RELEVANCE IN THE LANGUAGE PRODUCTION OF APHASIC PATIENTS

(RELEVANCIA EN LA PRODUCCIÓN DE LOS PACIENTES AFÁSICOS)

 

CONFERENCES
TOPIC: NEUROPSYCHOLOGY OF LANGUAGE

---

 
Abstract

Recent studies have shown that even mild aphasic patients of all types produce coherence violations in discourse, which though subtle remain consistent. Wernickeís aphasics in particular have demonstrated abnormal intrusions of irrelevant utterances when producing story narratives.

Using three tests tapping the ability of Brocaís, conduction, and Wernickeís aphasics to maintain coherence in production at the word, sentence, and discourse level, the current study questions whether the intrusion of irrelevant utterances might result from a general processing impairment affecting both local coherence between consecutive sentences and word associations in the lexicon, or whether it is related to only to discourse structure.

Results indicate that the coherence problems of the Wernickeís aphasics are probably limited to discourse, although more severe patients may experience difficulty at every level of language production. In addition, although language output was greatly reduced in Brocaís aphasics, they produced a range of essential and peripheral words or propositions roughly equivalent to the normal control subjects, indicating that they do not usually confine themselves to producing only the most crucial information to a given topic.
 

Resumen

Estudios recientes han mostrado que incluso los pacientes con afasia leve de todos los tipos producen violaciones de la coherencia en el discurso, lo que hace pensar que son restos sutiles consistentes. Los afásicos de Wernicke, en particular, han demostrado intrusiones anormales de producciones irrelevantes cuando producen narraciones de historias.

Usando tres tests que valoran la capacidad de los afásicos de Broca, de conducción y de Wernicke para mantener la coherencia en la producción de una palabra, oración y nivel de discurso, las preguntas de estudio actuales son si la intrusión de las producciones irrelevantes podrían resultar de un deterioro en el procesamiento general que afecta tanto a la coherencia local entre oraciones consecutivas como a las asociaciones de palabras en el léxico, o si está relacionado sólo con la estructura del discurso.

Los resultados indican que los problemas de coherencia de los afásicos de Wernicke se limitan probablemente al discurso, aunque los pacientes más severos pueden experimentar dificultades en cualquier nivel de la producción del lenguaje. Además, aunque el habla era muy reducida en los afásicos de Broca, ellos producían un rango de palabras esenciales y periféricas o proposiciones aproximadamente equivalentes a los sujetos control normales, indicando que ellos normalmente no se confinan a producir sólo la información más crucial de un determinado tema.

Coherence can be defined as the semantic connectedness of a text (Van Dijk, 1977; 1980). For a story narrative to be coherent, it should contain all of the propositions necessary for a listener to construct the correct macrostructure for the story, but should not include irrelevant and tangential propositions which would distract the listener from the main points (see Christiansen, 1995; Kintsch and Van Dijk, 1978; Van Dijk and Kintsch, 1983). Coherence has been thought to be relatively well-preserved in the discourse of aphasic patients (Glosser and Deser, 1990; Huber, 1990; Ulatowska, Allard, and Chapman, 1990; Ulatowska, Freedman-Stern, Doyel, Macaluso-Haynes, and North, 1983; Ulatowska, North, and Macaluso-Haynes, 1981), however, this may not be the case for all types of aphasic patients. Broca's aphasics, while producing few irrelevant propositions, have difficulty producing all of the propositions essential to relay complete stories (Goodglass, Christiansen, and Gallagher, 1994). In a propositional analysis of anomic, conduction, and Wernicke's aphasics, Christiansen (1995) found that some fluent aphasic patients, particularly Wernicke's aphasics, produced numerous irrelevant propositions in their narratives. Christiansen proposed that the Wernicke's aphasics had difficulty de-termining which proposi-tions were crucial to a particu-lar plot-. Because of this uncer-tainty, the Wer-nicke's patients may simply have pro-duced every proposi-tion which came to mind--essen-tial or irrelevant--as an adaptive strategy. This strate-gy ensured that their nar-ratives were complete, at the expense of dis-rupting coher-ence.
It is not clear whether the irrelevant propositions produced by Wernicke's aphasics represent a particular disturbance in discourse coherence or is secondary to an overall semantic impairment affecting language production at the level of words and sentences as well as paragraphs. By using a multiple case study design, the current study seeks to determine the relations between the ability of individual aphasic patients to maintain topic-relevant language output in narrative discourse, consecutive sentences, and word lists. Specifically, the following questions are asked:

1. Can Broca's, conduction, and Wernicke's aphasics produce concise and coherent narratives, i.e., without the intrusion of irrelevant propositions, despite their difficulties in formulating grammatical sentences and accessing appropriate lexical items?

2. Are difficulties in producing coherent discourse associated with particular aphasic syndromes?

3. If an aphasic patient produces an abnormal number of irrelevant propositions in discourse, is this intrusion of irrelevant propositions confined to the production of narrative-length discourse, or might it result from a more general processing impairment also affecting the production of smaller units such as single words or sentences?
 

The Function of Relevance in Discourse

Relevance has been shown to be a key tool in all realms of human communica-tion, including action sequences and verbal com-pre-hension, as well as verbal produc-tion (Sperber and Wilson, 1986). But how does a speaker know what is relevant to the current dis-course, and how can a listener use relevance to interpret the speaker's message? In Sperber and Wilson's (1986) relevance theory, each participant in a conversation begins with his/her cognitive environment, which is the set of all assumptions available to the individual at any particular time. These assumptions include facts about the current physical environment, encyclopedic facts, common experiences with the other participants, and so forth, as well as all inferences made from this knowledge. When an individual encounters new information, that information, along with all of the inferences made from it, cause a change in the individual's cognitive environment, referred to as contextual effects. When a proposition is relevant, as defined by Sperber and Wilson, it produces contextual effects on the listener's cognitive environment with a minimum of processing effort--the greater the effects, the greater the relevance.

To identify the main points, or macrostructures, of a given piece of discourse, the listener must first be able to parse the segment into a series of semantic propositions (Van Dijk and Kintsch, 1983), defined as the minimal semantic units represented therein. As specified in Kintsch (1991), propositions consist of a head and a number of slots for arguments, along with their precise relation to the head. Semantic connections are established between propositions by means such as bridging (Clark, 1975; Clark and Haviland, 1977; Haviland and Clark, 1974), argument overlap (Kintsch and Van Dijk, 1978), and centering (Grosz and Sidner, 1986; Gordon, 1993). Yet inherent throughout each of these means of extracting the main points of discourse is the presumption that the propositions presented are somehow relevant to the main points of the discourse, and that the listener is able to discern the degrees of relevance for each proposition, based on the presentation by the speaker and the current cognitive environment of the listener.

As relevance and macrostructure theories have focused primarily on discourse com-prehension, less is known about the processes involved in discourse production. Production processes cannot simply be the reverse of comprehension processes, as the formation of macrostructures in production is internally driven, while the extraction of macrostructures in comprehension must be deduced from external information in the discourse context. Yet the two processes must be closely relat-ed, as both draw upon the same rules of macro- to microstructure mapping. It seems reasonable to assume that in language production, the speaker begins by generating a global intent, or macroproposition (Levelt, 1989; Van Dijk and Kintsch, 1983). Once the global intent has been formed, the speaker must then specify the information for smaller, more local units (e.g., local utterances and micropropositions). Van Dijk and Kintsch (1983) have suggested three possible opera-tions which the speaker might use: adding de-tails, particu-lariz-ing general propositions, and analyzing (i.e., breaking down) complex actions into simpler compo-nents.
 

Relevance in the Discourse of Aphasic Patients

In past stud-ies, coherence problems in dis-course have usu-ally been as-cribed to stroke patients with right-hemisphere brain damage (RBD), who dem-onstrate strik-ing coherence and pragmatic deficits despite an in-tact ability to encode and de-code language at the surface lev-el (Myers, 1993; Weylman, Brownell, and Gardner, 1988). In contrast to the left-hemi-sphere brain-damaged (LBD) aphasic pa-tients, RBD stroke patients often have difficulty interpreting connotative and metaphorical meanings in words and sen-tences (Brownell, Pot-ter, Michelow, and Gardner, 1984; Brownell, Simpson, Bihrle, Potter, and Gardner, 1990; Van Lancker and Kempler, 1987; Winner and Gardner, 1977), as well as comprehending indi-rect requests (Foldi, 1987; Hirst, LeDoux, and Stein, 1984; Weylman, Brownell, Roman, and Gardner, 1989), and jokes (Bihrle, Brownell, Powelson, and Gardner, 1986; Brownell and Gardner, 1988; Brownell, Michel, Powelson, and Gardner, 1983). They also have difficulty inferencing from connected dis-course (Brownell, Pot-ter, Bihrle, and Gardner, 1983; Molloy, Brownell, and Gardner, 1990). However, the coherence diffi-cul-ties demon-strat-ed by RBD and LBD patients appear to be more divergent in comprehension than in produc-tion. In pro-duction, both LBD aphasic and RBD patients demon-strate reduced informa-tional content in their con-nected discourse (Gleason, Goodglass, Obler, Green, Hyde, and Weintraub, 1980; Joanette, Goulet, Ska, and Nespoulous, 1986; Joanette and Goulet, 1990; Ulatowska, et al., 1981; Ulatowska, et al., 1983). In addition, both Wernicke's aphasics and RBD patients produce numerous extra-neous utterances in their narra-tives (Christiansen, 1995; Myers, 1993). The quality of the extraneous ut-teranc-es, howev-er, seems to differ among the two groups. While RBD pa-tients may re-late a series of details, includ-ing tangential and irrelevant observations, without actually getting to the main point of the sto-ry, Wernicke's apha-sics can relate the essential elements of the story even though the story is infiltrated with numerous irrelevant de-tails (Christiansen, 1993; 1995). However, some aphasic patients have even been found to confabulate responses (Sandson, Albert, and Alexander, 1986).

In analyzing the LBD aphasic patients' abili-ty to incorpo-rate Sperber and Wilson's princi-ple of relevance into their narra-tives, it is necessary to consider both the lack of essential proposi-tions and the intrusion of extraneous prop-ositions. Most aphasic patients demonstrate some difficulty ei-ther in provid-ing all of the essential propo-sitions of a story or in carefully moni-toring the rele-vance of addi-tional proposi-tions to the story being told, with the possible excep-tion of very mild patients and some conduc-tion aphasics (Christiansen, 1994; 1995).

Story Content and Information Gaps. When com-pared to age-matched normal control subjects, many types of aphasic patients fail to produce the same range of propo-si-tions regarded by the normal controls as es-sential. In narra-tives elic-ited with cartoon pictures, both Broca's and Wernicke's apha-sics consistent-ly produced few-er main points than nonaphasic con-trols and tended to per-severate on the most sa-lient points in the story (Gleason, et al., 1980). Even patients who included most of the essential propositions of the story often showed a marked reduction in nonessential elaborative propositions as compared to the normal subjects (Ulatowska, et al., 1983). Earlier studies of coherence in aphasia tended neither to in-vestigate indi-vidual behaviors of the patient population nor to analyze pa-tient perfor-mance in rela-tion to narrow subtypes of aphasia. In a propo-sitional analysis of Broca's, amne-sic, and Wer-nicke's apha-sics, Huber (1990) found that the apha-sics produced a similar range of essential or optional propo-si-tions. Howev-er, from the individual data he presents, one can see that the Broca's and anomic aphasics in his study --also tended to omit essential story propositions. Ad-ditional studies of proposition-al content in narrative produc-tion have also revealed informa-tion gaps in the narra-tives of both Broca's aphasics (Goodglass, et al., 1994) and anomic apha-sics (Chris-tiansen, 1995).

Irrelevant Prop-ositions. Though they rarely omit essential information in narrative production, Wernicke's aphasics, as opposed to other types of apha-sics, seem to have the most diffi-culty maintain-ing relevance in their stor-ies (Christiansen, 1995). The anomic, conduc-tion, and Wer-nicke's aphasics in Christiansen's study all pro-duced distinc-tively different discourse pat-terns. Anomic aphasics tended to give very brief narratives, often omitting essen-tial proposi-tions and refraining from unnecessary propositional elaborations. Conduc-tion aphasics did not omit essential propo-sitions in their narratives, but reiterated propo-si-tions with abnormal fre-quency. Christiansen attributed the narrative pro-duction patterns of the anomic and conduction aphasics to the strate-gies they used to compen-sate for their lexical and syn-tactic deficits. The Wer-nicke's apha-sics, on the other hand, produced all of the essen-tial propo-sitions required for their narra-tives, but in addition, produced signifi-cantly more ir-rele-vant propo-sitions than either of the other aphasic groups or the normal controls. Christian-sen concluded that the produc-tion of all possible narrative propo-sitions, relevant and irrelevant, by the Wernicke's apha-sics may also be a strategy, but one that is used to compensate for a loss of macro-struc-tural knowledge. Because the Wernicke's could no longer judge the relevance of each of their propositions to the main points of the story, they produced everything which came to mind. Such a production strategy would provide com-plete-ness at the expense of co-herence.

In light of the fact that some aphasic patients exhibit difficulty in producing only those propositions that are most relevant to a particular story, the current study seeks to investigate whether those individual patients who include irrelevant propositions in their narratives also produce irrelevant responses when asked to produce only sentences or individual words in a particular context. Such behavior would indicate that the intrusion of irrelevant propositions into the patient's discourse might be connected to a difficulty with the conceptual relevance of words and propositions to a particular topic. On the other hand, it is possible, that the intrusion of irrelevant propositions in narrative is confined to discourse, in which case the inability to maintain coherence may result from an overload in discourse processing when the subject is required to plan lengthy connected speech.
 

METHODS

This current study presents three subtests from Christiansen's (1994) Relevance/ Coherence Battery (RCB). The RCB was originally designed to ana-lyze the aphasics' abili-ty to utilize commonly known scripts and frames to pro-cess language, with a primary emphasis on rel-evance and contextualization. The original battery consisted of nine subtests which focused on the subjects' ability to maintain relevance at three levels: 1) accessing situational frames to com-prehend and pro-duce single words, 2) ac-cessing scripts and frames both to produce and judge the rele-vance of sequen-tial utterances or events, and 3) interpreting and producing coherent narra-tive discourse. At each level, the subtests generated verbal production, ver-bal comprehen-sion, and non-verbal conceptu-al data. As this study fo-cuses solely on relevance in verbal produc-tion and the issue of processing load, only the three production subtests from the RCB are used--narrative production, sen-tence pair completion, and word list gener-ation--as de-scribed below.
 

Subjects

The subjects used in this study consisted of 13 mild to moderate aphasic patients (3 fe-male, 10 male), as described in appendix A, and exhib-ited various types of aphasia (4 Broca's, 4 conduc-tion, and 5 Wer-nicke's). Each sub-ject was evaluated at the time of testing with the oral production and auditory comprehension subtests of the Boston Diag-nos-tic Aphasia Exam (BDAE) (Good-glass and Kaplan, 1983a), the short form of the Token Test (DeRenzi and Faglioni, 1978), and the Boston Naming Test (BNT) (Goodglass and Kaplan, 1983b). All of the sub-jects had rela-tively good au-ditory compre-hension (aver-aging at least in the 70th per-centile on the BDAE auditory compre-hension sub-tests). All subjects had experi-enced a single, left hemi-sphere infarction or ruptured aneurysm, and none of the subjects had any other history of head injury, neuro-log-ical disease, or alcoholism.

It is important to note here that the classification of the aphasic patients was based on the original diagnosis made between one and six months post onset. All of the patients in this study have been followed clinically for years since their stroke. The Wernicke's patients presented here all presented early on with classic neologistic jargon, paragrammatic output, and poor auditory comprehension on clinical testing. They have recovered to the point where their auditory comprehension scores are now on a par with the conduction aphasics, and their language output is still empty and paragrammatic, though neologisms were rare. Unfortunately, it is logistically impossible to analyze the coherence of more severe Wernicke's aphasics with neologistic output.

In addition to the aphasic pa-tients, the study included 20 normal con-trol subjects (11 females, 9 males) with no history of neu-rolog-ical dis-ease, head inju-ry, or alcohol-ism. The control group was roughly matched with the aphasic patients for age and edu-cation. Among the aphasic patients, the mean age was 66.31 years (range 50-77 years) and the mean level of educa-tion was 11.92 years (range 8-18 years); among the normal control subjects, the mean age of was 67.75 years (range 61-81 years) and the average level of education was 12.95 years (range 12-18 years). All subjects, normal and aphasic, were native speakers of American English.
 

EXPERIMENT ONE: NARRATIVE PRODUCTION

The purpose of the narrative production subtest of the RCB was to gain insight into the ability of aphasic pa-tients to produce coherent stories, i.e. those which include all of the propositions necessary to make the story complete but avoid irrelevant embellishments. The intrusion of irrelevant prop-ositions in nar-rative produc-tion may indicate an in-accurate or underspe-cified macrostruc-ture in some of the aphasic sub-jects. The narrative production subtest in the current study is a partial replication of Christiansen (1995) with the additional inclusion of Brocaís aphasics, a more detailed analysis of the range of relevant propositions, and a more carefully controlled rating system.
 

Stimuli and Data Analyses

The stimuli used for the narrative pro-duction subtest consisted of four cartoon stories of five frames each (one practice cartoon and three test car-toons). While the cartoons each contained a clearly depicted story, they also contained many additional back-ground elements which were ap-propriate to the situa-tional frame but irrel-e-vant with re-gard to the sto-ry content. The purpose of these background items was to test the sub-ject's abili-ty to distin-guish which items in the cartoons were relevant to the story and which items were not. The cartoon frames were pre-arranged in the correct order and presented in a vertical ar-ray. Each sub-ject was asked to look at all of the pictures in the given cartoon, then tell the story that was depict-ed. All narra-tive samples were tape-re-cord-ed and tran-scribed verbatim for later analy-sis.

To determine the relevance of the propositions produced by both the aphasic and normal groups, each narrative was submitted to the following three stages of analysis. First, the raw transcriptions were edited by eliminating self-corrections and abandoned utterances. Phonemic and morphological errors were ignored for the rest of the propositional analysis. Second, the edited texts were parsed into a series of semantic propositions based on a version of Kintsch's (1974) "text base" format, which was adapted to analyzing aphasic speech (see Christiansen, 1995). In general, the semantic propositions consisted of: 1) verbs and their arguments; 2) predicates and their argu-ments; and 3) adjunct adverbials of time, manner, and place. Only the story-based propositions (i.e., those proposi-tions describing or elaborating on the cartoon pic-tures) were used for the rele-vance scoring. Person-al com-ments made to the examin-er about the story or task as well as repetitions and digressions were not scored. Third, all of the story-related propositions were ranked for degree of rele-vance to the main ideas of the stories by two indepen-dent rat-ers, as de-scribed below.

Relevance rank-ings for the story-related propositions were based on a 7-point Likert scale. For each of the three test cartoon stories, a com-posite text base was compiled from all of the propositions given by both the normal and the aphasic pa-tients, so that the raters were blind as to whether the propositions were given by an aphasic subject or a normal subject. The propositions on the master list were grouped according to picture frame and primary top-ic. The two independent rat-ers then gave each propo-si-tion an overall rank-ing as to its rele-vance, which was based on how closely the proposition was connected to the main charac-ters and plot of the story. A proposition ranked "7" was essential to the story; a proposition ranked "4" was peripheral; and a proposition ranked "1" was either completely irrele-vant or illogical.
After the raters had ranked all of the propositions, a master scoresheet was created for each cartoon story in which the rankings from the two raters were averaged for each proposition. The propositions in the text bases for each individual subject was then scored according to the master scoresheet. The scores for the propositions in each individualís story were averaged; the aver-age ranks for each of the three test narratives were then summed and divided by 21 (the total possible if all proposi-tions were ranked "7") to obtain a rel-evance score based on 100.
 

Results and Discussion

In general, the propositions produced by both the normal control group and the aphasic patients were ranked relatively high in relevance, however, some significant differences do emerge. The normal subjects produced an average of 14.1 propositions per narrative and averaged 89.7 (s.d.=4.28) in their relevance scores. Responses ranking low in relevance (1-4) averaged 9.6% of the propositions produced, ranging from 0-21% for each subject. Most of the aphasic subjects could produce as many propositions per story as the normal subjects, averaging 11.96. Only the Broca's aphasics demonstrated a reduced output, each averaging less than 6.0 propositions per story. But despite producing a similar number of propositions, the aphasics' stories were generally ranked lower in relevance than the normal control subjects (p=.02; two-tailed T-test). As a group, the relevance scores of the aphasic patients averaged 83.2, with 20.2% of their propositions ranked between 1 and 4.

Results of a one-way ANOVA reveals a main effect for aphasic type (F(2,10) = 22.03, p<.001), in which the Wernickeís aphasics scored significantly lower in relevance than either the Brocaís aphasics (p<.001, Tukey test) or the conduction aphasics (p<.01, Tukey test). When the individual performance of the aphasic patients is analyzed, as in Appendix B, it becomes obvious that the results of the group analysis for the production of relevant propositions in narrative holds for each individual patient. Each one of the Wer-nicke's aphasics scored at least three standard deviations below normal, but all of the Broca's and conduction apha-sics scored within normal limits, with the exception of one Broca's patient (RO), with a score of +2.41, indicating that he could only produce the most essential propositions for the story and lacked more peripheral elabora-tions.
 

EXPERIMENT TWO: SENTENCE PAIR COMPLETION

As a follow-up to the narrative production test, the purpose in sentence pair completion was to determine if the aphasic patients could produce a relevant response when limited to just a single sentence. If the Wernicke's aphasics, who produced numerous irrelevant propositions in narrative production, also exhibited reduced relevance in their single sentence responses, this would indicate that the Wernicke's aphasics have a particular difficulty with judging even the local coherence between two sentences and that their difficulty in maintaining relevance may not be limited to connected discourse.
 

Stimuli and Data Analyses

The stimuli for this subtest consisted of 20 sentences and one practice sentence, each describing a real life prob-lem. The exam-iner presented the sentence to the subject orally and asked the subject to respond by pro-duc-ing a logical follow-up sen-tence, as illus-trated below.
Exam-iner: James failed the math test, so ...
Subject: ... he studied harder for the next one.
All responses were tape-re-corded and tran-scribed for lat-er analysis. The tran-scribed responses were edited by remov-ing false starts and phonologi-cal errors, as well as correcting grammatical er-rors. Grammati-cal errors were corrected so that the raters would be blind as to whether the responses were made by an aphasic or nor-mal subject.

The edited re-sponses were then combined into a separate master list for each of the 20 stimulus sentences to be judged for their degree of relevance to the given situation. As in experiment one, the responses on each master list were put in random order, mixing both aphasic and normal respons-es, and each re-sponse was ranked by two inde-pendent rat-ers using a 7-point Likert scale. A ranking of "7" indicated that the response was fully logical and required little process-ing effort on the part of the rater (e.g., James failed the math test, so he studied harder next time). A ranking of "4" could be viewed as logical, but either forced the rater to increase his/her processing effort to bridge the stimulus with the response by creating some possible intermediary scenario (e.g., to bridge ÑJames failed the math test, so he missed the party Saturday night," the rater might imagine that James was punished for his failure), or failed to pro-vide adequate new information and simply re-phrased informa-tion already implicit in the stimulus item (e.g., James failed the math test, so he got zero). A rank-ing of "1" indi-cated that the rater could not make logical sense of the response in the context of the stimulus sen-tence (e.g., James failed the math test, so I'll be darned, he found it).

To attain a relevance score for the sentence pair completion subtest, each subject's indi-vidual response was ranked ac-cording to the master scoresheet. If a subject pro-duced more than one response for a particular stimulus item, each response was scored sepa-rately and aver-aged with the other responses for the stimulus item, so that each of the 20 items was given an average score between 1 and 7. The relevance scores for all of the items were then converted to a score based on 100, as in experiment 1.
 

Results and Discussion

Performance of the normal control subjects on this task was almost equal to that on the narrative production task. Their average rele-vance score was 89.7 (s.d.=3.23). The number of low-relevance re-sponses for the normal group aver-aged 5.25%, ranging from 0-20% for each subject. Among the apha-sic patients, several had diffi-culty producing relevant follow-up responses in the sen-tence pair com-pletion. Though the performance of the aphasic patients var-ied greatly as compared to the normal subjects, the aphasic group as a whole scored signifi-can-tly lower in rele-vance (p<.005, two-tailed T-test), averag-ing only 78.7. While some patients produced no low-relevance responses, up to 60% of other patients' responses were ranked low in relevance.

When analyzing the performance of individual subjects, as seen in Appendix B, it first appears that the production of relevant verbal sequences is related to aphasic syndrome. However, results of a one-way ANOVA reveal no significant differences among the aphasic groups (F(2,10) = 2.86, p=.10). Judging from performance on the narrative production task, one would suspect the Wernicke's aphasics to have difficulty in sentence pair completion; in fact, the opposite appeared to be true. Only one Wer-nicke's patient (RI) had difficul-ty producing relevant sequences, scoring more than 7 s.d.'s below normal; the others were all well within normal limits. On the other hand, all of the Bro-ca's and all but one of the conduction aphasics (PA) scored more than 2.5 standard deviations below normal.

Closer inspection, however, reveals that the apparent relation between relevance scores on sentence pair completion and aphasic syndromes may be misleading. Performance by the aphasic patients on the sentence pair comple-tion task correspond-ed exactly with the patients' BDAE severity ratings. All of those patients who were given a severity rating of 3.0 or below scored at least two standard deviations below the normal control subjects in producing relevant sentence pairs; all patients with a severity rating of 3.5 or above scored within normal limits. The Wernicke's aphasics, in general, had higher severity ratings than the other two groups, probably due to the selection criteria. Only mild Wernicke's aphasics had recovered enough in auditory compre-hension to score above the 70th percentile on the BDAE. Therefore, the subjects in the Wernicke's group performed well on sentence pair comple-tion, with the exception of R.I. who had a severity rating of only 2.0. The Broca's and conduction aphasics, on the other hand, had severity ratings between 1.5 and 3.0, except for PA who had a 3.5. Future research would probably reveal that milder Broca's and conduction aphasics could also produce relevant responses on sentence pair completion, and that more severe Wernicke's would produce more irrelevant responses.
 

EXPERIMENT THREE: WORD LIST GENERATION

The issue of processing load in maintaining topic relevance was most critically addressed in the word list generation subtest, as the subject only needed to produce single words without the added processing required to produce entire propositions and syntactic frames. In order to perform this task successfully, the subject had to evoke a mental image of a common setting or situation given by the examiner and verbally list items, events, or qualities usually associated with such a setting. The use of everyday settings such as schools or gas stations to elicit word lists, rather than lexical superordinates such as fruit or animals, was considered more closely related to narrative production, the only difference being that the subjects needed to access related lexical items arising from simple referring propositions rather than formulating complex propositions resulting in full sentences.
 

Stimuli and Data Analyses

In this subtest, the examiner presented the subject orally with a word or phrase describing a common script or frame (e.g., wedding, hospital, or gas sta-tion). The subject was then asked to say as many words as s/he could think of which would be associated with that particular setting. The responses could be nouns, verbs, adjectives or adverbs; in some cases subjects gave circumlocutory phrases (e.g., machine for changing tires for the gas station setting) and these responses were treated in the same way as single-word responses. Eight different topics were given in total. In order not to fatigue the subjects, only two topics were given in succession and interspersed among other comprehension or nonverbal tasks from the RCB. The subject was allowed ninety seconds to respond with as many words as possible, and all responses were tape-recorded for later transcription and analysis.

As in the first two experiments, to analyze the relevance of each subject's responses, a master list for each situation was constructed which combined all of the re-sponses given by both normal and aphasic sub-jects. Phone-mic errors were ig-nored and cir-cumlocutions (e.g., the thing you sit on and people push you for "wheel-chair") were given credit as the word the subject was try-ing to describe. All of the respons-es for each topic were simply listed alphabetically so that the raters were blind as to whether a par-ticular response came from a nor-mal or aphasic subject. As in experiments 1 and 2, each response on the master list was then ranked, by two independent rat-ers, on a 7-point Likert scale ac-cording to its centrality to the given setting. A "7" ranking meant that the item was a cru-cial component of the setting (e.g., doctor and nurse for "hospi-tal"). A "4" ranking in-di-cat-ed that the word was common-ly found within the setting, but unnec-essary or peripheral (e.g., chair, cafeteria, and gift shop, for "hospital"). A "1" ranking indicat-ed an item which was completely irrelevant or illogical within the given setting (e.g., tires for "hospital"). The rankings of the two raters were then averaged to create a master scoresheet. Individual scores were then calculated from the master list and converted to 100-point scale.
 

Results and Discussion

On word list gen-eration, the aphasic patients did not differ from the normal control subjects on rele-vance ratings (p=.56, two-tailed T-test), but did differ in the total number of words each group could generate. Normal per-formance on the word list genera-tion subtest, as with the other production tasks, did not show ceiling effects. The normal control sub-jects produced an aver-age of 22.14 words (range 10.75-30.625) for each of the eight topics given, and their rele-vance scores averaged 87.6 (s.d.=2.80). The overall percentage of words produced which ranked low in rele-vance (1-4) was 7.5%, ranging from 3%-13%. The aphasic patients produced significantly fewer responses than the normals (p<.001, two-tailed T-test), averaging only 11.94 words per topic, but were comparable to the normal controls in relevance. Their overall relevance scores averaged 86.8, and though they produced a slightly higher proportion of low-relevance words, averaging 12.9% (range 4%-26%), this difference was not significant.

As can be seen in Appendix B, the relevance scores of the aphasic patients on the word list gener-a-tion task were generally within normal limits, and individual difficulties were not related to any particular type of aphasia as indicated by the results of a one-way ANOVA (F(2,10) = 0.28, p=.76). One Broca's (MA), one conduction (CA), and one Wernicke's aphasic (RI) had z-scores less than -2.00, indicating that they produced words more peripheral to a given topic. In contrast, one Broca's aphasic (MP) had a z-score of +2.32, indicating that she produced only the very central core words for a given topic. Deviant performance on this task does not appear to be related to overall severity of aphasia, as seen in experiment 2, nor does it relate directly to naming difficulties as seen on the Boston Naming Test.

At first glance, the relevance scores of the aphasic patients on the word list generation task seem to indicate that their ability to construct situa-tional frames is intact, as few subjects scored outside of the normal range. But although they produced words within a normal range of relevance, the aphasic patients produced only about half as many words per topic as the normal control subjects. This was a consistent pattern even among the very fluent aphasic patients. The combina-tion of normal relevance with reduced lexical production would indicate that the aphasic subjects in general had not lost the more distally related concepts within a particular frame. Had these more peripheral concepts been lost, the aphasic patients would have produced fewer responses but would have demonstrated an increase in relevance ratings, as they would have only produced the most central features of the situational frame. Most of the aphasic patients in this study showed generally impoverished access to the necessary lexical items, regardless of their centrality to the situational frame.
 

DISCUSSION

It should be stressed that the original purpose of this study was exploratory in nature, and the specific questions posed were: 1) whether aphasic patients could produce concise and relevant narratives, 2) whether difficulties in producing relevant discourse were associated with particular aphasic syndromes, and 3) whether the intrusion of irrelevant propositions in discourse might result from a general processing impairment also affecting the production of sentences and single words. To produce a coherent narrative from the cartoon picture stimuli, the subject first had to determine which parts of the cartoon pictures were crucial to the action series depicted and which parts were irrelevant embellishments of the setting. The subject would then base his/her narrative on the crucial actions and consequences in the cartoon. Difficulty in maintaining coherence in narrative production appears to be explicitly connected with Wernickeís aphasia, as every one of the Wernicke's aphasics produced an abnormal number of low-relevance proposi-tions on the narrative production task, while none of the other aphasic patients scored outside of the normal range. These findings confirm Chris-tiansen's (1995) previous finding, indicating that the strategy of determining the relevance of story elements may not be available to the Wernicke's aphasics. The results of the narrative production task further revealed that the Broca's aphasics performed as well as the conduc-tion aphasics in terms of maintaining relevance, even though they produced fewer propositions.
 

Discourse and the Issue of Processing Load

If processing load has a direct effect on formulating relevant discourse at the micropropositional level, we would expect to find aphasic patients with difficulties either on all three tasks, on the sentence and discourse tasks but not on word list generation, or only on the narrative production task. The difficulty maintaining coherence in narrative discourse as exhibited by the Wernickeís aphasics appears to be limited to discourse, however, the ability to maintain coherence may interact with the severity of the Wernickeís aphasia. RI, the most severe Wernickeís aphasic in the study, was the only subject who produced an abnormal number of irrelevant responses on all three production tasks. It is possible that RI has lost part of his ability to monitor the relevance of his own utterances to the discourse context. Further research would have to determine whether RI's difficulty with producing relevant responses can be generally found in patients with a moderately severe Wernicke's aphasia or is an idiosyncratic disturbance.

An interesting and unexpected finding was that severity of aphasia affects the production of local coherence, as seen on the sentence pair completion task, but that this difficulty at the sentence level does not carry over to narrative production. The apparent relation between performance by the aphasic patients on the BDAE and their performance on the sentence pair completion test may seem logical at first, but one must stop and consider why a patient's ability to formulate language grammatically should affect his/her ability to produce a RELEVANT response. It is likely that patients with more severe aphasia produced irrelevant responses not because of their overt language errors, but because of the alternative strate-gies they used to communi-cate. It is possible that aphasic patients were able to generate a relevant response at the prelinguistic level, but could not formu-late it linguistically, so they attempted to generate a related response for which they could formulate a surface struc-ture. However, the alternative responses were often less precise and circumlocutory, causing them to be rated lower in relevance. Less relevant responses often tended to be either very general, either approach-ing automatic speech or lacking in new information.

Most of the aphasics' responses demonstrated some logical relation to the given sentence. Few patients gave responses which did not seem to fit the appropriate script being referred to by the examiner. Only RI (Wernicke's) and GE (conduction) produced illogical responses such as The hell with the dog! Who cares about a dog? in response to the stimulus sentence ÑMy friend got her car stolen in Boston yesterday", or Take the elevator in response to ÑParking in the city is simply impossible".

The obvious question that arises from the performance of the aphasic patients on sentence pair completion is: why should the severity of aphasia affect the production of local coherence between single sentences but not global coherence in narrative discourse? One possibility is that for communication purposes, more severe aphasic patients benefit from the freedom of a discourse context, since they are then allowed to produce as many utterances as they wish to get their point across. When restricted to a single response utterance, more severe aphasic patients do not seem to be able to embed enough information into one utterance, and therefore their responses were judged as less relevant than the milder aphasic patients who had a greater capacity to embed information. Such a notion would agree with Goodglass et al.ís (1994) previous finding that while normal age-matched subjects produced an average ratio of approximately two propositions per utterance in free speech, both Brocaís and conduction aphasics could only produce roughly a one-to-one ratio. While the Brocaís patients in Goodglass et al.ís study produced only about the same number of utterances as the normal subjects, thereby producing only half as many propositions, the conduction aphasics made up for their lack of embedding by producing twice as many utterances as the normal group, therefore producing approximately the same number of propositions. This same production behavior was found in the narrative production task of the current study. The Brocaís patients may also have difficulty producing coherent narratives, however, their lack of coherence results from a lack of essential information in their narratives, and not from a lack of relevance relation between their story propositions and the main points of the intended story, as discussed below.
 

Discourse Production in Brocaís Aphasia

From the beginning of the 20th century, aphasiologists supporting the economy of effort hypothesis espoused that agrammatic Brocaís aphasics restricted their language output to only the most essential elements of their intended message. Pick (1913) expressed the notion of ÑNotsprache" (emergency speech) as a strategy in the discourse of agrammatic Brocaís patients in which the patients try to convey the most core information in the shortest possible time. Although the idea of Notsprache was used to explain the telegraphic nature of agrammatic speech, it also related to the overall sparse language production of many Brocaís aphasics. In the current study, the use of Notsprache should be evident in the relevance scores of the Brocaís aphasics on narrative production and word list generation, where production was not limited to a single response. We would expect to find unusually high relevance ratings, more than two standard deviations above normal. Such a finding would indeed indicate that the Brocaís aphasics were attempting to convey only the words or propositions most central to a given topic.

Evidence for the economy of effort strategy by the Brocaís aphasics in the current study was far from overwhelming. Although two patients performed above the normal range on one task each, RO scoring +2.41 on narrative production and MP scoring +2.32 on word list generation, all other relevance scores by the Brocaís aphasics fell within the normal range, including both core words or propositions as well as concepts more peripheral to the topic. In narrative production, all of the conduction aphasics produced more propositions per story than any of the Brocaís aphasics, yet they were not less relevant. The Brocaís aphasics, however, tended to omit key propositions, but included peripheral propositions. In word list generation, even though the Brocaís patients produced fewer words per topic than any of the fluent aphasics, except for GO (conduction), they included such peripheral words as turnips (BL) or cold (MP) under the Restaurant topic, stones (BL) or parking the car (RO) under the City Park topic, and produced donkey but not cow (MA) under the Farm topic. The reduced output of the Brocaís aphasics in conjunction with their maintenance of a normal range of essential and optional information in their output would seem to rule out a production strategy based on economy of effort, or Notsprache. It remains an open question as to why the Brocaís aphasics leave out essential information in their discourse even at the propositional level, and this global reduction in output should be considered in future theories of agrammatic production.

Conclusion

To summarize the overall findings of the current study, Wernickeís aphasics, and only Wernickeís aphasics, consistently have difficulty producing coherent story narratives without the intrusion of irrelevant propositions. This difficulty appears to only affect production at the discourse level, however, there may be an interaction with the severity of the Wernickeís aphasia. Unfortunately, the interaction between Wernickeís aphasia and the severity of the aphasia will be extremely difficult to investigate, given the frequent neologistic output and attention limitations of more severe Wernickeís patients. For other patients, severity of aphasia only affects the relevance of their responses when their output is limited by task demands, such as restricting them to only one sentence. An additional finding is that even though the language output of the Brocaís aphasics is severely reduced as compared to normal control subjects, they nonetheless produce a similar proportion of essential and peripheral concepts when given a specific topic. It must be stressed that the current study was purely exploratory in nature and that the findings should form the basis for future research questions rather than providing definitive answers about aphasic patientsí ability to produce coherent language output.
 

Acknowledgements

This research was part of an unpublished doctoral dissertation by the author at Boston University. My special thanks to Sui Choi for drawing the cartoon stimuli, and to Lorrie Verplaetse, Pam Wendler-Shaw, Todd Brown, Susan Ward, Qian Hu, and Sulochana Naidoo for scoring the data. I would also like to thank Harold Goodglass and Herman Kolk for their comments and support.
 
 

BIHRLE, A.M., BROWNELL, H.H., POWELSON, J.A., and GARDNER, H. (1986) Comprehension of humorous and nonhumorous materials by left and right brain-damaged patients. Brain and Cognition, 5, 399-411.

BROWNELL, H.H. and GARDNER, H. (1988) Neuropsychological insights into humour. In J. Durant and J. Miller (Eds.) Laughing matters: A serious look at humour (John Wiley & Sons, Inc., New York), pp. 17-34.

BROWNELL, H.H., MICHEL, D., POWELSON, J., and GARDNER, H. (1983) Surprise but no coherence: Sensitivity to verbal humor in right-hemisphere patients. Brain and Language, 18, 20-27.

BROWNELL, H.H., POTTER, H.H., BIHRLE, A.M., and GARDNER, H. (1986) Inference deficits in right brain-damaged patients. Brain and Language, 27, 310-321.

BROWNELL, H.H., POTTER, H.H., MICHELOW, D., and GARDNER, H. (1984) Sensitivity to lexical  denotation and connotation in brain-damaged subjects: A double dissociation? Brain and Language, 22, 253-265.

BROWNELL, H.H., SIMPSON, T.L., BIHRLE, A.M., POTTER, H.H., and GARDNER, H. (1990) Appreciation of metaphoric alternative word meanings by left and right brain-damaged patients. Neuropsychologia, 28(4), 375-383.

CHRISTIANSEN, J.A. (1993) Coherence disturbance in Wernicke's aphasia. Paper presented at the Annual Meeting of the Linguistic Society of America, Los Angeles, California.

CHRISTIANSEN, J.A. (1994) Coherence disturbances in aphasia: The assessment of relevance. Unpublished doctoral dissertation, Boston University, Boston, Massachusetts.

CHRISTIANSEN, J.A. (1995) Coherence violations and propositional usage in the narratives of fluent aphasics. Brain and Language, 51, 291-317.

CLARK, H.H. (1975) Bridging. Reprinted in P.N. Johnson-Laird and P.C. Wason (Eds. Thinking: Readings in cognitive science (Cambridge University Press, Cambridge, England), 1977, pp. 411-420.

CLARK, H.H. and HAVILAND, S.E. (1977) Comprehension and the given-new contract. In R.O. Freedle (Ed.) Discourse production and comprehension (Ablex Publishing Corp., Norwood, New Jersey), pp. 1-40.

DERENZI, E. and FAGLIONI, P. (1978) Normative data and scoring power of a shortened form of the Token Test. Cortex, 15, 41-49.

FOLDI, N.S. (1987) Appreciation of pragmatic interpretations of indirect commands: Comparison of right and left hemi-sphere brain-damaged patients. Brain and Language, 31, 88-108.

GLEASON, J.B., GOODGLASS, H., OBLER, L., GREEN, E., HYDE, M.R., and WEINTRAUB, S. (1980) Narrative strategies of aphasic and normal-speaking subjects. Journal of Speech and Hearing Research, 23, 370-382.

GLOSSER, G. and DESER, T. (1990) Patterns of discourse production among neurological patients with fluent language disorders. Brain and Language, 40, 67-88.

GOODGLASS, H., CHRISTIANSEN, J.A., and GALLAGHER, R. (1994) Syntactic constructions used by agrammatic speakers: Comparison with conduction aphasics and normals. Neuropsychology, 8(4), 598-613.

GOODGLASS, H. and KAPLAN, E. (1983). Boston Diagnostic Aphasia Examination (Lea & Febiger, Hillsdale, New Jersey).

GOODGLASS, H. and KAPLAN, E. (1983b) Boston Naming Test (Lea & Febiger, Hillsdale, New Jersey).

GORDON, P.C. (1993) Computational and psychological models of discourse. In H. Brownell and Y. Joanette (Eds.) Narrative discourse in neurologically impaired and normal aging adults (Singular Publishing Group, Inc., San Diego, CA), pp. 23-46.

GROSZ, B.J. and SIDNER, C.L. (1986) Attention, intentions, and the structure of discourse. Computational Linguistics, 12(3), 175-204.

HAVILAND, S.E. and CLARK, H.H. (1974) What's new? Acquiring new information as a process in comprehension. Journal of Verbal Learning and Verbal Behavior, 13, 512-521.

HIRST, W., LEDOUX, J., and STEIN, S. (1984) Constraints on the processing of indirect speech  acts: Evidence from aphasiology. Brain and Language, 23, 26-33.

HUBER, W. (1990) Text comprehension and production in aphasia: Analysis in terms of micro-  and macroprocessing. In Y. Joanette and H. H. Brownell (Eds.) Discourse ability and brain damage (Springer-Verlag, New York), pp. 154-179.

JOANETTE, Y. and GOULET, P. (1990) Narrative discourse in right-brain-damaged right-handers. In Y. Joanette and H. H. Brownell (Eds.) Discourse ability and brain damage (Springer-Verlag, New York), pp. 131-153.

JOANETTE, Y., GOULET, P., SKA, B., and NESPOULOUS, J.-L. (1986) Informative content of narrative discourse in right-brain-damaged right-handers. Brain and Language, 29, 81-105.

KINTSCH, W. (1974) The representation of meaning in memory (Lawrence Erlbaum Associates, Hillsdale, New Jersey).

KINTSCH, W. (1991) The role of knowledge in discourse comprehension: A construction-integration model. In G. Denhiere and J.-P. Rossi (Eds.) Text and text processing (Elsevier Science Publishers B.V., Amsterdam).

KINTSCH, W., and VAN DIJK, T.A. (1978) Toward a model of text comprehension and pro-  duction. Psychological Review, 85(5), 363-394.

LEVELT, W. J. M. (1989) Speaking: From intention to articulation (MIT Press, Cambridge, Massachusetts).

MOLLOY, R., BROWNELL, H., and GARDNER, H. (1990) Discourse comprehension by right-hemisphere stroke patients: Deficits of prediction and revision. In Y. Joanette and H. H. Brownell (Eds.) Discourse ability and brain damage (Springer-Verlag, New York), pp. 113-130.

MYERS, P.S. (1993) Narrative expressive deficits associated with right-hemisphere damage. In H. Brownell and Y. Joanette (Eds.) Narrative discourse in neurologically impaired and normal aging adults (Singular Publishing Group, Inc., San Diego, CA), pp. 279-296.

PICK, A. (1913) Die agrammatischen Sprachstörungen (Karger, Berlin).

SANDSON, J., ALBERT, M., and ALEXANDER, M.P. (1986) Confabulation in aphasia. Cortex, 22, 621-626.

SPERBER, D. and WILSON, D. (1986) Relevance: Communication and cognition (Harvard University Press, Cambridge, Massachusetts).

ULATOWSKA, H.K., ALLARD, L., and CHAPMAN, S.B. (1990) Narrative and procedural discourse in aphasia. In Y. Joanette and H.H. Brownell (Eds.) Discourse ability and brain damage (Springer-Verlag, New York), pp. 180-198.

ULATOWSKA, H.K., FREEDMAN-STERN, R., DOYEL, A.W., MACALUSO-HAYNES, S., and NORTH, A.J. (1983) Production of narrative discourse in aphasia. Brain and Language, 19, 317-334.

ULATOWSKA, H.K., NORTH, A.J., and MACALUSO-HAYNES, S. (1981) Production of narrative and procedural discourse in aphasia. Brain and Language, 13, 345-371.

VAN DIJK, T.A. (1977) Text and context. Explorations in the semantics and pragmatics of discourse (Longman, London).

VAN DIJK, T.A. (1980) Macrostructures: An interdisciplinary study of global structures in discourse, interaction, and cognition (Lawrence Erlbaum Associates, Hillsdale, New Jersey).

VAN DIJK, T.A. and KINTSCH, W. (1983) Strategies of Discourse Comprehension (Academic Press, New York).

VAN LANCKER, D.R. and KEMPLER, D. (1987) Comprehension of familiar phrases by left- but not by right-hemisphere damaged patients. Brain and Language, 32, 265-277.

WEYLMAN, S.T., BROWNELL, H.H., and GARDNER, H. (1988) "It's what you mean, not what you say": Pragmatic language use in brain-damaged patients. In F. Plum (Ed.) Language, communication, and the brain (Raven Press, New York), pp. 229-243.

WEYLMAN, S.T., BROWNELL, H.H., ROMAN, M., and GARDNER, H. (1989) Appreciation of indirect requests by left- and right-brain-damaged patients: Effects of verbal context and conventionality of wording. Brain and Language, 36, 580-591.

WINNER, E. and GARDNER, H. (1977) The comprehension of metaphor in brain-damaged patients. Brain, 100, 717-729.

 

APPENDIX B