BRAIN AND LANGUAGE 49, 105—124 (1995) Agrammatic Sentence Production: The Use of a Temporal—Spatial Strategy Lynn M. Mauer,*t ANJAN CHATTERJEE, LESLIE J. GONZALEZ Rorut,**§ AnD KENNETH M. HEILMAN*** *Department of Neurology, University of Florida College of Medicine, Gainesville, Florida; +Research Service, ‘Neurology Service, and $Audiology/Speech Pathology Service, Veterans Affairs Medical Center, Gainesville, Florida; and tDepartment of Neurology, University of Alabama at Birmingham, Birmingham, Alabama We report the sentence production of a left-handed man with a right-hemisphere infarct. He demonstrated an inability to correctly map grammatical categories (subject, object) onto thematic roles (agent, patient) even for simple active sen- tences. The patient’s performance appears to be the result of selective damage to the functional level (Garrett, 1980) of sentence production. His failure could not be accounted for by theories of agrammatism that implicate memory deficits, phonologic processing impairments, or deficits in processing complex transforma- tions. The patient's performance revealed the consistent application of a tempo- ral-spatial strategy in sentence production, despite adequate lexical—-semantic abilities. © 1995 Academic Press, Inc. Agrammatic aphasic patients are typically characterized by halting and effortful fragmented language production in which syntactical complexity of sentences is reduced and content words are used more frequently than grammatical words (articles, prepositions, auxiliary verbs). Bound grammatical morphemes (i.e., verb endings, possessive markers, etc.) are Supported in part by a Rehabilitation and Development grant and the Medical Research Service of the Department of Veterans Affairs and NIH Grant 1ROINS2866S and the Memory Disorder Clinic of the Department of Health and Rehabilitative Service of the State of Florida. Portions of this paper were presented at the Annual Convention of the American Speech/Language/Hearing Association in November 1991, Atlanta, GA. The authors acknowledge Steve Nadeau, M.D., for his feedback on the initial work-up of this patient, Anne Crawford for manuscript preparation, Janet Wootten for manuscript editing, and especially WH for his patience and tolerance throughout the course of our study. Address correspondence and reprint requests to A. Chatterjee, M.D., Department of Neu- rology, University of Alabama at Birmingham, 1720 7th Avenue South, 454 Sparks Center, Birmingham, AL 35294-0017. 105 0093-934X/95 $6.00 Copyright © 1995 by Academic Press. Inc. All rights of reproduction in any form reserved. 106 MAHER ET AL. frequently omitted and single-word comprehension is relatively preserved (Goodglass, 1968). Various theories have been postulated to explain how such patterns of sentence production occur. For example, Goodglass noted that agrammatic patients omit unstressed grammatical morphemes. He invoked a psychological concept of ‘‘saliency’’ and thought that agrammatic patients focused on salient morphemes that are phonologi- cally stressed. Subsequently, Kean (1978) proposed a theory of agramma- tism making explicit reference to phonologic representations. She under- scored a distinction between phonological words (which provide stress in English) and clitics (which do not affect stress) and argued that phono- logic words were retained in agrammatic language, whereas clitics were relatively underused. In contrast, Bradley, Garrett, and Zurif (1980) and Zurif (1982) have argued that agrammatics are impaired in their ability to use ‘‘closed-class’’ or grammatical elements and thus are unable to mark the grammatical relationship between some phrasal constituents within a sentence. Others postulate that the morphemic abnormalities results from a more basic deficit (Saffran, Schwartz, & Marin, 1980). Finally, some theories, such as one proposed by Kolk, Grunsven, and Keyser (1985), implicate general processing defects such as a slowing down of ‘‘on-line”’ processing without necessarily the loss of syntactic knowledge. Kolk in- terprets the observed linguistic behavior such as the omission of gram- matical morphemes as strategic adaptations. Production deficits of agrammatic patients vary considerably. Dissocia- tions between the omission of grammatical morphemes and other word order abnormalities (Saffran et al., 1980; Kolk & Van Grunsven, 1985; Berndt, 1987), between the production of bound versus unbound gram- matical morphemes (Saffran, Berndt, & Schwartz, 1989), and between elements within classes of grammatical morphemes (Miceli, Silveri, Ro- mani, & Caramazza, 1989) have all been described (See Berndt, 1991, for a review). Such dissociations make generalizations about the core nature of agrammatism suspect, leading some investigators to question the theoretical underpinnings of a unifying concept of agrammatism (Badecker & Caramazza, 1985). The term ‘‘agrammatism’’ most likely refers to a variety of disorders rather than a homogeneous syndrome. Without making strong claims concerning the core nature of an idealized agrammatism, detailed analysis of the ways in which sentence processing may get disrupted in single case studies continue to inform us about the nature of sentence production. Models of sentence production date back at least to Pick’s (1931) dis- cussions of agrammatism. They delineate different levels of representa- tions involved in the formulation of a sentence starting with a preverbal communicative intent that is subsequently coded into its linguistic form in stages, prior to being articulated. In this vein, Garrett (1980) used a body of normal speech errors as his data base to develop a detailed model AGRAMMATIC SENTENCE PRODUCTION 107 of sentence production. The preverbal level of representation is named the message level in this model. Garrett provides little analysis of this level (see Levelt, 1989, for a more detailed discussion of this level). The message level triggers several operations that occur at the next level of representation, the functional level. A preliminary lexical search of open class words occurs, a predicate argument structure that defines thematic roles is created, and potential lexical items are inserted into the argument structure. The representation is still relatively abstract and may result in a variety of surface manifestations. At the position level, the lexical items are given their phonemic representations and assigned phrasal sites, which are constructed at this level. This frame provides the context for the selection and insertion of closed-class grammatical morphemes in the phrasal sites, resulting in a phonemically specified string prepared for phonetic processing. In such a model, deficits at any point or combination of points in the process could result in disturbances of sentence produc- tion. The use of such a model in analyzing data from agrammatic patients both helps to organize the investigation, as well as refine and redefine the model (Caramazza & Hillis, 1989). We report data from a well-educated, left-handed man with an exten- sive right-hemisphere lesion that resulted in an aphasia with grammatic deficits in production, as well as asyntactic comprehension. This report focuses on the nature of his production deficit. In the accompanying paper, we report investigations of his comprehension defect. The purpose of these papers is to describe the consistency of the patient's failure across tasks of sentence comprehension and production, to identify the level of impairment responsible for his deficits, and to report his perva- sive application of a temporal—spatial strategy in sentence processing. In this paper we focus on his production performance in order to learn what his deficit reveals about the structure of sentence production. This patient’s left-handedness and language disturbance with a right- hemisphere lesion indicates anomalous anatomic brain organization and could provide insights into functional dissociations of language compo- nents not usually seen in right-handed aphasics. CASE HISTORY WH was a 66-year-old, left-handed college professor who, following a few episodes of transient left upper extremity weakness, underwent a right carotid endarterectomy. Immediately following surgery he was found to have an aphasia and a left hemiparesis. Computer tomographic scan revealed an extensive right-hemisphere infarction in the distribution of the right middle cerebral artery. The large lesion involved the inferior, middle, and superior temporal gyri, the angular and supramarginal gyri, and the precentral and postcentral gyri, and the insular cortex, possibly 108 MAHER ET AL. extending into caudal portions of the inferior frontal gyrus, but sparing the pars triangularis. The testing reported here was initiated 6 weeks after his stroke and spanned the following year. At that time he did not demonstrate evidence of unilateral spatial neglect on line bisection, cancellation, or drawing tasks. He also did not show any evidence of auditory, visual, or tactile extinction. Digit span was six forward and five backward, suggesting adequate short-term memory. Standard Aphasia Testing As can be seen from Table 1, WH’s aphasia was tested periodically throughout the year and demonstrated a relatively stable aphasia profile as indicated by the Western Aphasia Battery (Kertesz, 1982) from 6 weeks to a little more than | year following his stroke. Performance on the Western Aphasia Battery was consistent with a transcortical sensory aphasia. However, the patient had flawless comprehension of single words. Naming was remarkably preserved, as can be seen by WH’s per- formance on the Boston Naming Test (Kaplan, Goodglass, & Weintraub, 1983) and the Action Naming Test (Obler & Albert, 1986). Nine of the 12 errors on the first administration of the Action Naming Test were the nominalized verb target, e.g., ‘“hand shake’’ for shaking hands, ‘‘conduc- tor’’ for conducting, etc., which suggests difficulty generating the correct morphological form for the task, rather than a failure of lexical retrieval. To assess his lexical/semantic system more completely, WH was asked to name 60 pictures of low- and high-frequency nouns belonging to six semantic categories, which he did flawlessly. He was also able to provide a specific functional definition for each item. In addition, he was able to name another 60 low- and high-frequency members of six other semantic categories based on their verbal definition. Furthermore, he scored 169 of 175 on the Peabody Picture Vocabulary Test (Dunn & Dunn, 1981) a test of single-word comprehension. Based on these results, we concluded that WH’s single-word lexical-semantic abilities were preserved. WH had fluent, prosodic speech output that was marked by poor syn- tactic organization and substitution and omission of free-standing and bound morphemes. The following is an excerpt from his initial interview. Examiner: ‘‘What kind of problems are you having with your speech?” WH: ‘Well, uh, essentially language abandon preposition, I telegraph. . . 1, 1, consciously, uh, continuity .. . 1, 1, uh, this subtle of prepositional phrases this simply cannot do, Under stress, under stress rapid I just flustered . . . but continue to do basically...” The following is a verbatim excerpt (except for the physician's name) from a letter he typed to a professional colleague approximately 1 year after his stroke: AGRAMMATIC SENTENCE PRODUCTION 109 TABLE 1 Results of WH on Standard Tests Testing intervals following WH’s stroke 6 Weeks 5 Months 9 Months 15 Months Western Aphasia Battery Spontaneous speech 13 15 15 15 Info. content & 9 9 9 Fluency 5 6 6 6 Comprehension 73 6.2 8.2 8.4 Repetition 9.2 10 9.4 9.8 Naming 9.8 10 10 9.7 AQ. 78.6 82.0 85.2 85.8 Boston Naming Test + 58/60 Action Naming Test + 43/55 + 55/55 “The diagnosis is Trans-cortico Syndrome. It was pure accident. The attending physician, XXXX XXXX, that a blood clot develops and I can’t stop it. Oh, well, cross your fingers. And ZAP! .. . That's all she wrote. So .. . I forgot the language. Hell, I CAN't even write a sentence. The relatively rare phenomenon: Agrammatism, aphasia. | damn near died. But I survive. On with the show.”' The patient’s speech profile does not fit that of the typical Broca’s aphasic (the classical agrammatic) in that he is not telegraphic, there is no evidence of phonemic struggle, phonetic disintegration, or phonemic paraphasia, and he repeats well. WH also does not fit the profile of a patient with transcortical sensory aphasia, in that his lexical-semantic system is remarkably preserved and his spontaneous speech is marked with significant syntactic errors. Specialized sentence production testing. The subsequent sections de- scribe the measures and results we used to isolate the nature of WH’s production deficit. Unconstrained but predictable sentence production. Saffran et al. (1989) have described a methodology by which speech production may be analyzed and compared across patients. Structural complexity and morphological competence is analyzed by having the subject tell the Cinderella story. The following is an excerpt from WH’s narrative speech. “Once upon a time Cinderella, a beautiful young daughter and two ugly selfish of the sister and brother and uh shes . . . Cinderella tends to uh shove aside and the preferred of parent uh of the sisters preferred uh but she shunted off and neglected uh Cinderella . . . and she does she she she jewelry and finery and she dominate the Cinderella of the sister and shes... uh... a a great ball of the prince and royalty and she expected to be the uh older sister to to to the ball and to royalty look forward but Cinderella she said no you chose to the Cinderella Pl stay home and we'll all very enjoy and shes all alone and very unhappy...” 110 MAHER ET AL. In this narrative WH made both syntactic and morphologic errors. Of the 22 utterances coded in the passage, 14 could be considered sentences, and only 4 of these were complete and grammatical. His sentence produc- tion was marked by morphologic omissions as reflected in his lower pro- portion of inflections, auxiliary verbs, and closed-class items. There was also evidence of morphologic substitutions. Despite this profound gram- matic disturbance he had a near normal rate of speech, use of deter- miners, pronouns, and main verbs and a relatively normal proportion of closed-class items (Table 2). He also did not have phonemic struggle or disintegration. Phonologic processing. Various theories have emphasized phonologi- cal deficts in agrammatism. The phonological implications of Goodglass’s (1968) emphasis on omissions of unstressed morphemes and Kean’s (1978) explicitly phonological theory led Schwartz, Marin, and Saf- fran (1979) to speculate that syntax and phonology are ‘‘tightly wired’ in the left hemisphere and perhaps represent a final common pathway in sentence production. To test his phonological abilities, we asked WH to repeat, read aloud, and write to dictation a list of 60 morphologically complex single words and a list of 40 six-word sentences of varying syntactic structures taken from the Assessment of Intelligibility of Dysarthric Speech (A.1.D.S.) (Yorkston & Beukleman, 1981). As shown in Table 3, WH did well on this task, without evidence of significant omissions or substitutions of unstressed morphemes. WH was also asked to read a list of 20 functors aloud, which he did with 100% accuracy. In reading the 100-word rainbow passage aloud, he made only two unstressed omissions: a plural ‘’s’’ and the article ‘‘a’’, both of which were self-corrected. He made two unstressed substitutions, the/his and a/the, one of which was self-corrected. Thus, under the constraints of a reading task, WH produced unstressed morphemes accurately. From these data it is difficult to attribute his agrammatic production, as exem- plified by the Cinderella story, to a deficit in phonological processing. He demonstrates a ‘‘morphosyntactic’’ deficit that is dissociated from phonology. We then examined his morphologic abilities in greater detail. Production of inflectional morphology. To directly test WH’s ability to produce accurate inflectional morphemes he was given the test de- signed by Goodglass and Berko (1960). WH listened to short sentences and was required to complete the last sentence with the correct word form (e.g., The millionaire bought a new horse. He now has a stable full of ). As can be seen from Table 4, WH performed poorly on this task, consistent with his narrative production. His poor performance could be the direct consequence of a deficit in processing of grammatical morphemes at the positional level of Garrett’s (1980) model (Bradley et al., 1980; Zurif, 1982). If he had a specific deficit in the processing of AGRAMMATIC SENTENCE PRODUCTION Hl TABLE 2 Comparison of WH’s Cinderella Story with Normal Control and Agrammatic Data Reported by Saffran et al. (1989) Normal WH control Agrammatic Words per minute 109 132.4 + 19 31.44 114 Proportion of closed class words 476 558 + .04 326 + .08 Obligatory use of determiners 1 1 372 + 24 Pronoun/noun index 1.764 1.60 + .85 8.93 + .27 Verb/noun index 1.764 1.18 + .30 2.93 + 1.92 Verb inflection index -66 -976 + .054 342 + .29 Auxiliary complexity index 6 1.43 + 31 S1 + 30 Proportion words in sentences -615 1 43 + 14 TABLE 3 Results of WH on Constrained Speech Production Tasks Morphologically Sentences complex single words (%) (%) Reading aloud 98 100 Repetition 100 100 Writing to dictation 95 100 Omission of unstressed morphemes 0 0 TABLE 4 Results of WH on Inflectional Morphology Task (Goodglass & Berko, 1960) Plural Past tense Present Possessive Comparative Superlative Testing intervals following stroke (No. correct/total possible) 6 Weeks 5 Months 3/12 8/12 1/12 6/12 1/12 52 O/12 O/12 0/6 3/6 0/6 0/6 12 Months VA/12 4/12 8/12 0/12 3/6 1/6 112 MAHER ET AL. grammatic morphemes, he should also do poorly on constrained tasks of morphemic processing. To assess WH’s ability to provide the correct morphology in a more constrained context than the Goodglass and Berko task, WH was asked to select the correct morphologic form, unaffixed, inflectional, or deriva- tional response from a choice of three alternatives that best completed a sentence (e.g., ‘‘He detect/detected/detection a small problem with the car.’’) (Badecker and Caramazza, 1986). WH did well on this task, cor- rectly completing 20 of 21 sentences, suggesting that he was able to choose between the inflected or derived form when provided with the sentence frame. Grammaticality judgment. To further assess WH’s ability to select the correct morphology within a specified context, he was asked to evaluate the grammaticality of 40 sentences, 20 of which were agrammatic. Of the 20 sentences with errors, 9 had morphologic ending errors (5 omissions, 4 substitutions), 4 had auxiliary verb form errors (3 omissions, | inappro- priate inclusion), 5 had free standing functor errors (2 omissions, 3 substi- tutions), and 2 had word-order errors. The sentences were presented orally and on a later date were presented in written form. WH success- fully judged the grammaticality of these sentences (90% accurate on both auditory and written presentations). Under the auditory condition he made errors in 1 sentence in which a past-tense inflectional ending was omitted and in | sentence in which there was a substitution of a preposi- tion. Under the reading condition he made an error on | sentence in which a verb auxiliary was inappropriately included and | sentence in which a bound adverbial inflection was appropriately included. Although a grammaticality task might be considered to be a comprehension task, these results are included in the discussion of WH’s production abilities because in addition to ‘‘sensing”’ that these agrammatic sentences were incorrect, he was able to then produce the correct form of 15/20 agram- matic sentences on the written version. Specifically, he appropriately corrected 8/9 bound morpheme errors, 3/4 auxiliary verb form errors, 2/5 free-standing functor errors, and 2/2 word-order errors. We thus have evidence that WH processes and produces grammatical morphemes more accurately when not required to provide the context on-line in which they occur. Anagrams, We then assessed if WH could correctly order sentential elements. WH was given each word of a sentence presented individually on cards. His task was to place the individual cards in the correct order to make a grammatically and semantically accurate sentence. Ten six-word sentences were selected from the previously described A.I.D.S. sen- tences. Seven of the 10 sentences generated by WH were both grammati- cally and semantically accurate. An additional sentence was grammati- cally accurrate, but the nouns in the sentence were incorrectly ordered, AGRAMMATIC SENTENCE PRODUCTION 113 resulting in a reversal of the correct thematic relationships in the sen- tence. Picture description. Thirteen pairs of pictures were presented to the patient in random order. Each picture depicted an action that included an agent (i.e., the actor) and a patient (i.e., the recipient) of the action. Six of the pictures were semantically reversible in that either noun could be the subject (e.g., a man pushing a woman, or a bus pulling a truck). The remaining 7 pictures were not reversible, always involving an inani- mate and an animate noun (e.g., a man painting a wall). The paired sen- tence for the reversible sentences presented the opposite thematic rela- tionship (e.g., a woman pushing a man), whereas for the nonrevers- ible sentences, the paired sentence was a lexical variation (e.g., a woman painting a wall). The agent of the action was either on the left side of the page, on the right side of the page, or near midline. The same 26 pictures were presented under three conditions: 1. UNCONSTRAINED. WH was asked to simply describe what he saw in the picture. Two pictures of the same pair were never presented in se- quence. WH was free to select any words and sentence type to describe the picture. When multiple attempts to describe the picture were pro- duced the final sentence was considered his response. WH exclusively responded with active sentences, almost always in the present perfect tense (e.g., “The man is painting the wall.”*). Of 26 sentences, WH pro- duced grammatical simple sentences on all but one (the addition of *‘at’’ for the sentence ‘‘The woman is sewing at her hat’’). However, on 3 of the reversible sentences WH assigned the incorrect thematic roles to the nouns. Errors in thematic role assignment occurred only on those sen- tences where the agent was placed either to the right of the page or at midline. Furthermore, multiple attempts at a target occurred on 42% of reversible sentences, but only on 6% of nonreversible sentences, sug- gesting that WH had more difficulty generating the correct sequence of words when not aided by animacy or the semantic plausibility of the sentence. If the agent was located to the left of the page, the sentence was produced flawlessly (see Table 5). 2. PASSIVE SENTENCES—CONSTRAINED BY NOUN. In the second condition WH was required to describe the pictures by using the thematic patient first in the sentence, in order to try to force him to use a passive construc- tion (e.g., for the picture of the man painting the wall, begin with ‘‘the wall’’). On these sentences, WH produced only 3 grammatically and the- matically correct passive sentences. The other 10 sentences were pro- duced either accurately in active voice, reflecting the opposite thematic relationship, or were an agrammatical mixture of active and passive sen- tences (e.g., ‘‘The bull is chasing by the boy’’) (see Table 5). 3. PaSsSIVE SENTENCES—CONSTRAINED BY VERB. In the third condition, WH was given the passive verb form to use in the sentence (e.g., “‘is 114 MAHER ET AL. painted’’) and asked to describe the picture. WH’s grammaticality im- proved in this condition, as he produced 11/13 grammatically correct passive sentences. However, the correct thematic relationship was main- tained in only 4 of them, as depicted in Table 5. In this condition, animacy had little effect on the correct sequencing of the nouns in the sentence, resulting in such responses as ‘‘The girl is painted by the wall’’ and ‘“The woman is sewn by the hat.’’ However, it was only on nonreversible sentences that WH generated multiple attempts at the same target, again indicating that on some level he was aware of the inadequacy of these responses. He generated inaccurately sequenced reversible sentences without hesitation. Because these conditions were presented in blocks, we were concerned that WH was influenced by the sentence form he had produced on the immediately preceding item. The entire set of pictures was repeated, randomly varying the active and passive targets by indicating which noun should occur first in the sentence. WH did not generate any grammati- cally or thematically correct passive sentences in this condition. He con- sistently used active sentences, which resulted in 100% reversal of the- matic role assignment for passive targets and 0% passive morphology. Conversely, he made no errors when the target was an active sentence. Errors of thematic role assignment were not observed on active sentences in this condition since WH was provided with the noun to appear first (i.e., in the agent position) in the sentence. Based on these performances we had evidence that under certain con- ditions WH was capable of generating sentences with correctly se- quenced words. WH produced grammatically and semantically accurate active sentences when aided by animacy and plausibility of thematic roles TABLE 5 Results of WH’s Picture Description Testing Agent on (No. correct/total possible) Left Right Middle 1, Unconstrained Plausibly reversible sentences: Nouns of equal animacy 5/5 25 0/2 Nonreversible sentences: Nouns of differing animacy 6/6 4/4 5/68 2. Passive sentences: constrained by noun Plausibly reversible sentences: Nouns of equal animacy 2/2 0/2 O/1 Nonreversible sentences: Nouns of differing animacy O3 W/2 0/3 3. Passive sentences: constrained by verb Plausibly reversible sentences: Nouns of equal animacy 0/2 1/2 VI Nonversible sentences: Nouns of differing animacy 0/2 1/3 1/3 * Error resulted from addition of ‘‘at’’ to the sentence. AGRAMMATIC SENTENCE PRODUCTION 115 of major lexical items. However, when he is not able to make use of animacy and semantic plausibility to guide the sequencing of nouns in the sentence, he had more difficulty. When placed in the situation in which the selection of a passive verb form is required to accurately reflect the thematic relations, WH not only failed to represent the meaning of the picture accurately but also at times failed to generate the correct morphology for either an active or passive sentence, supporting our hy- pothesis that his morphological failure may be a result of an inability to organize the sentence frame at the functional level of representation. This may be consistent with linguistic theories that stress the verb as the central element of sentence structure. WH provided further evidence of his morphologic competence once the sentence frame is structured, by his dramatic improvement in gram- maticality when given the verb form to be used in the sentence. However, he continued to fail at sequencing the nouns in the sentence to reflect the appropriate thematic relations. Not even animacy cues seemed to facili- tate performance on passive sentences. For example, when required to start a sentence with the inanimate object in order to force a passive construction, he produced sentences like ‘‘The wall is painting the girl.”” WH’s performance on the reversible active sentences suggests that his assignment of nouns within the sentence frame was not random, but could have been influenced by the spatial location of the items in the picture. His only failures in producing active sentences occurred when the subject of the sentence was not to the left of the page. Our observations suggest that WH might have been using a spatial strategy in his processing of sentences. To investigate WH’s possible use of a spatial strategy in sen- tence processing in the absence of animacy and plausibility cues, another sentence production task was employed. Picture description—circle and square. The task, adapted from the syntax comprehension task of Schwartz et al. (1980), uses stick figures of a circle and a square representing various actions (see Fig. 1). Since the circle figure and square figure are ‘‘thematically’’ interchangeable, all the target sentences were reversible. Half of the pictures depicted the action occurring from left to right, the other half depicted the action from right to left. For half of the sentences, WH was free to describe the picture any way he wanted. For the other half of the sentences, WH was given either the noun to use first or the verb form to use in the sentence, forcing a passive sentence to accurately describe the picture. We hypoth- esized that if WH was using a left-to-right, temporal-spatial strategy in picture description, then for active sentences, those that were drawn with the action proceeding from left to right would be described correctly and those that were drawn from right to left would be produced with the thematic roles reversed. For the passive sentences we predicted the op- posite effect: in using a left-to-right strategy, those pictures drawn with 116 MAHER ET AL. Fic. 1. Stick figures depicting actions from left to right (top) and from right to left (bottom). the action proceeding from right to left would be described accurately because in the passive sentence form the agent of the action occurs after the verb. As predicted, when free to describe the picture independently, WH generated only active sentences (see Table 6). The spatial direction of the depicted actions significantly affected the semantic accuracy of WH’s production (x? = 17.1, 3 df, p < .001). For those pictures with the action represented from left to right, with the exception of one omission of the preposition ‘‘to,"”» WH generated grammatically and semantically accu- tate sentences. For those sentences that were depicted with the action proceeding from right to left, WH produced only 2 of 12 semantically accurate sentences, even though they were all grammatically correct (on 1 of the 2 correct sentences he commented that the picture was ‘‘drawn backwards’’). Conversely, for those passive sentences where he was given the noun to use first, WH again generated gramatically correct active sentences, using the figure on the left side of the picture as the subject, regardless of the direction of the action. For the 6 sentences AGRAMMATIC SENTENCE PRODUCTION 117 TABLE 6 Results of WH's Active and Passive Sentence Production with Circle and Square Stick Figures No. correct semantic No, correct relations morphology Direction (No. correct/ (No. correct/ Sentence form of action total possible) total possible) Active Left > Right 12/12 1/42 Active Right — Left 2/12 12/12 Passive Noun provided Left — Right 0/6 6/6 Verb provided Left — Right 1/6 6/6 Passive Noun provided Right —> Left 0/6 6/6 Verb provided Right > Left 5/6 4/6 pictured from right to left for which WH was given the passive verb form, 3 were grammatically and semantically accurate, 2 more were semanti- cally accurate but had morphological omissions (both requiring an addi- tional functor, ‘‘is waved to by”’ and ‘‘is smiled at,’’ which were omitted), and the 6th was grammatical but had the thematic roles reversed. How- ever, for those pictures with the action proceeding from left to right, the passive sentences were all grammatically correct but only one sentence was semantically accurate. Our observations suggest that WH was using a temporal-spatial strat- egy in assigning linguistic thematic roles to major lexical items. Specifi- cally, in his sentence constructions he utters the grammatical subject first and matches this to the item on the left of the picture. This spatial strategy is mitigated by his propensity to spontaneously produce only active sen- tence forms. Thus, if shown a picture of a circle being kicked by a square, and asked to begin the sentence with ‘‘the circle’’ he always said ‘‘the circle is kicking the square”’ regardless of the spatial location of the circle and the square. However, if he is constrained to using the verb construc- tion ‘is kicked by,’” he is much more likely to be accurate if the action is depicted going right to left (5/6) rather than left to right (1/6), as shown in Table 6. To further test the hypothesis that WH is influenced by tem- poral and spatial characteristics in sentence construction under less demanding (nongenerative) conditions, another anagram task was con- structed. Additional anagrams. Six sentences of each of the following categories were administered: (1) active reversible sentences where either noun is a plausible subject, e.g., ‘The man is kissing the woman’’; (2) active reversible sentences that are semantically possible but improbable when 118 MAHER ET AL. reversed, e.g., ‘‘The reporter is interviewing the politician’; (3) active irreversible sentences, e.g., ‘‘The maid is cleaning the window’’; (4) pas- sive reversible sentences, where either noun could be a plausible agent, e.g., ‘The tiger is hurt by the lion’’; (5) passive reversible sentences that are semantically possible but implausible when reversed, e.g., ‘‘The hunter is mauled by the lion’; and (6) passive irreversible sentences, e.g., “‘The book is read by the student.’’ Given our previous observations, we made the following predictions regarding WH’s word order performance on these anagrams. Because of his competence in making grammaticality judgments and ability to do anagrams, WH would construct sentences that were grammatic regardless of the form (active or passive). He would then place the major lexical item that he thought was the logical subject to the left of the verb even if it was the grammatical object as determined by the argument structure provided by a passive verb form. It would follow that he would perform well on both active and passive forms of reversible sentences, since no specific predictions can be made about which noun is the logical subject. For active implausible reversible and active irreversible sentences he would also perform well because these constructions require the logical subject to be to the left of the verb. Because of his preserved single word retrieval abilities and preserved semantics, we expected WH to recognize which lexical item serves as the logical subject based on semantic plausibility. In the passive construc- tion, we predicted that he would do worse than chance on implausibly reversible and irreversible sentences since he would place the logical subject on the left of the verb. For all 36 of the anagram sentences, the nouns, verbs, and functors were correctly sequenced to generate a grammatical sentence 100% of the time (see Table 7). However, WH produced only one active, improbable thematic role reversal (i.e., ‘‘The politician is interviewing the reporter’), and, as predicted, he generated 5 of 6 implausibly reversible passive sentences (*‘The tiger is mauled by the hunter’’) and failed in 6 of 6 irreversible passive sentences (‘‘The girl is climbed by the stairs’’). These TABLE 7 Results of WH’s Performance on Anagram Tasks Grammatically correct Semantically correct (No. correct/No. possible) (No. correct/No. possible) Active/reversible 6/6 6/6 Active/semantically implausible 6/6 5/6 Active/irreversible 6/6 6/6 Passive/reversible 6/6 6/6 Passive/semantically implausible 6/6 1/6 Passive/irreversible 6/6 0/6 AGRAMMATIC SENTENCE PRODUCTION 119 results suggest that, even though WH is able to sequence the grammatical elements in the sentence, he had significantly more difficulty (x? = 12.50, 3 df, p < .05) ordering the nouns of the sentence to respect their thematic relationships in the passive form (i.e., when the subject must appear to the right or after the verb). In summary, WH demonstrated a word-order problem, specifically when the logical subject should be before or to the left of the verb. He made these errors despite correctly placing grammatical morphemes such as ‘‘by’’ after the verb in passive constructions. GENERAL DISCUSSION We present data from a well-educated, left-handed man with a large right-hemisphere stroke that resulted in a language disturbance. His anomalous language organization provided an opportunity to probe for a pattern of deficits that would uncover patterns of dissociations not usually seen in right-handed aphasic subjects. His lexical-semantic system, as tested by picture naming, naming to provide definitions, as well as provid- ing definitions for words, was exquisitely preserved. He also did well on tests of reading and repetition, demonstrating the virtual absence of omission or substitution of unstressed grammatical morphemes and sug- gesting that this phonological processing was also normal. Despite his normal performance at the single-word semantic level and his normal phonological processing, he demonstrated a marked disorder of sentential processing, both in producing and in comprehension (discussion in the accompanying paper). His spontaneous speech production was notable for both morphologic omissions and substitutions. His utterances were often fragmented, and when they contained elements of a sentence, they were frequently agrammatic. This production pattern was observed de- spite a near normal rate of speech, without any evidence of phonemic struggle or disintegration. WH therefore demonstrates a clear dissocia- tion of syntax (disrupted) from single-word semantics and phonology (both preserved) in ways not reported for right-handed aphasic patients. His presumed anomalous neural organization of language functions of- fered an unusual opportunity to investigate syntactic deficits in the ab- sence of commonly associated deficits of short-term memory, phonology, or lexical-semantics. To more closely analyze WH’s syntactic deficit (or set of deficits), we used Garrett’s (1980) model of sentence production. Our analysis sug- gests that WH for the most part had preserved positional level processing. We contrast his performance to patient ML described by Caramazza and Hillis (1989), who was interpreted as having a positional level deficit. ML demonstrated normal comprehension and agrammatic sentence produc- tion. She made both omission and substitution errors with bound and 120 MAHER ET AL. unbound grammatical morphemes in both spontaneous speaking and reading and repetition. Her production of single words was normal. She performed poorly on a sentence anagram task. This abnormal perfor- mance was interpreted as reflecting an inability to use local contexts to guide her performance. Similar to ML, WH also made omission and substitution errors of grammatical morphemes in spontaneous speech. However, unlike ML, he performed accurately on more constrained tasks of reading and repetition. Although both ML and WH were agrammatic, their deficits were quali- tatively different. Unlike ML, on constrained morphological tasks, WH performed well. He read and repeated grammatical morphemes accu- rately. When given either noun phrases or the verb to be used in picture description, WH usually produced well-formed grammatical sentences. On anagram tasks, he accurately constructed the sentence frame imposed by the presence of specific grammatic morpheme and also chose the cor- rect verb inflectional form. If he had a positional level deficit, he would have done poorly on these tasks. WH’s phonological and position level competence suggests that his deficit must lie in a prior level of representa- tion, and we have evidence that he has a deficit at the functional level. According to Garrett’s (1980) model, several processes occur at the functional level of sentence representation. First, there is a lexical re- trieval of open-class words. From WH’s naming abilities, we know that his ability to retrieve nouns and verbs was preserved. Since the predicate argument structure is also set up in the functional level, retrieval of verbs must play an important part of this process. The model as currently speci- fied does not provide an explanation for why some patients tend to nomi- nalize verbs and what the consequences of this tendency would be on the setting up of an argument structure. The fact that on picture descrip- tion tasks WH used a subject-verb-object ordering in his responses sug- gests that he used the basic form of a predicate argument structure. Where he clearly failed is in the mapping of logical thematic roles (such as agent and patient) onto grammatical roles (such as subject and object) in the sentence. His mapping of thematic roles were not random. Spatial characteristics of the stimuli influenced his performance. He preferen- tially placed the logical agent to the left of the verb regardless of the verb form. Thus, when given the first noun to be used in a picture description, he performed well when the picture depicted the action going from left to right. When the action to be described was drawn right to left, he performed poorly, reversing thematic roles. When given anagram tasks, he was able to construct both active and passive forms of the sentence. In this task he accurately processed the form of passive sentences, includ- ing accurate placement of verb auxiliaries and prepositions, although he never spontaneously uses the passive form. His performance on ana- grams demonstrated that, despite some retained ability to process pas- AGRAMMATIC SENTENCE PRODUCTION 121 sive verb forms, he did not extract the predicate argument structure cre- ated by this verb form. He treated both the active and the passive form in the same way and again placed the logical subject to the left of the verb, even when the results were semantically improbable (‘‘The lion is mauled by the hunter’’) or impossible (‘‘The student is read by the book’’). Two other cases with functional level abnormalities have been re- ported. Martin and Blossom-Stach (1986) reported a mild Wernicke’s aphasic who made role reversal errors despite spontaneous production with acceptable surface form. Caramazza and Miceli (1991) also reported a case of thematic role impairment with the spared ability to process the morphological structure of sentences. Our case is different from these cases in two important respects. First, unlike the other two cases, WH clearly had no short-term memory deficits, supporting the contention that such memory impairments are not a necessary feature contributing to thematic role assignment failures. Second, WH was severely impaired on both active and passive reversible sentences, unlike the other two cases that performed better on active constructions. Thus, any explanation for his thematic role assignment failure must take his clear deficits in simple active sentences into account. The reason that WH used a left-to-right spatial heuristic in mapping thematic roles is not clear. Right-handed patients with right-hemisphere lesions may have dramatic disturbances of directed attention and scan stimuli abnormally. WH did not show any evidence of spatial neglect on cancellation tasks, line bisection, drawing, or reading tasks. Additionally, a scanning disturbance as a consequence of right-brain damage would be expected to induce a right-to-left rather than a left-to-right pattern (Ishiai, Furukawa, & Tsukagoshi, 1989). Therefore, we think it is unlikely that his left-to-right spatial heuristic in assigning thematic roles is a conse- quence of a scanning disturbance caused by damage to right hemisphere attentional mechanisms. Two alternative (although not mutually exclu- sive) explanations are possible. The first, suggested by Kolk and Van Grunsven (1985), is that we are observing an adaptive strategy. The sec- ond possibility, along the lines suggested by Caplan (1986), is that be- cause of the breakdown of a linguistic process, a preverbal nonlinguistic process is being uncovered. Kolk and Van Grunsven (1985) have suggested that agrammatic pa- tients may have a breakdown of general processing resources. They point out that on-line performance is usually worse than ‘‘off-line’’ perfor- mance by agrammatics. He proposes that as a consequence of this pro- cessing deficit they resort to various positive and negative adaptive strate- gies. Economy of effort might be one such strategy. Because of the effort involved in production, patients use mostly major lexical items to max- imize the communication intent. Could WH’s spatial strategy be adaptive 122 MAHER ET AL. in the sense used by Kolk and Van Grunsven (1985)? Assigning the first produced noun or the left-sided noun to the role of agent may be adaptive in the sense that the active form, which has the logical and grammatical subject on the left, is thought to be canonical and is used more frequently in English. However, as Berndt (1991) has pointed out, ‘‘adaptive strate- gies’’ potentially explain any behavior, making the hypothesis difficult to verify experimentally. Thus, although one cannot disprove the hypothesis that WH is resorting to an adaptive strategy, we find this explanation less compelling. The second possible explanation has its roots in Hughlings Jackson’s (1932) concept of ‘‘dissolution.’’ Jackson proposed that there is a hierar- chy of processes with higher order processes imposing constraints, inhibi- tion, or modification on lower order processes. This principle is generally adopted by the nervous system in a variety of processes, including cogni- tive systems. The observed behavior in brain damage may be a reflection of lower order processes being released from the influence of higher order processing. Using such a conception of functional organization of behav- ior, we would interpret WH’s use of a spatial strategy as a lower order process emerging after the dissolution of a higher level ability to map logical thematic roles onto grammatic roles. Similarly, Caplan (1986) ana- lyzes performance of agrammatics by making reference to Chomsky’s (1981) idea of a ‘‘primitive basis of concepts.’’ Primitives are described as prelinguistic constructs that are epistemologically prior to the mapping of concepts by the language faculty onto grammar. Caplan’s argument, primarily derived from the analysis of asyntactic comprehension, sug- gests that some patients lose the ability to appreciate phrasal structure and use linear sequences of major lexical items. He states, ‘‘It is an intriguing thought that agrammatic patients, or some other group of lan- guage-impaired subjects, may retain this primitive basis and not the struc- tures complexly defined over it’’ (Caplan, 1983). Our observations of WH’s use of a spatial strategy to map logical roles onto thematic roles may be an example of such a primitive basis for linguistic concepts. We will return to this theme in our discussion of WH’s asyntactic comprehen- sion in the accompanying paper. A final comment needs to be made about the adequacy of our analysis thus far. WH’s inability to properly assign thematic roles in production was evident even in extremely constrained tasks. However, the reason that he omits or substitutes grammatical morphemes in spontaneous speech is not immediately obvious. A deficit of thematic role assignment does not necessitate omissions and substitutions of grammatical mor- phemes in spontaneous speech (Martin and Blossom-Stachs 1986, Cara- mazza and Miceli 1991). We are left with someone who does poorly in the ‘‘unconstrained’’ condition of spontaneous speech, but well on constrained tasks such as grammaticality judgments, anagram tasks, and AGRAMMATIC SENTENCE PRODUCTION 123 choosing between specific verb inflections. As currently specified, Gar- rett’s (1980) model does not make distinctions between sentence produc- tion elicited in unconstrained or constrained conditions. The data base upon which this model was developed is derived from samples of conver- sation of normal individuals, presumably an unconstrained condition. Its applicability to ‘‘constrained”’ tasks remains uncertain. It is possible that WH’s pattern represents a damaged linguistic system with limited re- sources, such that in the more difficult task of generating grammatical morphemes with its relatively unrestricted degrees of freedom he is over- whelmed. In contrast, when given a restricted choice such as in the con- strained task, he succeeds. Such an interpretation is consistent with the Kolk and Van Grunsven (1985) hypothesis that agrammatics may have limited on-line processing resources such that off-line performance may be significantly better than on-line production. Note that this hypothesis may explain why WH processes grammatical morphemes better in con- strained rather than spontaneous production tasks. However, it does not apply to WH’s thematic role assignment deficit, which followed the pre- dicted left-to-right pattern even in constrained off-line tasks. Perhaps it is this limitation of on-line processing resources that WH was referring to in his initial interview when he said, ‘‘Under stress, under stress rapid I just flustered . . . but continue to do basically . . .”’ REFERENCES Badecker, W., & Caramazza, A. 1985. On consideration of method and theory governing the use of clinical categories in neurolinguistics and cognitive neuropsychology: The case against agrammatism. Cognition, 20, 97-126. Badecker, W., & Caramazza, A. 1986. The Johns Hopkins University morphology battery, unpublished. Berndt, R. 1987. Symptom co-occurrence and dissociation in the interpretation of agramma- tism. In M. Coltheart, G. Sartori, & R. Job (Eds.), The cognitive neuropsychology of language. Hillsdale, NJ: Erlbaum. Berdnt, R. S. 1991, Sentence process in aphasia. In M. Sarno (Eds.}, Acquired aphasia. New York: Acacemic Press. Pp. 223-269. Bradley, D. C., Garrett, M. F., & Zurif, E. B. 1980. Syntactic deficits in Broca’s aphasia. In D. Caplan (Eds.), Biological studies of mental processing. Cambridge, MA: The MIT Press. Pp. 269-286. Caplan, D. 1983. A note on the ‘‘word-order problem” in agrammatism. Brain and Lan- guage, 20, 155-165. Caplan, D. 1986. Assignment of thematic roles to nouns in sentence comprehension by an agrammatic patient. Brain and Language, 27, 117-134. Caramazza, A., & Hillis, A. E. 1989. The disruption of sentence production: some dissocia- tions. Brain and Langauge, 36, 625-650. Caramazza, A., & Miceli, G. 1991. Selective impairment of thematic role assignment in sentence processing. Brain and Language, 41, 402~436. Chomsky, N. 1981. Lectures on government and binding. Dordrecht: Foris. Dunn, L. M., & Dunn L. M. 1981. Expanded manual for the Peabody picture vocabulary test—Revised. Minneapolis: American Guidance Service. 124 MAHER ET AL. Garrett, M. 1980. Levels of processing in sentence production. In B. Butterworth (Eds.), Language production. New York: Academic Press. Goodglass, H. 1968. Studies on the grammar of aphasics. In S. Rosenberg & K. Joplin (Eds.}, Developments in applied psycholinguistics research. New York: Macmillan. Goodglass, H., & Berko, J. 1960. Agrammatism and inflectional morphology in English. Journal of Speech and Hearing Research, 3, 257-267. Ishiai, S., Furukawa, T., & Tsukagoshi, H. 1989. Visuospatial processes of line bisection and the mechanisms underlying unilateral spacial neglect. Brain, 112, 1485-1502. Jackson, J. 1932. Selected writings of John Hughlings Jackson. London: Hodder and Stoughton. Kaplan, E., Goodglass, H., & Weintraub, S. 1983. The Boston naming test. Philadelphia: Lea and Febiger. Kean, M.-L. 1978. The linguistic interpretation of aphasic syndromes. In E. Walker (Ed.), Explorations in the biology of language. Montgomery, VT: Bradford Books. Kertesz, A. 1982. Western Aphasia Battery. San Antonio, TX: The Psychological Corpora- tion, Harcourt Brace. Kolk, H., & Van Grunsven, M. 1985. Agrammatism as a variable phenomenon. Cognitive Neuropsychology, 2, 347-384. Kolk, H. H., Grunsven, M. J. F. V., & Keyser, A. 1985. On parallelism between production and comprehension in agrammatism. In M.-L. Kean (Eds.), Agrammatism. New York: Academic Press. Pp. 165-206. Levelt, W. 1989. Speaking: From intention to articulation. Cambridge, MA: The MIT Press. Martin, R., & Blossom-Stach, C. 1986. Evidence of syntactic deficits in a fluent aphasic. Brain and Language, 28, 196-234, Miceli, G., Silveri, M., Romani, C., & Caramazza, A. 1989. Variation in the pattern of omissions and substitutions of grammatical morphemes in the spontaneous speech of so-called agrammatic patients. Brain and Language, 36, 447-492. Obler, L. K., & Albert, M. L. 1986. Action naming test. Unpublished, Pick, A. 1931. Aphasia (Brown, J.W., Trans.). Springfield: Thomas. Saffran, E. M., Berndt, R. S., & Schwartz, M. F. 1989. The quantitative analysis of agram- matic production: procedure and data. Brain and Language, 37, 440-479. Saffran, E. M., Schwartz, M. F., & Marin, O. S. M. 1980. The word order problem in agrammatism. II. Production. Brain and Language, 10, 263-280. Schwartz, M. F., Marin, O. S. M., & Saffran, E. M. 1979. Dissociations of language function in dementia. Brain and Language, 7, 277-306. Schwartz, M. F., Saffran, E. M., & Marin, O. S. M. 1980. The word order problem in agrammatism. I. Comprehension. Brain and Language, 10, 249-262. Yorkston, K., & Beckelman, D. 1981. Assessment of intelligibility of dysarthric speech. Oregon: C.C. Publications. Zurif, E. 1982. The use of data from aphasia in constructing a performance model of lan- guage. In M. Arbib, D. Caplan, & J, Marshall (Eds.), Neural models of language processing. New York: Academic Press.