Psychiatry Research 97 Ž2000. 107᎐118
Prosody and lexical accuracy in flat affect schizophrenia Murray Alpert a,U , Stanley D. Rosenberg b, Enrique R. Pouget a , Richard J. Shaw c a
Department of Psychiatry, HN 323, New York Uni¨ ersity Medical Center, 550 First A¨ enue, New York, NY 10016, USA b Department of Psychiatry, Dartmouth Medical School, Hano¨ er, NH, USA c Department of Psychiatry and Beha¨ ioral Sciences, Stanford Uni¨ ersity School of Medicine, Stanford, CA, USA Received 4 November 1999; received in revised form 2 October 2000; accepted 4 October 2000
Abstract To test the hypothesis that flat affect in schizophrenia involves a motor-expressive deficiency, but not an emotional deficiency, we compared the acoustic properties of speech that are used to express emotion with the emotional content of the words. DSM-III-R schizophrenic patients were divided into flat Ž N s 20. and non-flat affect Ž N s 26. groups on the basis of rating-scale scores. Twenty normal subjects also were included. Subjects were recorded on audio tape as they described a happy and a sad experience for about 10 min. The recordings were analyzed acoustically for fluency and for two types of prosody: inflection and emphasis. Words from transcriptions of the recordings were sorted by content analysis software into psychologically meaningful categories; we compared ‘pleasure’ and ‘distress’ word categories. Patients with flat affect spoke with less inflection, and were less fluent. However, they were similar to the other groups in the rate at which they used ‘pleasure’ words to describe happy experiences and ‘distress’ words to describe sad experiences. The behavioral deficiency in flat affect appears to be restricted to reduced activity in communicative motor channels. Other aspects of emotion processing seem intact. 䊚 2000 Elsevier Science Ireland Ltd. All rights reserved. Keywords: Flat affect; Emotions; Prosody; Fluency; Motor expression.
U
Corresponding author. Tel.: q1-212-263-5716; fax: q1-212-263-7513. E-mail address:
[email protected] ŽM. Alpert.. 0165-1781r00r$ - see front matter 䊚 2000 Elsevier Science Ireland Ltd. All rights reserved. PII: S 0 1 6 5 - 1 7 8 1 Ž 0 0 . 0 0 2 3 1 - 6
108
M. Alpert et al. r Psychiatry Research 97 (2000) 107᎐118
1. Introduction Flat affect is a cardinal sign of the negative syndrome of schizophrenia. In flat affect, facial, vocal and gestural channels of display are reduced. However, patients with flat affect appear similar to non-flat patients in regard other aspects of emotion ŽKnight and Valner 1993; Dworkin et al., 1996.. For example, flat affect schizophrenic patients report appropriate subjective experiences ŽKrause et al., 1989; Berenbaum and Oltmanns, 1992; Kring et al., 1994; Sison et al., 1996.. Also, they respond to emotion-eliciting stimuli with a level of physiological activation that is comparable to that shown by controls ŽKring et al., 1994; Sison et al., 1996; Curtis et al., 1999.. In addition, schizophrenic patients with or without flat affect are similar in their ability to perceive the emotions of others ŽBorod et al., 1993; Shaw et al., 1999.. Emotion is a construct that integrates psychological and physiological components into qualitative patterns such as anger or fear. Flat affect reflects a reduction in the quantity of emotional expression ŽAmerican Psychiatric Association, 1987; Alpert and Rosen, 1990.. Experimental studies of flat affect have focused, mainly, on facial expression. Vocal expression or gestures have been less studied, while the relations between the lexical channel and flat affect have not been examined. The lexical channel ŽBorod, 1993. includes processes that underlie word selection during free speech, as the word that expresses the intent of the communication is chosen from among a number of words of slightly different meaning or coloring in the speaker’s lexicon. Lexical selection reflects preconscious activity that occurs late in the course of speech generation ŽLevelt, 1989.. Among other factors that contribute to word choice, the selector is guided by the emotional context to select words of appropriate nuance ŽSeegmiller and Epperson, 1987; Oxman et al., 1988.. In parallel with the lexical selector is a controller of speech prosody ŽLevelt, 1989.. Prosody is introduced by varying voice level, fundamental frequency and speech timing. Voice level is the physical stimulus for impressions of loudness. Variation in loudness across syllables is perceived
as emphasis. The speaker’s fundamental frequency is the stimulus for perception of pitch. Variation in pitch across syllables is perceived as inflection. This provides for the coding of nonlexical information in the sounds of speech. Affective prosody refers to the vocal broadcasting of the speaker’s physiological state ŽFriedhoff et al., 1962; Alpert et al., 1963.. If, for example, anger affects muscle tension or fear inhibits respiration, the sounds of voice can be altered without changing the words being uttered. In addition to affective prosody, there is linguistic prosody. Linguistic prosody involves the application of vocal stress through emphasis or inflection or through subtle changes in speech timing to highlight target syllables for a linguistic or pragmatic purpose. Murphy and Cutting Ž1990. examined the ability of schizophrenics to stress designated words while reading, and found that they performed as well as manic, depressed, and normal subjects. Although flat affect was not an experimental factor in this study, case finding involved selection from a chronic population, suggesting that a number of flat affect subjects were included. In contrast with the Murphy and Cutting findings, however, non-flat schizophrenic patients and normal subjects apply greater stress to concrete high-imagery words, and flat affect patients do so less or not at all ŽAlpert and Anderson, 1977.. It appears that schizophrenic subjects are able to apply linguistic stress on direction, but flat affect renders some cognitive mechanisms less effective in controlling emphasis or inflection. Other prosodic mechanisms provide information about linguistic intent. This may be seen in terminal rising inflection that changes a statement into a question, or falling inflection and emphasis, as talkers inform their conversational partners that they are ready to yield the floor. Schizophrenic subjects with flat affect are able to form the interrogative but provide fewer end-ofturn cues Žunpublished observations.. In a study with schizophrenic patients, Gotheil et al. Ž1970. found that when judges heard natural speech Žboth the words and the prosody. they could correctly identify the emotional context, although the controls were rated as more emotionally in-
M. Alpert et al. r Psychiatry Research 97 (2000) 107᎐118
tense. However, when judges were provided only lexical but not prosodic information, they could not differentiate flat affect from non-flat affect groups ŽLevin et al., 1985.. Thus, it appears that flat affect interferes with selective aspects of prosody. Content analysis of speech samples has been used to establish lexical correlates of personality traits, mood states, and psychopathology ŽSeegmiller and Epperson, 1987; Gottschalk, 1979; Broehl and McGee, 1981; Viney and Tych, 1985; Stewart et al., 1988; Zullow et al., 1988; McDermott and Porter, 1989.. The present study employs an objective computerized word-count system, the Dartmouth adaptation of the General Inquirer Content Analysis Program ŽTucker and Rosenberg, 1975. and an expanded version of the Harvard III Psycho-Sociological Dictionary ŽStone et al., 1966.. This procedure yields highly acceptable test᎐retest and between-context reliability ŽSchnurr et al., 1986., as well as superior performance in discriminating mood states and personality ŽOxman et al., 1988; Rosenberg et al., 1994, 1996.. We examine the words used in a free speech task while subjects recount their memories of happy and sad experiences. We compare flat affect and non-flat affect schizophrenic subjects to each other and to a normal group in their use of two General Inquirer tags: Pleasure and Distress. The classification of subjects into flat and nonflat groups is based on clinical ratings. Such ratings may lack precision, and it will be useful to establish that the individuals in the clinically described flat affect group are, indeed, less expressive. If the flat affect group does not show reduced expressiveness, a failure to demonstrate objective lexical differences will be more difficult to interpret. Alogia is another negative sign that often confounds clinical impressions of expressiveness ŽAlpert et al., 1997.. Fluency, the reciprocal of alogia, is reflected in the rate and duration of utterances and silences. We will report the percent of available floor time that is used for talking. As a measure of prosody, we report the variance of voice level Žemphasis. and the variance of fundamental frequency Žinflection..
109
2. Methods 2.1. Subjects. The subjects were 46 male schizophrenic patients who were referred from two inpatient VA hospital services as they were being prepared for discharge to outpatient care. There may be gender differences in fluency and prosody, and we restrict this sample to males because there were not enough females to constitute a viable separate group. Based on self-report, in the flat affect group 12 patients were Caucasian, four were African-American, and four were Hispanic; in the non-flat group, 15 patients were Caucasian, eight were African-American, one was Hispanic and one Asian. All patients were clinically stable at the time of the study, and their medications had not been changed for at least a week. The diagnosis of schizophrenia was made according to DSM-III-R ŽAmerican Psychiatric Association, 1987. criteria. Trained raters used the Schedule for Affective Disorders and Schizophrenia ŽSADS; Endicott and Spitzer, 1978. or the Structured Clinical Interview for DSM-III-R ŽSCID-III-R; Spitzer et al., 1992.. Subjects were excluded if: their tested IQ was below 80; if there was a complicating neurological disorder; a toxic psychosis or a seizure disorder; if ECT had been used in the previous six months; or if there was evidence of serious recent alcohol or substance abuse Žby structured interview criteria .. Written consent was obtained from all subjects after the goals and methods of the study were described. We also include results for a sample of 20 normal males who were recruited from employees of the hospital dietary staff. They were paid a nominal fee for their participation. The normal subjects ranged in age from 25 to 70 years Žmean s 41.1, S.D.s 13.58., and had received an average of 13.1 years ŽS.D.s 1.89. of education. They were without evidence of psychiatric illness on interview or lifetime SCID. They did not differ from the patients in regard to age or education. Based on self-report, 10 of the normal subjects were Caucasian, five were African-American, and five were Hispanic.
110
M. Alpert et al. r Psychiatry Research 97 (2000) 107᎐118
2.2. Ratings of flat affect Subjects were rated with the Scale for the Assessment of Negative Symptoms ŽSANS; Andreasen, 1983, 1990. by raters trained to a criterion reliability of r G 0.80. A scale was formed with five SANS Flat Affect items, two concerning gesture and one each concerning facial expression, vocal inflection and emotional responsiveness. We omitted the inappropriate affect item, which was omitted by most workers, and the global item. By Cronbach Ž1951. coherence analysis, the five items yield an ␣ s 0.96 in the current data set. The patients who scored two or more on at least three of the five items were assigned to the flat affect group Ž N s 20.. The other 26 patients were in the non-flat group. 2.3. Ratings of extrapyramidal signs The measure of neuroleptic, induced extrapyramidal side effects ŽEPSE. is an average of the Dystonia, Tremor, Gait, Salivation and Akathisia items from the Simpson and Angus scale ŽSimpson and Angus, 1970., or the same items from the St. Hans Rating Scale ŽGerlach et al., 1993.. Bradykinesia was omitted because it seemed more likely to be confused by the presentation of flat affect. Because reduced expressive motor activity is central to flat affect, it is important to clarify the effects of neuroleptics on prosody. Many authors report neuroleptic dose Žusually in chlorpromazine-equivalent dose. as a measure of the potential for EPSE to confound clinical observations. Knowledge of the dose administered may not be informative since therapeutic effects, EPSE, or their interactions are not linear with dose. Haase and Janssen Ž1965. have shown that therapeutic actions do not increase beyond a certain dose of neuroleptic for chronic patients with a partial therapeutic response. The situation is even more complex for the relationship of EPSE to neuroleptic dose. Hollister ŽHollister, 1976, see also Alpert et al., 1978., in addressing the observation that increasing the dose of typical neuroleptics may produce a lower rate of EPSE, suggested that the production of EPSE is associ-
ated with the dopamine-blocking action of these drugs. However, these drugs also have anticholinergic actions. If the dopamine-blocking action saturates at a lower dose than does the anticholinergic blockade, increments of neuroleptic while without additional dopaminergic effect can increase anti-cholinergic action, producing a paradoxical decrease in EPSE. Nor is there a simple relation between therapeutic action and EPSE. To investigate this issue, Alpert et al. Ž1977, 1978. recruited groups of patients who were relatively homogeneous for prognosis and for vulnerability to EPSE. All subjects were pre-menopausal females who had shown at least a moderate therapeutic response to neuroleptic treatment on a prior admission to the hospital. For the first 5 days of treatment, all patients received the same neuroleptic at the same modest dose and did not receive anticholinergic drugs. EPSE were measured quantitatively with a sensitive device for detecting digital tremor. Then, over the next several weeks, neuroleptic dose was optimized for each patient and anticholinergic drugs added if indicated. Outcome was assessed with total BPRS score from which the baseline score had been statistically partialled. It was found that patients with a greater increase in tremor at treatment day 4 showed less therapeutic improvement at outcome. About a third of the variance in improvement could be explained by the magnitude of the increase in EPSE at day 4. The results were interpreted as suggesting that individual differences in response to neuroleptics reflect correlated processes that are expressed in both the patient’s vulnerability to EPSE and their therapeutic response. Thus, neuroleptic dose can be uninformative of therapeutic actions; may be misleading in relation to EPSE; and maybe irrelevant in regard to individual differences in treatment response. In this study, we will report the sum of ratings of dystonia, tremor, gait, akathisia and salivation as a measure of the impact on EPSE of typical neuroleptics. While bradykinesia ratings are coherent with these items, exclusion of bradykinesia may clarify the assessment of flat affect and EPSE. Demographic information and background rat-
M. Alpert et al. r Psychiatry Research 97 (2000) 107᎐118
111
Table 1 Patient characteristics for flat affect Ž N s 20. and non-flat affect Ž N s 26. schizophrenia subjects Characteristic
Age Žyrs. Education Žyrs. Duration of illness Žyrs. EPSEa Flat affectb a b
Flat Affect
Non-flat Affect
Mean
S.D.
Mean
S.D.
41.1 12.2 16.6 2.7 2.6
10.82 2.02 8.77 1.91 0.96
40.9 13.1 16.4 1.7 0.7
8.31 2.50 6.68 1.33 0.34
EPSE indicates five items from Extrapyramidal Symptom Rating Scale Ždystonia, tremor, gait, akathisia, salivation.. Student’s T s 9.6, d.f.s 44, P - 0.01, two-tailed test.
ings are presented in Table 1. Except for the flat affect ratings, Ž t s 8.6, d.f.s 44, P- 0.01., there were no significant group differences for the items in Table 1. Twelve subjects were receiving atypical neuroleptics Žfive flat and seven non-flat.. The other patients were receiving typical neuroleptics involving nine different drugs of varying potency; 27 also were receiving anticholinergic medication. Surprisingly, the patients on atypical neuroleptics had slightly higher EPSE scores Žmean s 2.4, S.D.s 2.12. compared with scores of patients on typical neuroleptics Žmean s 2.0, S.D.s 1.49.. This difference is not statistically significant. Assignment to type of neuroleptic medication, made by the attending psychiatrist, was likely influenced by the presence of EPSE. 2.4. Procedures The subject and the interviewer were recorded on separate tracks of a stereo tape-deck using head-mounted microphones to reduce background noise and improve separation. During the interview, subjects were asked to describe a happy and a sad experience from their life, with each narrative lasting from 3 to 5 min. They were instructed to describe the situation, to remember who was present, to reconstruct what had led to the situation, and what happened afterwards. The order of task valence was counter-balanced across subjects. The interviewer was instructed to help the patient to produce at least 3 min of speech by asking open-ended questions if the patient’s description was too brief, or if the patient strayed from the topic. The recordings were transcribed
and pairs of assistants reviewed the transcripts while listening to the recordings until they agreed on the accuracy of the transcripts. The audio recordings were analyzed with a hybrid analog-digital program ŽVOXCOM, Alpert et al., 1986.. In the analog stage, the signal from the patient’s track is full-wave rectified and demodulated to produce an amplitude waveform of speech and silence. In a parallel channel, a frequency to voltage converter produces a signal that is proportional to the fundamental frequency of the patient’s voice. The amplitude signal is sampled 100 timesrs. Through software logic, the sound pulses of syllables are located, and their duration and peak amplitude are noted. At the point of maximal amplitude, the fundamental frequency for that syllable is noted. The duration of all pauses ) 0.2 s are collected. The duration of a run of syllables not interrupted by pauses ) 0.2 s is measured as an utterance. The duration of switching pauses from the end of an interviewer’s turn to the beginning of the patient’s response is saved as the patient’s response latency and, similarly, for the interviewer’s latency. Duration Žpauses and utterances . is log-transformed for averaging and reported in s. Inflection is measured in semitones around the patient’s average fundamental frequency and reported in Hertz. Emphasis is measured in decibels referenced to a calibration signal that is recorded on each tape at the time of the interview Žsee Alpert et al., 1986, 1993.. In other work ŽAlpert et al., 1999. we have found that these transformations produce Gaussian distributions. Percent talk time
112
M. Alpert et al. r Psychiatry Research 97 (2000) 107᎐118
is calculated as the ratio of the patient’s utterance time to the patient’s floor time. Emphasis is calculated as the variance of the voice level peaks in the amplitude waveform. Inflection is calculated as the variance of the fundamental frequency measures noted for each syllable. The variance is used as a measure of dispersion around the mean since variances are additive and, in a Gaussian distribution, the variance is not correlated with the mean 2.5. General Inquirer program The typescripts of the narratives were analyzed using the Dartmouth Adaptation of the General Inquirer ŽOxman et al., 1988; Tucker and Rosenberg, 1975; Schnurr et al., 1986. content-analysis program. The 5000item Harvard III Psycho-Sociological Dictionary ŽStone et al., 1966. was used. The program reduces each word to a base form and then performs a dictionary search routine. Upon finding a match Ža ‘tag’., an increment is added to the category or categories to which the word has been assigned. The output also includes tallies of tags and words used, the proportion of all words tagged, and a list of unclassified tokens. These are, typically, proper nouns or uncommon words. The categories for the tags include personal ŽSelf, Selves, Other., evaluative ŽGood, Bad, Deviation., affective ŽAffection, Distress, Pleasure ., etc. Žsee Stone et al., 1966 for a complete description.. Each word or stem is assigned to at least one, but sometimes more, of the 83 categories. For example, the word ‘ache’ is assigned to the Distress category, whereas the word ‘trousers’ is assigned to both the Clothing and Male Theme categories. We used the results for Pleasure and Distress with the transcripts of the happy and sad narratives. The Pleasure category contains 49 words and word stems connoting positive affective experiences. Category entries include laugh, happy, enjoy and pleasant. The Distress category contains 129 entries including upset, tragic, insecure and sad. The unit for this measure is the number of tagged in-category words divided by the total number of tagged words times 100.
3. Results 3.1. Acoustic analysis The results for the acoustic analysis are presented in Table 2. The speech of subjects in the flat affect group had less inflection than that of the normal controls. The effect size for the difference in inflection between the flat and non-flat schizophrenic groups is 0.89, but this difference was not significant in the post hoc planned comparison because of the stringency imposed by the Bonferroni correction. The groups did not differ in regard to emphasis. To examine further the contributions of inflection and emphasis to the rating of flat affect, we performed a Multiple Regression Analysis ŽMRA. with flat affect as the dependent measure and inflection and emphasis as the predictors. Together inflection and emphasis explained Ž R 2 . 16% of the variance in flat affect ratings Ž F s 4.1; d.f.s 2,43; P - 0.05.. The regression coefficient for inflection was significant Ž s y0.37; T s y2.6; d.f.s 2,43; P - 0.05.; more inflection was associated with lower flat affect ratings. The regression for emphasis remained insignificant Ž s y0.09; T s y0.7; d.f.s 2,43; n.s.., further limiting the scope of emphasis in determining the clinician’s impression of flat affect. In addition, the flat affect group was less fluent; they used less of their floor time speaking than the non-flat affect and the normal groups. To pursue the association between alogia and ratings of flat affect, we performed an MRA with inflection Žas a measure of flat affect. and percent time talking Žas a measure of alogia. as predictors of flat affect ratings. Both inflection Ž s y0.30; t s y2.4; d.f.s 2,43, P- 0.05. and percent time talking Ž s y0.48; t s y3.9, d.f.s 2,43, P0.01. independently predicted flat affect ratings. The R 2 for this MRA is 0.38 Ž r s 0.61; F s 12.9; d.f.s 2,43; P- 0.01., over a third of the variance in flat affect ratings can be explained by the two acoustic measures. It appears that the rater’s impression of flat affect was confounded by the patient’s alogia.
M. Alpert et al. r Psychiatry Research 97 (2000) 107᎐118
113
Table 2 Acoustic and General Inquirer measures of the happy and sad narratives for flat affect Ž N s 20. and non-flat affect Ž N s 26. schizophrenia subjects and normal control Ž N s 20. subjects
Item A. Acoustic measures Prosody Inflection Žsemitones. Emphasis ŽdB. Fluency Percent time talking B. General Inquirer measures Number of words Percent of words tagged Number of sentences Average words per sentence
ŽA. Flat affect
ŽB. Non-flat affect
ŽC. Normal control
Mean
Mean
Mean
S.D.
0.40 16.3
0.162 7.15
0.62 17.5
S.D.
0.317 4.44
S.D.
F Žd.f .s 2,63.
Multiple comparisonc
0.84 14.8
0.502 7.60
7.76a 1.05
C )A
55.6
11.30
63.9
11.06
67.3
7.28
7.16a
B, C ) A
416.9 59.2 31.8 14.4
199.06 3.87 12.54 6.78
562.0 58.8 46.1 13.6
170.47 3.18 19.90 5.12
763.1 58.2 37.1 24.6
388.73 3.31 24.03 10.17
8.77a 0.46 3.20b 14.31a
C ) A, B B )A C ) A, B
a
P - 0.01. P - 0.05. c 0.05 criterion for Least Significant Difference multiple comparisons is Bonferroni corrected to equal 0.0167. b
3.2. General Inquirer content analysis Table 2 also presents the General Inquirer results. The normal subjects produced more words than the patients did, as well as more words per sentence. The rate of production of tagged words did not show any simple association with demographic characteristics of the patients or normal subjects. The flat affect group produced fewer sentences than the non-flat affect group, although this may reflect the confounding of flat affect and alogia. The normal subjects produced more words but not more sentences than the other subjects. If there is an isomorphism between thoughts and sentences, the patients and normal subjects do not differ in the number of thoughts elicited in the recall task. However, the normal subjects expressed their thoughts in more elaborate syntax. Fig. 1 reflects the precision with which subjects in the different diagnostic groups selected words appropriate to the context. The histogram depicts, in the left column for each subject group, the rate at which Pleasure words were produced during the happy narrative and in the right column the rate at which Distress words were produced during the sad narrative. The intervening columns
represent the rate of context inappropriate tags for the two narratives. The error bars present the standard deviation for each mean. We analyzed these results in a multivariate analysis of variance ŽMANOVA. with narrative type Žhappy or sad. and category ŽPleasure or Distress. as withinsubject effects and group Žflat, non-flat, and normal. as the between-subjects effect. The group effect was not significant, while the category effect was Ž F s 14.9, d.f.s 1,53, P- 0.01., indicating that Distress tags were more frequent than Pleasure tags, overall. The narrative type effect was significant Ž F s 4.3, d.f.s 1,53, P- 0.05., indicating a higher rate of tagged words during the sad task. The category by narrative type interaction was significant Ž F s 80.3, d.f.s 1,53, P0.01., indicating that the Pleasure category had higher values during the happy narrative and the Distress category had higher values during the sad narrative. There were no significant main effects or interactions for diagnostic group contrasts. Table 3 presents a matrix of the zero order correlations for the patient group Ž N s 46. above the diagonal, and for the control group Ž N s 20. below the diagonal. For the normal controls, the acoustic measures and the General Inquirer mea-
M. Alpert et al. r Psychiatry Research 97 (2000) 107᎐118
114
Fig. 1. Rate of tagged Pleasure and Distress category words during the happy and sad narratives for flat affect Ž N s 20. and non-flat affect Ž N s 26. schizophrenia subjects and normal control Ž N s 20. subjects.
Table 3 Correlations between acoustic measures, General Inquirer measures and ratings for the patients Ž N s 46. above the diagonal and for the normal control subjects Ž N s 20. below the diagonal Inflection Acoustic Inflection Emphasis Percent time talking Content Pleasure Distress Ratings Flat affect Alogia EPSE a b
P - 0.01. P - 0.05.
Emphasis
%TT
Pleasure
Distress
Flat affect
Alogia
EPSE
0.23
0.19 y0.13
y0.01 0.07 y0.18
y0.11 0.08 y0.08
y0.39a y0.18 y0.54a
y0.22 y0.08 y0.54a
0.18 0.31b y0.20
0.30b 0.08
0.32b 0.02
0.83a
0.35b 0.38a
y0.17 y0.25
y0.25
0.07 y0.11
0.04 0.24
y0.18 0.14
0.34b 0.01
0.29 0.09
M. Alpert et al. r Psychiatry Research 97 (2000) 107᎐118
sures do not show significant relations, either within or across domains. Similarly for the patients, except for the correlation between the rates of content-appropriate pleasure and distress words, the acoustic and General Inquirer measures are unrelated. There are a number of correlations between acoustic measures and SANS ratings. Inflection is uniquely related to flat affect while emphasis is unrelated to the SANS ratings but modestly related to EPSE ratings. Percent time talking is equally related to flat affect and alogia ratings, but not to EPSE ratings. Flat affect and alogia are strongly related to each other, and both are modestly related to EPSE ratings. Thus, there is a cluster of correlations among the clinical ratings.
4. Discussion A goal of this study was to examine different aspects of speech generation, simultaneously, in the same text. Levelt Ž1989. suggested that word choice is controlled by a lexical selector module which operates late in speech generation and in parallel with a prosody generator. In this study, subjects in the flat affect group showed less prosody and were less fluent. However, they were as precise in the selection of context-appropriate words as the non-flat subjects or, even, the controls. The flat subjects also were similar to the non-flat schizophrenic subjects in producing an excess of Distress words overall, and longer sad than happy narratives. The output of the lexical selector indicates that emotional valence influenced speech generation while failure of the prosody generator in the flat affect subjects implicates other factors associated with motor aspects of expressiveness in flat affect. Ratings of flat affect reflect reduced motor activity of face, voice and gesture. The affected activities involve display during speech communication. In this regard, it is instructive to observe a normal person talking on a telephone in a public place. Their conversational partner cannot see them but, often, they display a full range of expressions, uninhibited, even, by the notice of passers-by. The display is generated as part of the
115
speaker’s paralinguistic activity, which mobilizes gesture and facial expression as well as prosody. This display is muted in flat affect whereas subjective, autonomic and contextual aspects of emotion are intact. The reduced display arises from speech mechanisms distal to those responsible for lexical choice, resulting in reduced prosody in lexically intact speech. Thus, flat affect may be viewed as reduced speech-associated movements, analogous to the loss of walking-associated movements of Parkinson’s disease. Although Parkinson’s disease can share some of the characteristics of schizophrenic flat affect, there are a number of differences. Parkinson’s disease tends to involve walking and increased muscle resistance. These are not seen in flat affect. Also, in a discriminate function analysis, we were able to differentiate the expressive blunting of Parkinson’s patients from schizophrenic flat affect on the basis of acoustic measures of the patient’s speech ŽAlpert et al., 1989.. Both processes, however, likely involve hypodopaminergia ŽAlpert and Friedhoff, 1980 . with Parkinson’s focused in the basal ganglia and flat affect implicating forebrain mechanisms that integrate paralinguistic activities. Prosody and fluency provide cues for clinicians as they evaluate the negative syndrome ŽAlpert et al., 1997.. On a logical basis, one might expect that the two processes would dissociate, with inflection tied to impressions of flat affect, and percent talk time to alogia. In previous work ŽAlpert et al., 1994. it had been shown that, in free speech, alogic patients pause before low frequency words Žas do, but to a lesser degree, normal speakers.. Since the influence of word frequency on word finding would reflect cognitive slowing in the operation of the word selector module, alogia and flat affect seem attributable to different speech-generation modules. However, acoustic correlates of the two processes independently contributed to ratings of flat affect. The processes were confounded, either in the speech characteristics of the negative syndrome patient or in the perceptions of the clinician. For a number of reasons, it appears that the confounding occurred in the rater. Percent talk time was not correlated with the variance of fre-
116
M. Alpert et al. r Psychiatry Research 97 (2000) 107᎐118
quency in either the flat affect group Ž r s y0.01. or the non-flat group Ž r s 0.09., while ratings of Alogia and Flat Affect were highly intercorrelated in the flat affect group Ž r s 0.83.. Converging evidence that confounding occurs in the raters has been reported ŽAlpert et al., 1995.. In that investigation, fluency was manipulated in recordings by electronically editing pause duration. In one version of a recording of an interview, the patient’s pauses were lengthened and, in another version, they were shortened. The variance of the fundamental frequency was not changed by this manipulation. Clinicians, unaware of the editing, rated the extended pause versions as both flatter and more alogic, and vice versa for the version with increased fluency. Thus, several lines of converging evidence implicate the rating process in the observed confounding. Reduced expressiveness also may result from neuroleptics, and the assessment of flat affect in the presence of EPSE presents a clinical challenge ŽMayer et al., 1985; Carpenter et al., 1995.. Confusion can run in both directions. Bradykinesia may be seen as flat affect, or ‘decreased spontaneous movements’ ŽSANS n2. or a ‘paucity of expressive gestures’ ŽSANS n3. could confound ratings of bradykinesia. Bradykinesia constituted a third of the EPSE ratings in the data set. Even without the bradykinesia item, the EPSE ratings correlated with ratings of flat affect Žsee Table 3.. However, the EPSE measure was not significantly correlated with the acoustic inflection measure. EPSE was correlated with increased emphasis. In other studies with larger numbers of subjects, and where patients were randomized to treatment with either typical or atypical neuroleptics ŽAlpert et al., 2000., we had not observed this correlation. Interpretation of the correlation should be reserved, although it does appear that inflection and emphasis may be under the control of different mechanisms and probably should not be combined in evaluations of flat affect. In any case, raters may not be able to judge both processes while listening to the patient. It is important to be able to differentiate flat affect from alogia and from EPSE. Each has different implications for etiology and requires different interventions. Despite careful training
and ‘satisfactory’ reliability, the raters confounded the different conditions. One should bear in mind that inter rater reliability sets the upper but not the lower limit on validity. The objective acoustic measures provided unique insight into the ascertainment process. There is other evidence that ratings are often imprecise. Andreasen Ž1990. reported a psychometric analysis of SANS ratings and noted that a Cronbach analysis of the global ratings of the five SANS subscales yielded an ␣ s 0.85. This high coherence is consistent with a single negative dimension, not the five different parameters suggested by the separate subscales. The acoustic measures seem better able to separate flat affect from alogia. It might be useful to develop a clinical lab for the assessment of disturbed feelings. It has been noted that technical terms such as affect, mood and emotion are used imprecisely in the psychiatric literature ŽAlpert and Rosen, 1990.. The introduction of objective, operationally defined constructs may have a salutary influence on clinical practice and research.
Acknowledgements The authors thank Melissa Dong for help with data collection, and Dr. Robert Cancro for a critical reading of an earlier version of this report. References Alpert, M., Anderson, L.T., 1977. Imagery mediation of vocal emphasis in flat affect. Archives of General Psychiatry 34, 208᎐212. Alpert, M., Clark, A., Pouget, E., 1994. The syntactic role of pauses in the speech of schizophrenic patients with alogia. Journal of Abnormal Psychology 103, 750᎐757. Alpert, M., Diamond, F., Kesselman, M., 1977. Correlation between extrapyramidal and therapeutic effects of neuroleptics. Comprehensive Psychiatry 18 Ž4., 333᎐338. Alpert, M., Diamond, F., Weisenfreund, J., Taleporos, E., Friedhoff, A.J., 1978. The neuroleptic hypothesis: study of the covariation of extrapyramidal and therapeutic drug effects. British Journal of Psychiatry 133, 169᎐175. Alpert, M., Friedhoff, A.J., 1980. An undopamine hypothesis of schizophrenia. Schizophrenia Bulletin 6, 387᎐390. Alpert, M., Kotsaftis, A., Pouget, E.R., 1997. Speech fluency
M. Alpert et al. r Psychiatry Research 97 (2000) 107᎐118 and schizophrenic negative signs. Schizophrenia Bulletin 23, 171᎐177. Alpert, M., Kurtzberg, R.L., Friedhoff, A.J., 1963. Transient voice changes associated with emotional stimuli. Archives of General Psychiatry 8, 362᎐365. Alpert, M., Merewether, F., Homel, P., Martz, M.R., Lomask, M., 1986. Voxcom: a system for analyzing natural speech in real time. Behavioral Research Methods Instrumentation and Computing 18, 267᎐272. Alpert, M., Pouget, E.R., Shaw, R., Dong, M., Lim, K., 1999. A clinical lab for psychiatry: vocal acoustic measures of flat affect and alogia. Presented at NCDEU, March 1999, manuscript in preparation. Alpert, M., Pouget, E., Silva, R., 1995. Cues to the assessment of affects and moods: speech fluency and pausing. Psychopharmacology Bulletin 31, 421᎐424. Alpert, M., Pouget, E.R., Welkowitz, J., Cohen, J., 1993. Mapping schizophrenic negative symptoms onto measures of the patient’s speech: set correlational analysis. Psychiatry Research 48, 181᎐190. Alpert, M., Rosen, A., 1990. A semantic analysis of the various ways the terms ‘affect’, ‘emotion’, and ‘mood’ are used. Journal of Communication Disorders 23, 237᎐246. Alpert, M., Rosen, A., Welkowitz, J., Sobin, C., Borod, J.C., 1989. Vocal acoustic correlates of flat affect in schizophrenia: similarity to Parkinson’s disease and right hemisphere disease and contrast with depression. British Journal of Psychiatry 155, 51᎐56. Alpert, M., Smith, R., Allan, E., Infante, M., Sison, C., Pouget, E.R. Response of schizophrenic negative signs to treatment with haloperidol or olanzapine as reflected in measures of the patient’s voice. Presented at NCDEU, March 2000, manuscript in preparation. American Psychiatric Association, 1987. DSM-III-R: Diagnostic and Statistical Manual of Mental Disorders, 3rd ed. revised American Psychiatric Press, Washington, DC. Andreasen, N.C., 1983. The Scale for the Assessment of Negative Symptoms ŽSANS.. University of Iowa, Iowa City, Iowa. Andreasen, N.C., 1990. In: Andreasen, N.C. ŽEd.., Schizophrenia: Positive and Negative Symptoms and Syndromes. Modern Problems in Pharmacopsychiatry, 24. Karger, Basel, pp. 73᎐88. Berenbaum, H., Oltmanns, T.F., 1992. Emotional experience and expression in schizophrenia and depression. Journal of Abnormal Psychology 101, 37᎐44. Borod, J.C., 1993. Cerebral mechanisms underlying facial, prosodic, and lexical emotional expression. Neuropsychologia 7, 445᎐446. Borod, J.C., Martin, C.C., Alpert, M., Brozgold, A., Welkowitz, J., 1993. Perception of facial emotion in schizophrenic and right brain-damaged patients. Journal of Nervous and Mental Disease 181, 494᎐502. Broehl, W.G.J.r., McGee, V.E., 1981. Content analysis in psychohistory: a study of three lieutenants in the Indian Mutiny, 1857᎐58. Journal of Psychohistory 8, 281᎐306.
117
Carpenter, W.T., Conley, R.R., Buchanan, R.W., Breier, A., Tamminga, C.A., 1995. Patient response and resource management: another view of clozapine treatment of schizophrenia. American Journal of Psychiatry 152, 827᎐832. Cronbach, L.J., 1951. Coefficient alpha and the internal structure of tests. Psychometrika 16, 297᎐334. Curtis, C.E., Lebow, B., Lake, D.S., Katsanis, J., Iacono, W.G., 1999. Acoustic startle reflex in schizophrenia patients and their first-degree relatives: evidence of normal emotional modulation. Psychophysiology 36, 469᎐475. Dworkin, R.H., Clark, S.C., Amador, X.F., Gorman, J.M., 1996. Does affective blunting in schizophrenia reflect affective deficit or neuromotor dysfunction? Schizophrenia Research 20, 301᎐306. Endicott, J., Spitzer, R.L., 1978. A diagnostic interview: the Schedule for Affective Disorders and Schizophrenia. Archives of General Psychiatry 35, 837᎐844. Friedhoff, A.J., Alpert, M., Kurtzberg, R.L., 1962. An effect of emotion on voice. Nature 193, 357᎐358. Gerlach, J., Koorsgard, S., Clemmesen, P., Lund Lauersen, A.-M., Magelund, G., Noring, U., Povlsen, I.J., Bech, P., Casey, D.E., 1993. The St. Hans rating scale for extrapyramidal syndromes: reliability and validity. Acta Psychiatrica Scandinavica 87, 244᎐252. Gotheil, E., Paredes, A., Exline, R.V., Winkelmayer, R., 1970. Communication of affect in schizophrenia. Archives of General Psychiatry 22, 439᎐444. Gottschalk, L.A., 1979. The Content Analysis of Verbal Behavior. Further Studies. SP Medical Scientific Books, New York. Haase, H.J., Janssen, P., 1965. Action of Neuroleptic Drugs: Psychiatric, Neurologic and Pharmacological Investigation. Year Book Medical, Chicago, pp. 1᎐174 ŽTrans. from German by Wilkinson, M.L... Hollister, L.E., 1976 Personal communication. Knight, R.A., Valner, J.B., 1993. In: Costello, C.G. ŽEd.., Affective Deficits in Schizophrenia. Symptoms of Schizophrenia. Wiley, New York, pp. 145᎐200. Krause, R., Steiner, E., Sanger-Alt, C., Wagner, G., 1989. Facial expression of schizophrenic patients and their interaction with partners. Psychiatry 52, 1᎐12. Kring, A.M., Kerr, S.L., Smith, D.A., Neale, J.M., 1994. Flat affect in schizophrenia does not reflect diminished subjective experience of emotion. Journal of Abnormal Psychology 102, 507᎐517. Levelt, W.J.M., 1989. Speaking from intention to articulation. The MIT Press, Cambridge, Mass. Levin, S., Knight, R.A., Alpert, M., 1985. Verbal and nonverbal expression of affect in speech of schizophrenic and affective disorder patients. Journal of Abnormal Psychology 94, 487᎐497. Mayer, M., Alpert, M., Stastney, P., Perlick, D., Empfield, M., 1985. Multiple contributions to clinical presentation of flat affect in schizophrenia. Schizophrenia Bulletin 11, 420᎐428.
118
M. Alpert et al. r Psychiatry Research 97 (2000) 107᎐118
McDermott, J., Porter, D., 1989. The efficacy of poetry therapy: a computerized content analysis of the death poetry of Emily Dickinson. Psychiatry 52, 462᎐468. Murphy, D., Cutting, J., 1990. Prosodic comprehension and expression in schizophrenia. Journal of Neurology, Neurosurgery and Psychiatry 53 Ž9., 727᎐730. Oxman, T., Rosenberg, S., Schnurr, P., Tucker, G., Gala, G., 1988. The language of altered states. Journal of Nervous and Mental Disease 176, 401᎐408. Rosenberg, S., Blatt, S., Oxman, T., McHugo, G., Ford, R., 1994. Assessment of object relatedness through lexical content analysis of the TAT. Journal of Personality Assessment 631, 345᎐362. Rosenberg, S., Drake, R.E., Mueser, K., 1996. New directions for treatment research on sequela of sexual abuse in persons with severe mental illness. Community Mental Health Journal 32, 387᎐400. Schnurr, P., Rosenberg, S., Oxman, T., Tucker, G., 1986. A methodological note on content analysis: estimates of reliability. Journal of Personality Assessment 50, 601᎐609. Seegmiller, R.A., Epperson, D.L., 1987. Distinguishing thinking᎐feeling preferences through the content analysis of natural language. Journal of Personality Assessment 51, 42᎐52. Shaw, R.J., Dong, M., Lim, K.O., Faustman, W.O., Pouget, E.R., Alpert, M., 1999. The relationship between affect expression and recognition in schizophrenia. Schizophrenia Research 37, 245᎐250.
Simpson, G., Angus, J.W.S., 1970. A rating scale for extrapyramidal side effects. Acta Psychiatrica Scandinavica 46, 44᎐51. Sison, C., Alpert, M., Fudge, R., Stern, H., 1996. Constricted expressiveness and psycho-physiological reactivity in schizophrenia. Journal of Nervous and Mental Disease 184, 589᎐597. Spitzer, R.L., Williams, J.B., Gibbon, M., First, M.B., 1992. The Structured Clinical Interview for DSM-III-RŽSCID.. I. history, rationale, and description. Archives of General Psychiatry 49, 624᎐629. Stewart, A.J., Franz, D., Layton, L., 1988. The changing self: using personal documents to study lives. special issue: psycho-biography and life narratives. Journal of Personality 56, 41᎐74. Stone, P., Dunphy, D., Smith, M., Ogilivie, D., 1966. The General Inquirer: a Computer Approach to Content Analysis. The MIT Press, Cambridge, Mass. Tucker, G., Rosenberg, S., 1975. Computer content analysis of schizophrenic speech: a preliminary report. American Journal of Psychiatry 132, 611᎐616. Viney, L.L., Tych, A.M., 1985. Content analysis scales measuring psycho-social maturity in the elderly. Journal of Personality Assessment 49, 311᎐317. Zullow, H.M., Oettingen, G., Peterson, C., Seligman, M.E.P., 1988. Pessimistic explanatory style in the historical record. American Psychologist 43, 673᎐682.