Received 24 October 2005; received in revised form 16 February 2006; accepted 1 March 2006

Transcription

1 Psychiatry Research 149 (2007) Use of the process dissociation procedure to study the contextual effects on face recognition in schizophrenia: Familiarity, associative recollection and discriminative recollection Fabrice Guillaume a,b,, François Guillem a, Guy Tiberghien b, Flavie Martin b, Emilia Ganeva a, Martine Germain a, Tania Pampoulova a, Emmanuel Stip a, Pierre Lalonde a a Centre de Recherche F-Séguin, Hôpital Louis-H Lafontaine, 7331 rue Hochelaga Montréal, (Québec), Canada b Institut des Sciences Cognitives, UMR 5015 CNRS, 67 Bd Pinel, Bron, France Received 24 October 2005; received in revised form 16 February 2006; accepted 1 March 2006 Abstract Contextual effects were explored in schizophrenia patients and paired comparison subjects during a long-term face recognition task. The objective was to investigate the contextual effects on face recognition by manipulating, in the same experiment, the perceptual context of the face (intrinsic vs. extrinsic) and the task context (inclusion vs. exclusion instructions). The situation was derived from the Jacoby's [Jacoby, L.L., A process dissociation framework: separating automatic from intentional uses of memory. Journal of Memory and Language 30, ] process dissociation procedure. The results showed that schizophrenia patients (N=20) presented lower performances than healthy controls (N=20) in the inclusion but not in the exclusion task. This observation emphasizes the heterogeneity of recollection and suggests that the memory impairment in schizophrenia reflects an imbalance between two mechanisms. The first is a deficit in associative recollection, i.e., the failure to use efficiently associative information. The other is an enhanced discriminative recollection that impedes their capacity to process information separately from its perceptual context. In addition, correlation with symptoms suggest that the former is expressed in the loosening of associations characteristic of disorganization symptoms, whereas the latter reflects the lack of flexibility or the contextualization bias related to psychotic symptoms, i.e., delusions and hallucinations Elsevier Ireland Ltd. All rights reserved. Keywords: Context; Dual-process model; Exclusion; Face recognition; Familiarity; Inclusion; PDP; Recollection; Schizophrenia 1. Introduction Since it was first shown that patients with schizophrenia are impaired in categorizing facial stimuli of Corresponding author. Institut des Sciences Cognitives, UMR 5015 CNRS, 67 Bd Pinel, Bron, France. Tel.: ; fax: address: guillaume@isc.cnrs.fr (F. Guillaume). different emotional expressions (Shannon, 1971; Dougherty et al., 1974), research interest in face processing abilities has considerably increased. Subsequent investigations have confirmed that schizophrenia patients are impaired on varied tasks involving the analysis of faces, including familiar and unfamiliar face recognition and identity matching tasks (Berndl et al., 1986; Kring et al., 1993; Archer et al., 1994; Salem et al., 1996) /$ - see front matter 2006 Elsevier Ireland Ltd. All rights reserved. doi: /j.psychres

2 106 F. Guillaume et al. / Psychiatry Research 149 (2007) Considering recognition, we have all had the uncomfortable experience of recognizing a person as familiar, yet being unable to recollect any qualitative or contextual information about the person such as her name or where we met the person before. Such experience suggests that face recognition can be based on an acontextual sense of familiarity, or on the retrieval of contextual details about previous events. Motivated in part by these phenomenological experiences, dualprocess theories propose that recognition reflects the product of two distinct memory processes: familiarity and recollection (Mandler, 1980; Tulving, 1983; Gillund and Shiffrin, 1984; Jacoby, 1991; Hintzman and Curran, 1994; Yonelinas, 1994). Familiarity (F) is generally thought to reflect an assessment of global perceptual similarity between studied and tested items (Mandler, 1980; Murdock, 1982; Gillund and Shiffrin, 1984), whereas recollection (R) entails the retrieval of specific contextual information about studied items, such as physical attributes (Hintzman and Curran, 1994). The simplest method used to dissociate these two processes consists in asking the subject whether he remembers or knows actually remembering seeing an item previously. The first type of response implies the encoding episode, i.e., the context of having seen the item, hence recollection, whereas the second implies only knowing himself to have seen the item without being able to say when or where, hence familiarity. To date, most of the dual process studies of recognition in schizophrenia have used the remember/know procedure and the results have shown a reduced number of remember responses compared with healthy subjects identifying a recollection impairment (Huron and Danion, 2002; Tendolkar et al., 2002). Another method to explore the contribution of familiarity and recollection is the process dissociation procedure (PDP) developed by Jacoby et al. (1991, 1992). The PDP uses two tasks: an inclusion and an exclusion task. In the inclusion task, the subject is required to make a simple old/new recognition judgment. In this task, it is assumed that both familiarity and recollection contribute to the recognition decision. In the exclusion task, the subject has the specific instruction to recognize the test item as well as the context (e.g., the study list) in which it has been presented before. In this case, whereas item recognition can be based on familiarity, the context recognition requires conscious recollective processes; hence, familiarity and recollection act in opposition. Quantitative estimates of recollection and familiarity are then derived according to the assumption that familiarity and recollection contribute independently to recognition (see Section 2 for details) (Jacoby et al., 1993; Joordens and Merilke, 1993; Curran and Hintzman, 1995; Hay and Jacoby, 1996). Some studies using this procedure in schizophrenia patients have also confirmed an impairment in the recollective use of memory (Kazes et al., 1999; Linscott and Knight, 2001). Obviously, the distinction between familiarity and recollection is closely related to the notion of context. Recollection involves the retrieval of specific and contextual information about the studied episode (e.g., source information), whereas familiarity does not. The concept of context is also central in one of the main theories of the memory impairment in schizophrenia whereby patients suffer from a degraded ability to construct and maintain an internal representation of the context (Schwartz et al., 1991; Cohen and Servan- Schreiber, 1992). This view has been reformulated recently as impairment in the mechanism that binds together the separate aspects of an event into a cohesive representation that is consciously retrievable (Danion et al., 1999). If numerous studies have examined the effects of context during recognition, the term context has been given a number of different meanings across studies, such as clothing (Brutsche et al., 1981), encoding instructions (Baddeley and Woodhead, 1982), the background in which the face is presented (Davies and Milne, 1982; Klee et al., 1982; Péris and Tiberghien, 1984), or as a characteristic of the person whose face is presented. In these conditions, the context-processing approach of schizophrenia needs to better specify what is exactly understood by context. From the visual standpoint alone, one can make the distinction between two types of perceptual context (Murdock, 1982; Baddeley and Woodhead, 1982). The intrinsic context refers to the whole set of visual characteristics that are automatically processed as integral parts of the item (e.g., facial expression during face recognition). It is opposed to the extrinsic context, which refers to the environmental aspects that are independent of but visually associated with the item (e.g., the place where the person was seen). On another hand, Baddeley (1982) distinguished between two types of processing context depending on the type of information the subject was required to process in the task. The interactive context is defined by the fact that it affects the meaning or the interpretation of the target event in the context of the task. By contrast, independent context does not interfere with the meaning of the target event and is contingent to the task. The processing context (i.e., interactive vs. independent) and the perceptual context (i.e., intrinsic

3 F. Guillaume et al. / Psychiatry Research 149 (2007) vs. extrinsic) should not be confused one with each other. In fact, both the situational relationship between the target item and its context, i.e., the perceptual context, as well as the type of processing required to accomplish the task, i.e., the processing context, can affect recollection and familiarity in a differential but interactive manner (Atkinson and Juola, 1974; Tiberghien, 1986; Yonelinas, 1994). For instance, the processing of contextual information can be oriented (i.e., interactive context), either toward an intrinsic or an extrinsic aspect of the face picture. In addition, there is strong evidence that the distinctions between different contextual information are all highly relevant in regard to the memory impairment in schizophrenia. For instance, it is well established that schizophrenia patients present an inability to inhibit irrelevant information, whether this information is extrinsic (as in Peterson's type tests: Stuss et al., 1982; Randolph et al., 1992) or intrinsic (as in the Stroop test: Baxter and Liddle, 1998) to the target. Studies more specifically designed to investigate contextual effects in schizophrenia also revealed impairments related to elements of the experimental situation such as source retrieval instruction (i.e., task processing context: Waters et al., 2004) or relational information, such as recency or between-item associations (Vinogradov et al., 1997; Tendolkar et al., 2002; Achim and Lepage, 2003). In addition, there is an increasing amount of data to suggest that impairments in the processing of different types of contextual information are related to different aspects of the expression of the illness. For instance, the impairment in inhibiting irrelevant information has been associated with disorganization symptoms, e.g., thought disorders (Baxter and Liddle, 1998; Bazin et al., 2000; Guillem et al., 2001), whereas the deficit in expression recognition has been related to negative symptoms in schizophrenia (Schneider et al., 1995; Martin et al., 2005). Unfortunately, whereas increasing data suggest that memory impairment is limited to recollection (Frith, 1999; Bazin et al., 2000; Huron and Danion, 2002; Achim and Lepage, 2003), and in spite of the evidence that the contribution of recollection and familiarity can be affected by contextual information and the nature of its association with the target (Atkinson and Juola, 1974; Tiberghien, 1986; Yonelinas, 1994), the study of the interactions between different kinds of contextual information and recognition processes has been largely neglected. There is thus a special challenge to investigate the effects of cross-modifications in the perceptual (intrinsic vs. extrinsic) and task processing Table 1 Theoretical framework accounting for the influence of the various types of context on recollection (R) and familiarity (F) estimates Processing context Perceptual context Intrinsic Extrinsic Interactive R++ and F+ R++ and F Independent R+ and F++ R and F+ : tiny or nonexistent influence, +: weak influence, ++: major influence. context (independent vs. interactive) during recognition (McKenzie and Tiberghien, 2004). The present study was designed to address this issue in face recognition. To date, it seems that only one study has addressed this issue with a face recognition task in schizophrenia (Martin et al., 2004). Unexpectedly and contrary to other studies, the results of this study showed that schizophrenia patients present a familiarity deficit compared with healthy subjects, whereas recollection performance appears relatively spared. The experiment proposed here used a variant of the PDP (Jacoby, 1991) in a face recognition task in which either intrinsic (face expression) or extrinsic (background scene) perceptual aspects of the items were considered either as interactive (exclusion) or independent (inclusion), depending on the task instructions. The manipulation of perceptual context was expected to affect the estimates of recollection and familiarity in a differential manner as a function of both the nature of its relation with the face (intrinsic vs. extrinsic) and of its relevance in the task (independent vs. interactive). Table 1 summarizes the hypothetical influence of contextual manipulations on familiarity and recollection processes. According to this framework, if schizophrenia patients suffer from a recollection deficit (Frith, 1999; Bazin et al., 2000; Huron and Danion, 2002; Achim and Lepage, 2003), the recognition impairment should be more evident in the exclusion rather than in the inclusion task regardless of the intrinsic or extrinsic nature of the contextual change. On the other hand, if schizophrenia patients suffer from impairment in familiarity (Martin et al., 2004), the recognition impairment should be also evident in the inclusion task, particularly in the presence of intrinsic context changes. 2. Methods 2.1. Participants Twenty patients fulfilling the DSM-IV criteria for schizophrenia validated on Structured Clinical Interview

4 108 F. Guillaume et al. / Psychiatry Research 149 (2007) for Diagnosis (SCID) (Spitzer et al., 1990, revised 2002) participated in this study. All were clinically stable (mean illness duration: 6.4, S.D.: 4.9; mean total positive symptoms score: 19.5, S.D.: 23.3; mean total negative symptoms score: 21.4, S.D.: 12.6), suffered from paranoid schizophrenia and were receiving antipsychotic medication at fixed doses for at least 2 months at the time of testing. The global severity of their illness was assessed using the Brief Psychiatric Rating Scale (BPRS), and positive and negative symptoms were assessed with the Scale for the Assessment of Negative Symptoms (SANS) and the Scale for the Assessment of Positive Symptoms (SAPS), respectively (Andreasen, 1984a,b). These evaluations allowed us to exclude concomitant neurological or psychiatric disorders. Exclusion criteria included visual difficulty, history of neurological illness or trauma according to DSM-IV, and age older than 40 years. No correlation was observed between performance and age at onset or antipsychotic dose. Twenty healthy subjects paired with the patients for age, sex and parental socio-educational status were also recruited. The parental socio-educational level was defined on the basis of the occupation of the head of household (usually the father) according to the National Occupational Classification (NOC, 1993). In this classification, occupations are assigned to five categories on the basis of the skill level: I=management, II = professional, III = technical, paraprofessional and skilled occupations, IV = intermediate occupations, V = laboring and unskilled. Written informed consent was obtained from every subject. Information concerning their socio-demographic and clinical data is presented in Table 2. The patients with schizophrenia Table 2 Socio-demographic and clinical profiles a Control group (n=20) Schizophrenia group (n=20) Age (years) 29,4±7,8 27.9±6.5 Gender (male/female) 16/4 17/3 Parental SES 50% 55% (% in classes I III) Age at onset (years) 20.5±2 Illness duration (years) 6.4±4.9 Global symptomatology 9.5±8.4 (BPRS) Positive symptoms (SAPS) 19.5±23.3 Negative symptoms (SANS) 21.4±12.6 Antipsychotic dose (mg CPZeq/day) b 413±284 a Values indicate mean±s.d. b 19 patients received novel or typical antipsychotics, 1 patient was drug-free. and the controls did not differ significantly in sex (χ 2 =1.13) or age (t=0.99) Stimuli Stimuli consisted of black and white photographs of young Caucasian adults without distinctive facial features and carefully edited to maintain a standard brightness and contrast. Different sets of stimuli were constructed, each one comprising two series: one study series (40 faces) and one test series (40 old faces +40 new faces). The order of presentation was randomized for items in the study and the test lists. In the sets corresponding to the handling of the face expression, each face had two versions that differed by the expression, i.e., happy or neutral, so as to allow the manipulation of the intrinsic perceptual context between study and test. These facial expressions have been chosen for their low emotional connotation compared with sad faces. Previous studies have shown the relevance of using these two expressions during recognition in both schizophrenia patients (Loughland et al., 2002) and healthy subjects (Graham and Cabeza, 2001). In the other sets corresponding to the handling of the face's extrinsic background, each face (except the new ones) had two versions that differed by the background scene, so as to allow the manipulation of the extrinsic perceptual context between study and test. The landscapes used consisted of views of mountains or forests that can be easily differentiated one from the other. The stimuli were presented on the monitor of a computer driving the sequence. At both study and test, each trial began with a fixation cross (lasting 1500 ms) in the center of the screen. Then the picture was presented in the same place for 2500 ms. The intertrial interval was randomized between 2500 and 3000 ms at study and between 2750 and 3250 ms at test. Subjects' responses and their reaction times (RT) were recorded separately for each experimental condition and each response type on the computer disk for subsequent analysis Experimental design All participants performed both the inclusion and exclusion tasks involving either change in intrinsic or extrinsic features (Fig. 1). Thus, all participants performed four recognition tests with four different sets of stimuli: inclusion followed by exclusion with intrinsic context manipulation (expression) and inclusion followed by exclusion with extrinsic context manipulation (background). The task order was

5 F. Guillaume et al. / Psychiatry Research 149 (2007) Fig. 1. Intrinsic and extrinsic context manipulations in recognition tests. counterbalanced across participants according to the perceptual context (expression vs. background) manipulations. Each of the four tests consisted of a study followed, 5 min later, by the recognition test. The recognition test required the recognition of studied faces intermixed among new faces (see Fig. 1). Faces tested in the same context condition were paired with the same perceptual features (expression or background: physically identical picture) as in the study. In the different context condition, the faces were paired with a context (expression or background scene) different from the study. The following definitions of the test item categories were used: old items are faces that were in the study list with the same perceptual features (physically identical picture); similar items are faces that were on the study list but appeared with a different perceptual context (expression or background) in the test list; new items are faces that did not appear in the study list (and were presented either with different types of expression or background scene). In each recognition task, subjects were asked to respond at each trial YES or NO by pressing distinct keys on the computer keyboard Intrinsic context manipulation At study, the subjects had to memorize a list of 40 faces and were informed of the perceptual variations occurring in the pictures. During the intrinsic

6 110 F. Guillaume et al. / Psychiatry Research 149 (2007) recognition tests, all faces were superimposed on one landscape: mountain for half of the participants and forest for the others. These faces presented two different expressions (20 happy faces and 20 neutral faces). Two types of recognition instructions were given. During inclusion tasks, participants were asked to recognize any face regardless of changes in expression, i.e., respond YES to any previously presented faces (old and similar face picture) and NO to new faces. Responses to items with new expression, i.e., YES to similar and old expressions; YES to old expressions was scored separately. During exclusion tasks, participants viewed a different set of stimuli with faces that had never been seen before. They were instructed to make affirmative recognition judgments only for faces appearing in their original expressions, i.e., respond YES to old items only, and to reject any face with changes in expression, i.e., NO to similar items, or completely new face, i.e., NO to new items (Fig. 1). This resulted in 80 trials during each recognition test: 40 repeated faces (20 with the same perceptual context: old items; 20 with a new perceptual context: similar items) and 40 new faces (20 new faces with happy expressions and 20 new faces with neutral expressions). For each pair of pictures, face background scene and list presentation order were counterbalanced across participants. In this way, interactions between the intrinsic and the extrinsic context are neutralized Extrinsic context manipulation Subjects were informed of the perceptual variations occurring in the pictures. During the extrinsic recognition tests, faces presented only one expression (happy for half of the participants and neutral for the others) and were superimposed on two different landscapes (20 faces on a mountain background scene and 20 faces on a forest background scene). During inclusion tasks, participants were asked to recognize any face regardless of changes in background scene, i.e., respond YES to any previously presented faces (old and similar face picture) and NO to new faces. Responses to items with a new background scene, i.e., YES to similar, and responses to items with the same background scene, i.e., YES to identical, were scored separately. During exclusion tasks, participants viewed a different set of stimuli with faces that had never been seen before. They were instructed to make affirmative recognition judgments only for faces appearing in their original background scenes, i.e., respond YES to old items only, and to reject any face with changes in background scene, i.e., NO to similar items, or completely new face, i.e., NO to new items (Fig. 1). This resulted in 80 trials during each recognition test: 40 repeated faces (20 with the same background scene: old items; 20 with a new background scene: similar items) and 40 new faces (20 new faces with mountain background scene and 20 new faces with forest background scene). Each pair of pictures, face expression and the list presentation order were counterbalanced across participants. In this way, the interaction between the intrinsic and the extrinsic context is neutralized Data analysis Behavioral measures The rates of correct responses (global accuracy), hits ( yes to old items) and correct rejections ( no to new items) were analyzed using a three-factor analysis of variance (ANOVA) with repeated measures on the variables of group (patients vs. control), type of perceptual context manipulated (intrinsic/expression vs. extrinsic/background) and processing context (independent/inclusion vs. interactive/exclusion). To compare the influence of the perceptual modification in each group as a function of whether the context is independent (inclusion) or interactive (exclusion), a similar ANOVA was conducted on the frequencies of correct responses to similar items in both the inclusion ( yes to similar) and exclusion ( no to similar) tasks. To compare the performance obtained for the exclusion of similar faces, i.e., a condition that required the recollection of the perceptual context, an ANOVA was conducted on the rate of false recollections of similar items ( yes to similar) during the exclusion test. The ANOVA design was a two-factor 2 2 design with repeated measurements on the variables of group (patients vs. controls) and type of perceptual context manipulated (expression vs. background) in a withinsubject variable. An ANOVA was also conducted on the false alarm rate in the inclusion task ( yes to new) in order to assess whether a response bias was present between the inclusion and the exclusion tasks. The ANOVA was a two-factor 2 2 design with repeated measurements on the variables of group (patients vs. controls) and instructions (inclusion vs. exclusion) as a within-subject variable. Reaction times (RT) for correct responses were analyzed using a ANOVA on the variables of group (patients vs. controls), type of perceptual context manipulated (intrinsic/expression vs. extrinsic/ background), type of processing context (independent/ inclusion vs. interactive/exclusion) and recognition conditions (old vs. similar vs. new).

7 F. Guillaume et al. / Psychiatry Research 149 (2007) Two psychophysical measures (A and B ) were also quantified in the four recognition tests from hits ( yes to old) and false alarms ( yes to new) frequencies. A is an index of discriminability between old and new items. It ranges from 0 for null discriminability to 1 for maximal discriminability. B is an index of decision criterion. It ranges from 1 to + 1 with positive values of B reflecting conservative responding, negative values indicating lenient responding and 0 indicating the optimal decision. These psychophysical measures permit us to consider simultaneously hits and false alarms (Pollack and Norman, 1964; Macmillan and Kaplan, 1985). We have chosen A and B rather than d and β because A and B do not postulate normality of distribution. Hence, these psychophysical measures are particularly relevant in pathological or heterogeneous populations (Donaldson, 1996). Similar ANOVAs were conducted on both A and B measures PDP estimates of recollection and familiarity The process dissociation procedure (PDP) proposed by Jacoby provides estimates of recollection (R) and familiarity (F) by comparison of inclusion and exclusion task performances. The estimates of R and F were computed with the following equation of the PDP: manipulated (expression vs. background) and taskprocessing context (inclusion vs. exclusion). ANOVA results obtained on the correct response rate showed a significant group processing context interaction [F (1,19) = 5.86, P = 0.03]. Breaking down this interaction did not reveal any significant effect of the group on performance during the exclusion task. In contrast, in the inclusion task, the percentage of correct response was significantly lower in schizophrenia patients compared to control subjects [72% vs. 79%; F(1,19) = 11.2, P=0.003]. There was also a significant effect of the task instruction where performance being better in inclusion than in exclusion tasks [76% vs. 72%; F(1,19)=12.8, P = 0.002]. Considering only the hit responses rate ( yes to old), ANOVA confirmed that, in both groups, performances were significantly lower when contextual information was interactive to the task than when it was independent [exclusion = 65%, inclusion = 71%; F(1,19) = 13.3, P = 0.002]. As shown in R ¼ pðyes to OldÞInclusion pðyes to SimilarÞExclusion F ¼ pðyes to SimilarÞExclusion=ð1 RÞ: Correlation with symptomatology For each patient, SAPS and SANS item scores were collapsed to provide three dimensions, i.e., negative, disorganization and psychotic, according to the model of Andreasen et al. (1995). Fisher's z test was then used to correlate the SAPS and SANS global scores as well as the dimension scores with four behavioral measures: (1) percent correct response, (2) false alarms rate, (3) discriminability (A ) and (4) decision criterion (B ). In addition, the correlations were analyzed between symptomatology, dimensions scores and the yes responses to similar faces relative to the task instructions and the perceptual context manipulated. 3. Results 3.1. Accuracy Fig. 2 presents the mean percentage of correct response for each group, type of perceptual context Fig. 2. Correct responses and reaction times in schizoprenia and comparison subjects to presented faces under different conditions.

8 112 F. Guillaume et al. / Psychiatry Research 149 (2007) Table 3 Means of yes to old responses in the inclusion tests, yes to similar responses during the exclusion tests for the two groups and the corresponding estimates of familiarity (F) and recollection (R) according to the type of perceptual context manipulated Group Inclusion test ( yes/old ) Exclusion test ( yes/similar ) Schizophrenia patients Comparison subjects Schizophrenia patients Comparison subjects Expression change 0.66 (0.19) 0.79 (0.09) 0.41 (0.2) 0.43 (0.14) Background change 0.63 (0.2) 0.75 (0.14) 0.48 (0.14) 0.55 (0.2) R F Expression change Background change Table 3, correct responses to old faces were significantly lower in schizophrenia patients than in healthy subjects [63% vs. 74%; F(1,19) =7.8, P=0.01]. Finally, considering the false alarm rate ( yes to new), separate ANOVAs showed no significant difference between inclusion and exclusion tasks in each group and each condition (intrinsic or extrinsic). This lack of response bias suggests that the PDP equations may be used without any corrections (Yonelinas and Jacoby, 1996a; Buchner et al., 1997) Reaction time Reaction times (RTs) obtained for correct responses in each condition are presented in Fig. 2. ANOVA shows a significant group processing context (inclusion vs. exclusion) recognition conditions (old vs. similar vs. new) interaction [F(1,19) = 4.9, P = 0.01]. Breaking down this interaction revealed that schizophrenia patients, but not healthy subjects, had significantly slower responses to the similar faces than the old ones during the inclusion tasks [F(1,19)=8.2, P =0.007]. Conversely, healthy subjects, but not patients, had significantly slower responses to the similar faces than the old ones during the exclusion tasks [F(1,19)=15.8, P=0.0003]. This dissociation indicates that changes in the perceptual context disturbed schizophrenia patients when it was independent, whereas, when it was interactive, this manipulation disturbed healthy subjects Correct responses to similar items (influence of perceptual context) The ANOVA on the rate of correct responses to similar items also showed an interaction between group and processing context (inclusion vs. exclusion) [F (1,19) = 5.7, P = 0.03]. Breaking down this interaction revealed that comparison subjects performed significantly better when perceptual context was independent than when it was interactive [inclusion = 67%, exclusion = 51%; F(1,19) = 12.8, P = 0.002], whereas there was no significant difference relative to the processing context in schizophrenia patients (58% vs. 57%). A significant interaction between the perceptual context (expression vs. background) and the processing context was also observed on the correct responses rate to similar items: F(1,19) =12.7, P = Breaking down this interaction revealed that the rate of correct responses to similar items was significantly higher when the perceptual context to be recollected was intrinsic rather than extrinsic to the face [expression=59%, background= 48%; F(1,19)=9.1, P=0.007] Analysis of yes to similar responses in the exclusion test (influence of recollection) More specifically, ANOVA of the yes responses to similar items showed a main effect of group [F(1,19)= 4.6, P=0.04] and a main effect of perceptual context [F (1,19) =9.2, P=0.007]. Comparison subjects did more yes responses to similar items (false recollections) than schizophrenia patients in the exclusion tasks (50% vs. 43%). In addition, both groups had more yes responses to similar items when the interactive context was extrinsic rather than intrinsic to the face (background =51%, expression=42%) Psychophysical measures ANOVA on psychophysical measures showed no significant interactions on discriminability (A ), but three main effects were observed. Schizophrenia patients has a significantly lower discriminability than comparison subjects [0.87 vs. 0.82; F(1,19)=4.4, P = 0.05]. This difference could reflect the patients' difficulty to accept an item as old whatever the task

9 F. Guillaume et al. / Psychiatry Research 149 (2007) instruction (inclusion or exclusion). This difficulty is evidenced by their lower hit rate ( yes to old) compared with healthy subjects (62% vs. 74%), whereas false alarm rates ( yes to new) do not significantly differ between groups (16% in each group). Discriminability was also significantly higher during inclusion compared with exclusion task [inclusion = 0.86, exclusion = 0.83; F (1,19)=7.3, P =0.01] and when perceptual context manipulated was intrinsic rather than extrinsic to the face [expression = 0.85, background = 0.83; F(1,19) = 4.2, P=0.05]. Finally, the analysis of the decision criterion (B ) showed a main effect of the processing context, indicating that subjects adopted a stricter criterion when the perceptual context was interactive than when it was independent [exclusion = 0.3, inclusion = 0.2; F (1,19) = 4.3, P=0.05]. No group effect was observed on the decision criterion, indicating that all participants have respected task instructions PDP estimates of recollection and familiarity Means of yes responses to old items in the inclusion tests, yes responses to similar items during the exclusion tests, and the corresponding estimates of familiarity and recollection according to the type of perceptual context are presented in Table 3. First, the results show that both F and R estimates were lower in schizophrenia patients than in comparison subjects, but the analysis revealed a significant difference between the two groups for the F estimate only [F(1,19) =8.8, P=0.008]. Since it is the yes responses to similar items that establish the F estimate (see Section 2), the lower F estimate in patients could be explained by the fact that they make significantly fewer responses of this type during the exclusion test [comparison subjects=0.51, schizophrenia patients=0.43; F(1,19)= 4.6, P = 0.04]. Second, the analysis showed that in both groups the R estimate was smaller when the perceptual context manipulated was extrinsic rather than intrinsic to the face [F(1,19) = 7.45, P = 0.013]. This difference is partially explained by the increased false recollection probability ( yes to similar) when the extrinsic context is interactive (exclusion) Correlation with symptomatology Correlations between dimension scores and behavioral performance are summarized in Table 4. Regardless of the contextual manipulation, the results showed a significant negative association between discriminability (A ) and positive symptoms (SAPS: z= 2.2, r= 0.49, P = 0.03). More precisely, dimension scores showed negative correlations between discriminability (A ) and both the disorganization (z= 2.2, r= 0,48, P=0.03) and the psychotic dimensions (z= 2.4, r= Table 4 Correlations between schizophrenia patients' symptomatology and behavioral measures (percent correct responses: %RC, false alarms: FA, discriminability: A, yes responses to similar items: yes/similar) Perceptual context Processing context Measures Negative Disorganization Psychotic Expression Inclusion %RC FA A B Yes/similar z=2.4, P=0.017 Exclusion %RC FA z=2.4, P=0.016 A z= 2.7, P=0.008 z= 2.1, P=0.03 B Yes/similar Background Inclusion %RC z= 2.2, P=0.02 FA z=3.4, P= z=2.6, P=0.009 A z= 2.2, P=0.03 z= 2.5, P=0.014 B Yes/similar Exclusion %RC FA A B Yes/similar

10 114 F. Guillaume et al. / Psychiatry Research 149 (2007) , P = 0.015). Similarly, a negative correlation was observed between global accuracy and the positive symptoms (SAPS: z= 2.52, r= 0.55, P=0.01), and, more precisely, with the patients' disorganization (z = 2.4, r = 0.52, P = 0.02) and psychotic dimension scores (z= 2.75, r= 0.58, P=0.006). False alarms also correlated positively with positive symptoms (SAPS: z=2.02, r=0.45, P=0.04) and, more precisely, with the patients' disorganization score (z=2.1, r=0.47, P=0.04). In the sense that yes responses to similar items are produced by old faces and, consequently, are divergent on both the face familiarity assessment and the recollection potentiality, they should not be confused with false alarms (i.e., yes to new). The properties of the yes responses to similar items thus constitute a potentially interesting result to specify the recognition deficit observed in schizophrenia. The analysis showed a positive correlation between yes responses to similar items and the SAPS scores only in the condition where the perceptual context manipulated was intrinsic and independent of the task (i.e., inclusion intrinsic condition) (z = 2, r = 0.44, P = 0.05). This association was specifically related to disorganization symptoms (z=2.3, r=0.52, P=0.02). In summary, the correlations obtained emphasize that positive symptoms correlated with memory discriminability and false alarms both when intrinsic context was interactive and when extrinsic context was independent of the recognition judgment. In addition, results presented in Table 4 also indicate that the higher the patients' disorganization, the more they tended to accept similar items (old face with new expression) or new items (new faces) as old during the inclusion task. 4. Discussion The objective of the study was to investigate the influence of different types of contextual information on familiarity and recollection processes in schizophrenia and healthy subjects using the PDP (inclusion vs. exclusion tasks). Consistent with the general framework presented in Section 1, the results show that both groups have better performance in a simple inclusion recognition judgment compared with the exclusion condition when the perceptual context is interactive to the task. Whatever the nature of the perceptual context manipulated (intrinsic or extrinsic), both schizophrenia patients and healthy subjects present the same difficulty in separating similar items from old ones during the exclusion tasks. If these results are consistent with previous studies showing that recollection and source memory performances are more difficult than the simple judgment of prior occurrence (Mandler, 1980; Tulving, 1985; Kelley and Jacoby, 1998; Yonelinas, 2002), they also indicate that all participants were able to follow the PDP test instructions, i.e., to exclude similar items which perceptual form had changed during the exclusion task. By contrast, the analyses of the interaction between the factors of group and task instructions reveal that schizophrenia patients have poorer performance (i.e., more incorrect responses and slower RTs) in the inclusion tasks compared with healthy subjects. The absence of a group difference on the decision criterion (B ) further indicates that this difficulty is unrelated to non-specific processes of expectancy, motivation or fatigue but is related to recognition process per se. Following the theoretical framework presented in Section 1 (Table 1), these results indicate that the recognition deficit observed here is prominent in the conditions where familiarity is involved to a greater extent. Based on similar results, Martin et al. (2004) suggested that schizophrenia patients had a familiarity rather than a recollection deficit. Such findings challenge the general assumption that recollection, but not familiarity, is impaired in schizophrenia (Huron et al., 1995, 2003; Huron and Danion, 2002; Tendolkar et al., 2002). If schizophrenia patients actually had a recollection deficit, they would at least have shown poorer performance in the exclusion tasks that require the conscious retrieval of contextual information. This expectation is clearly not verified in the present experiment in which patients perform comparably to healthy subjects during the exclusion tasks. The idea of a familiarity impairment is further supported by the analysis of the PDP estimates that reveals that only the decrease of familiarity significantly dissociates schizophrenia patients from healthy subjects. As shown in Table 3, PDP results also reveal that, for each group, the effects observed on familiarity and recollection depend on the relation between the face and the perceptual context manipulated, thus confirming the results obtained on the direct behavioral measures. A major source of the discrepancies between studies showing a recollection deficit (Huron et al., 1995, 2003; Huron and Danion, 2002; Weiss et al., 2002) and those showing a familiarity deficit (present study and Martin et al., 2004) may lie in the designs of tasks in which PDP equations have been applied. In the former studies, the encoding list is used as the exclusion criterion and thus the old similar distinction is based on associative or inter-item information (Linscott and Knight, 2001). In this and in the Martin et al. studies, the

11 F. Guillaume et al. / Psychiatry Research 149 (2007) distinction is based on information encoded as an integral part of the recognition target, i.e., on intraitem's perceptual information. The most parsimonious interpretation of the discrepancies between studies would thus be that schizophrenia patients have poor recollection in tasks involving retrieval of associative or inter-item information but spared recollection and decreased familiarity in tasks requiring the conscious discrimination of intra-item's perceptual changes. It follows that any account of memory performance in terms of familiarity vs. recollection, in healthy or impaired subjects, needs not only to consider the processing context (e.g., inclusion vs. exclusion) and its phenomenological consequences, but also the nature of the contextual information manipulated and its relation to the target item. An alternative proposal could be that patients may not suffer from a familiarity deficit per se, but rather from the paradoxical effect of an over-retrieval of an item's perceptual features that interferes with and impedes the emergence, assessment or utilization of familiarity. Consistently, the analyses show that a patient's poorer accuracy in an inclusion task arises more from the misrecognition of similar faces ( no to similar) than from false alarms ( yes to new). At first, if one assumes that false recognition arises from an illusory familiarity (Schacter et al., 1998), the absence of a group difference in the false alarm rates ( yes to new) appears inconsistent with the idea of a familiarity deficit in schizophrenia. However, the results obtained on the discriminability index (A ) indicate that patients have more difficulties than healthy subjects in accepting an item as old when perceptual modifications have been made. Further, when assessed directly by the rate of no to similar responses, this difficulty is even more pronounced when the intrinsic (i.e., expression) rather than the extrinsic context (i.e., background) is manipulated. These observations raise the possibility that schizophrenia patients may actually base their recognition judgment on a perceptual item's familiarity but are unable to inhibit the processing of irrelevant perceptual changes. The failure to process items separately from their intrinsic perceptual details could well induce a loss of the phenomenological experience of face familiarity and lead to the rejection of similar items. This interpretation is very close to the proposal by Cutting (1985) that schizophrenia patients have an analogical processing excess that leads them to focus their processing on details (Hemsley, 1987), thus impairing their capacity to bind and organize information (Magaro, 1984). In the same line, a potentially useful distinction can be made between two distinct recollection mechanisms depending on the possibility to base conscious recollection either on associative information of the encoding episode (e.g., study list, recency, conceptual codes) or on perceptual familiarity only (intra-item perceptual organization). The first, hereafter termed associative recollection, can be defined as the conscious and conceptually driven retrieval of inter-item or interdomain information (e.g., name or occupation encoded with a face). It corresponds to what is generally understood by recollection and required associative processes between encoded memories. The second, termed discriminative recollection, can be defined as the conscious and perceptually driven retrieval of the information encoded as a whole with the target. It is thus restricted to the retrieval of intra-item or intra-domain information (e.g., details of a face picture). This distinction is reminiscent of the distinction between item and associative recognition made in recent neuropsychological studies in brain lesions (Vargha- Khadem et al., 1997; Mayes et al., 2002) and the theoretical propositions that the nature of the recollection process depends on the type of information to be recollected as well as on the nature of its relation with the target (e.g., the Source Monitoring Framework: Johnson et al., 1993). The distinction between associative and discriminative recollection also appears highly relevant to schizophrenia, particularly in that it may reconcile the discrepancies in the results of studies that used an intra-item's perceptual information (e.g., face expression) and those that used encoding list or any associative information (e.g., recency) as source information. The apparent impaired familiarity associated with the over-retrieval of an item's perceptual features in the former could in fact reflect preserved or enhanced discriminative recollection, whereas the decreased performance obtained in the latter could rely on deficient associative recollection. This distinction can further provide an account for the finding that in schizophrenia the number of remember responses, i.e., recollection, is larger with pictorial material than with words in schizophrenia (Huron et al., 1995). To the extent that pictures favor intra-item perceptual processing, whereas words favor inter-item conceptual processing, this result is consistent with the view of a preserved discriminative recollection vs. a deficient associative recollection. Previous face recognition studies have also reported difficulties for schizophrenia patients in processing one dimension without interference from the other (Whittaker et al., 2001; Baudouin et al., 2002). The enhanced discriminative recollection could well account for these patients' difficulties at dissociating perceptual features previously

12 116 F. Guillaume et al. / Psychiatry Research 149 (2007) encoded as a whole. Closer to the discriminative vs. associative recollection proposal is the observation that a patient's memory performance is poorer when the encoding involves separate elements from different representational domains, but preserved and robust when the encoding involves elements belonging to the same representational domains (Achim and Lepage, 2003; Titone et al., 2004; Pelletier et al., 2005). Many studies have shown that the hippocampus is associated with rapid automatic encoding or arbitrary associations between a to-be-memorized target and its context. The hippocampus may therefore act as a memory differentiator. The neocortex, on the other hand, is thought to associate traces by binding them to their different contexts and, as such, performs a generalization function (for a review, see O'Reilly and Rudy, 2001; O'Reilly and Norman, 2002). The cerebral locus of the memory deficit observed here could then result from a specific impairment of the interaction between the hippocampal system and the temporal lobe. Altogether, these observations foster the conclusion that the memory impairment observed in schizophrenia reflects an imbalance between two mechanisms. The first expressed as a deficient associative recollection could correspond to the failure to use efficiently associative information that is thought to underlie the loosening of associations and thought disorders, characteristic of the disorganization. The other, expressed as an enhanced discriminative recollection, could well correspond to lack of flexibility and to the abstraction or contextual bias that has been related to psychotic, i.e., delusion and hallucination, symptoms (Bentall, 1990). This view is very similar to the one developed in recent studies investigating memory in schizophrenia with eventrelated potentials (Guillem et al., 2001, 2003). These studies described a decrease in associative activation (posterior N400 effect) with a concomitant increased, but inappropriate, processing of intrinsic contextual information (FC or F-N400 effect). Further, the first anomaly has been associated with disorganization, and the second, interpreted as an attempt to compensate for the reduction in associative activation, was specifically related to psychotic symptoms. Besides, the present correlation with symptom dimension scores is also in agreement with this view. On the one hand, results showed that false alarms and yes responses to similar items (in an expression inclusion task) increase mainly, if not only, with disorganization scores. Assuming that these associations respectively reflect an increased illusory familiarity and a greater facility to accept an item as old, these results are consistent with the loosening of associations that characterizes disorganization (see above). On the other hand, a decreased discriminability (A ) and accuracy score (in background inclusion task) correlate either with both the disorganization and psychotic scores or with psychotic scores only. The feelings of depersonalization often observed in delusional syndromes are in line with this idea and are interpreted clinically as a loss of the feeling of familiarity with respect to a past perceptual experience (Weinstein, 1994). This is particularly evident in the conditions in which the enhanced discriminative recollection could impede subjects' capacity to dissociate perceptual features previously encoded as a whole, i.e., expression exclusion and background inclusion. In this case, the idea of Yonelinas and Jacoby (1996b) that patients can somehow compensate their deficits in recollection, i.e., namely associative recollection, by boosting familiarity, Table 5 Mean and standard deviation (S.D.) of correct-answer percent and corresponding reaction time (ms) for old, similar and new items during the inclusion and the exclusion face recognition time as a function of the perceptual context manipulation Perceptual context manipulation Intrinsic context manipulation (expression) Extrinsic context manipulation (background) Recognition task Subjects Old Similar New Correct response (%) Reaction time (ms) Correct response (%) Reaction time (ms) Correct response (%) Reaction time (ms) Inclusion Controls 0.79 (0.09) 1679 (303) 0.64 (0.11) 1826 (359) 0.85 (0.11) 1708 (342) Schizophrenia 0.66 (0.19) 2043 (484) 0.54 (0.17) 2179 (543) 0.84 (0.12) 2010 (596) patients Exclusion Controls 0.73 (0.15) 1939 (387) 0.55 (0.14) 2083 (337) 0.83 (0.12) 1720 (349) Schizophrenia patients 0.60 (0.16) 2148 (498) 0.62 (0.19) 2138 (551) 0.86 (0.09) 1927 (458) Inclusion Controls 0.75 (0.14) 1770 (464) 0.71 (0.13) 1745 (524) 0.84 (0.11) 1797 (436) Schizophrenia 0.63 (0.2) 1900 (517) 0.63 (0.2) 2010 (597) 0.83 (0.13) 1905 (547) patients Exclusion Controls 0.67 (0.15) 1999 (488) 0.45 (0.2) 2196 (426) 0.80 (0.15) 1865 (413) Schizophrenia 0.61 (0.12) 2329 (596) 0.52 (0.14) 2292 (616) 0.82 (0.14) 1996 (665) patients