an electrophysiological comparison of visual

Cognitive Affective amp Behavioral Neuroscience2002 2 (1) 1-18

There is a subtle distinction between the cognitiveprocesses involved in object categorization and those in-volved in recognition memory for specific objects Whereasobject categorization emphasizes the process of findingsimilarities that bind object exemplars into categories(eg Spot Fido and Rover are dogs) recognition mem-ory emphasizes the process of judging whether or not a par-ticular exemplar has been previously encountered within agiven context (eg ldquoThis dog is Spot who lives next doorrdquo)Previous research has highlighted both differences andsimilarities in the cognitive mechanisms thought to under-lie categorization and recognition

Neuropsychological research has emphasized differ-ences between recognition and categorization and has fos-tered the view that the two abilities may depend on differ-ent brain systems When required to classify novel dotpatterns according to whether they belonged in the samecategory as a set of training patterns (following themethod of Posner amp Keele 1968) amnesic patients cate-gorized as accurately as control subjects despite signifi-

cant recognition memory impairments for the training patterns (Knowlton amp Squire 1993 Squire amp Knowlton1995) Similar amnesic dissociations between explicit mem-ory for exemplars and categorization have been observedwith artificial grammar learning (Knowlton Ramus ampSquire 1992 Knowlton amp Squire 1996) probabilisticclassification learning (Knowlton Gluck amp Squire 1994Knowlton Mangels amp Squire 1996) and learning cate-gories of novel object-like stimuli (Reed Squire PatalanoSmith amp Jonides 1999) Knowlton Squire and colleagueshave generally interpreted these dissociations as indicatingthat categorization and recognition depend on separatememory systems Recognition memory depends on a de-clarative memory system including the hippocampus themedial temporal cortex and diencephalic brain regions typ-ically implicated in amnesia Categorization is thought todepend on implicit memory systems that are somewhattask dependent Most relevant to present visual categoriza-tion research learning of visual dot patterns is related toactivity within visual cortical areas as measured with func-tional magnetic resonance imaging (f MRI Reber Starkamp Squire 1998a 1998b)

In contrast to neuropsychological methods mathemat-ical modeling approaches have emphasized the similari-ties between categorization and recognition In particularseveral exemplar-based models can explain a variety of cat-egorization and recognition phenomena (eg Estes 1994Hintzman 1986 1988 Medin amp Schaffer 1978 Nosofsky1988 1991) In general a single memory system is positedto store exemplars of individual experiences Memory re-trieval involves computing the similarity between a test item

1 Copyright 2002 Psychonomic Society Inc

This research was supported by a grant from the McDonnell-PewFoundation Program in Cognitive Neuroscience a 21st Century Collab-orative Activity Award from the James S McDonnell Foundation (sup-porting the ldquoPerceptual Expertise Networkrdquo) National Science Founda-tion Grant 9729030 and National Institute of Mental Health GrantMH64812 Thanks to D Collins C Piatt M Pollack K Tepe and K Wai-mey for research assistance and Electrical Geodesics for technical sup-port Address correspondence to T Curran Department of PsychologyUniversity of Colorado 345 UCB Boulder CO 80309-0345 (e-mailtcurranpsychcoloradoedu)

An electrophysiological comparison of visual categorization and recognition memory

TIM CURRANUniversity of Colorado Boulder Colorado

and

JAMES W TANAKA and DANIEL M WEISKOPFOberlin College Oberlin Ohio

Object categorization emphasizes the similarities that bind exemplars into categories whereas recog-nition memory emphasizes the specific identification of previously encountered exemplars Mathe-matical modeling has highlighted similarities in the computational requirements of these tasks but neu-ropsychological research has suggested that categorization and recognition may depend on separatebrain systems Following training with families of novel visual shapes (blobs) event-related brain poten-tials (ERPs) were recorded during both categorization and recognition tasks ERPs related to early visualprocessing (N1 156ndash200 msec) were sensitive to category membership Middle latency ERPs (FN400effects 300ndash500 msec) were sensitive to both category membership and oldnew differences LaterERPs (parietal effects 400ndash800 msec) were primarily affected by oldnew differences Thus there wasa temporal transition so that earlier processes were more sensitive to categorical discrimination andlater processes were more sensitive to recognition-related discrimination Aspects of these results areconsistent with both mathematical modeling and neuropsychological perspectives

2 CURRAN TANAKA AND WEISKOPF

and the exemplars held in memory The same exemplar-based system underlies both recognition and classificationbut different decision rules are applied to each task (Nosof-sky 1988 1991) Categorization depends on comparing atest item with stored exemplars from various categoriesand classifying the item as belonging to the category withwhich it is most similar For example a novel animalwould be categorized as a dog if its similarity to stored dogexemplars is higher than its similarity to stored exemplarsfrom other categories (eg cats horses trees furniturecars etc) Recognition depends on comparing the testitem with all stored exemplars so that recognition proba-bility increases with the overall similarity of the test itemto all stored exemplars across all categories Thus a par-ticular dog would be recognized as having been previouslyencountered if its similarity to all stored exemplars is highenough to pass a personrsquos criterion for responding oldrather than new

Single-system exemplar models can account for aspectsof the neuropsychological results originally interpretedwithin the multiple-systems perspective (for a summaryand continuation of this debate see Knowlton 1999 Nosof-sky amp Zaki 1999) Nosofsky and Zaki (1998) have re-cently shown that an exemplar-based model can success-fully simulate the pattern of spared categorization andimpaired recognition exhibited by amnesic patients (Knowl-ton amp Squire 1993) Amnesic performance was simulatedby varying a sensitivity parameter that controls discrimi-nation among exemplars As has been observed for am-nesia decreased sensitivity led to poorer recognitionmemory without significantly impairing categorizationSimilarly amnesicsrsquo pattern of spared artificial grammarlearning and impaired recognition memory has been sim-ulated with a single-system connectionist network (Kinderamp Shanks 2001) Thus multiple systems may not be nec-essary to account for neuropsychological dissociations be-tween categorization and recognition On the other handthe multiple-systems perspective seems more consistentwith functional neuroimaging results showing that visualcategorization is associated with activity in visual corticalregions not typically related to declarative memory (Reberet al 1998a 1998b)

Research using event-related brain potentials (ERPs)can provide information that is relevant to understandingthe relationship between categorization and recognitionScalp-recorded ERPs are obtained by averaging elec-troencephalogram (EEG) activity across multiple trials thatare designed to engage specific cognitive processes Bytime-locking the average to stimulus onset ERPs reflectthe activity of brain processes that are regularly associatedwith stimulus processing under the experimental condi-tions of interest ERPs can be differentiated on the basis oftheir timing (with millisecond resolution) and scalp dis-tribution so different neurocognitive processes can beidentified with distinct spatiotemporal voltage patternsPrevious studies have examined ERPs in categorizationand recognition tasks so we will review the main findingsrelevant to each

Several studies have indicated that ERP effects relatedto the categorization of visual objects are observed within200 msec of stimulus onset The N1 is a negative polarityERP component peaking between 150 and 200 msec overposterior head locations that is thought to reflect visualdiscrimination processes (eg Vogel amp Luck 2000) TanakaLuu Weisbrod and Kiefer (1999) used a category verifi-cation task in which a category label referring to superor-dinate (eg ANIMAL) basic (eg BIRD) or subordinate(eg ROBIN) levels was followed by a picture and subjectsresponded true or false according to the correspondencebetween the label and the picture The N1 was more neg-ative for subordinate than for basic level classificationTanaka et al interpreted the N1 enhancement as reflectingincreased visual analysis associated with subordinate clas-sification The N1 has been observed to be more negativefor pictures of objects from natural than from artifactualcategories (Kiefer 2001) Assuming that perceptual in-formation is more important for categorizing natural (egdifferent birds or dogs look somewhat similar) than arti-factual (eg different tools or pieces of furniture look dis-similar) object categories Kiefer also attributed the ERPeffect to perceptual-processing differences Another re-cent study has shown that the N1 is enhanced when ex-perts categorize objects within their domain of expertise(Tanaka amp Curran 2001) Tanaka and Curran labeled the expertise-sensitive ERP component N170 because of its similarity to the N1 component that numerous stud-ies have found to be greater in response to faces than to other objects (eg Bentin Allison Puce Perez amp Mc-Carthy 1996 Bentin amp Deouell 2000 Boumltzel Schulze ampStodieck 1995 Eimer 1998 2000a 2000b GeorgeEvans Fiori Davidoff amp Renault 1996 Rossion Cam-panella et al 1999)

In contrast to the early occurring ERP effects related tocategorization numerous studies have suggested that ERPeffects relevant to recognition memory do not begin until300ndash400 msec after stimulus onset In a typical recogni-tion memory experiment subjects study a list of stimuli(words pictures etc) followed by a test list containing amixture of old (studied) and new (nonstudied) stimuliWhen ERPs are recorded during the test list electrophys-iological differences between correctly classified old andnew conditions emerge after about 300 msec The numberof component processes underlying such ERP oldnew ef-fects and their functional significance remain uncertainTwo recent reviews have suggested that ERP oldnew ef-fects can be broken into at least three different subcom-ponents (Friedman amp Johnson 2000 Mecklinger 2000)and this division corresponds well to results from our ownlaboratory (Curran 1999 2000 Curran Schacter John-son amp Spinks 2001) FN400 effects parietal effects andlate frontal effects The present study is most relevant tothe FN400 and parietal oldnew effects so they will bediscussed in more detail

The FN400 oldnew effect involves an N400-like com-ponent that is more negative for new than for old stimuliover frontal electrode sites between 300 and 500 msec

ERPS OF CATEGORIZATION AND RECOGNITION 3

(called the ldquofrontal oldnew effectrdquo by Mecklinger 2000and the ldquoleft medial prefrontal episodic memory effectrdquoby Friedman amp Johnson 2000) The standard N400 com-ponent is best studied in psycholinguistic research sug-gesting that it indexes semantic integration (Kutas ampIragui 1998 Kutas amp Van Petten 1994) The FN400nomenclature is used here to denote the observation thatthis oldnew effect is often more frontally distributed thanis the centro-parietal N400 typically observed in studies oflanguage although many recognition studies of recogni-tion memory and word repetition have observed centro-parietal N400 oldnew effects (eg Besson Kutas amp VanPetten 1992 Rugg amp Nagy 1989 Smith amp Halgren1989 Van Petten Kutas Kluender Mitchiner amp McIsaac1991) It is unclear whether the N400 and the FN400 aredistinct components but given their spatiotemporal simi-larity it is likely that they share at least some neural gen-erators1 The parietal oldnew effect is associated withgreater positivity over parietal regions to old than to newitems between about 400 and 800 msec (reviewed by AllanWilding amp Rugg 1998 Friedman amp Johnson 2000Mecklinger 2000) The parietal oldnew effect encom-passes the P300 ERP component (Spencer Vila Abad ampDonchin 2000)

Several ERP investigators (Curran 2000 Friedman ampJohnson 2000 Mecklinger 2000 Rugg Mark et al 1998)have suggested that the FN400 and parietal oldnew effectsare related to the separate processes of familiarity and rec-ollection posited within dual-process theories of recogni-tion memory (Brainerd Reyna amp Kneer 1995 Hintzmanamp Curran 1994 Jacoby 1991 Mandler 1980 Normanamp OrsquoReilly 2001) Familiarity is often thought to arise froman assessment of the total similarity between a test itemand all study-list information in memory This conceptu-alization of familiarity is consistent with the similarity-based matching processes in exemplar models of catego-rization and recognition (Estes 1994 Hintzman 19861988 Medin amp Schaffer 1978 Nosofsky 1988 1991)Recollection is a process that enables the retrieval of spe-cific information about individual items such as physicalattributes or associativecontextualsource information Inone relevant ERP study (Curran 2000) subjects studiedsingular and plural nouns (eg TABLES CUP ) followedby a test list with studied words (TABLES) similar lures thatwere in the opposite plurality of studied words (CUPS) andcompletely new words (BOOK) As would be expected of afamiliarity process that is merely sensitive to studyndashtestsimilarity the FN400 differed between similar (CUPS) andnew (BOOK) words but not between studied words (TABLES)and similar lures (CUPS) that were both highly familiar Aswould be expected of a recollection process that enablesthe retrieval of detailed information the parietal ERP ef-fect differed between studied (TABLES) and similar (CUPS)items

The hypothesized correspondence between the FN400and familiarity is relatively new (reviewed by Friedman ampJohnson 2000 Mecklinger 2000) but the relationshipbetween recollection and the parietal ERP oldnew effect

is more widely accepted When subjects are asked to in-trospectively differentiate words specifically rememberedfrom those they merely know to be old larger parietaloldnew effects are associated with remembering thanwith knowing (Duumlzel Yonelinas Mangun Heinze amp Tul-ving 1997 Rugg Schloerscheidt amp Mark 1998 Smith1993 but see Spencer et al 2000) The parietal oldneweffect is sensitive to variables thought to affect recollec-tion more than familiarity (Paller amp Kutas 1992 PallerKutas amp McIsaac 1995 Rugg Cox Doyle amp Wells1995) The parietal oldnew effect also is associated withthe recollection of specific information such as studymodality (Wilding Doyle amp Rugg 1995 Wilding ampRugg 1997b) speakerrsquos voice (Rugg Schloerscheidt ampMark 1998 Wilding amp Rugg 1996 1997a) and tempo-ral source (Trott Friedman Ritter amp Fabiani 1997)

In summary previous ERP research has identified at leastthree distinct ERP effects associated with categorizationor recognition Activity related to visual categorizationhas been associated with the N1 component (150ndash200 msec) that peaks over posterior visual brain areasRecognition memory has been related to the fronto-central FN400 (300ndash500 msec) associated with familiar-ity as well as with the parietal oldnew effect (400ndash800 msec) associated with recollection One might inter-pret these results as consistent with a multiple-systemsview because categorization and recognition are associ-ated with different spatiotemporal patterns of brain activ-ity but such a conclusion would be premature because itis overly dependent on interexperiment comparisons

The primary goal of the present experiment was to di-rectly compare ERPs related to categorization and recog-nition memory in a single experiment Subjects weretrained to recognize categories (or families) of visually sim-ilar novel objects called blobs Blobs rather than stimulisuch as dot patterns were used to examine conditionsmorelikely to be relevant to the categorization of real-world ob-jects After training with a given family of blobs both recog-nition and categorization tests were given under identicalconditions Test lists included old blobs from the traininglist that were in the trained family (oldin) old blobs fromthe training list that were out of the family (oldout) newblobs that were not training-list exemplars but were in thefamily (newin) and new blobs that were out of the fam-ily (newout) Subjects made inout judgments for catego-rization tests and oldnew judgments for recognition tests

Predictions were derived from the ERP research re-viewed above Given previous evidence that the N1 is en-hanced by experience with certain object categories (Tanakaamp Curran 2001) we predicted that the N1 would be en-hanced for blobs that were in trained families relative tothose that were out of trained families We did not expectoldnew effects on the N1 because oldnew effects are nottypically observed so early in recognition memory studiesGiven previous evidence that the parietal oldnew effect isrelated to the recollection of specific information we pre-dicted that it would show standard oldnew effects but nofamily membership effects The FN400 was predicted to


show inout effects because previous research has sug-gested that the FN400 varies with the similarity betweenstudied and tested items (Curran 2000 Nessler Meck-linger amp Penney 2001) and similarity should be higherfor blobs from within the trained family The FN400 oldnew predictions were less certain Curran (2000) showed thatthe FN400 does not discriminate between studied itemsand highly similar lures (eg opposite plurality words)The blob lures are highly similar in the present experimentbut possibly not as similar as plurality-reversed lures Fur-thermore the small number of trained items (eight perfamily) and larger number of training repetitions (10 peritem) may foster more highly differentiated encoding ofthe items (eg Shiffrin amp Steyvers 1997) Finally we hadno basis for predicting that the ERP old new or inout ef-fects would vary between recognition and categorizationtasks The FN400 and parietal oldnew effects for wordsdid not differ when a recognition test was compared witha lexical decision task (Curran 1999) so we did not ex-pect task differences on these oldnew effects in the pres-ent experiment

EXPERIMENT 1

MethodSubjects

The subjects were 37 right-handed students from Case WesternReserve University who participated for pay ($10 per hour)Twenty-four subjects were included in the final analyses 2 Each ofthe final 24 subjects completed two 2-h sessions on separate days

StimuliThe stimuli were computer-generated two-dimensional polygons

called blobs (see Figure 1) Twelve blob families were constructedby creating a prototype and making family members that were dis-tortions of the prototype Prototypes were created by dividing a cir-cle into a specified number of vertices (range 5 10ndash28 M 5 1683vertices per prototype) The possible distance of each vertex fromthe origin could fall within a specified range of the original circlersquosradius (range 5 30 ndash70 M 5 5333 ) An actual radius was

randomly selected within the specified range of each vertex Oncethe distances of the vertices were specif ied they were intercon-nected to form a closed polygon Exemplars (ie family members)were created around each prototype by randomly changing the ra-dius of the vertices 620 from the prototype Sixteen exemplarswere created around each prototype Four exemplars were inolditems that appeared in the training list and both test lists The other12 blobs were innew items that appeared only on the test lists (6 pertest list) Blobs from out of the trained families were constructed bycreating new prototypes that were matched to each in blob accord-ing to the number and distance of the vertices Each out blob was itsown prototype so out blobs did not form distinctive families (Fig-ure 1 bottom row) Each blob family and its matched out blobs had aunique color All blobs fit within a circle approximately 45 cm in di-ameter subtending a visual angle of approximately 322ordm whenviewed from the subjectrsquos distance of 80 cm Stimuli were presentedon a 15-in Apple Multiscan Color Monitor

Design and ProcedureAll variables were manipulated within subjects in a task (catego-

rization recognition) 3 family membership (in out) 3 oldnew de-sign Old blobs were presented on the training list and new blobswere not In blobs were members of the trained family and out blobswere not family members The subjects completed 12 experimentalblocks (6 per session) A practice training set was given at the be-ginning of the first session Sessions were run on separate days (M 5138 range 5 0ndash6 days intervening between sessions) Each blockincluded training and test phases for a different set of blobs Duringeach block the subjects were trained to categorize members of oneparticular family of blobs After training the subjects completed a cat-egorization test and a recognition test EEG was recorded during thetest lists Blob color was varied between blocks to minimize in-terblock interference but color was held constant within each blockso it could not be used as a basis for discrimination

Training began by showing the subject the prototype of one blobfamily for 10 sec Next the subjects were given a training list inwhich four exemplars from in the family were intermixed with fourexemplars from out of the family Each blob was repeated 10 timeswithin the training phase (80 total trials) Repetitions were spaced sothat each blob was presented once within Trials 1ndash8 9ndash16 and soforth Blobs were randomly ordered within each set of 8 trials

Each training trial started with a central f ixation cross for200 msec followed by the blob for 2000 msec followed by the in-struction to ldquoplease respondrdquo for 1000 msec The intertrial interval

Prototype

In

Out

Figure 1 Examples of blob stimuli The prototype for a single family is shownalong with training set exemplars that were in or out of the family


was 500 msec The subjects were instructed to press a key for anyblobs they considered to be in the family and to not respond to blobsthat were out of the family Responses were recorded only during the1-sec ldquoplease respondrdquo interval A computer beep sounded aftereach error (false alarms to out blobs or misses to in blobs) The sub-jects were given a self-paced rest break between Trials 40 and 41

After each training list the subjects completed two test listsTraining and test lists were separated by a period of at least 2 min forSensor Net adjustment Each test list contained the following num-ber of blobs in each condition four inold six innew four outoldsix outnew Pilot testing indicated that false alarm rates were highin the recognition task because of lure similarity so the design includedmore new than old items to obtain a sufficient number of accuratetrials in all conditions Each trial consisted of a fixation circle (ran-domly varying from 500 to 1000 msec) the test blob (2000 msec)and a question mark that was displayed until the subject respondedThe subjects responded with the first two fingers of their right handsand were told to withhold their responses until the question mark ap-peared The delayed response procedure was used to minimize response-related ERP differences between the two test tasks Uponresponding the subjects saw a central square during the 1000-msecintertrial interval

For recognition tests the subjects pressed an old key for any blobsthey remembered seeing in the training set or a new key for any blobsnot in the training set They were told to disregard family member-ship and were warned that the new blobs would be highly similar tothose in the training set For categorization tests the subjects pressedan in key for any blobs that were members of the trained family andan out key for blobs that were not family members They were toldto disregard oldnew status The order of the test lists was counter-balanced so that categorization occurred on the even-numberedblocks for half the subjects and on the odd-numbered blocks for theother half The same old blobs were tested in both the recognitionand the categorization tests but new blobs were different in each testThe subjects responded with the first two fingers of the right handsand assignment of keys to responses was counterbalanced acrosssubjects

EEGERP MethodsScalp voltages were collected with a 128-channel Geodesic Sensor

Net (Tucker 1993) connected to an AC-coupled 128-channel highinput impedance amplifier (200 MV Net Amps Electrical Geo-desics Eugene OR) Amplified analog voltages (01ndash100 Hz band-pass 23 dB) were digitized at 250 Hz Individual sensors were ad-justed until impedances were less than 50 kV

Trials were discarded from analyses if they contained eye move-ments (vertical EOG channel differences greater than 70 mV) or morethan five bad channels (changing more than 100 mV between sam-ples or reaching amplitudes over 200 mV) ERPs from individualchannels that were consistently bad for a given subject were replacedusing a spherical interpolation algorithm (Srinivasan Nunez Sil-berstein Tucker amp Cadusch 1996) The median number of excludedchannelssubject was 100 (M 5 142 mode 5 1 range 5 0 to 5)Subjects with fewer than 20 good trials in any condition were re-moved from the final analyses (n 5 5 as specified in note 2) Acrossall subjects and conditions the mean number of acceptable trials percondition per subject was 5196 (range 5 26ndash72)

ERPs were baseline corrected with respect to a 100-msec prestim-ulus recording interval and were digitally low-pass filtered at 40 HzAn average-refe rence transformation was used to minimize the effectsof reference site activity and accurately estimate the scalp topogra-phy of the measured electrical fields (Bertrand Perin amp Pernier1985 Curran Tucker Kutas amp Posner 1993 Dien 1998 Lehmanamp Skrandies 1985 Picton Lins amp Scherg 1995 Tucker LiottiPotts Russell amp Posner 1994) Average-referenced ERPs are com-puted for each channel as the voltage difference between that chan-nel and the average of all the channels Mastoid-referenced ERP

plots from representative locations from the International 10ndash20system (Jasper 1958) are presented in the Appendix to facilitatecomparison with results from other laboratories

ResultsBehavioral Results

The mean proportion correct from each condition isshown in Table 1 A task (recognition categorization) 3family (in out) 3 oldnew repeated measures analysis ofvariance (ANOVA) revealed several significant effectsAll main effects were significant (categorization recog-nition out in old new all ps 001) All interactionsexcept for the task 3 oldnew interaction (F 1) weresignificant culminating in a significant task 3 family 3oldnew interaction [F(123) 5 9065 MSe 5 0004 p 001] Both recognition and categorization accuracy werelowest for new blobs that were in trained families For blobsthat were out of the trained families performance wassimilar in all conditions except for the recognition of oldblobs being lower than the others

Reaction time (RT) measures were likely to be insensitivebecause the subjects withheld their responses until the testblob disappeared (2 sec after blob onset) Although failureto observe RT differences between conditions may be at-tributable to delayed responding significant RT differ-ences may be meaningful An ANOVA showed significantmain effects of family [F(123) 5 442 MSe 5 4047 p 05] and oldnew [F(123) 5 765 MSe 5 2231 p 05]The subjects responded faster to blobs that were in thanout of trained families and faster to old than to new blobsA significant task 3 family interaction [F(123) 5 1198MSe 5 2439 p 01] indicated that family RT differ-ences were observed only within the categorization taskThus despite the requirement to withhold responses RTwas meaningfully affected by the experimental condi-tions

ERP ResultsN1 results N1 analyses were conducted on ERPs aver-

aged across all artifact-free trials Both correct and incorrecttrials were included because large accuracy differencesbetween recognition and categorization could obscuretask-related differences on N1 amplitude

N1 analyses focused on scalp regions where the N1 am-plitude was maximal (most negative) The N1 peaked

Table 1Mean Accuracy and Reaction Time (RT in Milliseconds)

in Experiment 1

In Out

Old New Old New

Categorizationp(correct) 93 77 92 90RT 2369 2393 2427 2423

Recognitionp(correct) 88 52 77 91RT 2403 2434 2401 2425


within a left-hemisphere region including channels T557 63 64 65 69 and 70 and a right-hemisphere regionincluding channels T6 90 91 95 96 100 and 101 (see theposterior inferior regions in Figure 2) Mean ERP ampli-tude will be the primary dependent measure in all analy-ses but initial latency analyses were conducted for tworeasons (1) to assess the extent to which amplitude differ-ences between conditions may be compromised by latencydifferences and (2) to estimate the temporal window inwhich amplitude analyses should be conducted Peak la-tencies were identified as the time at which the N1 was mostnegative for each subject within each condition (from 100to 236 msec after stimulus onset) The peak latencies wereentered into a task (recognition categorization) 3 family(in out) 3 oldnew 3 hemisphere repeated measuresANOVA The only significant effect was a family 3oldnew interaction [F(123) 5 1255 MSe 5 5695 p 01] N1 latency for old items was shorter for blobs thatwere in (184 msec) than for those that were out (188 msec)of the trained families but N1 latency for new items didnot vary with family membership (in 5 187 msec out 5186 msec) Although reliable these differences appear toosmall (ie 4 msec 5 1 EEG sample) to compromise theamplitude analyses

Amplitude analyses were conducted within a 156ndash220 msec window that was defined around the overall

mean N1 latency of 186 msec (rounded to the nearest 4-msec sample) The upper and lower bounds were selectedto be two standard deviations (one SD 5 16 msec) aboveand below the mean The mean amplitudes within the tem-poral interval were entered into a task (recognition cate-gorization)3 family (in out) 3 oldnew3 hemisphere re-peated measures ANOVA The only significant effect wasthe main effect of family membership [F(123) 5 1641MSe 5 070 p 001] As is shown in Figure 3 the N1was enhanced (more negative) for family members (meanamplitude in 5 2217 mV out 5 2182) No other effectsor interactions were significant (all ps 16)

One particularly important result for understanding thenature of the family membership effect concerns whetherit generalized to new blobs that had never before been seenin the experiment A planned contrast focusing only onnew blobs verified that N1 amplitude was more negativefor blobs that were in (2213 mV) than for those that wereout of (2176 mV) the family [t (23) 5 396 SE 5 010p 001] A planned contrast focusing only on new blobsverified that N1 amplitude was more negative for blobsthat were in (2213 mV) than for those that were out of(2176 mV) the family [F(123) 5 689 MSe 5 092 p 01] Likewise the N1 was more negative for old blobsthat were in (2220 mV) than for those that were out of(2189mV) the family [F(123)5 466MSe 5 092 p 05]

Figure 2 Approximate channel locations on the Geodesic Sensor Net Locations from the International 10ndash20 system are shown for reference Theeight clusters of black channels depict the locations used for analyses (rightleft 3 anterior posterior 3 inferiorsuperior)


The relative magnitudes of the in out and old new dif-ferences were compared by computing the relevant differ-ences scores for each subject (averaged across hemi-spheres) The inout differences were significantly largerthan the oldnew differences [t(22) 5 215 SE 5 011p 05 two tailed]

FN400 results The FN400 (300ndash500 msec) is typi-cally observed as more negative amplitudes recorded tonew than to old items over superior frontal sites (Curran1999 2000 Friedman amp Johnson 2000 Guillem Bicu ampDebruille 2001 Mecklinger 2000 Rugg Mark et al1998) With the exception of studies from our own labo-ratory most of these studies have referenced their record-ings to the average of the two mastoid recording sites(channels 57 and 101 in Figure 2) In our own laboratoryusing the average-referenced technique we have foundthat the FN400 is associated with posterior inferior (PI)differences (new old) that have the opposite polarity tothe anterior superior (AS) differences (new old) that arenormally observed Thus the present FN400 analyses fol-low our previous methods of capturing oldnew differencesover both of these regions (Curran 1999 2000)

Curran (1999 2000) has observed the FN400 oldneweffect over AS and PI scalp regions between 300 and500 msec Mean amplitudes are more negative for newthan for old items over AS regions but new items are morepositive than old items over PI regions Thus with thepresently used ERP recording and measurement techniques

the FN400 can be quantified by a crossover interaction be-tween conditions (old new) and regions (AS PI) Meanamplitudes (300ndash500 msec) were entered into a task (recog-nition categorization)3 family (in out) 3 old new 3 re-gion (AS PI) 3 hemisphere repeated measures ANOVAThe old new 3 region interaction [F(123) 5 427 MSe 5227 p 5 05] captured differences consistent with typi-cal FN400 oldnew effects (see Figure 4) AS voltageswere more negative for new (2053 mV) than for old(2029 mV) blobs but PI voltages were more negative forold (2084 mV) then for new (2062 mV) blobs The fam-ily 3 region interaction was significant [F(123) 5 980MSe 5 161 p 01] AS voltages were more negative forblobs out of (2057 mV) than in (2026 mV) the trainedfamilies but opposite-going differences were observedover PI regions (out 5 2060 mV in 5 2086 mV) Thefour-way task 3 family 3 oldnew3 region interaction wasdifficult to interpret [F(123) 5 440MSe 5 102 p 5 05]To better understand this interaction separate ANOVAswere run for each task separately but they failed to revealany differences in the pattern of effects shown within eachtask The family 3 region interaction was significantwhen each task was considered separately but no other ef-fects were significant for either task alone In summarythe FN400 was affected by both family membership andoldnew differences

Parietal results The spatial distribution of the parietaloldnew effect is somewhat different with average-referenced

Figure 3 Mean average-referenced ERPs averaged within channels from the left and right posterior inferior regions in Experiment 1(see Figure 2 for locations) All artifact-free trials are included and averaged across tasks (categorization recognition) and old new


ERPs than is typically observed with mastoid-referencedERPs (as has already been explained with regard to theFN400) Relative to a mastoid reference parietal ERPs(posterior superior [PS] scalp regions) are more positivefor old than for new conditions between about 400 and800 msec (Friedman amp Johnson 2000 Mecklinger 2000Rugg 1995) The average-reference captures these PS dif-ferences (old new) as well as opposite polarity differ-ences (old new) over anterior inferior (AI) regions(Curran 1999 2000 Curran et al 2001) Thus condition(old new) 3 region (PS AI) interactions are indicative ofthe parietal oldnew effect

Mean amplitudes (400ndash800 msec) were entered into atask (recognition categorization)3 family (in out) 3 old

new 3 region (PS AI) 3 hemisphere repeated measuresANOVA The oldnew 3 region interaction was signifi-cant [F(123) 5 720 MSe 5 084 p 05] PS voltageswere more positive for old (204 mV) than for new(189 mV) blobs but AI voltages were more negative forold (2132 mV) than for new (2112 mV) blobs The fam-ily 3 region interaction approached significance [F(123) 5395 MSe 5 139 p 05] Interestingly this trend to-ward a parietal inout (out in) effect was of opposite po-larity as compared with the parietal oldnew (old new)effect That is the parietal effect was larger for the morefamiliar old items within the oldnew comparison but ittended to be larger for the less familiar outsiders in theinout comparison

Although the parietal oldnew effect was statisticallysignificant it was somewhat weaker than is typically ob-served However the parietal oldnew effect usually is ob-served with ERPs including only correct trials (hits vscorrect rejections) Therefore ERPs were recomputed toinclude only correct trials Across all subjects and condi-tions the mean number of accurate and artifact-free trialsper condition per subject was 4288 (range 5 20ndash69)Again the inout 3 region interaction was not significant[F(123) 5 333 MSe 5 184 p 05] The task 3 oldnew 3 region interaction was significant [F(123) 5452 MSe 5 127 p 05] Follow-up ANOVAs indicatedthat the oldnew 3 region interaction was significant forthe recognition task [F(123) 5 766 MSe 5 172 p 05] but not for the categorization task (F 1 MSe 5098) As was expected differences between old and newconditions were larger when only accurate trials on therecognition task were considered (see Figure 5 PS old[216 mV] new [191 mV] AI old [2152 mV] new[2103 mV]) These differences are still somewhat smallbut this is understandable given the difficulty the subjectsexperienced discriminating old blobs (hit rate 5 88)from highly similar new blobs (false alarm rate 5 48)

Topographic comparisons In summary of the previousanalyses ERPs were modulated by both family member-ship and oldnew differences Whereas inout categorizationexerted significant effects on the N1 and FN400 compo-nents oldnew recognition influenced the magnitude of theFN400 and parietal effects Topographic analyses weredone to better understand the relationship between the fourprimary effects N1 inout FN400 inout FN400 oldnewand parietal oldnew Might they reflect the activity of asingle process (or system) that initially responds to cate-gorical differences and then later discerns differences be-tween studied and nonstudied exemplars Might each ofthese effects map onto different neurocognitive processesrelated to categorization or recognition Although pin-pointing the anatomical location of the neural generatorsof scalp-recorded ERPs is difficult different topographicpatterns indicate that ldquothe underlying combination of ac-tivities at the brain sources must also be differentrdquo (Pictonet al 2000 p 147) Such topographic differences couldreflect either different underlying sources or commonsources activated with different relative strengths (Alain

Figure 4 Mean average-reference d ERPs averaged withinchannels from the anterior superior and posterior inferior regions in Experiment 1 (see Figure 2 for locations) All artifact-free trials are included and averaged across tasks (categorizationrecognition) and hemispheres


Achim amp Woods 1999) Thus topographic differencesbetween conditions would be consistent with separate un-derlying processes although they could also reflect dif-ferent patterns of activation across the same processes

The four primary experimental effects were comparedtopographically Figure 6 (left panel) shows topographicmaps of the inout (left) and oldnew (right) differenceswithin each temporal window Mean differences associatedwith each effect were computed within each of the regionsdepicted in Figure 2 The differences were normalizedprior to analysis so that differences in the overall magni-tude of the effects would not bias the topographic com-parisons (following the vector length method of McCarthyamp Wood 1985)

Four specific questions were addressed through pair-wise comparisons of the normalized differences (1) Is thetopography of the N1 inout effect different from thetopography of the FN400 inout effect (2) Are the FN400

inout and oldnew effects associated with distinct topo-graphic patterns (3) Is the topography of the FN400oldnew effect different from the topography of the pari-etal oldnew effect (4) Is the topography of the N1 inouteffect different from the topography of the parietal oldnew effect These questions are addressed in turn

Is the topography of the N1 inout effect different fromthe topography of the FN400 inout effect (Figure 6 1Avs 1C) Normalized N1 (156ndash220 msec) and FN400(300ndash500 msec) inout differences were entered into a dif-ference (N1 FN400) 3 hemisphere 3 anteriorposterior3 inferiorsuperior ANOVA The difference 3 hemisphere3 inferiorsuperior interaction was significant [F(123) 5453 MSe 5 002 p 05] The interaction captured thefact that the N1 and the FN400 inout differences had sim-ilar relative magnitudes over superior scalp regions butover inferior regions the N1 difference was larger over theright hemisphere (see Figure 3) and the FN400 differencewas larger over the left hemisphere Although topographicdifferences between the N1 and the FN400 inout effectswere significant visual inspection of Figure 6 revealsbroadly similar topographic patterns3

Are the FN400 inout and oldnew effects associatedwith distinct topographic patterns (Figure 6 1C vs 1D)Normalized inout and oldnew mean differences werecalculated within the temporal window of the FN400(300ndash500 msec) and were entered into a difference (inoutoldnew) 3 hemisphere 3 anteriorposterior 3 inferiorsuperior ANOVA No effects were significant so thisanalysis provided no indication of topographic differencesbetween the FN400 inout and the FN400 oldnew effects

Is the topography of the FN400 oldnew effect differentfrom the topography of the parietal oldnew effect (Fig-ure 6 1D vs 1F) Normalized FN400 (300ndash500 msec)and parietal (400ndash800 msec) oldnew differences were en-tered into a difference (FN400 parietal) 3 hemisphere 3anteriorposterior 3 inferiorsuperior ANOVA The four-way difference 3 hemisphere 3 anteriorposterior 3 inferiorsuperior interaction was significant [F(123) 5488 MSe 5 001 p 05] The primary difference be-tween these temporal intervals is the presence of left-PS(old new) and right-AI (old new) differences associ-ated with the parietal but not with the FN400 oldnew ef-fect Overall the 400ndash800 msec topography was somewhatdifferent than the typical parietal oldnew effect In partic-ular it was dominated by a medial frontal old new dif-ference that originated around 200 msec and lasted untilapproximately 1400 msec after stimulus onset It appearsto overlap with the AS aspect of the FN400 but is closerto the frontal poles than is the typical FN400 Similarlydistributed long-duration frontal oldnew effects havebeen previously described in studies of recognition mem-ory but are poorly understood (reviewed by Friedman ampJohnson 2000) In the present experiment this frontaloldnew effect clearly increased the similarity of the 300ndash500 and 400ndash800 msec topographies Despite these visualsimilarities the oldnew effect was associated with statis-

Figure 5 Mean average-reference d ERPs averaged withinchannels from the anterior inferior and posterior superior re-gions in Experiment 1 (see Figure 2 for locations) ERPs havebeen averaged across right and left hemispheres Only correct tri-als within the recognition task are included


tically different topographies during the FN400 effect(300ndash500 msec) and the parietal effect (400ndash800 msec)intervals (replicating Curran 1999 2000)

Is the topography of the N1 inout effect different fromthe topography of the parietal oldnew effect (Figure 6 1Avs1F) Normalized N1 inout (156ndash220 msec) and pari-etal oldnew (400ndash800 msec) differences were entered intoa difference (FN400 parietal) 3 hemisphere 3 anteriorposterior 3 inferiorsuperior ANOVA The difference 3anteriorposterior [F(123) 5 596 MSe 5 025 p 05]and four-way difference 3 hemisphere3 anteriorposterior3 inferiorsuperior [F(123) 5 693 MSe 5 004 p 01]interactions were significant These topographic differ-ences can be seen by comparing 1A and 1F in Figure 6

EXPERIMENT 2

Experiment 1 indicated that the N1 and the FN400 weresensitive to categorical differences among the stimuli (inouteffects) and that the FN400 and parietal effects were sensi-tive to recognition-related differences (oldnew effects)Thus we have some evidence for electrophysiological dif-ferences between category-related and recognition-relatedprocesses Oldnew differences can be unambiguously at-

tributed to memory for the training set because appearanceon the training list is the only factor that differentiates theseitems It is conceivable however that categorical differencesare more stimulus related and not necessarily dependent onmemory for the training experience By definition the fam-ily members are structurally more similar to one anotherthan are the outsiders Thus inout differences could be aproduct of similarity effects within the test lists themselvesFor example short-term priming effects may occur whentwo family members are presented in succession ERP dif-ferences between recognition and categorization would beconsiderably less interesting if only recognition effects werethe product of memory for the training set Relevant behav-ioral evidence has shown that prior training with categoryexemplars is not necessary for above-chance categoriza-tion of novel dot patterns (Palmeri amp Flanery 1999) In factPalmeri and Flanery showed that subjects who were testedin the absence of category training performed much likeamnesic patients in showing accurate categorization perfor-mance but chance recognition

Experiment 2 tested the possibility that effects observedin Experiment 1 did not depend on category training byomitting the training task but otherwise replicating themethod in Experiment 1

Figure 6 Topographic distributions of the ERP inndashout and oldndashnew differences within the three temporal windowsanalyzed in Experiments 1 (left panel) and 2 (right panel) All artifact-free trials are included and averaged across tasks

Experiment 1InndashOut OldndashNew

Experiment 2Inndash Out OldndashNew

N1Effects

(156ndash220 msec)

ParietalEffects

(400ndash800 msec)

FN400Effects

(300ndash500 msec)


MethodSubjects

The subjects were 28 right-handed students from the Universityof Colorado at Boulder who participated for introductory psychol-ogy credit Twenty-four subjects were included in the final analyses4

Stimuli Design and ProcedureApart from the following exceptions Experiment 2 replicated the

method in Experiment 1 in all major respects except that the train-ing phase was omitted from Experiment 2

Experiment 2 was run in one session instead of two because omit-ting the training phase reduced the overall length of the experimentconsiderably As in Experiment 1 two test lists were given for eachfamily of blobs Unlike Experiment 1 in which one test list was cat-egorization and one was recognition categorization was tested inboth lists in Experiment 2 Recognition testing would make no sensein the absence of a studytraining list The subjects were instructedthat they needed to try to discover the structure of the family duringthe test blocks No feedback was provided

ResultsGiven the hypothesis that the ERP inout and oldnew dif-

ferences observed in Experiment1 were attributable to train-ing we predicted no such differences in Experiment 2Because the training set was omitted from Experiment 2the oldnew variable is not truly meaningful in the presentexperiment However it was retained in all analyses to as-sess the possibility that the Experiment 1 oldnew differ-ences were attributable to stimulus confounds because oldand new blobs were not counterbalanced in Experiment 1Thus old blobs are those that would have been presentedin training if the training lists had actually been presentedOne true difference between old and new blobs in Exper-iment 2 however is that old blobs were presented in eachof the two test lists for each family but new blobs werepresented only in one test list

Behavioral ResultsMean accuracy and RT is shown in Table 2 Categoriza-

tion accuracy was above chance in all conditions Thus thesubjects were able to learn to differentiate between familymembers and outsiders without training and within theconfines of the test lists Accuracy was considerably lowerin Experiment 2 than in Experiment 1 with the exceptionof new family members that the subjects categorized atsimilar levels in each experiment In Experiment 2 accu-racy was uniformly low in all conditions but Experi-ment 1 accuracy fell in the innew condition Although thesubjects in Experiment 1 learned the blob families morethoroughly because of training they were more likely to

be duped by the within-experiment novelty of the newblobs This would not be a factor in Experiment 2 inwhich old and new blobs did not truly differ (because oldblobs were not seen in a previous training phase)

The subjects were consistently slower in Experiment 2than in Experiment 1 This is consistent with the fact thatthe subjects had to learn the category structure within thetest lists in Experiment 2

ERP ResultsAs was predicted none of the inout or oldnew effects

observed in Experiment 1 were replicated when trainingwas omitted from Experiment 2 Relevant analyses arepresented below Only topographic differences are shown(Figure 6 right panel) because they can be easily com-pared against the corresponding differences from Experi-ment 1 (Figure 6 left panel)

N1 results N1 analyses again focused on the PI re-gions shown in Figure 2 Mean amplitudes from 156 to220 msec were entered into a family (in out) 3 oldnew3 hemisphere ANOVA A similar ANOVA revealed sig-nificant inout differences in Experiment 1 In Experi-ment 2 N1 amplitude did not significantly differ betweenthe in (M 5 2122 mV) and the out (2112 mV) condi-tions [F(124) 5 133 MSe 5 036] Differences betweenold (2114 mV) and new (2120 mV) conditions werealso nonsignificant [F(124) 5 44 MSe 5 034] Like-wise no interactions involving the oldnew or inout con-ditions were significant

FN400 results Mean amplitudes (300ndash500 msec) wereentered into a family (in out) 3 oldnew 3 region (ASPI see Figure 2) 3 hemisphere repeated measuresANOVA A condition 3 region (AS PI) interaction in Ex-periment 1 indicated that the FN400 was significantly in-fluenced by both inout and oldnew differences Neithereffect approached significance in Experiment 2 [oldnew3 ASPI F(124) 5 008 MSe 5 104 inout 3 ASPIF(124) 5 097 MSe 5 123] No other effects involvingthe oldnew or the inout conditions were significant

Parietal results Mean amplitudes (400ndash800 msec)were entered into a family (in out) 3 oldnew 3 region(PS AI see Figure 2) 3 hemisphere repeated measuresANOVA A significant parietal oldnew effect was sub-stantiated with an oldnew 3 regions (PS AI) interactionin Experiment 1 (400ndash800 msec) The oldnew 3 PSAIinteraction did not approach significance in Experiment 2[F(124) 5 000 MSe 5 47] The inout 3 position inter-action was also nonsignificant [F(124) 5 052 MSe 5135] No other effects involving the oldnew or the inoutconditions were significant

GENERAL DISCUSSION

The purpose of the present research was to investigatethe relationship between categorization and recognitionmemory by using ERPs to specify the spatiotemporal dy-namics of the underlying brain processes ERPs related torelatively early visual processes (N1 156ndash200 msec) dif-

Table 2 Mean Accuracy and Reaction Time (RT in Milliseconds)

in Experiment 2

Categorization

In Out

Old New Old New

p(correct) 77 76 72 71RT 2694 2686 2706 2711


ferentiated between blobs that were members of trainedcategories (in) and those that were not (out) Middle latencyERPs (FN400 effects 300ndash500 msec) also were sensitiveto category membership (inout) as well as differentiatingbetween old (trained) and new exemplars Later ERPs(parietal effects 400ndash800 msec) were significantly sensi-tive only to oldnew differences These effects were ob-served after the subjects underwent categorical training(Experiment 1) but were not observed in the absence oftraining (Experiment 2) so the ERP differences can be at-tributed to memory for the training episode

The ensuing discussion interprets oldnew differencesas related to recognition and inout differences as relatedto categorization but one also might expect differencesaccording to whether subjects were actually performing arecognition or a categorization task For example a con-ceptually similar experiment compared categorization andrecognition memory of novel dot patterns with f MRI(Reber et al 1998a) Their primary finding was that ac-tivity within the posterior occipital cortex was greater forold than for new patterns in the recognition test but that itwas greater for noncategory members than for categorymembers in the categorization task Thus occipital activ-ity appeared to increase with familiarity in the recognitiontask but to decrease with familiarity in the categorizationtask Reber et al (1998a) chose to emphasize task differ-ences when interpreting these opposite polarity differ-ences but numerous stimulus-related differences couldhave influenced the comparison also (as was clearly dis-cussed by Reber et al 1998a) Task-related differenceswere minimal in the present research when stimuli wereequated across tasks

ERP oldnew and inout differences reveal the activityof brain processes that are capable of underlying category-based and recognition-based discriminations respectivelyWhether or not such differences interact with task instruc-tions may be related to the automaticcontrolled nature ofthe underlying processes The lack of task interactions forthe N1 and FN400 effects suggests that these are relativelyautomatic processes that discriminate between inout (N1and FN400) or oldnew (FN400) at similar levels regard-less of task instructions The observation that the parietaloldnew effect was enhanced when subjects were per-forming the recognition task as compared with the cate-gorization task suggests that it may reflect the activity ofa memory process that is more intentionally controlledCurran (1999) previously observed that neither the FN400nor the parietal oldnew effects were influenced by suchtask differences when words and pseudowords were giveneither recognition or lexical decision judgments Curranrsquos(1999) recognition task was considerably easier than thepresent task with similar lures so the present task depen-dency of the parietal oldnew effect may reflect the greatereffort needed to discriminate old from new blobs

N1 EffectsN1 amplitude was more negative for family members

than for outsiders This is the first demonstration to our

knowledge that the N1 is sensitive to experimentally in-duced long-term memory Previous research has shownthat the N1 is sensitive to immediate word repetition (Mc-Candliss Curran amp Posner 1993 1994 Posner amp Mc-Candliss 1999) The present research ruled out suchshort-term influences because N1 inout effects werepresent after categorical training (Experiment 1) but wereabsent without training (Experiment 2) The N1 did notdifferentiate between old and new blobs so it is unlikelyto reflect the activity of a process capable of supportingaccurate recognition judgments

The present finding that the N1 may be related to cate-gorization is consistent with previous research (Kiefer2001Tanaka amp Curran 2001 Tanaka et al 1999 ) Tanakaand Curran found that dog and bird experts exhibited anenhanced N1 when categorizing pictures of objects withintheir domain of expertise so N1 amplitude was modulatedby differential experience with particular object categoriesThe present results show that the N1 is similarly enhancedby moderate amounts of experimental training with cate-gories of visual objects (ie families of blobs) Not onlywas this enhancement observed for the specific exemplarsseen in the training phase but it also generalized to newblob exemplars that were family members The ability togeneralize from previous experience to classify new ob-jects is a fundamental property of categorization so theseresults strengthen the hypothesis that the N1 is related tovisual categorization

The N1 is considered to be related to visual identifica-tion processes (eg Vogel amp Luck 2000) The present re-sults along with results from real-world experts suggestthat the underlying identification processes can be shapedby visual experience These results are consistent withprevious f MRI research relating posterior occipital cor-tex activity to category learning (Reber et al 1998a1998b) Our results add important time course informa-tion to these results Because the N1 takes place at a rela-tively early stage of information processing it is likelythat categorical experience is influencing initial percep-tual identification processes rather than reflecting laterpossibly re-entrant or top-down visual processes

Tanaka and Curran (2001) speculated that their expertise-enhanced N1 may be related to the N170 component ob-served in studies of face recognition The N170 is morenegative when subjects view faces than when they viewother objects (eg Bentin et al 1996 Bentin amp Deouell2000 Boumltzel et al 1995 Eimer 1998 2000a 2000bGeorge et al 1996 Rossion Campanella et al 1999)The N170 is typically maximal for faces at mastoid (Ben-tin amp Deouell 2000) or T5T6 (Boumltzel et al 1995 Eimer2000b George et al 1996 Rossion Delvenne et al 1999Taylor McCarthy Saliba amp Degiovanni 1999) locationsof the International 10ndash20 System (Jasper 1958) Theselocations fall within the regions where the N1 was maxi-mal in the present experiment (left and right mastoids areChannels 57 and 101 see Figure 2) The N170 usuallypeaks around 170 msec but the present N1 peak (186 msec)is within the range of other published N170 studies with


faces (eg 189 msec George et al 1996) Thus the spa-tiotemporal distribution of the presently observed N1 issimilar to that of the N170 to faces

FN400 EffectsThe FN400 differentiated blobs on the basis of both

category membership (in vs out of trained families) andexemplar-specific memory (old vs new) It has been pre-viously argued that the FN400 oldnew effect is related tothe so-called familiarity component of dual-process theo-ries of recognition memory (Curran 2000 Friedman ampJohnson 2000 Mecklinger 2000 Rugg Mark et al1998) Although familiarity is a term with several differ-ent psychological meanings Curran (2000) specificallydefined familiarity as an assessment of the global simi-larity between studied and tested items This sense of fa-miliarity is consistent with the output of exemplar-basedmodels of categorizationrecognition (eg Estes 1994Hintzman 1986 1988 Medin amp Schaffer 1978 Nosof-sky 1991) as well as of several other models of recogni-tion memory (reviewed by Clark amp Gronlund 1996 egGillund amp Shiffrin 1984 Humphreys Bain amp Pike 1989Murdock 1982 Shiffrin amp Steyvers 1997) The sensitiv-ity of the FN400 to both inout and oldnew differencesprovides converging evidence that a familiarity-likeprocessconsistent with these models may exist within the brain

The FN400 results suggest the existence of a singleprocess contributing to both categorization and recogni-tion memory and this possibility is supported by the be-havioral interactions that were observed Categorizationwas more accurate for old and new items and recognitionmemory was more accurate for outsiders than for familymembers Thus our behavioral results suggest that processesunderlying categorization and recognition must interact atsome level to influence decision accuracy Such interac-tions may have been obscured in previous neuropsycho-logical studies that have tested categorization and recogni-tion memory under quite different conditions (see Nosofskyamp Zaki 1999) Dissociations between amnesic and con-trol subjects may be less likely if the tasks were comparedunder otherwise identical conditions as in the presentstudy

The FN400 results clearly show that a process sensitiveto categorical differences between stimuli can also differ-entiate between old and new exemplars but the N1 was sen-sitive only to category membership and not to oldnewdifferences Why would categorization and recognition bedissociated early on (N1) but associated at later stages ofprocessing One possibility is that the N1 reflects purelyvisual aspects of memory that are sufficient for catego-rizing the blobs (on the basis of visual similarity to trainedcategories) but insufficient for recognizing particular ex-emplars The FN400 on the other hand reflects a laterstate of processing that is likely to draw on different sortsof information to assess the overall similarity betweentraining and test items Assuming that the FN400 is re-lated to the N400 that is often observed in ERP studies oflanguage comprehension (Kutas amp Iragui 1998 Kutas amp

Van Petten 1994) it would be sensitive to semantic infor-mation (eg Olichney et al 2000) Other recognitionmemory research has shown the FN400 to be related tothe semantic similarity between old and new items (Nessleret al 2001) Such semantic information would not neces-sarily be available to the visual processes underlying theN1 The blobs used in the present experiment were se-mantically impoverished but it is conceivable for exam-ple that the subjects (who saw each exemplar 10 times dur-ing training) used a naming strategy that would fostersemantic processing (eg the prototype in Figure 1 mayremind a subject of Texas)

Both the N1 and the FN400 may reflect cortical processesthat are sensitive to stimulus familiarity (ie the similar-ity between a test item and previously stored experiences)but the processes may be sensitive to different types of in-formation Does this constitute evidence for separate sys-tems The specific criteria needed for postulation of amemory system are debatable (Schacter amp Tulving 1994Sherry amp Schacter 1987) and a single experiment is un-likely to be sufficient but aspects of our results seem rel-evant We detected significant differences in the topogra-phy of the N1 and the FN400 inout effects so differentneural populations may contribute to these effects How-ever as can be seen in Figure 6 (A vs C) the similarity ofthese patterns is more conspicuous than any differencesso the present experiment does not provide particularlycompelling evidence for the anatomical separability of theunderlying processes Furthermore the ERP waveformspresented in Figure 4 suggest that the inout differencesare temporally continuous between the N1 and the FN400epochs

Even if we were to emphasize the slight topographic dif-ferences observed between N1 and FN400 inout effectstwo different systems with such properties are unlikely toaccount for previously discussed amnesic dissociationsbetween categorization and recognition Damage to eithersystem would be more likely to impair categorizationwhereas it is recognition that is selectively impaired byamnesia A recent study showed that amnesic patientscould discriminate between semantically congruous (baby-animalndashcub) and incongruous (water-sportndashkitchen) cate-gorical pairs and showed normal N400 differences be-tween congruous and incongruous conditions (eg Olich-ney et al 2000) Another recent recognition memoryexperiment has shown that the FN400 oldnew effect wasnormal in an amnesic patient with selective hippocampaldamage but the parietal oldnew effect was abolished(Duumlzel Vargha-Khadem Heinze amp Mishkin 2001) Be-cause the FN400 is spared by amnesia the underlyingprocesses are unlikely to be the source of amnesic pa-tientsrsquo difficulties with recognition memory

Parietal EffectsThe standard parietal ERP oldnew effect was replicated

Previous research has related the parietal oldnew effect tothe ability to recollect specific information from a previ-ous study episode (Allan et al 1998 Curran 2000 Fried-


man amp Johnson 2000 Mecklinger 2000) Several lines ofevidence including ERP studies with amnesic patients(eg Duumlzel et al 2001) or with intracranially recordedERPs have suggested that the hippocampus andor me-dial temporal cortex may contribute to the parietal oldnew effect (reviewed by Friedman amp Johnson 2000 Meck-linger 2000) Thus the neural mechanisms underlying theparietal oldnew effect may be the same as those damagedto cause neuropsychological dissociations between cate-gorization and recognition memory (Knowlton et al1994 Knowlton et al 1996 Knowlton et al 1992 Knowl-ton amp Squire 1993 1996 Squire amp Knowlton 1995)

The multiple memory systems perspective (eg Knowl-ton 1999) would be bolstered if our results showed thatthe parietal oldnew effect was uniquely related to recog-nition memory However we cannot conclusively rule outthe presence of parietal inout effects in the present ex-periments We observed a nonsignificant trend indicatinglarger parietal voltages for outsider than for family mem-bers Thus in contrast to the oldnew effect in which theparietal effect was larger for the more familiar class ofitems (old new) the parietal inout effect was larger forthe less familiar class of items (out in) These oppositepolarity differences are reminiscent of the opposing pat-terns of f MRI activation observed in comparing catego-rization and recognition tasks (Reber et al 1998a) Giventhe evidence linking the parietal oldnew effect to the hip-pocampus we doubt that our parietal ERP results are en-tirely attributable to the posterior occipital areas impli-cated in Reber et alrsquos (1998a) f MRI study although theseareas could contribute to the ERP effects Other functionalimaging research has shown that the polarity of hippocam-pal activation changes between old and new recognitionmemory conditions can vary somewhat unpredictably Forexample using very similar PET recognition memory par-adigms with novel objects Schacter and colleagues havefound old new hippocampal activation differences insome experiments (Schacter et al 1995 Schacter et al1997) but new old differences in others (Heckers et al2000 Schacter et al 1999) Thus the relationship be-tween stimulus familiarity and hippocampal activity is notentirely straightforward

ConclusionsThe clearest general result to emerge from the present

experiments is a temporal sequence of events involving atransition between early sensitivity to category membership(inout differences) and later sensitivity to differential ex-perience with particular exemplars (oldnew differences)This temporal sequence suggests that the information nec-essary for categorization may become available earlierthan that necessary for recognition memory These tem-poral differences should be interpreted cautiously be-cause ERPs do not provide an exhaustive measure of brainactivity so our methods may be insensitive to earlierrecognition-related processes A more conservative inter-pretation of our results is that among the memory-related

effects we observed (N1 FN400 and parietal ERP com-ponents) categorical sensitivity appeared earlier thanrecognition-related sensitivity Even this more conserva-tive interpretation provides important information aboutthe temporal dynamics of the underlying memory effectsand their ERP correlates These empirical time course re-sults are particularly important now that models of cate-gorization are being extended to account for temporal dy-namics (Lamberts 2000 Nosofsky amp Alfonso-Reese1999)

One way to conceptualize the temporal transition maybe in terms of a shift from a gross level of sensitivity to stim-ulus similarity (N1 inout effects) to intermediate levels(FN400 inout and oldnew effects) to fine-grained dif-ferences (parietal oldnew effects) In the present experi-ment categorically different blobs (in vs out) were morevisually dissimilar than old and new blobs We believe thissituation is often true of real-world differences betweencategorization and recognition memory For example catsand dogs are categories that are easy to discriminate visu-ally as compared with the visually difficult problem ofrecognizing differences between your sisterrsquos dog Fido andyour brotherrsquos dog Rover Future research could explorethis issue more systematically by contriving situations inwhich different categories are visually similar yet recog-nition discriminations are visually dissimilar

The present results are generally consistent with theldquocomplementary learning systemsrdquo framework recentlyadvanced by OrsquoReilly and colleagues (McClelland Mc-Naughton amp OrsquoReilly 1995 Norman amp OrsquoReilly 2001OrsquoReilly amp Rudy 2001) According to this frameworkcortical networks learn by slowly integrating informationacross previous experiences in a manner that leads to dis-tributed representations of the statistical structure of envi-ronmental input Such representations are well suited forcategorization tasks that require generalization to new ex-emplars The hippocampus on the other hand is able toquickly learn about the specifics of individual events in amanner that fosters separate distinctive representationsThese hippocampal representations are thought to under-lie conscious recollection of prior episodes The categoricalsensitivity of the N1 is consistent with the operation of thehypothesized cortical networks whereas the exemplar-specific discrimination of the parietal oldnew effect isconsistent with the hypothesized characteristics of the hip-pocampus The FN400 being sensitive to both inout andoldnew differences seems to fall somewhere in betweenMuch like mathematical models of recognition and cate-gorization (eg Estes 1994 Hintzman 1986 1988 Medinamp Schaffer 1978 Nosofsky 1991) such cortical memorynetworks are capable of discriminating old from newitems in recognition memory tasks (Norman amp OrsquoReilly2001) The N1 and the FN400 sources may reflect the ac-tivity of formally similar neuronal networks but thegreater sensitivity of the FN400 to oldnew effects may berelated to differences in the type of information processedby each (eg purely perceptual N1 vs the FN400 which


is additionally sensitive to semantics as was discussedearlier)

Overall aspects of our results are consistent with boththe mathematical modeling (ie single-system) and the neu-ropsychological (ie multiple-systems) views reviewed inthe introduction The sensitivity of the FN400 (300ndash500 msec) to both categorical and recognition discrimina-tions is consistent with a single familiarity-based processcapable of contributing to both tasks However the N1 andthe parietal ERP effects seem more suited to categoriza-tion and recognition respectively The N1 (156ndash200 msec)was clearly affected by inout differences more than byoldnew differences and so may reflect a process (or sys-tem) that contributes to categorization but not to recogni-tion Parietal ERP effects (300ndash500 msec) were less clearbut appeared to be more sensitive to oldnew than to inoutdifferences Given other evidence linking the parietaloldnew effect to conscious recollection processing in-volving the hippocampus this may reflect the activity ofa process that is more likely to underlie recognition mem-ory than categorization The present results suggest thatthe brain was sensitive to inout categorical differencesprior to oldnew recognition differences so theories ad-dressing the relationship between categorization andrecognition (of either the single- or multiple-system vari-eties) should account for these temporal dynamics

REFERENCES

Alain C Achim A amp Woods D L (1999) Separate memory-relatedprocessing for auditory frequency and patterns Psychophysiology 36737-744

Allan K Wilding E L amp Rugg M D (1998) Electrophysiologi-cal evidence for dissociable processes contributing to recollectionActa Psychologica 98 231-252

Bentin S Allison T Puce A Perez E amp McCarthy G (1996)Electrophysiological studies of face perception in humans Journal ofCognitive Neuroscience 8 551-565

Bentin S amp Deouell L Y (2000) Structural encoding and identifica-tion in face processing ERP evidence for separate mechanisms Cog-nitive Neuropsychology 17 35-54

Bertrand O Perin F amp Pernier J (1985) A theoretical justifica-tion of the average reference in topographic evoked potential studiesElectroencephalography amp Clinical Neuroscience 62 462-464

Besson M Kutas M amp Van Petten C (1992) An event-related po-tentials (ERP) analysis of semantic congruity and repetition effect insentences Journal of Cognitive Neuroscience 4 132-149

Boumltzel K Schulze S amp Stodieck R G (1995) Scalp topographyand analysis of intracranial sources of face-evoked potentials Exper-imental Brain Research 104 135-143

Brainerd C J Reyna V F amp Kneer R (1995) False-recognitionreversal When similarity is distinctive Journal of Memory amp Lan-guage 34 157-185

Clark S E amp Gronlund S D (1996) Global matching models ofrecognition memory How the models match the data PsychonomicBulletin amp Review 3 37-60

Curran T (1999) The electrophysiology of incidental and intentionalretrieval ERP oldnew effects in lexical decision and recognitionmemory Neuropsychologia 37 771-785

Curran T (2000) Brain potentials of recollection and familiarityMemory amp Cognition 28 923-938

Curran T Schacter D L Johnson M K amp Spinks R (2001) Brainpotentials reflect behavioral differences in true and false recognitionJournal of Cognitive Neuroscience 13 201-216

Curran T Tucker D M Kutas M amp Posner M I (1993) Topog-raphy of the N400 Brain electrical activity reflecting semantic ex-pectation Electroencephalography amp Clinical Neurophysiology 88188-209

Dien J (1998) Issues in the application of the average reference Re-view critiques and recommendations Behavior Research MethodsInstruments amp Computers 30 34-43

Duumlzel E Vargha-Khadem F Heinze H-J amp Mishkin M (2001)Brain activity evidence for recognition without recollection after earlyhippocampal damage Proceedings of the National Academy of Sci-ences 98 8101-8106

Duumlzel E Yonelinas A P Mangun G R Heinze H-J amp Tul-ving E (1997) Event-related potential correlates of two states ofconscious awareness in memory Proceedings of the National Acad-emy of Sciences 94 5973-5978

Eimer M (1998) Mechanisms of visuospatial attention Evidence fromevent-related brain potentials Visual Cognition 5 257-286

Eimer M (2000a) Attentional modulations of event-related brain poten-tials sensitive to faces Cognitive Neuropsychology 17 103-116

Eimer M (2000b) Event-related brain potentials distinguish process-ing stages involved in face perception and recognition Clinical Neu-rophysiology 111 694-705

Estes W K (1994) Classification and cognition New York OxfordUniversity Press

Friedman D amp Johnson R Jr (2000) Event-related potential (ERP)studies of memory encoding and retrieval A selective review Mi-croscopy Research amp Technique 51 6-28

George N Evans J Fiori N Davidoff J amp Renault B (1996)Brain events related to normal and moderately scrambled faces Cog-nitive Brain Research 4 65-76

Gillund G amp Shiffrin R M (1984) A retrieval model for bothrecognition and recall Psychological Review 91 1-67

Guillem F Bicu M amp Debruille J B (2001) Dissociating mem-ory processes involved in direct and indirect tests with ERPs to unfa-miliar faces Cognitive Brain Research 11 113-125

Heckers S Curran T Goff D Rauch S L Fischman A JAlpert N M amp Schacter D L (2000) Abnormalities in the thal-amus and prefrontal cortex during episodic object recognition inschizophrenia Biological Psychiatry 48 651-657

Hintzman D L (1986) ldquoSchema abstractionrdquo in a multiple-tracememory model Psychological Review 93 411-428

Hintzman D L (1988) Judgments of frequency and recognition mem-ory in a multiple-trace memory model Psychological Review 95528-551

Hintzman D L amp Curran T (1994) Retrieval dynamics of recogni-tion and frequency judgments Evidence for separate processes of fa-miliarity and recall Journal of Memory amp Language 33 1-18

Humphreys M S Bain J D amp Pike R (1989) Different ways to cuea coherent memory system A theory for episodic semantic and pro-cedural tasks Psychological Review 96 208-233

Jacoby L L (1991) A process dissociation framework Separating au-tomatic from intentional uses of memory Journal of Memory amp Lan-guage 30 513-541

Jasper H A (1958) The tenndashtwenty system of the international feder-ation Electroencepholography amp Clinical Neurophysiology 10 371-375

Kiefer M (2001) Perceptual and semantic sources of category-specificeffects Event-related potentials during picture and word categoriza-tion Memory amp Cognition 29 100-116

Kinder A amp Shanks D R (2001) Amnesia and the declarativenondeclarative distinction A recurrent network model of classifica-tion recognition and repetition priming Journal of Cognitive Neuro-science 13 648-669

Knowlton B J (1999) What can neuropsychology tell us about cate-gory learning Trends in Cognitive Sciences 3 123-124

Knowlton B J Gluck M A amp Squire L R (1994) Probabilisticclassification learning in amnesia Learning amp Memory 1 106-120

Knowlton B J Mangels J A amp Squire L R (1996) A neostri-atal habit learning system in humans Science 273 1399-1402

Knowlton B J Ramus S J amp Squire L R (1992) Intact artificial


grammar learning in amnesia Dissociation of classification learningand explicit memory for specific instances Psychological Science 3172-179

Knowlton B J amp Squire L R (1993) The learning of categoriesParallel brain systems for item memory and category knowledge Sci-ence 262 1747-1749

Knowlton B J amp Squire L R (1996) Artificial grammar learningdepends on implicit acquisition of both abstract and exemplar-specificinformation Journal of Experimental Psychology Learning Mem-ory amp Cognition 22 169-181

Kutas M amp Iragui V (1998) The N400 in a semantic categorizationtask across 6 decades Electroencephalography amp Clinical Neurophys-iology 108 456-471

Kutas M amp Van Petten C (1994) Psycholinguistics electrifiedEvent-related brain potential investigations In M Gernsbacher (Ed)Handbook of psycholinguistics (pp 83-143) New York AcademicPress

Lamberts K (2000) Information-accumulation theory of speeded cat-egorization Psychological Review 107 227-260

Lehman D amp Skrandies W (1985) Spatial analysis of evoked po-tentials in man A review Progress in Neurobiology 23 227-250

Mandler G (1980) Recognizing The judgment of previous occur-rence Psychological Review 87 252-271

McCandliss B D Curran T amp Posner M I (1993) Repetition ef-fects in processing visual words A high density ERP study of lateral-ized stimuli Society for Neuroscience Abstracts 19 1807

McCandliss B D Curran T amp Posner M I (1994 November) Ex-ploring the time course of word recognition for identical and cross-caserepetitions Poster presented at the 35th Annual Meeting of the Psy-chonomic Society St Louis

McCarthy G amp Wood C C (1985) Scalp distributions of event-relatedpotentials An ambiguity associated with analysis of variance modelsElectroencepholography amp Clinical Neurophysiology 62 203-208

McClelland J L McNaughton B L amp OrsquoReilly R C (1995)Why there are complementary learning systems in the hippocampusand neocortex Insights from the successes and failures of connec-tionist models of learning and memory Psychological Review 102419-457

Mecklinger A (2000) Interfacing mind and brain A neurocognitivemodel of recognition memory Psychophysiology 37 565-582

Medin D L amp Schaffer M M (1978) Context theory of classifi-cation learning Psychological Review 85 207-238

Murdock B B (1982) A theory of the storage and retrieval of item andassociative information Psychological Review 89 609-626

Nessler D Mecklinger A amp Penney T B (2001) Event relatedbrain potentials and illusory memories The effects of differential en-coding Cognitive Brain Research 10 283-301

Norman K A amp OrsquoReilly R C (2001) Modeling hippocampal andneocortical contributions to recognition memory A complementarylearning systems approach (ICS Tech Rep 01-02) Boulder Univer-sity of Colorado Institute of Cognitive Science

Nosofsky R M (1988) Exemplar-based accounts of the relations be-tween classification recognition and typicality Journal of Experi-mental Psychology Learning Memory amp Cognition 14 700-708

Nosofsky R M (1991) Tests of an exemplar model for relating per-ceptual classification and recognition memory Journal of Experi-mental Psychology Human Perception amp Performance 17 3-27

Nosofsky R M amp Alfonso-Reese L A (1999) Effects of similar-ity and practice on speeded classification response times and accura-cies Further tests of an exemplar-retrieval model Memory amp Cogni-tion 27 78-93

Nosofsky R M amp Zaki S (1998) Dissociations between categoriza-tion and recognition in amnesic and normal individuals An exemplar-based interpretation Psychological Science 9 247-255

Nosofsky R M amp Zaki S (1999) Math modeling neuropsychologyand category learning Trends in Cognitive Sciences 3 125-126

Olichney J M Van Petten C Paller K A Salmon D PIragui V J amp Kutas M (2000) Word repetition in amnesia Elec-trophysiological measures of impaired and spared memory Brain123 1948-1963

OrsquoReilly R C amp Rudy J W (2001) Conjunctive representations in

learning and memory Principles of cortical and hippocampal func-tion Psychological Review 108 311-345

Paller K A amp Kutas M (1992) Brain potentials during memory re-trieval provide neurophysiological support of the distinction betweenconscious recollection and priming Journal of Cognitive Neuroscience4 375-391

Paller K A Kutas M amp McIsaac H K (1995) Monitoring con-scious recollection via the electrical activity of the brain Psycholog-ical Science 6 107-111

Palmeri T amp Flanery M (1999) Learning about categories in theabsence of training Profound amnesia and the relationship betweenperceptual categorization and recognition memory PsychologicalScience 10 526-530

Picton T W Bentin S Berg P Donchin E Hillyard S AJohnson R Jr Miller G A Ritter W Ruchkin D S RuggM D amp Taylor M J (2000) Guidelines for using human event-related potentials to study cognition Recording standards and publi-cation criteria Psychophysiology 37 127-152

Picton T W Lins O G amp Scherg M (1995) The recording andanalysis of event-related potentials In F Boller amp J Grafman (Eds)Handbook of neuropsychology(Vol 10 pp 3-73) Amsterdam Elsevier

Posner M I amp Keele S W (1968) On the genesis of abstract ideasJournal of Experimental Psychology 77 353-363

Posner M I amp McCandliss B D (1999) Brain circuitry during read-ing In R Klein amp P McMullen (Eds) Converging methods for under-standing reading and dyslexia (pp 305-337) Cambridge MA MITPress

Reber P J Stark C E amp Squire L R (1998a) Contrasting corti-cal activity associated with category memory and recognition mem-ory Learning amp Memory 5 420-428

Reber P J Stark C E amp Squire L R (1998b) Cortical areas sup-porting category learning identified using functional MRI Proceed-ings of the National Academy of Sciences 95 747-750

Reed J M Squire L R Patalano A L Smith E E amp Jonides J(1999) Learning about categories that are defined by object-like stim-uli despite impaired declarative memory Behavioral Neuroscience113 411-419

Rossion B Campanella S Gomez C M Delinte A Deba-tisse D Liard L Dubois S Bruyer R Crommelinck M ampGueacuterit J-M (1999) Task modulation of brain activity related to fa-miliar and unfamiliar face processing An ERP study Clinical Neu-rophysiology 110 449-462

Rossion B Delvenne J-F Debatisse D Goffaux V Bruyer RCrommelinck M amp Gueacuterit J-M (1999) Spatio-temporal local-ization of the face inversion effect An event-related potentials studyBiological Psychology 50 173-189

Rugg M D (1995) ERP studies of memory In M D Rugg amp M G H Coles (Eds) Electrophysiology of mind (pp 132-170) NewYork Oxford University Press

Rugg M D Cox C J C Doyle M C amp Wells T (1995) Event-related potentials and the recollection of low and high frequencywords Neuropsychologia 33 471-484

Rugg M D Mark R E Walla P Schloerscheidt A M Birch C S amp Allan K (1998) Dissociation of the neural corre-lates of implicit and explicit memory Nature 392 595-598

Rugg M D amp Nagy M E (1989) Event-related potentials and recog-nition memory for words Electroencephalography amp Clinical Neuro-physiology 72 395-406

Rugg M D Schloerscheidt A M amp Mark R E (1998) An elec-trophysiological comparison of two indices of recollection Journal ofMemory amp Language 39 47-69

Schacter D L Curran T Reiman E M Chen K Bandy D Jamp Frost J T (1999) Medial temporal lobe activation during episodicencoding and retrieval A PET study Hippocampus 9 575-581

Schacter D L Reiman E Uecker A Polster M R Yun L S ampCooper L A (1995) Brain regions associated with retrieval of struc-turally coherent visual information Nature 376 587-590

Schacter D L amp Tulving E (1994) What are the memory systemsof 1994 In D L Schacter amp E Tulving (Eds) Memory systems 1994(pp 1-38) Cambridge MA MIT Press

Schacter D L Uecker A Yun L S Bandy D Chen K


Cooper L A amp Curran T (1997) Effects of size and orientationchange on hippocampal activation during episodic recognition A PETstudy NeuroReport 8 3993-3998

Sherry D F amp Schacter D L (1987) The evolution of multiplememory systems Psychological Review 94 439-454

Shiffrin R M amp Steyvers M (1997) A model for recognitionmemory REMmdashretrieving effectively from memory PsychonomicBulletin amp Review 4 145-166

Smith M E (1993) Neurophysiological manifestations of recollectiveexperience during recognition memory judgments Journal of Cogni-tive Neuroscience 5 1-13

Smith M E amp Halgren E (1989) Dissociation of recognition mem-ory components following temporal lobe lesions Journal of Experi-mental Psychology Learning Memory amp Cognition 15 50-60

Spencer K M Vila Abad E amp Donchin E (2000) On the searchfor the neurophysiological manifestation of recollective experiencePsychophysiology 37 494-506

Squire L R amp Knowlton B J (1995) Learning about categories inthe absence of memory Proceedings of the National Academy of Sci-ences 92 12470-12474

Srinivasan R Nunez P L Silberstein R B Tucker D M ampCadusch P J (1996) Spatial sampling and filtering of EEG with spline-Laplacians to estimate cortical potentials Brain Topography 8 355-366

Tanaka J W amp Curran T (2001) A neural basis for expert objectrecognition Psychological Science 12 43-47

Tanaka J [W] Luu P Weisbrod M amp Kiefer M (1999) Track-ing the time course of object categorization using event-related po-tentials NeuroReport 10 829-835

Taylor M J McCarthy G Saliba E amp Degiovanni E (1999)ERP evidence of developmental changes in processing of faces Clin-ical Neurophysiology 110 910-915

Trott C T Friedman D Ritter W amp Fabiani M (1997) Itemand source memory Differential age effects revealed by event-relatedpotentials NeuroReport 8 3373-3378

Tucker D M (1993) Spatial sampling of head electrical fields The ge-odesic sensor net Electroencephalography amp Clinical Neurophysiol-ogy 87 154-163

Tucker D M Liotti M Potts G F Russell G S amp Posner M I(1994) Spatiotemporal analysis of brain electrical f ields HumanBrain Mapping 1 134-152

Van Petten C Kutas M Kluender R Mitchiner M ampMcIsaac H (1991) Fractionating the word repetition effect with event-related potentials Journal of Cognitive Neuroscience 3 131-150

Vogel E K amp Luck S J (2000) The visual N1 component as anindex of a discrimination process Psychophysiology 37 190-203

Wilding E L Doyle M C amp Rugg M D (1995) Recognition mem-ory with and without retrieval of context An event-related potentialstudy Neuropsychologia 33 743-767

Wilding E L amp Rugg M D (1996) An event-related potential studyof recognition memory with and without retrieval of source Brain 119889-905

Wilding E L amp Rugg M D (1997a) Event-related potential and therecognition memory exclusion task Neuropsychologia 35 119-128

Wilding E L amp Rugg M D (1997b) An event-related potentialstudy of recognition memory for words spoken aloud or heard Neu-ropsychologia 35 1185-1195

NOTES

1 Using the average reference may help explain the consistent frontalfocus of the FN400 in our laboratory but similar frontal effects havebeen observed relative to a mastoid reference (Friedman amp Johnson2000 Guillem Bicu amp Debruille 2001 Mecklinger 2000 Rugg Market al 1998)

2 Thirteen subjects were excluded for the following reasons Six sub-jects failed to attend the second session Five subjects did not have suf-ficient trials in all conditions because of low accuracy (n 5 3) or exces-sive blinking (n 5 2) The data from 2 subjects were not usable becauseof equipment problems

3 The primary analyses concentrated on only the negative PI aspectsof the N1 but the co-occurrence of AS positive potentials is commonThe face-enhanced N170 recorded near the mastoids is often accompa-nied by a vertex positive potential (VPP Bentin et al 1996 Boumltzel et al1995 Eimer 1998 2000a 2000b George et al 1996 Rossion Cam-panella et al 1999) The VPP and N1 may reflect activity within a com-mon generator process (George et al 1996) but others have argued fordifferent sources (Boumltzel et al 1995)

4 Four subjects were excluded Two subjects had excessively noisyEEG A computer crash prematurely terminated the experiment for 1 sub-ject One subject blinked excessively

(Continued on next page)


APPENDIX Mean Masatoid-Referenced ERPs From Channels Nearest to Several Locations

of the International 10ndash20 System (ERPs Have Been Averaged Across Tasks)

(Manuscript received July 13 2001revision accepted for publication December 20 2001)


and the exemplars held in memory The same exemplar-based system underlies both recognition and classificationbut different decision rules are applied to each task (Nosof-sky 1988 1991) Categorization depends on comparing atest item with stored exemplars from various categoriesand classifying the item as belonging to the category withwhich it is most similar For example a novel animalwould be categorized as a dog if its similarity to stored dogexemplars is higher than its similarity to stored exemplarsfrom other categories (eg cats horses trees furniturecars etc) Recognition depends on comparing the testitem with all stored exemplars so that recognition proba-bility increases with the overall similarity of the test itemto all stored exemplars across all categories Thus a par-ticular dog would be recognized as having been previouslyencountered if its similarity to all stored exemplars is highenough to pass a personrsquos criterion for responding oldrather than new

Single-system exemplar models can account for aspectsof the neuropsychological results originally interpretedwithin the multiple-systems perspective (for a summaryand continuation of this debate see Knowlton 1999 Nosof-sky amp Zaki 1999) Nosofsky and Zaki (1998) have re-cently shown that an exemplar-based model can success-fully simulate the pattern of spared categorization andimpaired recognition exhibited by amnesic patients (Knowl-ton amp Squire 1993) Amnesic performance was simulatedby varying a sensitivity parameter that controls discrimi-nation among exemplars As has been observed for am-nesia decreased sensitivity led to poorer recognitionmemory without significantly impairing categorizationSimilarly amnesicsrsquo pattern of spared artificial grammarlearning and impaired recognition memory has been sim-ulated with a single-system connectionist network (Kinderamp Shanks 2001) Thus multiple systems may not be nec-essary to account for neuropsychological dissociations be-tween categorization and recognition On the other handthe multiple-systems perspective seems more consistentwith functional neuroimaging results showing that visualcategorization is associated with activity in visual corticalregions not typically related to declarative memory (Reberet al 1998a 1998b)

Research using event-related brain potentials (ERPs)can provide information that is relevant to understandingthe relationship between categorization and recognitionScalp-recorded ERPs are obtained by averaging elec-troencephalogram (EEG) activity across multiple trials thatare designed to engage specific cognitive processes Bytime-locking the average to stimulus onset ERPs reflectthe activity of brain processes that are regularly associatedwith stimulus processing under the experimental condi-tions of interest ERPs can be differentiated on the basis oftheir timing (with millisecond resolution) and scalp dis-tribution so different neurocognitive processes can beidentified with distinct spatiotemporal voltage patternsPrevious studies have examined ERPs in categorizationand recognition tasks so we will review the main findingsrelevant to each

Several studies have indicated that ERP effects relatedto the categorization of visual objects are observed within200 msec of stimulus onset The N1 is a negative polarityERP component peaking between 150 and 200 msec overposterior head locations that is thought to reflect visualdiscrimination processes (eg Vogel amp Luck 2000) TanakaLuu Weisbrod and Kiefer (1999) used a category verifi-cation task in which a category label referring to superor-dinate (eg ANIMAL) basic (eg BIRD) or subordinate(eg ROBIN) levels was followed by a picture and subjectsresponded true or false according to the correspondencebetween the label and the picture The N1 was more neg-ative for subordinate than for basic level classificationTanaka et al interpreted the N1 enhancement as reflectingincreased visual analysis associated with subordinate clas-sification The N1 has been observed to be more negativefor pictures of objects from natural than from artifactualcategories (Kiefer 2001) Assuming that perceptual in-formation is more important for categorizing natural (egdifferent birds or dogs look somewhat similar) than arti-factual (eg different tools or pieces of furniture look dis-similar) object categories Kiefer also attributed the ERPeffect to perceptual-processing differences Another re-cent study has shown that the N1 is enhanced when ex-perts categorize objects within their domain of expertise(Tanaka amp Curran 2001) Tanaka and Curran labeled the expertise-sensitive ERP component N170 because of its similarity to the N1 component that numerous stud-ies have found to be greater in response to faces than to other objects (eg Bentin Allison Puce Perez amp Mc-Carthy 1996 Bentin amp Deouell 2000 Boumltzel Schulze ampStodieck 1995 Eimer 1998 2000a 2000b GeorgeEvans Fiori Davidoff amp Renault 1996 Rossion Cam-panella et al 1999)

In contrast to the early occurring ERP effects related tocategorization numerous studies have suggested that ERPeffects relevant to recognition memory do not begin until300ndash400 msec after stimulus onset In a typical recogni-tion memory experiment subjects study a list of stimuli(words pictures etc) followed by a test list containing amixture of old (studied) and new (nonstudied) stimuliWhen ERPs are recorded during the test list electrophys-iological differences between correctly classified old andnew conditions emerge after about 300 msec The numberof component processes underlying such ERP oldnew ef-fects and their functional significance remain uncertainTwo recent reviews have suggested that ERP oldnew ef-fects can be broken into at least three different subcom-ponents (Friedman amp Johnson 2000 Mecklinger 2000)and this division corresponds well to results from our ownlaboratory (Curran 1999 2000 Curran Schacter John-son amp Spinks 2001) FN400 effects parietal effects andlate frontal effects The present study is most relevant tothe FN400 and parietal oldnew effects so they will bediscussed in more detail

The FN400 oldnew effect involves an N400-like com-ponent that is more negative for new than for old stimuliover frontal electrode sites between 300 and 500 msec











EXPERIMENT 1

MethodSubjects









Prototype

In

Out


















in Experiment 1

In Out

Old New Old New



































EXPERIMENT 2







N1Effects

(156ndash220 msec)

ParietalEffects

(400ndash800 msec)

FN400Effects

(300ndash500 msec)


MethodSubjects
















GENERAL DISCUSSION



in Experiment 2

Categorization

In Out

Old New Old New

p(correct) 77 76 72 71RT 2694 2686 2706 2711

































REFERENCES










































































































NOTES




















EXPERIMENT 1

MethodSubjects









Prototype

In

Out


















in Experiment 1

In Out

Old New Old New



































EXPERIMENT 2







N1Effects

(156ndash220 msec)

ParietalEffects

(400ndash800 msec)

FN400Effects

(300ndash500 msec)


MethodSubjects
















GENERAL DISCUSSION



in Experiment 2

Categorization

In Out

Old New Old New

p(correct) 77 76 72 71RT 2694 2686 2706 2711

































REFERENCES










































































































NOTES


























in Experiment 1

In Out

Old New Old New



































EXPERIMENT 2







N1Effects

(156ndash220 msec)

ParietalEffects

(400ndash800 msec)

FN400Effects

(300ndash500 msec)


MethodSubjects
















GENERAL DISCUSSION



in Experiment 2

Categorization

In Out

Old New Old New

p(correct) 77 76 72 71RT 2694 2686 2706 2711

































REFERENCES










































































































NOTES










































EXPERIMENT 2







N1Effects

(156ndash220 msec)

ParietalEffects

(400ndash800 msec)

FN400Effects

(300ndash500 msec)


MethodSubjects
















GENERAL DISCUSSION



in Experiment 2

Categorization

In Out

Old New Old New

p(correct) 77 76 72 71RT 2694 2686 2706 2711

































REFERENCES










































































































NOTES











MethodSubjects
















GENERAL DISCUSSION



in Experiment 2

Categorization

In Out

Old New Old New

p(correct) 77 76 72 71RT 2694 2686 2706 2711

































REFERENCES










































































































NOTES










































REFERENCES










































































































NOTES














































































NOTES










an electrophysiological comparison of visual

Documents