflected familiarity ‘old–new’ repetition effects that were of similar magnitude for all three face types, indexing the accurate recognition of all faces. Thus, while structural encoding was disrupted by inversion and contrast-reversal, the learning phase was sufficient to abolish the effects of these configural manipulations behaviourally; all three face types were equally well recognised and this was reflected as equally large parietal old–new effects.

Introduction

Face inversion is well known to impair face perception and recognition by disrupting configural information at the encoding stage of face processing Bartlett and Searcy, 1993, Freire et al., 2000, Leder and Bruce, 2000, Rhodes et al., 1993, Searcy and Bartlett, 1996. Similar to this “inversion effect”, the “contrast-reversal effect” derived from reversing the contrasts in face pictures also impairs face recognition at the encoding level Bruce and Langton, 1994, Hole et al., 1999, Kemp et al., 1990, Lewis and Johnston, 1997, White, 2001. Although contrast-reversed (negative) faces seem to be processed holistically like upright faces (Hole et al., 1999), contrast-reversal also disrupts configural information Kemp et al., 1990, Lewis and Johnston, 1997, Liu and Chaudhuri, 1998, like inversion. Additive effects of inversion and contrast-reversal on performances have suggested that they impair different face processes Bruce and Langton, 1994, Kemp et al., 1990, Lewis and Johnston, 1997.

Numerous electrophysiological studies recording event-related potentials (ERPs) from the scalp have also investigated inversion effects. P1, an early visual endogenous ERP component appearing around 100–120 ms, seems sensitive to faces Halit et al., 2000, Itier and Taylor, 2002, Itier and Taylor, 2004a, Linkenkaer-Hansen et al., 1998, Taylor, 2002, Taylor et al., 2001, being shorter and sometimes smaller for upright than inverted faces and larger to faces than to objects categories in some studies (e.g. Itier and Taylor, 2004a). Much more frequently measured, the negative component N170, occurring between 140 and 200 ms after stimulus onset, is systematically larger and often earlier to faces than to other object categories (e.g. Bentin et al., 1996, George et al., 1996, Itier and Taylor, 2004a) and is considered a neurophysiological correlate of face perception. N170 is usually delayed and larger for inverted than upright faces Bentin et al., 1996, de Haan et al., 2002, Eimer, 2000a, Itier and Taylor, 2002, Itier and Taylor, 2004a, Linkenkaer-Hansen et al., 1998, Rossion et al., 1999, Rossion et al., 2000, Sagiv and Bentin, 2001, Séverac-Cauquil et al., 2000, Taylor et al., 2001, and these effects are thought to reflect the processing disruption of face configuration by inversion. Similarly delayed latencies and enlarged amplitudes of N170 were found for negative faces compared to upright faces (Itier and Taylor, 2002), also reflecting face configural disruption by contrast-reversal. N170 is often thought to reflect automatic face processing as some studies failed to find familiarity (e.g. Bentin and Deouell, 2000, Eimer, 2000b, Henson et al., 2003, Schweinberger et al., 2002a) or repetition effects (e.g. Eimer, 2000b, Pfütze et al., 2002, Schweinberger et al., 2002a, Schweinberger et al., 2002b) on N170. However, others have found familiarity effects under the form of smaller N170 amplitude for familiar faces compared to faces seen for the first time George et al., 1997, Jemel et al., 2003 while others found the opposite effect, that is, larger amplitude for familiar faces (Caharel et al., 2002). Repetition effects on N170 with unfamiliar faces have been reported, restricted to the right hemisphere (Campanella et al., 2000), or bilateral Guillaume and Tiberghien, 2001, Itier and Taylor, 2002; this suggests that this component could be influenced by priming and learning. This discrepancy in the literature could come from the stimuli (Mooney vs. greyscale pictures of faces, same or different pictures of the face between first and second presentation, neutral vs. emotional faces), and/or from the paradigm used (immediate or delayed repetitions, one vs. several repetitions of the same face, etc.). Using a study phase-test phase paradigm, we wished to investigate these repetition effects on the N170 and see whether they would differ according to face type and to the number of target presentations.

Early ERP components are the focus of the majority of face studies. However, in repetition paradigms, late ERP components are also a reliable neurophysiological index of cognitive events as they reflect memory-related processes associated with the repeated items. A negative component found between 200 and 300 ms at temporal parietal sites, the N250, has been shown to be sensitive to immediate Begleiter et al., 1995, Schweiburger et al., 1995 and short lag face repetitions (Pfütze et al., 2002) but not to long lag repetitions of faces (Schweinberger et al., 2002a). N250 is more negative for repeated than new face stimuli and is a good index of repetition priming (Schweinberger et al., 2002b). However, it has not been investigated with inverted and negative faces. In the present study, it was used to compare the repetition effects of upright, inverted and negative faces. Longer-latency repetition effects are also commonly analysed in the ERP literature. Some experiments using study phase–test phase paradigms have found that during the study (or learning) phase, more positive amplitudes were found for items that would be successfully recognised later in the test phase compared to items that would not be recognised. This effect was called the Dm (difference in subsequent memory) by Paller et al. (1987). Measured during encoding (learning) trials, it is predictive of memory performances at test, and was found for words (Paller et al., 1987) and also for faces Sommer et al., 1995, Sommer et al., 1997. Other studies have measured ERPs during the test phase or recognition trials and have found an ERP correlate of repetition differentiating new and old items. Larger amplitudes are typically found for repeated or studied (old) items compared to new, unstudied items. These so-called “old–new” repetition effects have been found at frontal and parietal sites for words (for a review, see Rugg and Allan, 2000), objects Curran and Cleary, 2003, Nielsen-Bohlman and Knight, 1994 and faces (Paller et al., 1999, Paller et al., 2000; Schweinberger et al., 1995; Schweinberger et al., 2002a, Schweinberger et al., 2002b). Although the exact timing and localisation of the effects varies between studies, recent research suggests that an early old–new effect occurring from 300 to 500 ms and situated frontally could be a neural correlate of familiarity, while a later centro-parietal component, from 500 to 800 ms, indexed recollection. In these memory tasks, familiarity usually refers to a global similarity in memory between new and old items while recollection refers to the retrieval of specific information concerning studied items (e.g. context, source, associative information). In an n-back task where upright, inverted and contrast-reversed (negative) faces repeated only once with 0 or 1 intervening face, Itier and Taylor (2002) found frontal and parietal repetition effects that varied in amplitude depending on face type. They hypothesised that the larger old–new effects found for upright faces reflected a better recognition than the other two face types for which the old–new effects appeared weaker. In the present study, we wished to investigate whether sufficient learning of a target face would lead to similar repetition effects between upright, inverted and negative faces. This would suggest that faces were encoded well enough to be recognised similarly between face types despite configural disruption by inversion and contrast-reversal. We investigated early (P1, N170) and late ERP components as neurophysiological correlates of face processes, while subjects were engaged in a learning phase–test phase paradigm involving upright, inverted and negative faces. As inversion and contrast-reversal impair the encoding stage of face processing, shown in the behavioural literature, we wished to determine if longer exposure to inverted or negative target faces would diminish these effects at the behavioural level, and if so, how this was indexed neurophysiologically. The structural encoding in the ERP face literature, which is different from the general term “encoding” in memory paradigms, refers to the early stages of perception where the facial features are integrated to generate a face representation. This process is believed to occur at the level of the P1–N170. However, note that regardless of whether the trial is a learning trial (in a study phase) or a recognition trial (in a test phase), these early processing stages are still necessary. Accordingly, if the hypothesis of configural disruption at the encoding step of face processing is correct, the disruption of P1 and N170 components by inversion and contrast-reversal should still be seen, even after many repetitions of a target face. However, what should vary with learning are the later repetition effects related to memory processes. In the case of diminished or no behavioural inversion and negative effects, which would suggest learning of these faces despite configural disruption, repetition effects should be similar across face types, or the difference in repetition effects between face types should be small. Subjects were presented with blocks of upright, inverted and negative faces and had to recognise target faces intermixed with nontarget new faces, after having learned the targets in a prior learning (study) phase. We thus had several goals: (1) to compare the effects of inversion and contrast reversal on encoding and recognition processes in this paradigm, (2) to investigate whether a long exposure to the target before test would diminish the inversion and contrast-reversal effects at the behavioural level, and the neurophysiological correlates of these effects; (3) to determine the magnitude of the repetition effect with few vs. many repetitions of the target. This last point was investigated by analysing the ERPs recorded in the first half and in the last half of target presentations in the test phase separately.