28
BRAIN AND LANGUAGE 56, 183–210 (1997) ARTICLE NO. BL961741 Referential Cohesion and Logical Coherence of Narration after Right Hemisphere Stroke G. ALBYN DAVIS University of Massachusetts at Amherst THERESE M. O’NEIL-PIROZZI Spaulding Rehabilitation Hospital, Boston AND MARIBETH COON University of Massachusetts at Amherst A group with right hemisphere dysfunction was compared to neurologically intact controls regarding the referential cohesion and logical coherence of narrative pro- duction. A somewhat varied sample of six stories was obtained with tasks of cartoon- elicited story-telling and auditory–oral retelling. We found deficits in the patient group with respect to referential cohesion, logical coherence, and accuracy of narra- tion, but the occurrence of deficits depended on the condition in which narration was produced and, to some extent, on the particular story used in each condition. The primary implications of this study pertain to the attention given by researchers to the feature of discourse production being studied. 1997 Academic Press There are two fundamental components of the clinical assessment of dis- course production. One is the method used to elicit discourse, which varies with respect to the task (e.g., spontaneous, interview, picture elicitation) and the type of discourse elicited (e.g., narration, exposition). The other funda- mental component is the analysis of an elicited discourse. Levels of analysis are indicative of the multiple levels of meaning involved in processing dis- course (Frederiksen & Stemmer, 1993). A task-related factor can be studied The authors are grateful for the assistance of Dr. Jerome L. Myers and Dr. Richard Freyman regarding crucial features of this research. Address correspondence and reprint requests to G. Albyn Davis, Ph.D., Communication Disorders, Arnold House, University of Massachusetts, Box 30410, Amherst, MA 01003- 0410. Fax. 413-545-1264. E-mail: [email protected]. 183 0093-934X/97 $25.00 Copyright 1997 by Academic Press All rights of reproduction in any form reserved.

Referential Cohesion and Logical Coherence of Narration after Right Hemisphere Stroke

Embed Size (px)

Citation preview

BRAIN AND LANGUAGE 56, 183–210 (1997)ARTICLE NO. BL961741

Referential Cohesion and Logical Coherence of Narrationafter Right Hemisphere Stroke

G. ALBYN DAVIS

University of Massachusetts at Amherst

THERESE M. O’NEIL-PIROZZI

Spaulding Rehabilitation Hospital, Boston

AND

MARIBETH COON

University of Massachusetts at Amherst

A group with right hemisphere dysfunction was compared to neurologically intactcontrols regarding the referential cohesion and logical coherence of narrative pro-duction. A somewhat varied sample of six stories was obtained with tasks of cartoon-elicited story-telling and auditory–oral retelling. We found deficits in the patientgroup with respect to referential cohesion, logical coherence, and accuracy of narra-tion, but the occurrence of deficits depended on the condition in which narrationwas produced and, to some extent, on the particular story used in each condition.The primary implications of this study pertain to the attention given by researchersto the feature of discourse production being studied. 1997 Academic Press

There are two fundamental components of the clinical assessment of dis-course production. One is the method used to elicit discourse, which varieswith respect to the task (e.g., spontaneous, interview, picture elicitation) andthe type of discourse elicited (e.g., narration, exposition). The other funda-mental component is the analysis of an elicited discourse. Levels of analysisare indicative of the multiple levels of meaning involved in processing dis-course (Frederiksen & Stemmer, 1993). A task-related factor can be studied

The authors are grateful for the assistance of Dr. Jerome L. Myers and Dr. Richard Freymanregarding crucial features of this research.

Address correspondence and reprint requests to G. Albyn Davis, Ph.D., CommunicationDisorders, Arnold House, University of Massachusetts, Box 30410, Amherst, MA 01003-0410. Fax. 413-545-1264. E-mail: [email protected].

1830093-934X/97 $25.00

Copyright 1997 by Academic PressAll rights of reproduction in any form reserved.

184 DAVIS, O’NEIL-PIROZZI, AND COON

TABLE 1Levels in the Analysis of Discourse Production

General levels Sublevels Types of measures

Within-sentence Lexical Lexical errorsMicrolinguistic

Syntactic CompletenessComplexity index

Lexical-semantic Content/information analysis

Between-sentence Local (microstructure) Referential cohesionMacrolinguistic Global (macrostructure) Logical coherence

Thematic coherence ratingsStory structure analysisGeneral attributes

with respect to its effect on production of words at one level or on productionof a narrative structure at another level. Our study explores the analysis ofstories produced by persons with right hemisphere dysfunctions (RHD)

LEVELS OF DISCOURSE PRODUCTION

Persons with RHD have been contrasted to aphasic persons with left hemi-sphere dysfunctions (LHD) according to two levels of language processing.‘‘Whereas the linguistic structure of the left-brain-damaged subjects gener-ally becomes fragmented at the word and sentence level, the breakdown inright-brain-damaged subjects’ discourse lies in processing larger languageunits. . .’’ (Bloom, 1994, p. 86). Brownell (1988) portrayed comprehensiondisorders similarly by suggesting that aphasia is a problem within sentencesand RHD is an impairment between sentences. Like difficulties in ‘‘gettingthe point’’ of a discourse, individuals with RHD have been described as‘‘wandering from the point’’ when conversing and telling stories (Gardner,1982).

There have been several attempts to characterize wandering from the pointaccording to frameworks appropriate for the discourse level of language pro-cessing. Table 1 summarizes the levels of language that comprise discourseproduction. Terminology can be momentarily confusing. For example, mi-crolinguistic analysis does not address the microstructural level of discourse.The following review explains these levels further and addresses the studyof RHD at the within-sentence and between-sentence levels.

Within-Sentence Analyses

‘‘Microanalysis’’ focuses on lexical and syntactic features of productionand is employed primarily in studies of aphasia (e.g., Haravon, Obler, &

NARRATIVE PRODUCTION 185

Sarno, 1994; Menn, Ramsberger, & Helm-Estabrooks, 1994). In a study ofRHD, Glosser found no deficits with respect to lexical errors and syntacticcomplexity when subjects were asked to describe family and work experi-ences. She concluded that people with RHD are generally not deficient atthe microlinguistic level of discourse production (Glosser, 1993; Glosser,Deser, & Weisstein, 1992). A similar conclusion had been reached in otherstudies (Metzler & Jelinek, 1977; Rivers & Love, 1980).

Several investigations have employed versions of a content analysis origi-nally developed for the study of aphasia (Yorkston & Beukelman, 1980;Nicholas & Brookshire, 1993). Content analysis is intended to evaluate theinformation conveyed at a level of individual concepts. Information is repre-sented in semantic units, called content units, that are usually no larger thana word or phrase. Most studies involve the presentation of pictured scenesto be described. For the study of RHD, Myers (1979) distinguished literalcontent (e.g., a ‘‘woman’’ in a scene) and inferred content (e.g., ‘‘themother’’). RHDs were deficient in the number of inferred concepts producedor in a ratio of inferred concepts relative to total concepts produced (Macki-sack, Myers, & Duffy, 1987; Myers, 1979). Trupe and Hillis (1985) reviewedclinical data from 62 RHDs and found 10 to be normal, 10 to be deficient,and 2 to be ‘‘verbose’’ in production of content units.

Joanette and his colleagues (1986) assessed informativeness of narrationaccording to the number of micropropositions elicited with a picture-storysequence. Micropropositions are hierarchical units representing meaning ata deep structure level of a sentence, and a sentence may contain multiplepropositions (Kintsch, 1994). A proposition contains a predicate and its argu-ments. The phrase ‘‘small horse’’ means the horse is small with small asthe predicate and horse as the argument (Joanette, Goulet, Ska, & Nespou-lous, 1986). Joanette found that 36 RHDs were equal to controls in produc-tion of words and syntactic units but were deficient in the number of proposi-tions used to tell the story. Other studies have also exposed informationdeficiencies in production of narrative and procedural discourse (Bloom,Borod, Obler, & Gerstman, 1992; Cherney & Canter, 1993).

The discovery of deficits has not been consistent. In a study of scene de-scription, 26 RHDs were not differentiated from LHDs and controls ac-cording to Myers’ inferential content analysis (Tompkins, Boada, McGarry,Jones, Rahn, & Ranier, 1993). In a study of story-telling, Davis and Turner(1995) found that 8 RHDs were deficient in number of content units butwere not deficient in information efficiency measured according to a ratioof units essential to the stories relative to the total number of content unitsproduced. Tompkins and her colleagues suggested that ‘‘our coding systemmay not have included some important characteristics. In certain cases, wehad the impression that there was something qualitatively unusual about theRHD samples that our analyses did not capture’’ (p. 120). Diagnosis of dis-

186 DAVIS, O’NEIL-PIROZZI, AND COON

course impairment with content analysis appears to depend on size of thepatient group, the type of content measured, and whether the measure is atotal or a proportion.

Between-Sentence Analyses

What was unusual in Tompkins’ samples may have occurred at the so-called between-sentence level of language production. Communicative dis-course depends on an interrelatedness between and among statements at whatGlosser (1993) has called the local and global levels of discourse structure(Table 1). The local or microstructural level exists between statements andis usually said to possess cohesion whereby two words (e.g., a pronoun andits antecedent) ‘‘co-refer’’ to the same entity (Halliday, 1985). The globalor macrostructural level exists across more than two statements. It containsthe essential or thematic information traversing a discourse and is the upperlimit of structural organization where we can identify the structural distinc-tiveness of exposition and narration (Britton, 1994; Kintsch, 1994).

There have been three investigations of RHD at the local level (Bloom,1994; Glosser et al., 1992; Uryase, Duffy, & Liles, 1991). The most fre-quently studied types of cohesion have been pronominal and lexical corefer-ence, the latter occurring when the definite article the precedes a noun tosignal that the noun refers to an entity mentioned previously (e.g., the car,the student). The research has shown that RHDs are likely to produce moreincomplete cohesive ties than a control group (i.e., a pronominal or lexicalelement without a clear antecedent) and are likely to be deficient in propor-tions of complete ties to total words or clauses.

The global level of discourse is most often illustrated by the narrativeform. Joanette had shown that many of his RHDs produced fewer core propo-sitions or those that were essential to telling the story that was studied (Joan-ette et al., 1986). Problems with utilizing narrative structure have been indi-cated in studies of sentence sequencing, and the presence of an initialthematic statement was not as helpful for RHDs as it was for other subjectgroups (Delis, Wapner, Gardner, & Moses, 1983; Schneiderman, Mura-sugi, & Saddy, 1992). Knowledge of story-telling conventions has been char-acterized by story grammars that specify functional categories of statements(e.g., setting, resolution) and an organization of these categories (Stein &Glenn, 1979; Thorndyke, 1977; see Gordon, 1993, for criticisms). Analysisof RHDs’ productive narrative structure may have started with Rivers andLove’s (1980) general finding that patients were incomplete relative to thebeginning and end of stories. Uryase and others (1991) examined complete-ness of episodes and production of story grammar categories such as setting,initiating event, direct consequence, and reaction. RHDs were deficient innumber of episodes and complete episodes, and they conveyed less informa-tion than controls in each functional category. Joanette and others (1986)

NARRATIVE PRODUCTION 187

found that their informationally deficient RHDs tended to omit content in acomplication category. Glosser and others (1992) found no deficit with re-spect to a general rating of global coherence and decided that RHDs areimpaired locally but not globally.

Some global-level problems may be uncovered in attempts to characterizewhat RHDs are doing when they ‘‘say as much, but convey less’’; the infor-mational void is often filled by ‘‘trivial, tangential, and seemingly unrelatedinformation’’ (Myers, 1993, p. 283). Analyses have included sorting behav-ior into general categories such as confabulation, digression, and embellish-ment (Myers, 1979; Rivers & Love, 1980; Trupe & Hillis, 1985; Wapner,Hamby, & Gardner, 1981). Cherney and Canter (1993) looked for propor-tions of essential, irrelevant, redundant, and other types of content units.RHDs produced fewer essential and elaborative units and more irrelevantand off-topic units than a control group.

One feature of discourse has received relatively little attention in clinicalresearch. Statements in a narration are interrelated by logical connections inwhich an action causes another action or a mental state motivates an action(Trabasso & Sperry, 1985). The presence of these ‘‘causal relations’’ hasbeen shown to influence the normal processing of discourse (Fletcher &Bloom, 1988; Myers, Shinjo, & Duffy, 1987). Moreover, causal relationsappear to be inferred during comprehension. Let us consider the followingstatements: Lawrence took the aspirins. His headache went away. Evidenceindicates that a causal connection (i.e., aspirin cures headaches) is automati-cally inferred as a ‘‘bridge’’ connecting the two statements (Singer, Halldor-son, Lear, & Andrusiak, 1992). In a study of story-retelling by one subjectwith RHD, Fredericksen and Stemmer (1993) observed causal relations asone component of a ‘‘narrative frame structure’’ which also included thetemporal sequencing of events. Their patient tended to preserve this framestructure. Our study applies Trabasso’s analysis of logical coherence to nar-ratives produced by a small group of RHDs.

ROLE OF THE TASK IN NARRATIVE PRODUCTION

In summarizing the research, Table 2 shows the types of discourse andtasks that have been employed in 21 studies of discourse production by per-sons with RHD. Narration has been examined more frequently than descrip-tive and procedural discourse. Picture-elicitation has been utilized muchmore than auditory–oral retelling, conversation, and spontaneous elicitation(i.e., instruction to produce a discourse). Usually one task was used. Typeof discourse appears to be associated with a particular task. In one study,an instruction to tell a story from a single picture-elicited description instead(Tompkins et al., 1993). This may be one reason for the tendency to preferpicture sequences to elicit narration, whereas procedural discourse is oftenelicited with instruction to tell how to fix a flat tire or bake a cake (e.g.,

188 DAVIS, O’NEIL-PIROZZI, AND COONT

AB

LE

2C

ompa

riso

nof

21St

udie

sof

Dis

cour

sePr

oduc

tion

with

RH

DSu

bjec

ts

Ana

lysi

s

Dis

cour

sety

peSt

imul

us/T

ask

With

in-s

ente

nce

Bet

wee

n-se

nten

ce

Des

cN

arr

Proc

Spon

Pic

Ret

ell

Con

vIn

foL

exSy

nC

ohes

Mac

roA

ttr

Met

zler

&V

elin

ek(1

977)

XX

XX

XM

yers

(197

9)X

XX

XX

Riv

ers

&L

ove

(198

0)X

XX

XW

apne

ret

al.

(198

1)X

XX

Tru

pe&

Hill

is(1

985)

XX

XX

XX

Joan

ette

etal

.(1

986)

XX

XX

Mac

kisa

cket

al.

(198

7)X

XX

Prut

ting

etal

.(1

987)

Xco

nvR

oman

etal

.(1

987)

XX

XX

Bry

an(1

988)

XX

Cim

ino

etal

.(1

991)

XX

XU

ryas

eet

al.

(199

1)X

XX

XB

loom

etal

.(1

992)

XX

XX

Glo

sser

etal

.(1

992)

XX

XX

XX

Blo

omet

al.

(199

3)X

XX

XC

hern

eyet

al.

(199

3)X

XX

XX

XX

Tom

pkin

set

al.

(199

3)X

XX

XFr

eder

ikse

n&

Stem

mer

(199

3)X

XX

XB

loom

(199

4)X

XX

XK

enne

dyet

al.

(199

4)X

conv

Mye

rset

al.

(199

4)X

XX

Not

e.D

isco

urse

type

sin

clud

ede

scri

ptio

n,na

rrat

ion,

and

proc

edur

es.T

asks

incl

ude

spon

tane

ous

prod

uctio

n,pi

ctur

e-el

icita

tion,

rete

lling

,and

conv

ersa

-tio

n.A

naly

ses

are

atle

vels

with

in-s

ente

nces

(inf

orm

atio

n/co

nten

t,le

xico

n,an

dsy

ntax

)an

dbe

twee

n-se

nten

ces

(loc

alco

hesi

on,g

loba

lm

acro

stru

ctur

e,an

dge

nera

lat

trib

utes

that

may

belo

cal

orgl

obal

and

incl

udes

conv

ersa

tiona

lan

alys

is).

NARRATIVE PRODUCTION 189

Cherney & Canter, 1993). As a result, comparing types of discourse mayentail additional variations in tasks used to elicit these types. Of the studiesin Table 2, only Bloom’s series of analyses employed picture sequences toelicit both narration and procedural discourse (Bloom et al., 1992, 1993).

There are indications that the task may influence characteristics of dis-course exhibited by normal and brain-damaged subjects (Cherney & Canter,1993; Shadden, Burnette, Eikenberry, & DiBrezzo, 1991). One reason forconcern is that diagnosis of discourse production deficit may be more relatedto task characteristics than to discourse ability. Regarding RHD, visuospatialimpairments have been among several disorders considered to be responsiblefor discourse performance. RHDs have had problems with interpreting theemotional significance of pictured scenes (Cicone, Wapner, & Gardner,1980; Dekoskey, Heilman, Bowers, & Valenstein, 1980), recognizing im-plicit information in pictures (Myers, Linebaugh, & Mackisack-Morin,1985), and organizing cartoon frames into their narrative sequence (Huber &Gleber, 1982). Using content analysis of descriptive discourse, Myers andBrookshire (1994) compared the influence of visually and inferentially sim-ple and complex illustrations. The investigators found that picture complex-ity had little influence on information content. The impact of pictured stimuliremains unclear without additional analysis of its influence on between-sen-tence levels of discourse production. Bloom’s studies indicate that emotionalcontent is more detrimental to discourse production at different levels thanvisuospatial information (Bloom et al., 1992, 1993).

THE PRESENT STUDY

Table 2 also indicates that analysis discourse production with RHD hastended to involve either counting information units the size of words or shortphrases, or tracking anticipated general categories of behavior. Most of thelarge group studies have been done at the within-sentence level. There havebeen only 3 studies of referential cohesion, whereas there have been 10 stud-ies of informational content. Thus, we have relatively little information basedon systematic analysis of coherence concerning a clinical population pre-sumed to have difficulties at the between-sentence level of language produc-tion. Moreover, when theoretical questions are raised, the explanatory factoris often not clearly related to the parameter of language being analyzed. Whatis in the nature of discourse formulation and RHD that would lead us tobelieve that emotional content might have an influence on referential cohe-sion?

Although not designed to test theoretical issues, the presented study wasintended to contribute to the development of measures of the between-sen-tence level of language production. Developing such measures should beimportant for future research that relates the measurement to theoretical as-sumptions underlying the choice of independent variables. We posed two

190 DAVIS, O’NEIL-PIROZZI, AND COON

TABLE 3Summary of RHD Subject Characteristics

YearsSex Age educ Lesion location

1 F 41 12 Frontal-posterior2 M 67 13 Frontal-parietal3 M 77 12 Frontal4 M 41 14 Temporal-parietal5 F 67 8 Frontal-posterior6 F 70 12 ‘‘Large’’7 M 73 18 Parietal8 F 50 12 Frontal-parietal

Note. Focal lesions were caused by thromboembolic infarction ex-cept for RHD-8 who had a hemorrhage.

main questions. (1) Do RHDs have deficits that are unique to between-sen-tence features of discourse production? We were primarily interested in twofeatures, namely, (a) referential cohesion between pronominal and lexicalelements and their antecedents and (b) logical coherence as indicated bycausal relations between propositional statements in a narration. There havebeen few studies of local cohesion with RHD, and we were aware of onlya case study regarding logical coherence. (2) Do referential cohesion andlogical coherence vary according to the task or condition used to elicit narra-tion? Our comparison was confined to tasks that would elicit one type ofdiscourse as opposed to mixing task with type of discourse. There were twobasic tasks, namely, narrative elicitation from picture-story sequences andretelling a short story. We created three conditions by introducing a memorycomponent to the picture-elicitation procedure that was comparable to thememory component of the retelling task.

METHOD

Subjects

The patient group was selected from in-patients and out-patients at rehabilitation hospitalsin New England (Table 3). This group consisted of eight individuals with focal right hemi-sphere lesions in varied locations confirmed by CT scans. Seven subjects had single hemi-sphere thromboembolic cerebrovascular accidents, and one had a right hemisphere hemor-rhage. The experiment was conducted at slightly over a year postonset for most subjects, butone was 5 months postonset and another was studied at 5 years postonset. The four malesand four females had a mean age of 61 years (41–77) and a mean education of 12.5 years(8–18). Medical records were reviewed to eliminate history of mental illness or dementia,and routine audiological examinations revealed normal hearing.

A picture-matching pretest was administered to maximize the likelihood that experimentalsubjects had adequate perception of visual stimuli similar to drawings used in the study. Tolook for an influence of left visual neglect, the correct match appeared equally on the left,

NARRATIVE PRODUCTION 191

right, or center. The eight subjects had a minimum score of 8/10, and seven subjects scoredat least 9 points without displaying a preference for choices on the right side.

The eight neurologically intact control subjects (NC) included hospital staff and other indi-viduals in the region. The three males and five females had a mean age of 55 years (39–72)and a mean education of 13.5 years (8–20).

Materials

The visual-elicited conditions employed two sets of cartoon sequences shown in previouspublications. One set, called the Car Repair, contains three frames which we arranged vertically(see Rivers & Love, 1980). In the story, a group of men are trying to repair a stalled car, anda woman comes along and fixes the car. The other set, called the Flower Pot story, containssix frames arranged in two rows (see Huber & Gleber, 1982). A man is walking his dog andgets hit on the head by a falling flower pot. He stomps into an apartment building but discoversthat the culprit is a nice woman, and the man’s attitude changes. For the auditory-retellingcondition, two short folktales were borrowed from literature on story grammar and causalrelations (Stein & Glenn, 1979; Trabasso, Secco, & van den Broek, 1984). One story, calledthe Tiger’s Whisker, is about a wife who needs to acquire a tiger’s whisker to use as medicinefor her sick husband. The other folktale, called Epaminondas, is about a little boy who doeserrands for his mother and grandmother, and he becomes confused about their instructions.Examples of each story, as told by control subjects, are presented in the Appendix.

Eliciting Narratives

Six narratives were generated in three experimental conditions. Narratives were elicited bypicture sequences in two conditions. In Elicited-A, a cartoon sequence was placed before asubject. The examiner oriented the subject by pointing to each picture in order. Subjects weretold that the pictures tell a story and then were asked to tell the story depicted, Car Repairfirst and Flower Pot second. The picture sequences were also in view of the examiner whilethe subject formulated a story. For Elicited-B, subjects were asked to retell each story withthe pictures absent. They were shown the cartoon briefly as a reminder, and the picture wasremoved from view. Then, the Retelling condition was administered. Subjects repeated theTiger’s Whisker first and then Epaminondas. An examiner read a story aloud with naturalprosody, and subjects were instructed to repeat the story as completely as possible. This orderof conditions was the same for all subjects. The six stories were usually obtained in a 20-minsession.

Analyzing Narratives

Two types of transcripts were developed from audio recordings. One was a standard para-graph format that was used for referential cohesion analysis. The second was derived fromthis transcript by identifying propositional statements and then listing them. Each transcriptwas coded so that the experimenters would not know a subject’s group membership duringthe analyses.

Referential cohesion. The paragraph transcripts were used for a cohesion analysis basedmainly on Halliday’s (1985) functional grammar. A simplified procedure focused on the mostcommon cohesive devices, namely, personal pronouns (e.g., he, her), demonstrative pronouns(e.g., this, here), and lexical reference signaled by the before a noun (e.g., the dog, referringto previous introduction of a dog). In a departure from previous research, certain pronounswere not automatically tagged for analysis. That is, a few demonstratives were considered tobe figures of speech not requiring an explicit antecedent (e.g., There was a lady . . .), and these

192 DAVIS, O’NEIL-PIROZZI, AND COON

terms were labeled as ‘‘neutral.’’ The basis for cohesion analysis became the identification of‘‘clear elements’’ which excluded the relatively few neutral items.

The first mention of an explicit referent in a narration was identified by drawing a squarearound it and numbering it. A complete tie was labeled by assigning a referent’s number toall related clear cohesive elements. A cohesion ratio was computed as the number of completeties relative to the number of clear elements produced. Reliability between two investigatorswas computed for two subjects in each group. Agreement was 100% for number of clearelements and 86% for being within 0.05 for the cohesion ratio.

Logical coherence. To prepare for analysis of logical coherence, each paragraph transcriptwas transformed into a numbered list of propositions in a manner consistent with proceduresof Stein and Glenn (1979) and Trabasso and van den Broek (1985). According to their proce-dures, a proposition contains an explicitly stated predicate (i.e., action or state) and its relatedarguments (e.g., agent, recipient, instrument). Such propositions may be referred to as state-ments (van den Broek, 1994) and differ from the hierarchical deep structural units employedby Kintsch (1994; also Joanette et al., 1986). In the following example, propositional state-ments are listed from the first part of the Flower Pot story provided in the Appendix:

A man is walking along.He walks under a balcony in a building.A flower pot falls offhits him on the head.He starts screaming back into the areawhere the flower pot came fromhis dog is barking.

Interexaminer agreement was 90% between two experimenters.The analysis of logical coherence was based on identifying causal relations between proposi-

tions according to Trabasso and Sperry’s (1985) ‘‘criterion of necessity’’ which states thatproposition B would not have occurred if proposition A had not existed or occurred. A causalconnection could be one of Trabasso and Sperry’s four types of relations considering commonknowledge and the context of the story: physical (mechanical causality between objects and/or people, such as getting hit on the head causing a bump or the hot sun causing butter tomelt), motivational (volitional, generally between a character’s goal and an action, such asthe need for medicine causing the search for a tiger’s whisker), psychological (involuntary,involving a non-goal-directed action or effect, such as getting hit on the head causing the manto scream), or enablement (actions/states that are necessary but not causal, such as the womanwalking her dog enabling her to encounter a disabled car).

Once a relation was identified, an arrow was drawn from the cause (proposition A) to itsconsequence (proposition B). One event may have multiple consequences, and multiple eventsmay have one consequence. For example, in Epaminondas, butter would not have melted ifthe boy had not put it on his head and the sun was not hot. A transcript was analyzed on itsown terms by minimizing consideration of what the story was supposed to be.

Two measures were used. The total number of causal connections is indicative of the extentto which logical relations are produced. A connectivity score is the average number connectionsfor a proposition either to or from other propositions is indicative of the density of connectionsor a story’s ‘‘tautness.’’ These scores were based on acknowledging that logical relations area variable product of a listener’s or reader’s comprehension; that is, there can be no absolutenumber of ‘‘correct’’ relations that can be inferred from a narration. Connectivity scores forthe experiment were derived from resolution of disagreements between the first and secondauthors after having scored each narrative independently.

Accuracy. Because an incorrect narration could be cohesive or coherent, we added a roughmeasure of accuracy to our analysis. Accuracy was judged relative to a story’s theme andpoint. A completely accurate theme or point was assigned 2 points based on containing two

NARRATIVE PRODUCTION 193

TABLE 4Summary of Results for the Principal Measures in the Referential Cohesion Analysis,

Including Number of RHDs Beneath the Range for Neurologically Intact Controls

Elicited-A Elicited-B Retelling

Cohesive Cohesion Cohesive Cohesion Cohesive Cohesionties ratio ties ratio ties ratio

NCMean 23.4 0.85 22.3 0.87 30.1 0.96Range 14–39.5 0.75–0.95 14–43 0.76–0.96 26–36.5 0.94–1.00

RHDMean 17.6 0.74 14.8 0.81 27.6 0.91Range 2–29 0.18–0.91 4–28 0.4–0.97 20.5–35.5 0.82–0.96

Below NCs 3 2 4 1 3 5

components. Incomplete responses were assigned 1 point, and completely inaccurate or omit-ted responses were assigned 0 points. Overall accuracy was established by adding scores fortheme and point so that the best score was 4 points. Agreement between two examiners was94% for three randomly selected subjects.

Criteria were established for completeness of theme and point. A complete theme of theCar Repair story is that it is about a disabled car and men are trying to fix it, and a completepoint is that the men are surprised because a woman fixed the car. The Flower Pot story isabout a man getting hit on the head with a pot and becoming very angry, and the point isthat he discovers the culprit is a woman and his attitude changes. The theme of the Tiger’sWhisker is that a woman seeks a whisker to make medicine for her husband, and the pointis that she gets the whisker leaving the tiger alone and sad. The theme of Epaminondas isthat a boy was asked to take cake to his grandmother and was warned to be careful, and thepoint is that his butter melts on a return trip and he gets scolded again. In defining each point,we selected consequences that were pictured or stated in the stimuli.

RESULTS

Effects of group, condition, and story were determined with nonparametriccomparisons (Norusis, 1993). Between-group comparisons were conductedwith Mann–Whitney U tests of independent measures to determine deficits,and within-group comparisons were conducted with Wilcoxon signed-rankstests of related measures to determine effects of tasks or conditions. Thecriterion for significance of differences was p , .05. Deficits within the pa-tient group are also indicated by the number of RHDs with scores beneaththe range of the control group.

Referential Cohesion

Table 4 summarizes results for the main parameters of referential cohe-sion. RHDs did not differ significantly from the control group regarding thenumber of cohesive ties produced in each condition. Effects of conditiondiffered slightly according to subject group. For the RHDs, Retelling resulted

194 DAVIS, O’NEIL-PIROZZI, AND COON

FIG. 1. Referential cohesion ratios for each subject group in each experimental condition.

in significantly more ties than both Elicited conditions (z 5 22.1004, p 5.0357; z 5 22.5205, p 5 .0117), and the two Elicited conditions did notdiffer significantly from each other (z 5 2.4901, p 5 .6241). For the NCgroup, Retelling resulted in more ties than Elicited-B (z 5 22.1004, p 5.0357) but not Elicited-A (z 5 21.6803, p 5 .0929), and the two Elicitedconditions did not differ (z 5 2.2801, p 5 .7794). For the RHD group,Flower Pot elicited more ties than Car Repair in the Elicited-B condition (z5 22.1974, p 5 .0280), and Epaminondas resulted in more ties than Tiger’sWhisker in the Retelling condition (z 5 22.5205, p 5 .0117). Story hadthe same effect in Retelling for the control group (z 5 22.5205, p 5 .0117).

Regarding cohesion ratios, RHDs were significantly lower than the NCgroup for the Retelling condition (U 5 11.5, p 5 .0281) but not for Elicited-A (U 5 26.5, p 5 .5737) nor Elicited-B (U 5 27.5, p 5 .6454) (Fig. 1).Whereas the differences between means were larger for the two Elicitedconditions, the number of RHDs beneath the normal range was much greaterin the Retelling condition (Table 4). The group effect for Retelling occurredwith the Epaminondas story (U 5 9.5, p 5 .0148) but not with the Tiger’sWhisker (U 5 22.5, p 5 .3282). Among conditions, Retelling had the highestcohesion ratio. It differed significantly from the Elicited-A condition for theRHD group (z 5 22.2404, p 5 .0251) and from both Elicited conditionsfor the NC group (z 5 22.3805, p 5 .0173; z 5 22.2014, p 5 .0277).When stories were compared for RHDs, Flower Pot had a significantly higherratio than Car Repair in Elicited-A (z 5 22.1704, p 5 .0300), and Epami-nondas had a significantly higher ratio than Tiger’s Whisker in Retelling (z5 22.2404, p 5 .0251). For the control group, the same effect occurred inRetelling, but Car Repair was found to have a significantly higher ratio thanFlower Pot in Elicited-A (z 5 22.1004, p 5 .0357). There was no significantdifference between stories in Elicited-B within each subject group.

NARRATIVE PRODUCTION 195

TABLE 5Summary of Results for Number of Propositional Statements in Narrations in Each Story

and Each Condition, Including Number of RHDs Beneath the Range for Neurologically IntactControls

Elicited-A Elicited-B Retelling

CR FP Total CR FP Total CR FP Total

NCMean 18.1 18.9 37.0 16.1 20.4 36.5 16.8 19.4 36.1Range 26–54 23–66 30–39

RHDMean 13.8 13.9 27.6 10.9 13.6 24.5 14.3 17.5 31.8Range 9–43 11–41 20–38

Below NCs 4 4 2

Logical Coherence

Narratives were divided into propositional statements as a foundation forthe causal relations analysis. Summaries of performance are shown in Table5. As indicated in Fig. 2, RHDs produced significantly fewer propositionsthan the NC group only in the Elicited-B condition (U 5 11.0, p 5 .0281).This difference occurred mainly with the Flower Pot story (U 5 13.5, p 5.0499). Table 5 shows that four RHDs were beneath the normal range in theElicited conditions, and this was twice the number beneath the normal rangein Retelling. There were no significant differences among conditions for ei-ther group. Only the NC group varied in number of statements producedaccording to story. Flower Pot elicited more statements than Car Repair in

FIG. 2. Number of propositional statements (‘‘p units’’) produced by each subject group ineach experimental condition.

196 DAVIS, O’NEIL-PIROZZI, AND COON

TABLE 6Summary of Causal Relations Analysis Consisting of Measures of Number of Connections

and the Connectivity Score, Including Number of RHDs Beneath the Range for NeurologicallyIntact Controls

Elicited-A Elicited-B Retelling

Connectivity Connectivity ConnectivityConnections score Connections score Connections score

NCMean 13.4 1.43 12.5 1.41 15.9 1.79Range 8.0–19.5 1.15–1.60 7.5–20.5 .95–1.77 12.5–19.5 1.31–2.01

RHDMean 8.3 1.22 7.2 1.20 12.4 1.59Range 2.5–13.5 1.00–1.88 3.0–12.5 .72–1.70 9.5–19.0 1.17–2.02

Below NCs 3 5 4 2 4 2

Elicited-B (z 5 22.1704, p 5 .0300), and Epaminondas elicited more state-ments than Tiger’s Whisker in Retelling (z 5 22.2404, p 5 .0251).

Logical coherence was analyzed with respect to causal relations betweenpropositions. Table 6 summarizes results for number of logical connectionsand the connectivity score. RHDs produced significantly fewer connectionsthan controls in each condition, namely, Elicited-A (U 5 11.0, p 5 .0281),Elicited-B (U 5 8.0, p 5 .0104), and Retelling (U 5 11.0, p 5 .0281). Theeffect occurred for both stories in Elicited-A and for the Flower Pot in Elic-ited-B. Condition produced an effect for the RHD group but not the NCgroup. For the patient group, Retelling was found to have significantly moreconnections than Elicited-A (z 5 22.3805, p 5 .0173) and Elicited-B (z 522.3664, p 5 .0180. The difference between Elicited-A and Elicited-B wasnot significant (z 5 21.8869, p 5 .0592). The NC group did not have sig-nificant differences between conditions. The effects of story differed betweensubject groups. There were no significant differences between stories forRHDs. For the control group, Flower Pot elicited significantly more connec-tions than Car Repair in Elicited conditions (z 5 22.3805, p 5 .0173; z 522.5205, p 5 .0117).

The connectivity score is a measure of causal relatedness relative to thenumber of propositions produced. The effect of subject group differed ac-cording to condition (Fig. 3). RHDs had significantly lower connectivity thanthe NC group only in the Elicited-A condition (U 5 12.0, p 5 .0379). Table6 shows that five RHDs were beneath the normal range of connectivity inthis condition, whereas two were beneath the normal range in Elicited-B andRetelling. There were effects of condition in each subject group. For RHDs,Retelling resulted in significantly higher connectivity than the Elicited-Bcondition (z 5 22.5205, p 5 .0117). For the control group, Retelling resultedin significantly higher connectivity than Elicited-A (z 5 22.3805, p 5

NARRATIVE PRODUCTION 197

FIG. 3. Logical connectivity scores for each subject group in each experimental condition.

.0173) and Elicited B (z 5 22.5205, p 5 .0117). The only significant effectof story was that Flower Pot produced higher connectivity than Car Repairin the NC group in Elicited-A (z 5 22.5205, p 5 .0117).

Accuracy

A summary of accuracy scores is shown in Table 7. A Mann–Whitneytest of total accuracy scores showed that subject group differences variedaccording to condition (Fig. 4). RHDs were significantly less accurate thanthe NC group in Elicited-A (U 5 6.5, p 5 .0047) and Elicited-B conditions(U 5 5.0, p 5 .0030) but not in the Retelling condition (U 5 22.5, p 5.3282). The decrease in accuracy with RHD in the Elicited conditions is alsoindicated by the four or five patients who were beneath the range of controlsubjects in contrast to one RHD beneath the control range in Retelling. TheWilcoxon test showed that condition made a difference for the experimental

TABLE 7Summary of Results for Accuracy of Narration, Including Number of RHDs Beneath the

Range for Neurologically Intact Controls

Elicited-A Elicited-B Retelling

Theme Point Total Theme Point Total Theme Point Total

NCMean 4.0 2.9 6.9 4.0 3.0 7.0 3.6 3.5 7.1Range 4 1–4 5.0–8.0 4 1–4 5.0–8.0 3–4 3–4 6.0–8.0

RHDMean 3.0 1.0 4.0 2.9 0.9 3.8 3.3 3.3 6.5Range 2–4 0–2 2.0–7.0 1–4 0–2 1.0–7.0 2–4 2–4 4.0–8.0

Below NCs 5 4 5 5 4 4 2 1 1

198 DAVIS, O’NEIL-PIROZZI, AND COON

FIG. 4. Total accuracy scores for each subject group in each experimental condition.

group but not the control group. RHDs were significantly more accurate withthe Retelling condition than Elicited-A (z 5 22.5205, p 5 .0117) and Elic-ited-B conditions (z 5 22.5205, p 5 .0117). For each subject group, thestories within each condition were equivalent with respect to global accuracy.

RHDs were significantly less accurate than the NC group in conveyingthe theme and point in both Elicited conditions. Mann–Whitney U’s rangedfrom 5.5 to 12.0 (p 5 .0030 to .0379). Table 4 indicates that both subjectgroups had more difficulty conveying the point than the theme in the Elicitedconditions. Within the patient group, the point was significantly less accuratethan the theme in Elicited-A (z 5 22.5205, p 5 .0117) and Elicited-B (z5 22.5205, p 5 .0117). This difference for the point between conditionsis further indicated by the one RHD who was beneath the control range forthe point in Retelling contrasted with the five RHDs beneath the range inElicited-A and four RHDs beneath the range in Elicited-B. Within the controlgroup, the point was also significantly less accurate in Elicited-A (z 522.0226, p 5 .0431) and Elicited-B (z 5 22.2014, p 5 .0277).

Individual Subjects

Table 8 summarizes results of the principal measures for each RHD sub-ject compared to the mean and lowest scores for the control group. The RHDscores that are beneath the NC range of performance shall be referred to asdeficit scores. Elicited-A, with the pictures present, contained the most deficitscores with 16. Elicited-B had 11 deficit scores, and Retelling contained10. Accuracy and number of propositional statements had 10 deficit scores.Logical coherence had 9, and referential cohesion had 8. The number ofthese scores tended to differ according to condition; that is, 9 deficit scoresfor accuracy occurred in Elicited conditions, whereas only 1 occurred in

NARRATIVE PRODUCTION 199

TA

BL

E8

Indi

vidu

alR

HD

Subj

ects

Com

pare

dto

NC

Mea

nsan

dL

owes

tSc

ore

for

the

Prin

cipa

lM

easu

res

Elic

ited-

AE

licite

d-B

Ret

ellin

g

acc

ref

coh

pun

itslo

gco

hac

cre

fco

hp

units

log

coh

acc

ref

coh

pun

itslo

gco

h

NC

mea

n6.

90.

8537

.01.

437.

00.

8736

.51.

417.

10.

9636

.11.

79N

Clo

w5.

00.

7526

.01.

155.

00.

7623

.00.

956.

00.

9430

.01.

31R

HD

-12.

00.

8120

.01.

882.

00.

8423

.01.

496.

00.

9636

.01.

67R

HD

-25.

00.

8233

.01.

015.

00.

9621

.01.

706.

00.

8820

.02.

00R

HD

-34.

00.

9142

.01.

285.

00.

8941

.01.

227.

00.

9236

.01.

45R

HD

-46.

00.

9043

.01.

125.

00.

7622

.01.

298.

00.

9438

.02.

02R

HD

-52.

00.

189.

01.

001.

00.

4011

.00.

977.

00.

9532

.01.

43R

HD

-64.

00.

6417

.01.

283.

00.

8521

.00.

726.

00.

9032

.01.

18R

HD

-72.

00.

7725

.01.

052.

00.

9734

.00.

834.

00.

8232

.01.

17R

HD

-87.

00.

9132

.01.

147.

00.

7923

.01.

398.

00.

8828

.01.

80

Not

e.T

heac

cura

cysc

ore

isth

eto

talo

fbo

thst

orie

sin

each

cond

ition

.The

refe

rent

ialc

ohes

ion

ratio

isth

eav

erag

efo

rth

etw

ost

orie

sin

each

cond

ition

.P

units

are

the

num

ber

ofpr

opos

ition

alst

atem

ents

tota

led

for

both

stor

ies

inea

chco

nditi

on.

The

logi

cal

cohe

sion

scor

eis

the

conn

ectiv

ityav

erag

edfo

rth

etw

ost

orie

sin

each

cond

ition

.

200 DAVIS, O’NEIL-PIROZZI, AND COON

Retelling. This pattern was similar for propositions. On the other hand, 3deficit scores occurred in the two Elicited conditions for referential cohesion,and 5 deficit scores occurred in Retelling.

RHD-5, RHD-6, and RHD-7 had the most deficit scores (i.e., 7 or 8). Anapparent gap exists between these subjects and the others whose number ofdeficit scores ranged from 2 to 4. RHD-3 and RHD-4 tended to be withinthe control group’s range and had the fewest deficit scores with 2. Subjectsappear to have differed with respect to parameters. Among those with thefewest deficit scores, RHD-1 had a pronounced difference from the NC groupin accuracy in Elicited conditions. However, RHD-8 had no difference fromNC performance in accuracy and was close to the normal range in deficitscores. In the group with the most deficit scores, all of RHD-5’s deficit scoreswere in the two Elicited conditions. However, RHD-6 and RHD-7 had somedeficit scores in Retelling, also. RHD-6 and RHD-7, who had the most deficitscores, were the only subjects with deficit scores in connectivity in Elicited-B and Retelling.

DISCUSSION

This study examined referential cohesion and logical coherence of narra-tive production after right hemisphere stroke. Significant between-group dif-ferences indicate that right hemisphere stroke caused deficits in narrativeproduction. However, identification of impairment depended on the featureof discourse measured and on the task used to elicit narration. Also, perfor-mance of both groups occasionally depended on the particular story beingelicited. The following discussion begins by addressing the group and taskeffects for each type of discourse analysis.

Referential Cohesion

RHDs as a group were not impaired in the number of cohesive ties pro-duced but were impaired according to the cohesion ratio, indicating thatfinding a deficit depends on how referential cohesion is measured. Consistentwith our results, Glosser and others (1992) found RHDs were impaired ac-cording to a ratio of incomplete ties to total words produced in an interviewsituation. Contrary to our result with number of cohesive ties, Bloom (1994)found a deficit according to number of incomplete ties for narration to picturesequences. A ratio of cohesiveness appears to be a more consistent indicationof problems with referential cohesion than the number of cohesive ties.

The deficit of cohesiveness occurred in the Retelling condition but not inthe Elicited conditions. Uryase and others (1991) found a deficit for a similarratio when subjects recalled a video story. The common feature of deficitwhen retelling indicates that cohesiveness becomes a problem for RHDswhen narration is evaluated in a recall condition. However, Uryase’s finding

NARRATIVE PRODUCTION 201

is inconsistent with our result in Elicited-B or the picture-story recall condi-tion. Our controls achieved ratios in Elicited conditions that were similar toUryase’s controls, but our RHDs scored 0.81 in Elicited-B after scoring 0.74in Elicited-A with the pictures in view. Because our conditions were pre-sented in the same order for each subject, our RHDs’ cohesion ratio in Elic-ited-B may have benefited from practice. The influence of stimulus recallon referential cohesion may need to be targeted for future study in whichthe same content is used for each condition and order of conditions is coun-terbalanced.

Logical Coherence

Narrations were divided into propositional statements in order to analyzelogical coherence. Because length of our propositions varied widely in num-ber of words and noun phrases, number of propositions should be consideredto be only coincidentally indicative of amount of formulation produced. TheRHD group was impaired in number of propositions only for the Flower Potstory in the Elicited-B condition. Table 8 indicates that only RHD-5 had apronounced deficiency of propositions. Our RHDs generally produced anumber of propositional statements that was sufficient for examining logicalcoherence.

Our patient group was deficient in the number of causal connections de-tected in each condition. RHDs also differed from the control group by pro-ducing fewer connections in both Elicited conditions than in Retelling,whereas conditions did not differ within the control group. Regarding theconnectivity score, RHDs were impaired only in Elicited-A in which thepictures were displayed. These findings contrast with those of referentialcohesion where number of ties was not deficient and cohesion deficit oc-curred in Retelling rather than Elicited conditions. Therefore, referential co-hesion and logical coherence may be influenced by different types of phe-nomena in the discourse formulation process.

An example of low connectivity (i.e., 0.33) comes from RHD-6’s CarRepair narration in Elicited-B:

(1) There was people goin out in an automobile,and the automobile broke down.And looks like the oil filter was the trouble.And they got outand started off, off again on their tripwhereever they were goin.And there was a lady there with her dog,walking her dog.And she’s in on this little safarior whatever it is.And they seem to be havin offto have a good time somewhere, the whole bunch of them.

202 DAVIS, O’NEIL-PIROZZI, AND COON

The only causal relations in this example are (a) from the third to secondproposition, namely, that the oil filter caused the car to break down, and (b)between the last two propositions, namely, that having a good time motivatesthe people to ‘‘be havin off.’’ Otherwise, two investigators agreed that noother causal connections could be reasonably inferred.

A standard of story-telling, which appeared early in the literature on causalrelations (e.g., Trabasso & Sperry, 1985), is that a concise logical sequencewould be indicated by a connectivity score close to 2.00. That is, each propo-sitional statement would have causal input from another statement and beconnected to another statement. A more densely coherent story might havea salient statement connected to several other statements, and this would beindicated by a connectivity score above 2.00. A striking finding of this studyis that connectivity of 2.00 or higher was achieved for only 6 of the 48 storiesproduced by the neurologically intact control group. The controls averaged1.43 in Elicited-A, 1.41 in Elicited-B, and 1.79 in Retelling. In the NC group,21 of the 32 cartoon-elicited stories were below 1.50, and 2 of the 16 narra-tions in Retelling were below this level. One NC subject was below 1.00for both stories produced in Elicited-B. Thus, it could be said that the controlgroup was less coherent than might have been expected, especially in theElicited conditions. It could also be said that RHDs were more coherent thanexpected.

Accuracy

RHDs’ accuracy was impaired in both picture-elicited conditions. Four orfive RHDs were beneath the normal range of accuracy when telling storiesfrom pictures, whereas the patient group was not deficient in retelling folk-tales accurately. Only one subject was below the normal range in retelling.Overall accuracy was equivalent between the two Elicited conditions, indi-cating that it did not matter whether the pictures were in view or removedfor subsequent recall. Moreover, the three conditions were equivalent for thecontrol group, indicating that the effect of conditions was unique to righthemisphere dysfunction.

The following examples show a subject who was accurate for theme butomitted the point in the Elicited-B condition with pictures absent (RHD-2)and another subject who was erroneous for both in the Elicited-A condition(RHD-1):

(2) RHD-2 (Car Repair) A car became disabled, and then a bunch of people camearound the car. And they were trying to diagnose what the problem with theengine was. So they lift up the hood, and they were lookin in the engine. Andthen apparently whatever was wrong was fixed, because then the car drove off.

(3) RHD-1 (Flower Pot): It looks like the man is out walking his pet dog, and itlooks kinda like he’s lost, and he’s looking for help. So he goes banging on one

NARRATIVE PRODUCTION 203

of the doors, and a lady opens her door, and out runs her pet. Looks like he’sasking her for directions, and she gives ’em to him.

Example (3), in particular, demonstrates that accuracy of story-telling is in-dependent of logical cohesion. The connectivity score for this story was 2.25.When comparing Figs. 3 and 4, it appears that RHDs were distinctively morelike the control group in logical coherence than in accuracy, especially inthe Elicited conditions.

The RHD group was impaired in the Elicited conditions for both the theme(e.g., initiating event) and the point (e.g., consequences, characters’ reac-tion), but there was no impairment of the theme or point in Retelling. Thelow scores for conveying the point are of particular interest because of theimpression from previous research that RHDs have impairments in remem-bering the point or moral of stories (e.g., Wapner et al., 1981). Becauseaccuracy for conveying the point was significantly lower than the theme inboth groups in the Elicited conditions, greater difficulty conveying the pointdoes not appear to be unique to right hemisphere dysfunction. Reformulatinga point appears to be generally more difficult than a theme, perhaps, becauseformulating a point relies more on inference than formulating a theme. Thelack of difficulty with the point in Retelling may be a reflection of the generalsuperiority of paraphrasing stories presented in an auditory–verbal mode.All someone has to do is repeat what he or she has heard.

Individual Patterns and Problems

A wide variation of performance levels by RHDs is indicated by the rangesof scores. The NC group was also quite variable. Joanette and others (1986)emphasized the heterogeneity of their subject groups in number of proposi-tions. Uryase and her colleagues (1991) found wide-ranging cohesion ratiosin their RHD and NC groups with considerable overlap between groups.Trupe and Hillis (1985) divided their RHDs into five distinct subgroups withone considered to be normal in discourse production. In our study, a gapexisted between a minimally impaired subgroup with two to four principaldeficits and a subgroup with seven or eight principal deficit scores. Severalstudies indicate that, like aphasias with LHD, people with RHD are not ahomogeneous group. However, contrary to the frequent comparison of apha-sic syndromes, investigators have found it difficult to identify syndromes ofRHD for similar comparisons.

Let us consider the five members of the less impaired RHD subgroup (seeTable 8). Subjects 3 and 4 had the minimum of two deficit scores across thethree conditions. Subjects 1, 2, and 8 had three deficit scores. Only RHD-1appeared to have a singular impairment. Her pattern was similar to that ofthe RHD group as a whole. That is, she was relatively low in accuracy forpicture-elicited stories and relatively good in referential cohesion and logicalcoherence across tasks. In Elicited-B, she repeated the ‘‘lost man’’ version

204 DAVIS, O’NEIL-PIROZZI, AND COON

of Flower Pot that she had produced in Elicited-A (see Example 3), indicatingthat this version was the mental model of the story retained in memory.

Three RHDs were beneath the normal range on more than half of theprincipal scores across the three conditions. RHD-6 and RHD-7 had deficitsin eight principal scores. These subjects were unique in that their connectiv-ity was relatively low in Retelling, which may be indicative of increasedoverall severity of narrative deficit. RHD-7, in particular, had a pronouncedproblem with accuracy and logical coherence across all conditions. All sevenof RHD-5’s deficit scores were in the Elicited conditions, which is like RHD-1 regarding impaired conditions but unlike RHD-1 regarding scope of im-pairment within conditions. In general, it appears that an RHD individualwith more severe narrative production deficits can have either multiple diffi-culties with picture-elicited stories or multiple difficulties across differenttasks.

RHD-5 was especially low in amount of language produced from the pic-ture sequences, indicated by the contrast between 9.0 propositions in Elic-ited-A and 32.0 propositions in Retelling. She produced only 2 complete tiesin Elicited-A, whereas four of the eight RHDs produced 21 to 29 ties. InElicited-B, she produced 4 ties. Thus, RHD-5 was exceptional with respectto the patient group regarding referential cohesion. The following examplesare illustrative of her low cohesion in the Elicited conditions. The first exam-ple is a ratio of 0.29, and the second is a ratio of 0.36

(4) RHD-5 (Car Repair, Elicited-B): Well, they had troubles with their car. In thesecond one, the girl tried to help repair it. In the last one, they seemed to givedirections, seemed to be the girl walking the dog.

(5) RHD-5 (Flower Pot, Elicited-A): The first one it looks like he’s returning homewith a stray dog. He takes him in. Third one. Fourth one, he’s banging on thedoor. Fifth one, he’s giving the dog a bone. Sixth one, he seems to be pleasedwith him, and the dog is taking off with his bone.

There is a small number of cohesive ties because of the absence of a specificantecedent early in each story. In these examples, they, he, and the girl haveno antecedent, whereas the dog was introduced earlier with a stray dog.These examples are also indicative of problems with accuracy and logicalcohesion. In example (5), the only causal relation was detected between thefirst and second statements resulting in a connectivity score of 0.50. RHD-5 was unique with respect to her education level of 8 years (Table 3), whichpoints to the value of controlling for education in studies of discourse.

Stories

The comparison of individual stories was intended to provide a clue towhether one methodological aspect of narrative assessment could be prob-lematic. Sometimes only one or two stories are used to generate conclusionsabout narrative production in general. In our study, story had a complex

NARRATIVE PRODUCTION 205

effect on the cohesion ratio. Flower Pot had a higher ratio than Car Repairin the RHD group only in Elicited-A, and a reversed effect occurred in theNC group in Elicited-A. Epaminondas had a higher ratio than Tiger’sWhisker in Retelling for the RHD group. Regarding logical coherence,Flower Pot had a higher connectivity than Car Repair in only the NC group.The findings with propositional statements indicated that story effects weremore prominent for productivity than the ratios. For example, Flower Pottended to elicit more cohesive ties and more logical connections than CarRepair, especially in the control group. Of course, this may have been dueto the greater number of frames in the Flower Pot story than in the Car Repairstory. Stories were equivalent within each condition for both subject groupswith respect to accuracy. RHD-1 and RHD-5 appear to have been excep-tional, because thematically incorrect stories with scores of 0.0 were uniqueto Flower Pot (i.e., ‘‘lost man,’’ ‘‘stray dog’’).

We also wondered if our data would point to an influence of the individualstory in the diagnosis of discourse deficit with RHD. The deficit of referentialcohesion in Retelling was restricted to Epaminondas, and the only deficit innumber of statements was with Flower Pot. Regarding logical coherence,both stories in the Elicited conditions tended to produce deficits in numberof connections and connectivity with one exception being number of connec-tions in Elicited-B. Stories were equivalent in exposing deficits of accuracy.Myers and Brookshire (1994) indicated that picture complexity has little im-pact when content analysis is employed. Bloom suggested that emotionalcontent is more likely to expose deficits than visuospatial content (Bloom,1994). Thus, pictures and story content may have some impact on diagnosinga deficit, but diagnosis may also depend on the level of measure used. Pre-senting more cartoon-sequences might be helpful in determining the breadthor consistency of difficulty.

Theoretical Implications

Our study was not designed to test a theory of discourse or discoursedysfunction. We were tempted to draw some conclusions regarding visuospa-tial influences on narration, because logical coherence and accuracy weredistinctly impaired in the picture-elicitation conditions. However, such con-clusions are strained, because we did not use the same narrative contentacross conditions and the order of tasks was not balanced. We mainly wantedto elicit four stories in a variety of ways in order to focus on certain between-sentence parameters of narrative production.

Several explanations of narrative production deficit with RHD have beenproposed in the literature. Most explanations seem to focus on contributionsof stimulus processing and task demands, such as requirements of visuospa-tial integration and inference, emotional content, and attention (Bloom, 1994;Myers, 1993). Some conclusions have been based on manipulating task vari-

206 DAVIS, O’NEIL-PIROZZI, AND COON

ables and then determining effects on familiar or general measures of dis-course production. However, it is not always clear why a particular indepen-dent variable would have an impact on content units, for example. The abilityto make predictions in theory-driven research may be enhanced by greaterattention to the possible relationships between task variables and the parame-ter of discourse production selected for analysis. A factor may be more likelyto affect referential cohesion than logical coherence or vice versa.

An illustration of this point may be drawn from our study. Elicited condi-tions produced lower cohesion ratios than Retelling for both groups. Thus,low ratios in the Elicited conditions may have been indicative of a normalcommunicative phenomenon. When pictures are in view, a speaker may as-sume that the listener can see a referent so that antecedents need not bestated. So-called exomorphic pronouns point outward to entities in the situa-tion. With our scoring system, we looked for antecedents in narration anddid not assume that subjects were referring to mutually known referents inthe pictures. Exomorphic pronouns would not have been counted as ties,thereby deflating the cohesion ratio (see Examples 4 and 5). In the study byUryase and others (1991), subjects told stories to naive listeners, therebymaking it communicatively necessary to state referents explicitly. Ratiosmay not have been deflated by a speaker’s assumptions about listener-knowl-edge, and yet, the RHD group’s cohesion ratio was still deficient. Uryase’sexperimental condition may have increased the likelihood of finding a refer-ential cohesion deficit over our picture-elicited conditions. The more basicpoint is that the communicative situation presented to subjects can be saidto have an explicit relationship to a measure of local coherence.

Summary and Conclusions

In this study, a group with right hemisphere dysfunction was compared toneurologically intact controls regarding the referential cohesion and logicalcoherence of narrative production. A somewhat varied sample of six storieswas obtained with tasks of cartoon-elicited story-telling and auditory–oralretelling. We found deficits in the patient group with respect to referentialcohesion, logical coherence, and accuracy of narration. The most consistentand pronounced difficulty occurred with accuracy of narration elicited bypicture-sequences. Impairment of referential cohesion was found with Retell-ing, whereas impairment of logical coherence occurred only with the Elicitedcondition in which pictures were present. Thus, the occurrence of deficitsdepended on the condition in which narration was produced and, to someextent, on the particular story used in each condition. The primary implica-tions of this study pertain to the attention given by researchers to the featureof discourse production being studied. Discourse production is not a unitaryphenomenon but is accomplished at multiple levels of processing. Researchthat attempts to provide an explanation for discourse production deficits

NARRATIVE PRODUCTION 207

should consider explicit relationships between possible primary dysfunctionswith RHD and the particular measure of discourse chosen as a dependentvariable.

APPENDIX

Examples of Each Story Produced by a Few Control Subjects: EachExample Is Among the Best According to the Scoring Systems Used in

the Study

Car Repair

A lady is walking along the street nicely dressed with a little dog, and she encounters a carwith a whole bunch of guys standing around watching the repairman repair the car. Next one,the lady is repairing the car. The repairman is holding the dog. The rest of the guys are standingaround laughing except for the old guy who’s not too happy about, with a chauvinist attitudeabout her fixing the car. The last picture, she’s got the car all fixed. She waves goodby to thehappy guy inside the car. The rest of the guys are standing around, say ‘‘I’ll be darned aboutthat.’’

Flower Pot

A man is walking along. He walks under a balcony in a building, and a flower pot fallsoff and hits him on the head. He starts screaming back into the area where the flower potcame from, and his dog is barking. And then he goes into the building and starts knockingon the door where I guess he thinks that the flower pot has come from. The woman comesto the door, gives the dog a bone, and then the dog is very happy, stops barking. The womanis very nice to the dog. The man winds up kissing the woman’s hand and tipping his hat toher and everything seems to be forgiven.

Tiger’s Whisker

A lady needed a tiger’s whisker to make some medicine for her husband who was verysick. She thought and thought as to how she could get a tiger’s whisker. So she knew thattigers liked food and they like soft music. So she went to a cave where she knew there wasa lonely tiger, and she set down a dish of food and played soft music. And the tiger cameout and ate the food and went up to the, up to the lady to thank her for the food. She got thewhiskers and ran down the hill and there was still, the tiger was very lonely.

Epaminondas

Once there was a little boy who lived in a very hot country, and his mother wanted himto take a piece of cake to his grandmother. And she told him to be very careful that, so notto make, make it into crumbs. So he carried it on a leaf under his arm to make sure he carriedit very carefully, and he got to his grandmother’s house and the cake was all crumbs. Andhis grandmother told him he was a very silly boy, he should have carried it on top of his headbecause so it wouldn’t get crumbly. And so she gave him a pad of butter to take back to hisgrandmother, ah mother, and he put it on top of his head so to be very careful of it. And itwas very hot, and the sun shone down on his head. And when he got home, the butter hadall melted. His mother told him he was a very silly boy, and I forgot the end.

208 DAVIS, O’NEIL-PIROZZI, AND COON

REFERENCES

Bloom, R. L. 1994. Hemispheric responsibility and discourse production: Contrasting patientswith unilateral left and right hemisphere damage. In R. L. Bloom, L. K. Obler, S. DeSanti, & J. S. Ehrlich (Eds.), Discourse analysis and applications: Studies in adult clinicalpopulations. Hillsdale, NJ: Erlbaum. Pp. 81–94.

Bloom, R. L., Borod, J. C., Obler, L. K., & Gerstman, L. J. 1992. Impact of emotional contenton discourse production in patients with unilateral brain damage. Brain and Language,42, 153–164.

Bloom, R. L., Borod, J. C., Obler, L. K., & Gerstman, L. J. 1993. Suppression and facilitationof pragmatic performance: Effects of emotional content on discourse following right andleft brain damage. Journal of Speech and Hearing Research, 36, 1227–1235.

Britton, B. K. 1994. Understanding expository text: Building mental structures to induce in-sights. In M. A. Gernsbacher (Ed.), Handbook of psycholinguistics. San Diego, CA: Aca-demic Press. Pp. 641–674.

Brownell, H. H. 1988. The neuropsychology of narrative comprehension. Aphasiology, 2, 247–250.

Bryan, K. L. 1988. Assessment of language disorders after right hemisphere damage. BritishJournal of Disorders of Communication, 23, 111–125.

Cherney, L. R., & Canter, G. J. 1993. Informational content in the discourse of patients withprobable Alzheimer’s disease and patients with right brain damage. In M. L. Lemme(Ed.), Clinical aphasiology. Austin, TX: Pro-Ed. Vol. 21. Pp. 123–134.

Cicone, M., Wapner, W., & Gardner, H. 1980. Sensitivity to emotional expressions and situa-tions in organic patients. Cortex, 16, 145–158.

Cimino, C. R., Verfaellie, M., Bowers, D., & Heilman, K. M. 1991. Autobiographical memory:Influence of right hemisphere damage on emotionality and specificity. Brain and Cogni-tion, 15, 106–118.

Coelho, C. A., Liles, B. Z., & Duffy, R. J. 1994. Cognitive framework: A description ofdiscourse abilities in traumatically brain-injured adults. In R. L. Bloom, L. K. Obler, S.De Santi, & J. S. Ehrlich (Eds.), Discourse analysis and applications: Studies in adultclinical populations. Hillsdale, NJ: Erlbaum. Pp. 95–110.

Davis, G. A., & Turner, C. 1995. Content analysis and cohesion measures of narrative dis-course by nonaphasic clinical groups. Paper presented at the Clinical Aphasiology Con-ference, June, Sunriver, OR.

Dekoskey, S., Heilman, K. M., Bowers, D., & Valenstein, E. 1980. Recognition and discrimi-nation of emotional faces and pictures. Brain and Language, 9, 206–214.

Delis, D. C., Wapner, W., Gardner, H., & Moses, J. A., Jr. 1983. The contribution of the righthemisphere to the organization of paragraphs. Cortex, 19, 43–50.

Fletcher, C. R., & Bloom, C. P. 1988. Causal reasoning in the comprehension of simple narra-tive texts. Journal of Memory and Language, 27, 235–244.

Frederiksen, C. H., & Stemmer, B. 1993. Conceptual processing of discourse by a right hemi-sphere brain-damaged patient. In H. H. Brownell & Y. Joanette (Eds.), Narrative dis-course in neurologically impaired and normal aging adults. San Diego, CA: Singular.Pp. 239–278.

Gardner, H. 1982. Art, mind, and brain: A cognitive approach to creativity. New York: BasicBooks.

Gardner, H., Brownell, H. H., Wapner, W., & Michelow, D. 1983. Missing the point: Therole of the right hemisphere in the processing of complex linguistic materials. In E. Perec-man (Ed.), Cognitive processing in the right hemisphere. New York: Academic Press.Pp. 169–191.

Glosser, G. 1993. Discourse production patterns in neurologically impaired and aged popula-tions. In H. H. Brownell & Y. Joanette (Eds.), Narrative discourse in neurologicallyimpaired and normal aging adults. San Diego, CA: Singular. pp. 191–212.

Glosser, G., Deser, T., & Weisstein, C. C. 1992. Structural organization of discourse produc-

NARRATIVE PRODUCTION 209

tion following right-hemisphere damage (abstract). Journal of Clinical and ExperimentalNeuropsychology, 14, 40.

Gordon, P. C. (1993). Computational and psychological models of discourse. In H. H.Brownell & Y. Joanette (Eds.), Narrative discourse in neurologically impaired and nor-mal aging adults. San Diego, CA: Singular. Pp. 23–46.

Halliday, M. A. K. 1985. An introduction to functional grammar. London: Edward Arnold.Haravon, A., Obler, L. K., & Sarno, M. T. 1994. A method for microanalysis of discourse in

brain-damaged patients. In R. L. Bloom, L. K. Obler, S. De Santi, & J. S. Ehrlich (Eds.),Discourse analysis and applications: Studies in adult clinical populations. Hillsdale, NJ:Lawrence Erlbaum. Pp. 47–80.

Huber, W., & Gleber, J. 1982. Linguistic and nonlinguistic processing of narratives in aphasia.Brain and Language, 16, 1–18.

Joanette, Y., & Goulet, P. 1994. Right hemisphere and verbal communication: Conceptual,methodological, and clinical issues. In M. L. Lemme (Ed.), Clinical aphasiology. Austin,TX: Pro-Ed. Vol 22, pp. 1–24.

Joanette, Y., Goulet, P., Ska, B., & Nespoulous, J-L. 1986. Informative content of narrativediscourse in right-brain-damaged right-handers. Brain and Language, 29, 81–105.

Kennedy, M. R. T., Strand, E. A., Burton, W., & Peterson, C. 1994. Analysis of first-encounterconversations of right-hemisphere-damaged adults. In M. L. Lemme (Ed.), Clinical apha-siology. Austin, TX: Pro-Ed. Vol. 22, Pp. 67–80.

Kintsch, W. 1994. The psychology of discourse processing. In M. A. Gernsbacher (Ed.), Hand-book of psycholinguistics. San Diego, CA: Academic Press. Pp. 721–740.

Levelt, W. J. M. 1989. Speaking: From intention to articulation. Cambridge, MA: MITPress.

Mackisack, E. L., Myers, P. S., & Duffy, J. R. 1987. Verbosity and labeling behavior: Theperformance of right hemisphere and non-brain-damaged adults on an inferential picturedescription task. In R. H. Brookshire (Ed.), Clinical aphasiology. Minneapolis: BRK.Vol. 17, pp. 143–151.

Menn, L., Ramsberger, G., & Helm-Estabrooks, N. (1994). A linguistic communication mea-sure for aphasic narratives. Aphasiology, 8, 343–359.

Mentis, M., & Prutting, C. A. 1987. Cohesion in the discourse of normal and head-injuredadults. Journal of Speech and Hearing Research, 30, 88–98.

Metzler, N. G., & Jelinek, J. E. M. 1977. Writing disturbances in patients with right cerebralhemisphere lesions. In R. H. Brookshire (Ed.), Clinical aphasiology conference proceed-ings. Minneapolis: BRK. Pp. 214–225.

Myers, J. L., Shinjo, M., & Duffy, S. A. 1987. Degree of causal relatedness and memory.Journal of Memory and Language, 26, 453–465.

Myers, P. S. 1979. Profiles of communication deficits in persons with right cerebral hemispheredamage: Implications for diagnosis and treatment. In R. H. Brookshire (Ed.), Clinicalaphasiology conference proceedings. Minneapolis: BRK. Pp. 38–46.

Myers, P. S. 1993. Narrative expressive deficits associated with right hemisphere damage. InH. H. Brownell & Y. Joanette (Eds.), Narrative discourse in neurologically impaired andnormal aging adults. San Diego, CA: Singular. Pp. 279–298.

Myers, P. S., & Brookshire, R. H. 1994. The effects of visual and inferential complexity onthe picture descriptions of non-brain-damaged and right-hemisphere-damaged adults. InM. L. Lemme (Ed.), Clinical aphasiology. Austin, TX: Pro-Ed. Vol. 22, pp. 25–34.

Myers, P. S., Linebaugh, C. W., & Mackisack-Morin, L. 1985. Extracting implicit meaning:Right versus left hemisphere damage. In R. H. Brookshire (Ed.), Clinical aphasiologyMinneapolis: BRK. Vol. 15, pp. 72–82.

Nicholas, L. E., & Brookshire, R. H. 1993. A system for quantifying the informativeness andefficiency of the connected speech of adults with aphasia. Journal of Speech and HearingResearch, 36, 338–350.

Norusis, M. J. 1993. SPSS for windows: Base system user’s guide, release 6.0. Chicago: SPSSInc.

210 DAVIS, O’NEIL-PIROZZI, AND COON

Obler, L. K., Au, R., Kugler, J., Melvold, J., Tocco, M., & Albert, M. L. 1994. Intersubjectvariability in adult normal discourse. In R. L. Bloom, L. K. Obler, S. De Santi, & J. S.Ehrlich (Eds.), Discourse analysis and applications: Studies in adult clinical populations.Hillsdale, NJ: Erlbaum. Pp. 15–27.

Rivers, D. L., & Love, R. J. 1980. Language performance on visual processing tasks in righthemisphere lesion cases. Brain and Language, 10, 348–366.

Roman, M., Brownell, H. H., Potter, Seibold, M. S., & Gardner, H. 1987. Script knowledgein right hemisphere-damaged and in normal elderly adults. Brain and Language, 31, 151–170.

Schneiderman, E. I., Murasugi, K. G., & Saddy, J. D. 1992. Story arrangement ability in rightbrain-damaged patients. Brain and Language, 43, 107–120.

Shadden, B. B., Burnette, R. B., Eikenberry, B. R., & DiBrezzo, R. 1991. All discourse tasksare not created equal. In T. E. Prescott (Ed.), Clinical aphasiology. Austin, TX: Pro-Ed.Vol. 20. Pp. 327–342.

Singer, M., Halldorson, M., Lear, J. C., & Andrusiak, P. 1992. Validation of causal bridginginferences in discourse understanding. Journal of Memory and Language, 31, 507–524.

Stein, N. L., & Glenn, C. G. 1979. An analysis of story comprehension in elementary schoolchildren. In R. O. Freedle (Ed.), New directions in discourse processes. Norwood, NJ:Ablex. Pp. 53–120.

Thorndyke, P. W. 1977. Cognitive structures in comprehension and memory of narrative dis-course. Cognitive Psychology, 9, 77–110.

Tompkins, C. A., Boada, R., McGarry, K., Jones, J., Rahn, A. E., & Ranier, S. 1993. Connectedspeech characteristics of right-hemisphere-damaged adults: A re-examination. In M. L.Lemme (Ed.), Clinical aphasiology. Austin, TX: Pro-Ed. Vol. 21, pp. 113–122.

Trabasso, T., Secco, T., & van den Broek, P. 1984. Causal cohesion and story coherence. InH. Mandl, N. Stein, & T. Trabasso (Eds.), Learning and comprehension of text. Hillsdale,NJ: Lawrence Erlbaum.

Trabasso, T., & Sperry, L. L. 1985. Causal relatedness and importance of story events. Journalof Memory and Language, 24, 595–611.

Trabasso, T., & van den Broek, P. 1985. Causal thinking and the representation of narrativeevents. Journal of Memory and Language, 24, 612–630.

Trupe, E. H., & Hillis, A. 1985. Paucity vs. verbosity. Another analysis of right hemispherecommunication deficits. In R. H. Brookshire (Ed.), Clinical aphasiology. Minneapolis:BRK. Vol. 15, pp. 83–96.

Uryase, D., Duffy, R. J., & Liles, B. Z. 1991. Analysis and description of narrative discoursein right-hemisphere-damaged adults: A comparison to neurologically normal and left-hemisphere-damaged aphasic adults. In T. E. Prescott (Ed.), Clinical aphasiology. Austin,TX: Pro-Ed. Vol. 19, pp. 125–138.

van den Broek, P. 1994. Comprehension and memory of narrative texts: Inferences and coher-ence. In M. A. Gernsbacher (Ed.), Handbook of psycholinguistics. San Diego, CA: Aca-demic Press. Pp. 721–740.

Wapner, W., Hamby, S., & Gardner, H. 1981. The role of the right hemisphere in the apprehen-sion of complex linguistic materials. Brain and Language, 14, 15–33.

Yorkston, K. M., & Beukelman, D. R. 1980. An analysis of connected speech samples ofaphasic and normal speakers. Journal of Speech and Hearing Disorders, 45, 27–36.