Semantic context-processing deficit in thought-disordered ...

COGNITIVE NEUROPSYCHIATRY, 2003, 8 (3), 173±189

Semantic context-processing deficit in thoughtdisordered schizophrenic patients: Evidence from new semantic priming paradigms Chrystel Besche-Richard UniversiteÂ de Bourgogne, Dijon, France

Christine Passerieux Centre hospitalier de Versailles, Le Chesnay Cedex, France Introduction. Disorders in the processing of the semantic context are now a wellestablished phenomenon in thought-disordered (TD) schizophrenic patients, and have been revealed especially well by studies that have made use of the experimental paradigm of lexical decision tasks coupled with semantic priming. The main question addressed by this study was the evaluation of the experimental conditions under which TD schizophrenic patients are able to deploy cognitive strategies for semantic context processing. Methods. We studied semantic priming in two double lexical decision tasks (i.e., involving the explicit processing of the prime word) using a sequential presentation of words (stimulus onset asynchrony; SOA 500 ms) with a different proportion of related words (Experiment 1 with 25% vs. Experiment 2 with 15%) in 15 TD schizophrenic and 15 normal participants. Results. The results showed no significant differences between the semantic priming of TD schizophrenic and normal participants for Experiment 1, unlike Experiment 2 in which we observed a significant reduction of the amplitude of semantic priming in TD schizophrenic patients. The results of Experiment 1 contrast with those obtained previously (Besche et al., 1997) using a classical lexical decision task (implicit processing of the prime word) which also contained 25% related words. Conclusions. Experimental variables, such as the cognitive processing required by the prime word or the proportion of related words or the way they are manipulated, all seem to influence the emergence of semantic priming abnormalities in TD schizophrenic patients. Correspondence should be addressed to Chrystel Besche-Richard, UniversiteÂ de Bourgogne, PoÃle AAFE, Esplanade Erasme, e-mail: [email protected] We are grateful to Marie-Christine Hardy-BayleÂ, Juan Segui, Jean-Paul Laurent, Sophie Kecskemeti, and GeÂrald Mesure for their assistance during this research. This paper is based on a doctoral dissertation (Chrystel Besche) conducted in the Equipe de Recherche en Psychologie Clinique (ERPC) (Director, Professor Alain Blanchet), UniversiteÂ Paris VIII, Saint-Denis, France. # 2003 Psychology Press Ltd http://www.tandf.co.uk/journals/pp/13546805.html DOI:10.1080/13546800244000274

174

BESCHE-RICHARD AND PASSERIEUX

Cognitive disorders in people diagnosed with schizophrenia have been described for many years (for a synthesis, see Frith, 1992). These cognitive abnormalities have been observed in several fields: attentional capacities, executive functions, language abilities, memory performances, communication, and theory-of-mind competencies. One cognitive difficulty, for schizophrenic patients, is to process the contextual linguistic information (Kuperberg, Mc Guire, & David, 1998). These abnormalities seem to be related to thought and language disorders in schizophrenic patients as Kuperberg et al. (1998) have shown. The most commonly used experimental paradigms for the evaluation of context processing consist of lexical decision tasks or pronunciation tasks with semantic priming. Using different experimental paradigms, research into word recognition in people diagnosed with schizophrenia has given rise to a large volume of data. In a lexical decision task, subjects are presented with letter string targets and have to decide whether each is a word or not. In pronunciation tasks, they have to pronounce the target. Normal subjects more rapidly identify that a target word like nurse is a word when it is immediately preceded by an associatively related word (the context word), such as doctor, rather by an unrelated word like bread (Meyer & Schvanevedt, 1971). This priming effect has been interpreted in the framework of attentional theories as the result of two levels of processes: an automatic process, the spread of activation in the semantic network, and at least two controlled or strategic processes: expectancy and semantic matching (de Groot, 1984; de Groot, Thomassen, & Hudson, 1986; Neely, 1977; Seidenberg, Waters, Sanders, & Langer, 1984; West & Stanovich, 1982). The presentation of a word activates the corresponding word node in semantic memory and, by spreading activation into this network, activates words that are related to the word's meaning. As a consequence, these words need less time for subsequent processing. This mechanism is automatic, fast acting, of short duration, requires no attentional capacity (Marcel, 1983), and is more predominant in pronunciation tasks than in lexical decision tasks. The second mechanism is expectancy-induced priming (Becker, 1980; Neely, 1976, 1977). This is a prediction strategy in which subjects use the information provided by the prime to generate an expectancy set for related target words. If the target is included in the set, it will be recognised more quickly. Other authors have proposed a different controlled process which they have termed ``postlexical semantic matching'' (de Groot, 1984; de Groot et al., 1986, Seidenberg et al., 1984; West & Stanovich, 1982): Lexical decision times are faster for related words than unrelated words because, if there is a semantic relation between prime and target, this indicates that the second string of letters is a real word (de Groot, 1984; Lorch, Balota, & Stamm, 1986; Shelton & Martin, 1992). These effects increased in proportion to the number of semantically related pairs and the percentage of nonwords (de Groot, 1984; den Heyer, 1985; den Heyer, Briand, & Dannenbring, 1983; Neely, 1991; Neely, Keefe, &

SEMANTIC PRIMING IN SCHIZOPHRENIA

175

Ross, 1989; Tweedy, Lapinski, & Schvaneveldt, 1977). This semantic matching is thought to be involved in postlexical semantic integration (Neely, 1991). Data from lexical decision or pronunciation tasks in schizophrenic patients has remained controversial. Hypotheses of the information-processing deficit have involved the two types of processes: the automatic spread of activation, on the one hand, and the controlled processes, and especially semantic matching, on the other. The hypothesis of an automatic hyperactivation of semantically related words and a slower decrease in this hyperactivation is supported by findings of a hyperpriming in schizophrenic individuals (Kwapil, Hegley, Chapman, & Chapman, 1990; Manschreck et al., 1988; Moritz et al., 2001; Spitzer, 1993; Spitzer, Braun, Hermle, & Maier, 1993a; Spitzer, Braun, Maier, Hermle, & Maher, 1993b; Spitzer et al., 1994; Weisbrod, Maier, Harig, Himmelsbach, & Spitzer, 1998). However, this hypothesis has been contested on the basis of studies that directly test automatic processes in verbal information processing and provide evidence of their integrity: Barch et al. (1996); Ober, Vinogradov, and Shenaut (1995); and Vinogradov, Ober, and Shenaut (1992) found normal priming effects in schizophrenic patients completing pronunciation tasks; on the other hand, Aloia et al. (1998), using pronunciation tasks in thought-disordered (TD) schizophrenic patients, found a reduction in their semantic priming. Henik, Priel, and Umansky (1992) reported the preservation of repetition priming in lexical decision task which they considered to be an automatic phenomenon. In addition, Henik, Nissinov, Priel, and Umansky (1995) observed a hyperpriming in schizophrenic patients completing lexical decision task and a reduction of this priming when the processing of a distracter stimulus was introduced or when they introduced a cognitive load. These authors suggested that hyperpriming might be an evasive phenomenon due, like the reduction of priming, to a deficit in controlled processes. The alternative hypothesis of a dysfunction in the integrative and controlled processing of the semantic information is supported by observations of a reduction in priming effects in schizophrenic patients (Besche et al., 1997; Henik et al., 1992, 1995; Ober et al., 1995; Ober, Vinogradov, & Shenaut, 1997; Passerieux, Hardy-BayleÂ, & WidloÈcher, 1995; Passerieux et al., 1997; Vinogradov et al., 1992). However, other authors have found priming effects in these patients that are equal to normal values (Blum & Freides, 1995; Chapin, McCown, Vann, Kenney, & Youssef, 1992; Chapin, Vann, Lycaki, Josef, & Meyendorff, 1989). The first explanation of this heterogeneity of data evoked the cognitive and clinical heterogeneity of schizophrenic individuals. The best supported cognitive-clinical correlation is between priming effect abnormality and thought disorders (Aloia et al., 1998; Besche et al., 1997; Kwapil et al., 1990; Henik et al., 1992, 1995, Passerieux et al., 1995, 1997; Spitzer, 1993; Spitzer et al., 1993a, 1993b, 1994; Manschreck et al., 1988; Moritz et al., 2001). Only one

176


study performed using thought-disordered schizophrenics failed to find any reduction or increase in semantic priming (Blum & Freides, 1995). This is why the current study focuses on a clinical population of TD schizophrenic patients. The second level of explanation relates to the experimental factors in lexical decision tasks: Are there any experimental factors affecting the informationprocessing modalities that might help schizophrenic patients reduce their cognitive deficit and thus perform normally? We have already seen that, in normal subjects, the proportion of semantically related pairs1 influences the priming effect. This factor is probably crucial for explaining findings about performances in schizophrenia: In the literature, we have noted that perturbations of semantic priming in schizophrenic patients have been reported in studies that used a low proportion of semantically related pairs (less than or equal to 20%) (Henik et al., 1992; Manschreck et al., 1988; Ober et al., 1995, 1997; Passerieux et al., 1995, 1997; Vinogradov et al., 1992; Weisbrod et al., 1998). In contrast, a higher proportion of semantically related pairs (greater than 20%) would allow schizophrenic individuals to use the semantic relationships (Barch et al., 1996; Blum & Freides, 1995; Chapin et al., 1989, 1992). However, this variable probably interacts with the presence of thought disorders in schizophrenic patients. As we shall present later in greater detail, we have shown that, in a lexical decision task with a relatively high proportion of related words (25%), TD schizophrenic patients, unlike normal participants, non-TD schizophrenic patients and depressed patients, exhibit a significant reduction in the semantic priming effect (Besche et al., 1997). Blum and Freides (1995), using a similar paradigm, did not find any abnormalities in semantic priming in schizophrenic patients with less pronounced thought disorder than exhibited by the participants in our study. The third level of explanation relates to the nature of the experimental tasks. We should note that, with the exception of the study conducted by Aloia et al. (1998), studies involving pronunciation tasks have revealed no abnormalities in semantic priming in schizophrenic individuals (Barch et al., 1996; Ober et al., 1995; Vinogradov et al., 1992). In the same way, when a double lexical decision task with simultaneous presentation is used, schizophrenic patients exhibit normal semantic priming (Chapin et al., 1989, 1992). But, two differences are clear between sequential lexical decision tasks (classic paradigm) and double lexical decision tasks with simultaneous presentation of letter strings. First, the presentation of the stimuli: sequential in the classic paradigm and simultaneous in the double lexical decision used by Chapin et al. (1989, 1992). Second, the nature of the cognitive processing of the prime word required by the task instructions: In the classic paradigm, the participants

1 To calculate the proportion of semantically related words: (Number of semantically related pairs of words/Total number of pairs of words) 6 100.


177

were asked to decide if only the second string of letters (the target) was a word. In the double lexical decision task, participants have to decide if the two strings of letters (the prime word and the target words) are real words or not, and thus process the prime word explicitly. Consequently, in double lexical decision tasks participants find it easier to process the relation between the two words than they do in the classic paradigm. We supposed that the instructions in the double lexical decision task would improve the performances of schizophrenic patients. One of the aims of this study is to test this hypothesis in a first experiment (Experiment 1). We therefore chose to use a new paradigm which we named the double sequential lexical decision task. In this task, two string of letters are presented sequentially and participants have to decide if the prime and target are words or not after the presentation of target. This type of paradigm makes it possible to elicit the explicit processing of the prime word without confounding this factor with the simultaneous presentation of the two words. The proportion of semantically related pairs of words was 25%. Our hypothesis was that the explicit processing of the prime word required by the task instructions should allow the mobilisation of semantic integration processes in schizophrenic patients. To support this hypothesis, we shall report the results of a published study involving normal participants and TD schizophrenic patients (Besche et al., 1997) which used the same lexical decision task (same verbal material, same SOA) but that utilised standard instructions requiring the implicit processing of the prime word. As the literature suggests, the second question addressed in this study concerns the effect of the proportion of related words on the performances of schizophrenic patients. We dealt with this question by manipulating the proportion of related words present in the lexical decision tasks. Thus, in order to evaluate the effect of the variation of the proportion of related words, we performed a second double sequential lexical decision task with a low proportion of related words (15%). If our observation, based on the literature, concerning the role of this experimental factor is correct, we should again obtain the classic response in TD schizophrenic patients: A reduction in priming effects. Finally, it is necessary to define the type of schizophrenic individuals involved in these two experiments: This was a schizophrenic subgroup in which context processing deficits have been regularly observed (i.e., symptomatic schizophrenic patients with formal thought disorder).

METHODS Participants We analysed the results of 15 schizophrenic patients and 15 normal controls. All the participants were native-French speakers and gave their informed consent. They all had a normal vocabulary level (Vocabulary Test; Binois & Pichot 1947). The schizophrenic participants were recruited according to Diagnostic

178


and Statistical Manual of Mental Disorders-Revised (DSM-III-R, American Psychiatric Association; APA, 1987) criteria. None of the patients presented any neurological disease, alcohol or drug abuse and none had had electroconvulsive therapy within the last six months, but all were on neuroleptic treatment. They were recruited from a number of different psychiatric departments. There were no differences on sociocultural variables (sex, age, socioeducational level, and vocabulary level) between the schizophrenic and normal participants. The sociocultural data and their statistical comparisons are presented in Table 1. The schizophrenic patients were clinically evaluated by an independent psychiatric clinician using two clinical scales. First, the Thought, Language and Communication Scale (TLC; Andreasen, 1979) which evaluates disorders of language, thought, and communication. A high total score ( 7) reflects thought disorders in patients. All our patients had a total score over 7. We evaluated the psychotic symptoms using the Positive And Negative Symptoms Scale (PANSS; Kay, Fiszbein, & Opler, 1987) and distinguished between the scores on the positive and the negative scales. The clinical variables are presented in Table 2.

Materials We used two lexical decision tasks with semantic priming: the first one, with a high proportion of related words (25%), used the same verbal material as the experiment published in Besche et al. (1997). Only the instructions were different: The experiment took the form of a double lexical decision task with sequential presentation of letter strings. In Experiment 2, we again used a double lexical decision task with sequential presentation of letter strings but with a low proportion of related words (15%). These two tasks were matched for word

TABLE 1 Demographic characteristics: means (and standard deviations) of groups and statistical analyses between controls and schizophrenic patients. Controls (N = 15) M Sex Age Years of education Vocabulary level a

12 31.5 11.9 25.6

(SD) Ð (7.1) (2.3) (3.7)

Schizophrenics (N = 15) M 11 34.9 12.2 25.8

(SD) Ð (8.3) (2.6) (3.8)

Chi-square test and Student test for independent group.

Statistical testsa 2

w = 0.19 y = 71.2 t = 70.29 t = 70.14

df

Significance

1, 28 1, 28 1, 28 1, 28

n.s. n.s. n.s. n.s.


179

TABLE 2 Clinical variables: means (and standard deviations) for the schizophrenic group

TLC scale M Schizophrenic 19.2 patients a

PANSS, total scale

PANSS, positive scale

PANSS, negative scale

Time course of Equiv. illness (yrs) chlorpromazinea

(SD)

M

(SD)

M

(SD)

M

(SD)

M

(SD)

M

(SD)

6.6

80.8

17.2

18.5

5.1

23.9

6.2

9.9

6.7

830

544

Equivalents of chlorpromazine in milligrams/day.

frequency (Centre National de la Recherche Scientifique' CNRS, TreÂsor de la Langue francËaise, 1971) and number of letters. Experiment 1: Double lexical decision task with sequential presentation and a high proportion of related words (25%). This task was consisted of 20 pairs of related words (e.g. mother-father), 20 pairs of unrelated words (e.g., oarmoon) and 40 pairs with a nonword (13 pairs of word-nonword, 13 pairs of nonword-word, and 14 pairs of nonword-nonword). Half of the correct responses were yes and half no, and there were 66% words and 34% nonwords. For the related condition, there were synonyms, antonyms or category-related stimuli. The nonwords were constructed from words other than those used in the related and unrelated conditions and respected the orthographical and phonological rules of the French language. In order to counterbalance the presentations there were two lists of stimuli, with the result that each target word appeared in different list and for different subjects in the related and unrelated conditions. The word frequency and the number of words were identical for the two lists. Word frequency was equal to 242 occurrences per 1 million, and the number of letters varied from 3 to 8. The target word order was fixed in each list and every target word was associated with a related prime in one list and with an unrelated prime in the other list. The item lists were randomised when we constructed the lists of stimuli. The task started with practice stimuli (N = 61), and each list started with 5 noninterpreted pairs of letter strings. For the validation of the task, 40 control participants were included and their results were compared for the two lists of trials. However, only 15 matched controls were compared with the schizophrenic participants. In the case of reaction times, neither the main effect for lists, F(1, 38) = 1.12, p < .29; nor the Group 6 List interaction for semantic priming were significant, F < 1. The results were the same for error rates: neither the main effect for lists, F(1, 38) = 1.3, p < .26; nor the Group 6 List interaction for semantic priming were

180


significant, F < 1. These results discount the existence of a list effect for the results obtained using the main analyses (ANOVA). Experiment 2: Double lexical decision task with sequential presentation and a low proportion of related words (15%). The same design was used in the double lexical decision task with sequential presentation and a low proportion of words. The pairs of presented words corresponded to a semantic sequence. This task consisted of 134 pairs of letter strings with 20 pairs of related words (e.g. summer-winter), 20 pairs of unrelated words (e.g., triangle-cloth), 27 filler items corresponding to unrelated pairs of words which were not interpreted in the results, and 67 pairs with a nonword subdivided into 22 word-nonword pairs, 22 nonword-word pairs and 23 nonword-nonword pairs (i.e., 66% words and 34% nonwords). Half of the correct responses corresponded to yes and half of no. The presence of filler items decreases the proportion of semantically related pairs of words. As in Experiment 1, the nonwords were constructed from words other than those used in the prime-target pairs and respected French orthographical and phonological rules. Two lists of stimuli were constructed and were matched for word frequency (CNRS, TreÂsor de la Langue francËaise, 1971) and number of letters. In this task, word frequency was 234 occurrences per 1 million and the number of letters varied from 3 to 10. A counterbalanced design was used in which each word target appeared in a related or unrelated condition in different lists. As in Experiment 1, the order of the target words was fixed in each list and the words in one list were associated with related primes and, in the other list, with the same primes but in an unrelated condition. Each list started with five nonanalysed item pairs in order to avoid variations in reaction times. A practice list consisting of 46 different pairs preceded the experimental lists. As in Experiment 1, this task was validated through the participation of 40 control participants. However, only 15 matched controls were compared with the schizophrenic participants. In the case of reaction times, neither the main effect for lists, F < 1; nor the Group 6 List interaction for semantic priming were significant, F(1, 38) = 2.5, p < .12. The results were the same for error rates: neither the main effect for lists nor the Group 6 List interaction for semantic priming were significant, F < 1. These results discount the existence of a list effect for the results obtained using the main analyses (ANOVA). The two tasks used similar linguistic material with an equal percentage of words and nonwords, with equivalent word frequency and number of letters but no common word or nonword.

Procedure Apart from the change of instruction, in the two experiments presented here we followed the same procedure as in our initial study presented in Besche et al. (1997). We used a Macintosh Powerbook portable computer, model 165,


181

interfaced with a Macintosh 14 inch monitor for stimulus presentation and data collection. All the strings of letters appeared in lower case letters in 36 point Geneva font (7 mm in height). For each lexical decision task, we used a 500 msSOA and an intertrial interval of 1500 ms. Thus, the first string of letters was presented for 250 ms, followed by a 250 ms blank interval, and then by the presentation of the target. The subjects responded by pressing one of the two keys on the computer keyboard after the presentation of the target on the computer screen. The participants gave a yes response using their dominant hand and a no response using their nondominant hand. The target item stayed on the screen until the subject made a response. The participants sat in front of the computer and were given the following instructions by the experimenter: On this screen you're going to see strings of letters flashing on and off. They will come in pairs, one after the other. Sometimes, the strings of letters will be real French words, and sometimes they will be nonsense words. Your task is to decide whether or not the two strings of letters in each pair are really words. Press the yes key if and only if the two strings of letters are both words; press the no key if one of the two strings of letters is a nonword. You should try to do this as quickly and accurately as possible. Remember, you have to decide whether the two strings of letters are real words.

The participants gave their responses after the presentation of the prime and target stimuli. The reaction times (in milliseconds, ms) and error rates (in percentage, %) were the dependant variables in each task. Reaction times that differed from the mean by more than two standard deviations were discarded. Sixteen of the participants (eight schizophrenics and eight controls) took part in Experiment 1 first and 14 (seven schizophrenics and seven controls) of the participants took part in Experiment 2 first in order to counterbalance the experimental sequence.

Data analysis For each task, we conducted a repeated measures analysis of variance (ANOVA) both for average reaction times (RTs) for correct responses and for average error rates (ERs). Condition (related-unrelated) was the within-group factor and Group (controls, schizophrenic patients) was the between-group factor. Moreover, we calculated a percentage of gain2 in order to control for the reaction time effect on the size of the semantic priming effect.

2

To control the effect of reaction times on the amplitude of semantic priming, we calculated a percentage gain in the same way as Spitzer et al. (1993a.): 17(RTRC/RTUC) 6 100, where RTRC is the reaction time in the related condition and RTUC the reaction time in the unrelated condition.

182


TABLE 3 Reaction times (RTs), means (standard deviations), and error rates (ERs) for related and unrelated prime conditions in Experiments 1 and 2 Related condition RT M

(SD)

Unrelated condition

ER M

RT

ER

Semantic priming (ms)

(SD)

M

(SD)

M

(SD)

M

(SD)

1 1

2.8 2.1

62 37

13.2 19.9

2.6 4.9

60 11

10.7 22.1

Experiment 1 Normal controls Schizophrenic patients

486 608

72.8 1.33 111.4 0.67

2.96 1.76

548 645

88.7 117.6

Experiment 2 Normal controls Schizophrenic patients

489 719

75.2 1.01 279.8 2

2.1 4.1

549 730

96.6 2.03 255.4 3.3

Note: Reaction times are in milliseconds (ms) and represent main mean reaction times of correctly recognised words. Error rates are in percentages.

Results We present the mean reaction times and their standard deviations for each task in Table 3. Experiment 1. The repeated-measures ANOVA conducted on reaction times revealed a significant group effect, F(1, 28) = 10.31, p < .003 (the global reaction times of the controls were faster than those of the schizophrenic patients; 517 ms vs. 626.5 ms); a significant condition effect, F(1, 28) = 17.2, p < .0003 (the reactions in the related condition were faster than in the unrelated condition; 547 ms vs. 596.5 ms); but no significant interaction between these two factors, F(1, 28) = 1.07, p < .31; thus indicating that semantic priming was not significantly different between schizophrenics and controls. These results were confirmed by the analysis of the percentage of gain, F(1, 28) = 2.7, p < .11 (see Figure 1). We obtained no further information from the analyses of error rates (ERs). For ERs, there was no significant group effect, F < 1 (the global ERs were comparable between schizophrenics and controls: 0.835% vs. 1.165%), condition effect, F < 1 (the unrelated condition did not provoke any more errors than the related condition: 1% vs. 1%) and no interaction effect, F < 1. Experiment 2. In the case of reaction times, the repeated measures ANOVA showed a main significant group effect, F(1, 28) = 8.24, p < .008. This result means that schizophrenics have longer global reaction times than controls (724.5 ms vs 519 ms). We obtained a significant condition effect, F(1, 28) = 8.2, p