ERP Assessment of Visual and Auditory Language Processing in Schizophrenia

Journal of Abnormal Psychology 1997, Vol. 106, No. 1, 85-94 In the public domain ERP Assessment of Visual and Auditory Language Processing in Schizophrenia M. A. Niznikiewicz, B. F. O'Donnell, P. G. Nestor, L. Smith, S. Law, M. Karapelou, M. E. Shenton, and R. W. McCarley Harvard Medical School and Veterans Affairs Medical Center, Brockton Language disturbance in schizophrenia has been recently attributed to disturbed priming mechanisms. In the present study, event-related potentials (ERPs), were recorded to final words in sentences presented to 13 chronic patients with schizophrenia and 12 normal controls. Half of the final words fit a sentence context and another half did not. The N400 (the ERP sensitive to language) latency was prolonged, and its amplitude was more negative to both correct and incorrect sentence endings in the group wilh schizophrenia relative to the group of normal controls. The early ERP components, N100 and P200, were similar in both groups. These results suggest that language abnormalities in schizophrenia are related to a dysfunction in the language system and not to a general cognitive dysfunction, and may be related to poor use of context in patients with schizophrenia. Language in Schizophrenia Clinical studies of schizophrenia have identified poverty of content, tangentiality, distractable speech, derailment, and incoherence as features of schizophrenic language (Andreasen, 1979). Schizophrenic discourse has been found to be poorly organized (Harvey, 1983; Rochester & Martin, 1979), and thought-disordered (TD) schizophrenia patients have been found to lack effective means (as conveyed by, e.g., appropriate pronouns; Harvey, 1983) to provide reference and cohesion to their discourse. These observations are in accord with Bleuler's descriptions of cognitive difficulties found in schizophrenia patients. On the basis of the samples of speech of his patients, he concluded In the normal thinking process, the numerous actual and latent images combine to determine each association. In schizophrenia..., instead, thinking operates with ideas and concepts which have no, or a completely insufficient, connection with the main idea and should therefore be excluded from the thought-process. The result is M. A. Niznikiewicz, B. F. O'Donnell, P. G. Nestor, L. Smith, S. Law, M. Karapelou, M. E. Shenton, and R. W. McCarley, Department of Psychiatry, Harvard Medical School, and Veterans Affairs Medical Center, Brockton. P. G. Nestor is now at University of Massachusetts, Boston, and acknowledges the university for enabling him to complete his work. This research was partially supported by a Clinical Research Training Program, National Alliance for Research on Schizophrenia "Vbung Investigator Award; National Institute of Mental Health Grants MH KO2 MH0110, MH IR29 50747, and RO1 MH-40-799, Brockton Schizophrenia Center of the Department of Veterans Affairs, and by the Commonwealth of Massachusetts Research Center. Correspondence concerning this article should be addressed to M. A. Niznikiewicz or R. W. McCarley, Psychiatry Service 116A, 940 Belmont Street, Brockton, Massachusetts 02401. Electronic mail may be sent via Internet to mnizniki@warren.med.harvard.edu. that thinking becomes confused, bizarre, incorrect, abrupt. (Bleuler, 1911/1950, p. 22) Experimental approaches have been aimed largely at uncovering cognitive and, more recently, neural mechanisms for language difficulties in schizophrenia. For example, Chapman, Chapman, and Miller (1964) reported schizophrenia patients' bias toward the most usual (strong) meaning of a word (e.g., interpreting pen as a writing utensil in spite of the context pointing to a less usual, weak meaning of an enclosure for chicken). In a further study by Chapman, Chapman, and Daut (1976), this tendency was related to schizophrenia patients' inability to screen their responses for appropriateness (i.e., to use available sentence context to help select a proper word in a given linguistic situation). Recent Hypotheses of Language Dysfunction in Schizophrenia The tendency to produce loose and irrelevant associations has been linked recently to dysfunctional processes in lexical memory (Kwapil, Hegley, Chapman, & Chapman, 1990; Maher, 1991; Manschreck et al., 1988; Spiteer et al., 1994). This hypothesis has been informed by models of lexical memory (Anderson, 1983; Morton, 1969), which assume that knowledge about words is stored in a semantically organized associative network and that words in the network can be activated (or primed) to a different degree. Given that in studies using word pairs at short (under 200-ms) interstimulus intervals (ISIs), TD patients showed a priming effect superior to those of normal controls and non-td patients (Kwapil et al., 1990; Manschreck et al., 1988; Spitzer et al., 1994); it has been argued that language dysfunction in schizophrenia is mediated by overactivation of semantic networks (hyperpriming). However, understanding and producing language involves more than a spread of activation from a word to its semantic 85

86 NIZNIKIEWICZ ET AL. neighbors. For example, sentence comprehension involves building a discourse representation or schema of what a sentence is about. As the sentence unfolds, the number of words fitting a particular sentence fragment becomes smaller because the word candidates are inhibited by the preceding sentence context (see Neely, 1991, for his review of priming phenomena). Thus, the use of sentences as stimuli may provide a more complete picture of language difficulties in schizophrenia. For example, it is possible that errors observed in schizophrenic speech may be due to (a) dysfunctional early processes of spreading activation (lexical level processes), (b) dysfunctional late processes of context integration (postlexical and sentence-level processes), or (c) a combination of both types of dysfunction. In this study, eventrelated potentials (ERPs) rather than response times (KB) were used to assess these late processes. The use of KTs as a dependent measure leaves many questions unresolved about temporal and qualitative differences in processing language between normal and schizophrenic individuals. For example, it is often difficult to conclude whether RT differences among patients and normal controls are due to the early, more sensory aspects of word processing or to the later, more semantic aspects of word processing. Also, it is not clear whether the differences between normal and schizophrenic patients' language processing consist solely of a temporal difference in otherwise similar processes or whether the differences are both temporal and qualitative. Event-Related Potentials as Measure of Cognitive Processes The ERP technique provides an on-line measure of cognitive events as they happen in real time by recording a scalp electrical potential produced by brain areas involved in processing specific stimuli. In this technique, electrical responses time-locked to stimulus presentation are averaged over many trials to construct a waveform characterized by positive and negative peaks. Their latency and amplitude reflect stages of stimulus (e.g., word) processing and allow inferences about the nature of these stages (Donchin & Coles, 1988; McCarthy & Donchin, 1981). The ERP technique has been used successfully in the past to demonstrate abnormal processing of sensory information and memory updating in patients with schizophrenia (e.g., McCarley, Shenton, O'Donnell, & Nestor, 1993). More recently, this technique has been used to study language difficulties in schizophrenia. N400 The ERP component sensitive to certain asnects of language processing is the N400. The N400 is a negative waveform deflection, peaking around 400 ms after a word presentation (Kutas & Hillyard, 1980). It has been recorded to target words in word pairs and sentences and in both auditory and visual modality (e.g., Holcomb, Coffey, & Neville, 1992; Holcomb & Neville, 1990). The N400 amplitude is inversely proportional to the degree of the predictability of the target word by its preceding word or sentence fragment (i.e., its context). The less a target word is predicted by the context, the larger the N400 it elicits (Kutas & Hillyard, 1980). The functional significance of the N400 has been addressed in several studies. While it was suggested that the N400 indexes automatic processes of spreading activation through semantic networks (e.g., Fischler & Raney, 1989; Kutas & Hillyard, 1989), the more recent evidence suggests that the N400 may reflect later processes (e.g., those related to the integration of a word into a sentence context; Holcomb, 1993). The Van-Petten and Kutas study (1990) underscored the important role of experimental design for the functional interpretation of the N400. In the study, the N400 amplitude varied as a function of word frequency in the lexicon (a lexical effect) if it appeared at the beginning of sentences. The less frequent a word was in the lexicon, the larger the N400 it evoked. This effect was superseded by contextual influences when words, regardless of frequency, were in a final position of the sentence (semantic, contextual effect). The N400 amplitude depended on how well a word fit the context. Taken together, the evidence from studies of normal population suggests that the N400 can be used as an alternative measure of verbal memory function and organization in schizophrenia. P600 In the paradigms using language stimuli, the N400 is often followed by a positive waveform appearing in the latency range of 600-800 ms. There is considerably less agreement on the functional significance. Some authors believe it is a P300 appearing late (e.g., Friedman et al., 1975; Kramer & Donchin, 1987), and others prefer to view it as the P600 indexing syntactic processes (Osterhout & Holcomb, 1992). It is possible that the P300 and the P600 share some characteristics, such as indexing some aspects of memory updating. However, the P300 is recorded in "odd-ball" paradigms that call for processing of relatively simple, nonlinguistic information. The P600 is found in waveforms recorded to language stimuli. Therefore, it is unlikely that the two components have the same functional significance. In this context, the finding of diminished P300 amplitude recorded to simple, mostly auditory stimuli in patients with schizophrenia (e.g., Bruder et al., 1996) offers few insights about the expected P600 morphology recorded in language paradigms in this group of people. The N400 as a Measure of Language Dysfunction in Schizophrenia Several studies have recently used the N400 paradigm to explore language difficulties in schizophrenia. To date, all of these studies have focused on the visual modality. Koyama et al. (1991) and Grillon, Ameli, and Glazer (1991) used word pairs to study lexical processes, whereas Mitchell, Andrews, Fox, Catts, and McConghy, (1991), and Adams et al. (1993) used sentences to study attentional and priming processes. The most consistent finding across these studies was N400 latency prolongation, indicating that the processes indexed by the N400 started later in schizophrenic patients than in normal participants. Most studies reported reduced N400 amplitude measured from a difference waveform (Mitchell et al., 1991, is an exception). However, the difference waveform is obtained by pointby-point subtraction of waveform to a congruent word, from a waveform to an incongruent word. This procedure may not be appropriate in clinical studies where the groups may differ in

LANGUAGE IN SCHIZOPHRENIA 87 their ERP to the baseline, congruent condition. For example, one reason for the smaller N400 difference in amplitude may be the presence of an N400 recorded to congruent sentences. The examination of the raw waveforms from those studies suggests that the following may be the case: Schizophrenic patients in those studies produced an N400 to congruent or primed targets, unlike normal controls, who tended to have an absent or negligible N400 to congruent endings. In fact, Koyama et al. (1991) cited the more negative waveform to congruent words as a possible reason for their results. In all those studies, long ISIs were used, suggesting that larger N400s to both congruent and incongruent sentence endings might reflect less efficient processes of contextual integration, an impairment long suspected by clinical researchers. Hypotheses of the Present Study The present study was designed to address two major questions. The first question was whether the N400 differences between normal participants and schizophrenia participants found in the visual modality also exist in the auditory modality. If language difficulties in individuals with schizophrenia are related to a dysfunction in an amodal language system, the group differences would be similar in the two modalities. Thus, longer N400 latencies in schizophrenia patients and similar amplitude group differences would be predicted in both visual and auditory modality. However, if impairment for processing linguistic stimuli was similar to that found for the nonlinguistic stimuli, such as simple tones (e.g., see Ford et al., 1994), where the processing was more affected in the auditory than in the visual modality, then more impaired processing would be found for auditory sentences. The second question was whether the use of long ISIs in sentences presented one word at a time (i.e., in the paradigm more sensitive to processes of context integration) will demonstrate schizophrenia patients' poor use of context. As reviewed above, the N400 recorded to sentences, presented one word at a time seems to index contextual integrative processes. If people with schizophrenia can not integrate context efficiently, more negative N400 amplitude to both congruous and incongruous sentence endings would be expected in schizophrenia patients relative to normal controls in both the visual and auditory modalities. If that were the case, it would suggest that at least two different processes, that is, spreading activation (behavioral evidence from previous studies) and context integration (ERP evidence from this study) may be involved in generating the errors found in schizophrenic speech. Participants Method A total of 18 right-handed men with diagnoses of chronic schizophrenia (Diagnostic and Statistical Manual of Mental Disorders; DSM- IH-R, 3rd ed., revised, American Psychiatric Association, 1987) and between the ages of 25 and 55, and 16 male normal controls matched for age, handedness, and parental socioeconomic status (SES) participated in the experiment. The participants whose accuracy was below 75% were excluded from further analysis. On the basis of their response accuracy data, 13 patients with schizophrenia and 12 controls were selected for further analysis. Two control participants were excluded because of the lack of accuracy data. All participants spoke English as their first language. For patients with schizophrenia, the inclusion criteria were (a) no substance or alcohol abuse in the past year and no lifetime history of substance dependence, (b) no medication that would influence electroencephalogram (EEC) responses, (c) no accompanying diagnosis of affective disorder, (d) no developmental disorder, and (e) no loss of consciousness or brain injury. The average duration of the illness was 21 years, and the average stay in the hospital was 4.1 years. On average the medication received equalled 527.5 mg of chlorpromazine. For normal control participants, the additional criterion was no history of psychiatric disorder in oneself or in one's first-degree relatives. Stimuli Two hundred sentences with an average number of five to eight words were developed in order to be presented in the auditory and visual modalities. Following Kutas and Hillyard's (1980) paradigm, half of the sentences ended with words semantically congruent with the previous sentence context (e.g., Mary has never been to Boston), and half of them ended with incongruent endings (e.g., John wanted to eat one more sleeve). The mean cloze probability for the congruent endings was.75, and.02 for the incongruous endings. The final words were mid to high frequency (Kucera & Francis, 1967). Their mean length was two syllables and 5.4 letters per word. In the visual modality, the sentences were presented on the computer screen, one word at a time, with 300 ms exposure time, and 800 ms 1SI. The width of the words subtended about 1.9. The time between the offset of one sentence and the onset of another sentence was 2 s. In the auditory modality, the sentences spoken by a male voice were delivered one word at a time over Etymotic insert earphones at the level of 60 sound level (SL). The length of a word was 245-750 ms. The ISI and the time between two consecutive sentences were identical to those in the visual modality. The long ISI, apart from serving its theoretical role, also ensured that all participants had more than adequate time to read the sentences. To eliminate the effects of the order of presentation, two sequences of sentences were developed for each modality. Each set of sentences was prepared for presentation in both the visual and auditory modality. Thus, if exposed to Set 1 in the visual modality, the participant was exposed to Set 2 in the auditory modality. For the next participant, Set 2 was presented in the visual modality while Set 1 was presented in the auditory modality. The sequence of sentence presentation in the two modalities was counterbalanced across participants. Thus, no participant saw the same set of sentences twice, and, at the same time, the same set of sentences was presented in the visual and auditory modality across participants. This design accomplished two objectives: (a) it ruled out repetition effects as the basis of differences observed across modalities, and (b) it ruled out contextual effects as the basis of differences across modalities. Because one can argue that no two sentence contexts are exactly the same in terms of building expectations about the target word, the use of identical sentence sets in the two modalities across participants made it possible to interpret potential modality differences as due to an input modality rather than to different contexts. Task Procedures Seated in a reclining chair in a sound-attenuating chamber at the distance of 65 cm from the monitor screen, the participants were given instructions to judge which sentences made sense and which did not by pressing a response button. The use of right and left hands for making yes and no responses was counterbalanced across participants. Prior to the experimental procedure, the participants were exposed to an average of 5 practice sentences to make sure they understood the instructions. The participants responded after they saw a star appear on the screen

NIZN1KIEWICZ ET AL. (i.e., 800 ms after the offset of the terminal word in a sentence). The responses emitted after 1 s were treated as nonresponses. Thus, KTs did not reflect the speed of processing and were not used in the statistical analyses. The accuracy data were recorded along with the RT data and served to select a group of patients who did not statistically differ in their accuracy from normal controls. The errors were denned as a number of incorrect decisions about a sentence semantic status made within 1 s after the star appeared on the screen. The patients did nol differ from normal control participants in the number of incorrect semantic decisions about the sentence (p <. 1), EEC Recording Procedures EEC was recorded from 28 tin plate electrodes referred off-line to linked ears using the Electrocap (Electrocap Co., Dallas, Texas) system. The set of 28 electrodes included all electrodes in the international 10-20 system and the following additional interpolated electrodes: FTC! and FTC2, located at the intersection between F3 and F4 and T3 and T4 and C3 and C4 and F7 and F8; TCP112 located between C3 and C4 and T5 and T6 and P3 and P4 and T3 and T4; Cpl/2 was placed midpoint at the diagonal between Cz. and P3 and P4; and P01 and P02 were placed midpoint between Pz and Ol and O2. Additional electrodes were used to record vertical eye movements by placing electrodes at the right eye supra- and infraorbital sites. An electrode placed on the left earlobe served as a reference. Horizontal eye movements were recorded using the right and left external canthi electrodes. The impedance of electrodes was below 5 ki I and matched within 1 ki'l The contribution of vertical electrooculogram deflection to the scalp recorded EEG was removed using participant-unique weighing coefficients (Semlitsch, Anderer, Schuster, & Presslich, 1986). The EEG was acquired using a bandpass of DC to 40 Hz (Neuroscan, Inc., El Paso, New Mexico, EEG amplifiers: 24 db/octave low-pass slope, and 24 db/octave high-pass slope). The recording epoch was 924 ms with 100 ms prestimulus baseline, sampled at the rate of 256 samples per second. Single-trial epochs were stored on disk for further analysis. Single-trial epochs with voltages over ± 75 mv at any electrode site were excluded. Averages were constructed separately from the EEG epochs recorded to congruent and incongruent sentence endings and filtered using an 8-Hz low-pass and.01 high-pass filter to eliminate excessive noise. ERP Data Analysis Grand averages across participants and conditions were constructed to assist in identification of ERP components in each condition. In both groups, and in the two modalities, the N400 was a well-defined peak. As reported in the studies of normal participants (Holcomb, 1993; Hoicomb & Neville, 1990), the N400 in the auditory modality had an earlier onset and longer latency than in the visual modality. In order to evaluate group differences at different points in the temporal course of verbal processing, four peaks (N100, P200, N400, and P600) were selected for analysis. The N100 and P200 ERP components preceded the N400. Those components were measured to test whether the group differences observed in the N400 amplitude and latency were related to differences in those early components. The N100 peak amplitude and latency were measured from raw waveforms within the latency window of 60-130 ms in the auditory modality and within the latency window of 110-200 ms in the visual modality. The P200 peak amplitude and latency were measured within the latency window of 140-220 nis in the auditory modality and within the latency window of 200-300 ms in the visual modality. The N400 peak amplitude was measured as the most negative voltage within a 250- to 380-ms latency range in the auditory modality and within a 350- to 450-ms latency range in the visual modality. These latency windows spanned the range of N400 peaks in controls' and patients' ERPs. The P600 peak amplitude was measured as the most positive voltage within a 450 to 650-ms latency range in auditory modality in the congruent condition and in a 650 to 800-ms range in the incongruent condition; it was measured in a 500 to 700-ms range in the visual modality in both congruent and incongruent conditions. These latency windows were selected to span the range of component latencies across the two groups and two modalities. Results N100 and P200 In the visual modality, the N100 and P200 were measured at parietal locations (Ol, Oz, O2), and in the auditory modality the components were measured at Fz, Cz, F3/4, and C3/4. Given the temporal and scalp distribution differences between auditory and visual modalities in the two components, the analyses were conducted separately for the two modalities. All statistical analyses reported here are based on peak amplitude measurements. The N100 and P200 amplitude and latency data were subjected to four overall mixed-model, multiple-level multivariate analyses of variance (MANOVAs), with group as a betweensubjects factor (2), and condition (2) and electrode sites as within-subject factors. No significant group differences were found for the amplitude and latency of the two components in either of the two modalities. 00 Three sets of electrodes were selected to provide the most comprehensive analysis of the differences between the two groups of participants in the two modalities. The midline chain (Fz, Cz, Pz), the lateral chain, spanning from the front to the back of the head (FP1/2, F3/4, C3/4, P3/4, and O1/2), and the temporal chain (T3/4, T5/6) were selected for analysis. The amplitude and latency from these electrode sites were subjected to an overall, multiple-level MANOVA with diagnosis (Nc and Sz) as a between-groups factor and modality (2), condition (2), and electrode (number of levels dependent on the electrode chain analyzed) as within-group factors. Amplitude Midline chain. As seen in Figures 1-4, at midline, a more negative amplitude was found in patients with schizophrenia than in normal controls as indicated by the main effect of group, F(22, 1) = 4.2, /; <.05. As Figures 1-4 illustrate, in both groups, the amplitude was more negative in the auditory modality than in the visual modality, F(22, 1) - 23.4, p <.0001, and in the incongruent than in the congruent conditions, F(22, 1) = 64.4, p <.0001. There was also a significant three-way interaction among group, modality, and electrode site, F(44, 2) = 4.6, p <.02. In relation to normal controls, patients with schizophrenia showed enhanced negativity for frontal electrodes in the auditory modality and for parietal electrodes in the visual modality (see Figures 1 and 3). To better understand the differences in the group effect between the two modalities, the overall MANOVA was followed by a mixed-model MANOVA for each modality, with group as a between-groups factor, and condition (2) and electrode site (3) as within-group factors. In the visual

LANGUAGE IN SCHIZOPHRENIA 89 VISUAL CONGRUENT CONDITION Fz 1-100 100 300 500 700 900-100 100 300 500 700 9CO -100 100 300 500 700 90! MILLISECONDS Figure 1. Grand average waveforms of normal control and schizophrenic participants to final congruent word completions in visual modality. modality, there was a significant interaction between group and electrode site, F(22, 2) = 5.3, p <.008. As shown in Figures 1 and 2, the largest group difference was observed at Pz, (t(22) =.05, whereas the two groups did not differ at Cz, /(22) =.1, orfz, ((23) =.3. In the auditory modality, the group differences existed at three electrode sites, resulting in the main effect of group, F(22, 2) = 6.7, p <.02 (see Figures 3 and 4). Lateral chain. As shown in Figures 1-4 at the lateral chain, more negative amplitudes were recorded in the auditory modality than in the visual modality in both groups, F(22, 1) = 21.5, p <.0001, and they were more negative in the incongruent than in the congruent condition, F(22, 1) = 33.7, p <.0001. There was also a significant interaction between modality and electrode, F( 198, 9) = 30.5, p <.001, and a significant three-way interaction between group, condition, and electrode site, F( 198, 9) = 2.3, p <.02. Planned comparison t tests (two-tailed) were conducted to better understand the source of this three-way interaction, and specifically the possibility that N400 amplitude was more negative in the congruent condition in the schizophrenia patients' waveform relative to that found in normal controls. These analyses suggested that the N400 was in fact more negative in patients at P3 and P4 in the auditory congruent condition (a trend toward a group difference at P3), ((20) =.08, and P4, ((21) =.08. The tendency for a larger N400 in the congruent condition in patients was also evident in comparing the differences between the N400 amplitude in the congruent and incongruent conditions within the same group. The N400 amplitude in the congruent and incongruent conditions was not significantly different at P3 and P01, ((12) =.1, and at Ol, f(12) =.09, in patients, whereas in normal controls the N400 amplitude was significantly more negative in the incongruent relative to the congruent condition for all lateral electrodes. Temporal chain. No significant group differences were found in the temporal electrode chain. In the auditory modality,

90 NIZNIKIEW1CZ ET AL. VISUAL INCONGRUENT CONDITION ft F4-100 300 500 900 300 500 700 MILLISECONDS 900-100 100 300 500 700 Figure 2. Grand average waveforms of normal control and schizophrenic participants recorded to final, incongruent sentence completions in visual modality. congruent condition, the N400 amplitude for T3, T4, T5, and T6, was, respectively,.06, -.7, -.3, and -.8 JJV in normal controls and.2,.1, 1.4, and 1.9 fn in patients. In the incongruent condition, the amplitude was respectively, 1.6, -2.3, 2.1, and -2.3 fn in normal controls and 1.5, 3.1, -2.1, and 3.3 ^V in patients. In the visual modality, congruent condition, the N400 amplitude for the same electrode chain was, respectively, 1.5, 1.5,.9, and 1.1 //V in normal controls and.7,.6, 1.5, and 1.4 //V in patients; in the incongruent condition, the amplitude was respectively, -.4,.4,.4, and.5 ^V in normals, and -.003, -1.5, -2.1, and -2.1 [N in patients. Summary. The N400 amplitude in schizophrenia participants was more negative than in control participants to both congruent and incongruent sentence endings in the two modalities. Also, more negative N400 was recorded to congruent than to incongruent sentences in both groups. At the same time, the group differences interacted differently with specific factors at different electrode chains: At midline, group differences depended on modality, and at lateral sites, they depended on condition and electrode site. Latency Midline chain. At midline, there were significant main effects of group, F(22, 1) = 22.4, p <.0001; modality, F(22, 1) = 99.8, p <.0001; and condition, F(22, 1) = 35.4, p <.0001; and a significant interaction between group, modality, and electrode site, F(44, 2) = 3.2, p <.03. Longer latencies were found in patients with schizophrenia relative to normal participants. In both groups the latency peaked later in the visual than in the auditory modality and to incongruent than to the congruent sentence endings. Separate mixed-model MA- NOVAs on the N400 latency in the auditory and visual modalities, with group as a between-groups factor and condition and electrode site as within-group factors revealed that, as seen in Figures 1-4, schizophrenia participants had prolonged laten-

LANGUAGE IN SCHIZOPHRENIA 91 10 F3 AUDITORY CONGRUENT CONDITION Fz 8 10 K8 o 22 0-2 100 300 500 700 900-100 10O 3OO 500 700 9OO MILLISECONDS 100 300 500 700 900 Figure 3. Grand average waveforms of normal control and schizophrenic participants recorded to final, congruent sentence completions in auditory modality. cies across all electrodes in the visual modality, main effect of modality, F(23, 1) = 10.3, p <.004, whereas in the auditory modality there was a significant interaction between the group and the electrode site, F(44, 2) = 6.2, p <.004. The latency was prolonged in the patient group at Pz, f(23) =.03, but not Cz or Fz. Lateral chain. At the lateral chain, there were significant main effects of group, f (22, 1) = 20.5, p < 0001; modality, f (22,1) = 134.8, p <.0001; and condition, F(22, 1) = 43.8, p <.0001, and a significant three-way interaction between group, modality, and electrode site, F(198, 9) = 2.7, p <.006. As shown in Figures 1 4, longer latencies were recorded in schizophrenic participants than in normal participants in both modalities. In both groups longer latencies were recorded in the visual than in the auditory modality and in the incongruent than congruent sentence condition. In the auditory modality, the group differences were more prominent at parietal sites (P3 and P4), P3, r(22) =.01, and P4, r(22) =.05, than at centra] and frontal sites (F3/4, C3/4). Temporal chain. The same main significant effects were observed for the temporal chain as for the lateral chain: a main effect of group, F(22, 1) = 16.1, p <.001; modality, F(22, 1) = 122.1, p <.0001; and condition, f(22, 1) = 58.6,^ <.0001. Longer latencies were recorded in patients with schizophrenia relative to normal controls in both modalities and in both conditions. In both groups, longer latencies were found in the visual than in the auditory modality and in the congruent than in the incongruent condition. Summary. Across most of the electrode sites, N400 latency was longer in patients compared with control participants in both modalities, for both congruent and incongruent sentence endings. In both groups, the N400 latency in the visual modality was later than in the auditory modality, and latency was later to the incongruent than to the congruent sentence endings.

92 NIZNIKIEWICZ ET AL. AUDITORY INCONGRUENT CONDITION normal schizophreni a P600-100 100 300 500 700 900-100 100 300 500 MILLISECONDS Figure 4. Grand average waveforms of normal control and schizophrenic participants recorded to final, incongruent words in auditory modality. P600 The same three chains of electrodes selected for the analysis of N400 peak amplitude and latency were used to evaluate the P600 component. The amplitude and latency for these separate electrode chains were submitted to overall, mixed-model, multiple-level MANOVAs with group (diagnosis: Sz and Nc), modality (2), condition (2), and electrode (number of levels dependent on the number of electrodes in the chain) as within-group factors. Amplitude. In spite of substantial amplitude differences between the two groups on the visual inspection of the waveforms, with grand average waveforms in the schizophrenia patients less positive than in the normal controls within the P600 latency window, no significant group differences were found for the midline and lateral electrode chains. For the temporal chain, a significant three-way interaction was found between group, condition, and electrode site, F(66, 3) = 3.15, p <.03. The post hoc analyses revealed significant group differences in the visual modality, congruent condition, at T3,1(23) 2.27, p <.03, and T5, r(23) = 2.6, p <.02. At both electrode sites, the waveforms in schizophrenia participants showed lower amplitude than in normal controls. Latency. For the midline chain, a significant three-way interaction was found between group, modality, and condition, F( 22, 1) = 4.97, p <.04. P600 peak latency was longer in the visual modality, congruent condition, and in patients compared to control participants. For the lateral chain, there was a suggestion of a three-way interaction between group, condition, and electrode, F(198, 9) = 1.69, p <.09. but no significant group effects were found. No significant group differences were found for the temporal chain. Discussion In this study of ERPs elicited during sentence processing, four peaks (N100, P200, N400, and P600) were analyzed. Significant group differences that spanned several electrode sites were found for the N400, a component associated with language operations. As demonstrated by the main effect of group, at the

LANGUAGE IN SCHIZOPHRENIA 93 midline, more negative N400 amplitude was found in schizophrenia patients than in normal controls in both visual and auditory modality to both the congruent and incongruent sentence endings. In contrast, cognitive processes documented in this study were not associated with abnormal earlier (N100 and P200) and later (P600) components. The absence of significant group differences in the latency and amplitude of the N100 and P200 indicates that language processing in schizophrenia is not compromised by dysfunctional early sensory processes. This result should not be taken to suggest that sensory processes in schizophrenia remain intact. Rather, it suggests that language dysfunction in schizophrenia is not related to sensory aspects of language processing. Also, in spite of apparent group differences in P600 on visual inspection of the waveforms, statistical analyses revealed few group differences in the amplitude and latency of this late, positive component. These were confined to the visual, congruent condition. As mentioned in the introduction, the P600 recorded in this paradigm probably does not index the same processes as the P300 recorded in the odd-ball paradigms designed to study aspects of sensory memory update. Therefore, the lack of group differences in this component should be considered as a finding specific to sentence processing rather than in relation to the reports of diminished P300 amplitude in the oddball paradigms. The results of the study helped to answer specific questions formulated at its outset: (a) Do language processing difficulties in patients with schizophrenia, previously documented by the N400 in the visual modality, also exist in the auditory modality? and (b) Do patients with schizophrenia show more negative N400 amplitude to both congruent and incongruent sentences? Similar, although not identical, group differences were found in both the visual and auditory modalities, demonstrating, for the first time, that language dysfunction documented by the ERPs in the visual modality also exists in the auditory modality. This result suggests that language difficulties are related to an amodal language system affected by schizophrenia. More negative N400 amplitude and prolonged latency was found in patients with schizophrenia than in normal controls in both auditory and visual modality. Furthermore, more negative N400 amplitude was recorded to both congruent and incongruent sentence endings in schizophrenia participants relative to normal controls. As seen in Figures 1 4, in the normal group, the waveform to congruent endings was characterized by the absence of (or a small) N400-like deflection, and in patients with schizophrenia, a substantial N400-like deflection was found to be both in congruent and incongruent sentence endings. In the auditory modality, there was a suggestion of significant group differences in the congruent condition. In the majority of the previous studies, a diminished N400 difference amplitude was reported. As mentioned in the introduction, such a result might be expected if both a waveform recorded to congruent and incongruent sentences included a negative-going component within the N400 latency range. Given that raw (unsubtracted) waveforms were not analyzed in these studies, it is difficult to form a final judgment on the matter. The use of raw waveforms in this study highlighted one of the possible causes for language dysfunction in schizophrenia. Also, the analysis of the late positive component in the 600- ms latency range underscored the fact that the group amplitude effect was largely confined to the area of the waveform within the 400-ms latency range, as only few sites (T3 and T5 in the visual, congruent condition) showed statistically significant group differences. A possibility remains that the group amplitude differences observed might be due to a slow negative component that develops around 200 ms poststimulus and spans the remainder of the recording epoch. Future studies should examine this possibility. The N400 recorded to target sentences presented one word at a time at long ISIs is especially sensitive to processes of context integration. In this study, both groups were exposed to target words preceded by identical context. Thus, the enhanced N400 amplitude to congruent and incongruent sentence endings observed in schizophrenia participants might indicate that context was used differently in the two participant groups. At the same time, one cannot make an argument that the context was simply not available to the schizophrenia group. The ERP data were measured only from patients whose accuracy was above 75% and whose performance did not differ significantly from the accuracy of their matched controls. Also, the N400 amplitude differentiated between the congruent and incongruent sentences. There are a number of possibilities why context was not used efficiently by schizophrenia patients. It has been suggested that an N400 amplitude reflects the extent of search through the mental lexicon; the more extensive the search, the larger the N400 amplitude. It also has been suggested that N400 amplitude reflects the difficulty of integrating a word into a sentence context. Thus, the larger N400 observed in patients with schizophrenia to both congruent and incongruent sentence endings might reflect either (a) the more extensive search through a larger number of word candidates lo fit them into the context or (b) the greater difficulty of fitting a terminal word into a context that is not readily available to patients with schizophrenia. Distinguishing between these mechanisms will be an important issue in future psycholinguistic studies. The concept of greater difficulty in utilizing the context is also supported by the N400 latency differences. The N400 latency was prolonged in patients with schizophrenia to both congruent and incongruent sentence endings and in both modalities. This effect was quite robust (an average of 21 ms latency difference in the auditory congruent, 12 ms difference in the auditory incongruent condition, and 31 ms difference in the visual modality). Assuming that latency indexes the speed of the cognitive processes, the later N400 latency in patients relative to controls may very likely indicate greater difficulty in performing these operations. This result corroborates the findings from earlier studies in which the N400 latency prolongation was consistently reported. Finally, the longer N400 latency in the absence of the delay in the earlier components helps to clarify the issue of slower RTs in patients with schizophrenia. The prolonged N400 latency suggests that the slower KIs in patients are partially due to slower lexical operations. The slowing of motor responses may further contribute to the RT effects reported in schizophrenia patients. Taken together, these results suggest that one aspect of language dysfunction in schizophrenia may be related to deficient context integration processes. Because context integration is a complex operation, further studies are needed to clarify which aspects of this operation are affected in schizophrenia. Considered in conjunction with the results of behavioral studies dis-

94 NIZNIKIEWICZ ET AL. cussed earlier, the present study allows constructing a preliminary model of language dysfunction in schizophrenia. The loose associations, tangentiality, and derailment observed in schizophrenia may be related both to overactivation of the lexical networks and to inefficient context integration. At present, these different possibilities are somewhat speculative. Future studies, reaching across experimental paradigms and approaches and integrating the evidence from different research areas, need to be conducted to further distinguish between the factors contributing to language dysfunction in patients with schizophrenia. References Adams, J., Faux, S. R, Nestor, P. G., Shenton, M. E., Marcy, B., Smith, S., & McCarley, R. W. (1993). ERP abnormalities during semantic processing in schizophrenia. Schizophrenia Research, JO, 247-257. American Psychiatric Association. (1987). Diagnostic and statistical manual of mental disorders (3rd edition, revised). Washington, DC; Author. Anderson, J. R. (1983). A spreading activation theory of memory. Journal of Verbal Behavior, 22, 261-295. Andreasen, N. C. (1979). Thought, language, and communication disorders. I. Clinical assessment, definition of terms, and evaluation of their reliability. General Archives of Psychiatry, 36, 1315-1321. Bleuler, E. (1950). Dementia praecox or the group of schizophrenias. New \brk: International Universities Press. (Original work published 1911) Bruder, G. E., Tenke, C. E., Rabinowicz, E., Towe, G. P., Malaspina, D., Amadoi; X., Kaufman, G. A., & Gorman, G. (1996). Elec trophy siological studies of brain activities in schizophrenia. In Schizophrenia: New directions for clinical research and treatment (pp. 17-33). New York: Mar Jeanne-Li ebert. Chapman, L. J., Chapman, J. P., & Daut, R. (1976). Schizophrenic inability to disattend from strong aspects of meaning. Journal of Abnormal Psychology, 85, 35-40. Chapman, L. J., Chapman, J. R, & Miller, G. A. (1964). A theory of verbal behavior in schizophrenia. In B. A. Maher (Ed.), Progress in experimental personality research (Vol. 1, pp. 49-77). New York: Academic Press. Donchin, E., & Coles, M. G. {1988). Is the P300 component a manifestation of context updating? Behavioral and Brain Sciences, II, 357-374. Fischler, L, & Raney, G. E. (1989). Language by eye: Behavioral, autonomic and cortical approaches to reading. In J. R. Jennings & M. G. Coles (Eds.), Handbook of cognitive psychology: Central and autonomous nervous system (pp. 511-577). New "\brk: Wiley. Ford, J. M., White, P. M., Csernasky, J. G., Faustman, W. O., Roth, W. T.. & Pfefferbaum, A. (1994). ERPs in schizophrenia: Effects of anti-psychotic medication. Biological Psychiatry, 36,, 153-170. Friedman, D., Simson, A., Ritter, W., Rapin, Y. (1975). The late positive component (P300) and information processing in sentences. Electroencephalography and Clinical Neuropsychology, 31, 255-262. Grillon, C., Ameli. R., & Glazer, W. M. (1991). N400 and semantic categorization in schizophrenia. Biological Psychology, 29, 467-480. Harvey, P. D. (1983). Speech competence in manic and schizophrenic psychoses: The association between clinically rated thought disorder and cohesion and reference performance. Journal of Abnormal Psychology, 92, 368-377. Holcomb, P. (1993). Semantic priming and stimulus degradation: Implications for the role of the N400 in language processing. Psychophy.iiology,.50, 47-61. Hnlcomb, P. J., Coffey, S., & Neville, H. J. (1992). Visual and auditory sentence processing: A developmental analysis using event-related brain potentials. Developmental Neuropsychology, 8, 203-241. Holcomb, P J., & Neville, H. J. (1990). Auditory and visual semantic priming in lexical decision: A comparison using evoked potentials. Language and Cognitive Processes, 5, 281 312. Koyama, S., Nageishi, Y, Shimokochi, M., Hokama, H., Myiazato, Y,, Miyatani, M., & Ogura, C. (1991). The N400 component of eventrelated potentials in schizophrenic patients: A preliminary study. Electroencephalography and Clinical Neurophysiology, 78, 124-132. Kramer, A., &. Donchin, E. (1987). Brain potentials as indices of orthographic and phonological interaction during word matching. Journal of Experimental Psychology: Learning, Memory, and Cognition, 13, 76-86. Kucera, H., & Francis, W. N. (1967). Computational analysis of the present day American English. Providence, RI: Brown University Press. Kutas, M., & Hillyard, S. A. (1980). Reading senseless sentences: Brain potentials reflect semantic incongruity. Science, 207, 203-205. Kutas, M., & Hillyard, S. A. (1989). An electrophysiological probe of incidental semantic association. Journal of Cognitive Science, }, 38-49. Kwapil, T., Hegley, D. C., Chapman, L. J., & Chapman, J. P. (1990). Facilitation of word recognition by semantic priming in schizophrenia. Journal of Abnormal Psychology, 99, 215 221. Maher, B. A. (1991). Language and schizophrenia. In R. Steinhauer, J. H. Gruzelier, & J. Zubin (Eds.), Handbook of schizophrenia: Neuropsychology. psychophysiology and information processing (Vol., pp. 437-463). New York: Elsevier Science. Manschreck, T, Maher, B., Milavetz, J. J., Ames, D., Wesstein, C. C., & Schneyer, M. (1988). Semantic priming in thought disordered schizophrenic patients. Schizophrenia Research, I, 61-66. McCarley, R. W., Shenton, M. E., O'Donnell, B. E, & Nestor, P. G. (1993). Uniting Kreapelin and Bleuler: The psychology of schizophrenia and the biology of temporal lobe abnormalities. Harvard Review of Psychiatry, I, 36 56. McCarthy, G., & Donchin, E, (1981). A metric of thought: A comparison of P300 latency and reaction time. Science, 211, 77-80. Mitchell, P E, Andrews, S., Fox, A,, Catts, P. B., & McConghy, N. (1991). Active and passive attention in schizophrenia: An ERP study of information processing in a linguistic task. Journal of Abnormal Psychology, 32, 101-124. Morton, J. (1969). Interaction of information processing in word recognition. Psychological Review, 76, 163-178. Neely, J. H. (1991). Semantic priming effects in visual word recognition: A selective review of current findings and theories. In D. Besner & R, Humphreys (Eds.), Basic processes in reading: Visual word recognition (pp. 264-336). Hillsdale, NJ: Erlhaum. Osterhrmt, L. t & Holcomb, P. (1992). Event related brain potential elicited by syntactic anomaly. Journal of Memory and Language, 31, 1-22, Rochester, S. R., & Martin, J. R. (1979). Crazy Talk: A study of the discourse of schizophrenic speakers. New York: Plenum. Semlitsch, H. V, Anderer, P., Schuster, P., & Presslich, O. (1986). A solution for reliable and valid reduction of ocular artifacts applied to the P300 ERP. Psychophysiology, 23, 695-703. Spitzer, M., Weisken I, Winter, M., Maier, S., Hermle, L., & Maher, B. (1994). Semantic and phonological priming in schizophrenia. Journal of Abnormal Psychology, 3, 485-495. Van-Petten, C., & Kutas, M. (1990). Interactions between sentence context and word frequency in event-related brain potentials. Memory and Cognition, 18, 380-393. Received June 19, 1995 Revision received March 28, 1996 Accepted June 14, 1996