C. Finocchiaro / Journal of Neurolinguistics 15 (2002) 433±446
433
Journal of Neurolinguistics 15 (2002) 433±446 www.elsevier.com/locate/jneuroling
Sensitivity to the verb [^agentive] feature: the case of an aphasic subject Chiara Finocchiaro* Laboratorio di Linguistica, Scuola Normale Superiore, Piazza dei Cavalieri 7, 56126 Pisa, Italy
Abstract Available data show that semantic categories of nouns can be selectively impaired in aphasia. This study investigates whether category-speci®c effects can also be obtained with verbs. The [^agentive] feature is addressed, drawing on the case of an aphasic subject (CAN). CAN showed a clear dissociation between [1agentive] verbs vs. [2agentive] verbs (signi®cantly more impaired). This pattern of performance was consistent across a battery of verb production tests (both in isolation and in context) and grammaticality judgements. The results provide evidence for the importance of the agentivity feature in the representation of verbs. q 2002 Elsevier Science Ltd. All rights reserved. Keywords: Verb semantics; [^agentive]; [^living]; Dissociation
1. Introduction Dissociations among semantic categories of nouns are not uncommon in the neuropsychological literature. The most common dissociation is between `living things' and `non-living things' (Caramazza & Shelton, 1998; Damasio, Grabowski, Tranel, Hichwa & Damasio, 1996; De Renzi & Lucchelli, 1994; Hillis & Caramazza, 1995; Samson, Pillon & De Wilde, 1998; Sartori & Job, 1988; Silveri & Gainotti, 1988; Warrington & Shallice, 1984). Even ®ner-grained dissociations have been reported and it has been proposed that different semantic categories of nouns are separately represented in the brain. Are verbs also categorically organized in the brain? If so, what might the principles of organization be? It is theoretically conceivable to classify verbs with respect to the parameters cognitively valid in the nominal domain (living vs. non-living). However, such an extension from the nominal to the verbal domain does not seem to be equally relevant linguistically: it does not make much sense to deal with verbs concerning living or non-living things, and verbs concerning animals, humans, plants and artifacts. Neither the type of grammatical subject nor the semantic relation among the verb and its arguments are satisfying criteria. * Fax: 139-050-563513. 0911-6044/02/$ - see front matter q 2002 Elsevier Science Ltd. All rights reserved. PII: S 0911-604 4(01)00033-1
434
C. Finocchiaro / Journal of Neurolinguistics 15 (2002) 433±446
Moreover, most verbs can be predicated both of living and non-living things, in a proper or metaphorical sense. 1 In this study the [^agentive] feature is addressed. There are many arguments in support of the relevance of the opposition [1agentive] vs. [2agentive]. First of all, it parallels, to a large extent, the opposition [^living]. 2 In fact: (1) an agent is necessarily animate; (2) an agentive verb has an agent subject. According to Hoepelman (1981, 1986), an individual is said to act if, under precisely the same historical preconditions, he could have refrained from acting, and if ªhe brings about a change which would not have taken place but for his action, or prevents a change which would have taken place but for his actionº (Hoepelman, 1986, p. 3). Furthermore, the agentivity feature has a quite solid linguistic tradition, and may be established on the basis of a set of syntactic tests. A linguistic expression should be considered agentive if it: (a) may occur as a complement of verbs such as to persuade, to command; (b) is compatible with manner adverbs, such as enthusiastically, cleverly and adverbial forms such as for love of, for sake of; (c) is compatible with the command imperative; (d) may be preceded by the phrase What X did was. Moreover, agentivity guides the syntactic behaviour of verbs with two agents (to ®ght, to quarrel, to play chess). These verbs cannot occur in the passive form (e.g. *It was quarrelled by Mary and John), in so far as they have not any non-agentive arguments, and the passive typically emphasizes the non-agentive argument of a given verb. 3 But perhaps a more compelling reason for the cognitive relevance of the opposition [1agentive] vs. [2agentive] has to do with the different thematic structure of [1agentive] and [2agentive] verbs. Grammatical subjects of [1agentive] verbs map onto the thematic role of agent, while grammatical subjects of [2agentive] verbs map onto thematic roles lower in the hierarchy (for instance, experiencers or themes). The ability in assigning verbs their thematic structure has proved to be important for aphasic subjects (Caplan, Baker & Dehaut, 1985; Caramazza & Berndt, 1985; Jones, 1984; Martin & Blossom-Stach, 1986; Saffran, Schwartz & Marin, 1980a,b; Schwartz, Saffran & Marin, 1980). Caramazza and Miceli (1991) described an Italian brain-damaged subject, EB, showing a dissociation between impairment in processing thematic roles and spared ability to process the morphological structure of sentences. EB's dif®culties were restricted, both in comprehension and in production, to semantically reversible sentences. Semantically reversible sentences are typically agentive sentences with two animate arguments (e.g. Il bambino carezza la nonna, `The boy caresses the grandmother'). Because of the presence of two animate arguments, it is impossible to rely on semantic information (the animacy contrast) to assign thematic roles. 4 On the other hand, in the case 1
Comrie (1981) attributes the animacy feature only to NPs, and the control feature to the relation between NP and VP. To be more precise, the opposition [^agentive] in the verbal domain is closer to the opposition [^animate] than [^living] in the nominal domain. 3 As a further con®rmation (drawn to my attention by Pier Marco Bertinetto), it cannot be said *to quarrel s.one, but to quarrel with s.one. Thus, verbs such as to quarrel are intransitive, their arguments being agents. 4 As Bever (1970) points out, sentences in which the semantic relations are unique are relatively easy to comprehend. He suggests that semantic cues are dominant in sentence comprehension, ªsince structural factors (such as the reversal of the canonical actor- action- object order in passive sentences) do not affect psychological complexity when the semantic relations are uniqueº (Bever, 1970, p. 302). 2
C. Finocchiaro / Journal of Neurolinguistics 15 (2002) 433±446
435
of non-reversible sentences, EB could take advantage of the difference in animacy between the subject and the object nouns and circumvent the de®cit in processing thematic roles. Note that EB's de®cit cannot be due to an impairment in the mapping procedure of grammatical functions onto thematic roles. Such an account would fail to explain the dissociation between reversible and non-reversible sentences in production (see Caramazza & Miceli, 1991 for further discussion on this point). This observation gives substantial support to the hypothesis that problems with verb thematic structure, although usually syntactically driven, can also be determined by semantic factors. In this connection, it would be of particular interest to investigate whether or not differences in the type of verb thematic structure can correspond to differences in performance. The opposition [1agentive] verbs vs. [2agentive] verbs allows us to address this question, by verifying whether the thematic role of the grammatical subject can in¯uence subjects' performance (recall that [1agentive] verbs take agent subjects, while [2agentive] verbs take subjects mapping onto thematic roles lower in the hierarchy). As will be shown, subject CAN made far more errors on [2agentive] than on [1agentive] verbs in all production tasks, this effect being hardly reducible to lexical (e.g. frequency) or contextual variables (e.g. the presentation of stimuli in isolation or in context). 2. Case study CAN is a 32 year-old man, premorbidly right-handed, with a high school education. Three months prior to the study, he presented with an acute right hemiparesis with speech dif®culties. A brain CT scan and a cerebral arteriography documented an aneurysm's rupture in the left middle cerebral artery. The subject then underwent a neurosurgical operation for clippage of the aneurysm. A new CT scan performed after the operation revealed the presence of ischaemic areas in the left internal capsule and in the pallido-putaminal region, with an ex-vacuo dilatation of the left ventriculum. Three months post-onset, the subject was submitted to the ENPA (Capasso & Miceli, 2001, in press). CAN's spontaneous speech was non-¯uent but meaningful. Repetition and reading of single words and non-words was accurate (he scored, respectively, 15/15 and 14/15). On repetition and reading of sentences, he performed equally well. He made no errors in comprehension and naming tasks, both on auditory and on visual presentation, without any difference between nouns and verbs. CAN was able to generate lists of words beginning with the same letter, or belonging to the same category (¯uency test). However, the subject tended to produce action verbs rather than non-action verbs. Although this result could re¯ect the prototypical status of action verbs, CAN's behaviour on verbs was further investigated by means of the following tests: Picture comprehension; Picture naming; Naming from de®nition (oral input); Sentence completion; Grammaticality judgements. This battery of tests was also administered to a group of 22 control subjects matched for age and education to CAN.
436
C. Finocchiaro / Journal of Neurolinguistics 15 (2002) 433±446
3. Materials Stimuli were controlled for frequency of occurrence (Bortolini, Tagliavini & Zampolli, 1971; De Mauro, Mancini, Vedovelli & Voghera, 1993) and syntactic complexity. The average frequency of [1agentive] and [2agentive] verbs was comparable in all tasks but in the sentence completion, where [2agentive] verbs were signi®cantly more frequent than [1agentive] verbs (Picture comprehension and Picture naming, [1ag.]: 73.29, [2ag.]: 75.38, unpaired T-test, two-tails: T 20.08, P 0.9; Naming from de®nition, [1ag.]: 100.9, [2ag.]: 73.05, unpaired T-test, two-tails: T 0.9, P 0.3; Sentence completion, [1ag.]: 100.9, [2ag.]: 289.3, unpaired T-test, two-tails: T 23.5, P 0.0001; Grammaticality judgements, [1ag.]: 129.2, [2ag.]: 187.8, unpaired T-test, two-tails: T 20.75, P 0.4). The degree of syntactic complexity was de®ned on the basis of three main parameters: (i) possibility of transitive and intransitive constructions; (ii) classi®cation of intransitive verbs as unaccusative and/or unergative; 5 (iii) valence. In order to control for the effect of the concrete/abstract dimension, a group of 20 subjects rated the verbs using a scale from 1 (very abstract) to 5 (very concrete). 6 [1Agentive] and [2agentive] verbs yielded comparable ratings ([1agentive] verbs: 3.2; [2agentive] verbs: 2.9; P . 0.05). 4. Results 4.1. Picture comprehension Stimuli corresponded to 64 black-and-white drawings of concrete verbs (31 [1agentive], e.g. to read, to sing, 13 [2agentive], e.g. to be born, to fall, 20 [éagentive], e.g. to break, to enter). 7 Each drawing was presented four times, each time paired to a different word pronounced by the examiner (the correct match, a semantic foil: e.g. morire,`to die'Ðnascere,`to be born', a phonological foil: e.g. morire,`to die'Ðmurare, `to wall up', an unrelated foil: morire,`to die'Ðapplaudire,`to applaud'). The subject had to say whether or not word and picture matched. The 256 ( 64 £ 4) stimuli were randomly distributed in four sub-lists, so that a picture was presented only once in each sub-list. The same number of correct matches, and of semantically, phonologically and unrelated foils (N 16) was presented in each sub-list. Controls produced on average 92% (SD 1.17) correct responses. Their accuracy was unaffected by agentivity ([1agentive] verbs: 93.5%, SD 1.7; [2agentive] verbs: 92.3, SD 2.4; [Bagentive] verbs: 90%, SD 1.7). 5
The opposition, within intransitive verbs, between unaccusatives and unergatives is based on the assumption that the single (inner) argument of an unaccusative verb is an object at the deep structural level, while the single (external) argument of an unergative verb is a subject (Cennamo, 1998; Levin & Rappaport Hovav, 1995; Perlmutter, 1978; Sorace, 1995). For a semantic approach to the unaccusativity, see Van Valin (1990). 6 Breedin, Saffran and Schwartz (1998) showed that the concrete/abstract dimension might play an important role also on verb retrieval. 7 The label [Bagentive] ([zero agentive] in the ®gures) refers to verbs that can be predicated both of agents and non-agents.
C. Finocchiaro / Journal of Neurolinguistics 15 (2002) 433±446
437
Fig. 1. CAN's performance on [1agentive], [2agentive], and [zero agentive] verbs in the picture naming task.
CAN correctly comprehended 59/64 pictures (92.1%). His performance did not seem to be affected by the agentivity feature ([1agentive] verbs: 29/31, or 93.5% correct; [2agentive] verbs: 12/13, or 92.3% correct; [éagentive] verbs: 17/20, or 85% correct; P . 0.05). Errors consisted in rejecting correct verb-picture pairs (2/5), or in accepting semantically related ones (3/5). 4.2. Picture naming The same 64 pictures as in the previous task were used. In this and in the following task subjects were asked to respond by producing only one wordÐa verb. Semantically incorrect responses, nominalizations, neologisms, failures to respond were scored as errors. Phonetic and phonemic distortions of the target were ignored. The average accuracy of the control group was 97.1% (SD 1.22), with no signi®cant difference among [1agentive] (97.7%; SD 2.4), [2agentive] (96.1%; SD 5.4) and [Bagentive] verbs (97.5%; SD 3.9). CAN correctly named 55/64 pictures (85.9%), with a performance below the normal standard. The subject made far more errors on [2agentive] (6/13, or 46% correct) than on [1agentive] verbs (30/31, or 96.7% correct). The dissociation was signi®cant: x 2 12.56, P 0.0004 (see Fig. 1). CAN performed very well on [Bagentive] verbs (19/20, or 95% correct), namely, on verbs that can be predicated both of agents and non-agents (e.g. to destroy, to enter). Errors consisted of semantic paraphasias (7/10) and nominalizations (2/10). 4.3. Naming from de®nition (oral input) This task included 85 verbs. In addition to the 64 verbs used in the previous task, 21 abstract and non-picturable verbs were introduced. For each verb, a de®nition was read aloud by the examiner. De®nitions were short (no longer than a line), simple (e.g. sposarsi, `to marry': diventare marito e moglie,`to become husband and wife'), and contained concrete and very familiar terms, used in everyday life. The same scoring criteria used for the picture naming task were adopted. The average accuracy of the control group was
438
C. Finocchiaro / Journal of Neurolinguistics 15 (2002) 433±446
Fig. 2. CAN's performance on [1agentive], [2agentive], and [zero agentive] verbs in the naming from de®nition task.
97.9% (SD 2.14), with no signi®cant difference among [1agentive] (97.9%; SD 3.3), [2agentive] (97.7%; SD 2) and [Bagentive] verbs (98.2%; SD 3.5). CAN correctly produced 56/85 correct responses (65.8%). The dissociation between [1agentive] and [2agentive] verbs was con®rmed (x 2 6.14, P 0.01): thus, [2agentive] verbs (12/26, or 46%) were signi®cantly more impaired than [1agentive] verbs (28/35, or 80%). As to [éagentive] verbs, CAN's accuracy level fell half-way between [1agentive] and [2agentive] verbs (16/24 [Bagentive] verbs, or 66.6% correct, see Fig. 2). Errors consisted of semantic paraphasias (18/29) and nominalizations (7/29) and failures to respond (4/29). 4.4. Sentence completion Simple and short sentences (157) lacking the main verb were presented. Subjects were asked to provide the missing verb, that could be easily inferred from the context. The gap occurred in one of the initial positions in the sentence, and for most sentences the context unambiguously induced perfective morphology (simple or compound past; e.g. Claudia [ha letto] il libro in un solo pomeriggio, `Claudia [read] the book in only one afternoon'). With those few verbs (denoting states) incompatible with perfective morphology, the context suggested the present tense (e.g. Il ®ume Po [scorre] lungo la Pianura Padana, `The Po river [¯ows] along the Padana plain'). Eighty-nine verbs were used as targets. Some verbs (N 27) were used only once. The remaining 62 verbs were used in two (N 56) or three (N 6) different contexts. Contexts were [1agentive] (N 104) or [2agentive] (N 53). The distribution of [1agentive] and [2agentive] verbs across the three sets (verbs used once, verbs used two times, verbs used three times) was roughly the same. Sentences were randomly distributed in two sub-lists. The two sub-lists were administered on different days. The same criteria as in the naming tasks were adopted in the evaluation of the responses. The average accuracy of the controls was 97.8% (SD 2.66),
C. Finocchiaro / Journal of Neurolinguistics 15 (2002) 433±446
439
Fig. 3. CAN's performance on [1agentive] and [2agentive] contexts in the gap-®lling task.
with no signi®cant difference between [1agentive] (98.3%; SD 1.5) and [2agentive] contexts (97.5%; SD 3.1). CAN provided the correct verb to 136/157 sentences (86.6%), with a performance slightly below the normal range. CAN correctly provided the lacking verb to 99/104 [1agentive] contexts (95.1%) and to 37/53 [2agentive] contexts (69.8%): x 2 17.39, P 0.00003 (Fig. 3, Table 1). Errors consisted of semantic paraphasias (14/21) and failures to respond (7/21). 4.5. Grammaticality judgements Both verbs (24) and nouns (24) were used as stimuli. Each stimulus was inserted in two different contexts, only one of which was correct. For each sentence, subjects were requested to provide a grammaticality judgement, answering to questions like: `Is this sentence grammatically correct'? The sentences were orally presented by the examiner. Noun-contexts were count (N 24) or mass (N 24). Verb-contexts were [1agentive] (N 34) or [2agentive] (N 14). In order to construct incorrect sentences, count-nouns were used with mass-quanti®ers (e.g. *Il mio vicino mi ha af®dato un po' di cane `*My neighbour entrusted a little of dog to me') and mass-nouns were used with count-quanti®ers (e.g. *Giulio e io abbiamo comprato i latti al supermercato `*Giulio and I bought milks at the store'). As to verb-contexts, the compatibility between each verb with temporal adverbials was Table 1 Percentage of CAN's correct responses for [1agentive], [2agentive] and [Bagentive] verbs in the production tests
[1agentive] verbs [2agentive] verbs [Bagentive] verbs
Picture naming
Naming from de®nition
Gap-®lling
30/31 (96.7%) 6/13 (46%) 19/20 (95%)
28/35 (80%) 12/26 (46%) 16/24 (66.6%)
99/104 (95.1%) 37/53 (69.8%)
440
C. Finocchiaro / Journal of Neurolinguistics 15 (2002) 433±446
Fig. 4. CAN's performance on [1agentive] and [2agentive] contexts in the grammaticality judgement task.
manipulated (e.g. Gli operai hanno lavorato sodo per due giorni `Workers worked hard for 2 days' but *Gli operai hanno lavorato sodo in due giorni `*Workers worked hard in 2 days'). Perfective morphology (namely, the compound past) was used, in both correct and incorrect sentences. The sentences were divided into two sub-lists, in such a way that: (a) a given stimulus appeared only once in each sub-list; (b) each sub-list contained an equal number of correct and incorrect sentences. Each sub-list contained 48 sentences. Sub-lists were submitted on different days. The average accuracy of the control subjects was 96.5% (SD 3.50). Their performance signi®cantly varied with respect to: 1. word class: nouns (98.9%; SD 1.32) vs. verbs (94.2%; SD 6.07, paired T-test: T 3.4, P 0.0039); 2. count-nouns (100%) vs. mass-nouns (97.9%; SD 2.65, paired T-test: T 3.16, P 0.006); 3. [1agentive] (91.7%; SD 3.1) vs. [2agentive] contexts (96.8%; SD 6.3, paired T-test: T 22.63, P 0.02). CAN correctly evaluated 86/96 sentences (89.5%), with a performance within the normal range. The subject did not seem to be sensitive to the word class: he correctly evaluated 41/48 verb-contexts (85.7%), and 45/48 noun-contexts (93.7%). No signi®cant difference emerged between count- (23/24 96%) and mass-nouns (22/24 91.6%). However, the subject's performance did not follow the control pattern as far as the agentivity feature is concerned. In contrast to normal people (signi®cantly better on [2agentive] as opposed to [1agentive] contexts), CAN was signi®cantly better on [1agentive] (32/34, or 94.1% correct) as opposed to [2agentive] contexts (9/14, or 64.2% correct; x 2 4.89, P 0.02; see Fig. 4). The errors consisted in accepting the incorrect contexts (9/10, or 90%), or in rejecting the correct ones (1/10, or 10%).
C. Finocchiaro / Journal of Neurolinguistics 15 (2002) 433±446
441
5. Discussion The main ®nding of the present study allows us to claim that the agentivity feature is a relevant parameter for the organization of verbs in the brain. Subject CAN showed a dissociation between [1agentive] (relatively preserved) and [2agentive] verbs (signi®cantly more impaired). This pattern was consistent across all the production tasks (with no difference between verbs in isolation and in context) and the grammaticality judgements. It is not obvious to me how different explanations could account for the effects observed. [1Agentive] and [2agentive] verbs were matched for frequency and concreteness in all tests except that in the sentence completion task, where [2agentive] verbs were even more frequent than [1agentive] verbs. If CAN were sensitive to frequency, his performance should have improved with [2agentive] verbs in this task. The subject's behaviour with [Bagentive] verbs is particularly intriguing. In the naming from de®nition task, as one would expect, the accuracy level with these verbs fell midway between the accuracy level with [1agentive] and [2agentive] verbs. However, in the picture naming task, CAN performed at ceiling level with [Bagentive] as well as with [1agentive] verbs. This apparent discrepancy can be accounted for by the different task demands. In the picture naming, since all [Bagentive] verbs were depicted as actions accomplished by agents, the subject might have caught only their [1agentive] interpretation. On the other hand, the naming from de®nition did not encourage such a strategy. The results of the grammaticality judgements are fully consistent with those emerging from the production tests. CAN followed an opposite pattern to that of the control subjects, showing a signi®cant dissociation between [1agentive] (relatively preserved) vs. [2agentive] contexts (signi®cantly more impaired). As to the picture comprehension task, no dissociation between [1agentive] and [2agentive] verbs emerged, probably because correct responses in this test can be based on a residual amount of semantic knowledge. Perhaps, an impairment to the semantic system that is of suf®cient magnitude as to undermine the retrieval of a speci®c entry for production, does not necessarily abolish semantic information. Such a de®cit may continue to allow suf®cient semantic activation as to provide a correct YES/NO answer in a word/picture matching task. For these reasons, some sort of disruption of semantic competence is not at all ruled out by the failure to demonstrate the [1agentive] vs. [2agentive] dissociation in the picture comprehension task. As said in the introduction, one of the most relevant differences between [1agentive] and [2agentive] verbs consists in their argumental structure. More speci®cally, grammatical subjects of [1agentive] verbs map onto the thematic role of agent, while grammatical subjects of [2agentive] verbs map onto thematic roles lower in the hierarchy. The ability in assigning thematic roles has been investigated in several neurolinguistic studies (see for instance Martin & Blossom-Stach, 1986). The case of EB (Caramazza & Miceli, 1991) is particularly interesting, as this subject showed a striking dissociation between impaired ability to assign thematic roles in both comprehension and production in the context of a spared ability to process morphosyntactic properties of sentences in both comprehension and production. However, the hypothesis of a selective dif®culty in assigning thematic roles to the verb
442
C. Finocchiaro / Journal of Neurolinguistics 15 (2002) 433±446
arguments is still too loose to account for CAN's performance, as it would fail to explain the selective impairment for [2agentive] verbs as opposed to [1agentive] verbs. This hypothesis would in fact predict great dif®culties with those sentences where the grammatical subject and the grammatical object of the verb do not differ in animacy. On the other hand, CAN's performance should substancially improve both when the grammatical subject and the grammtical object of the verb differ in animacy (in this case, the subject is supposed to take advantage of the difference in animacy to assign thematic roles), and when the verb has only one argument. This pattern should be observed irrespective of the agentivity value of the sentences. As it happens, CAN's performance is better on [1agentive] than on [2agentive] verbs irrespective of the presence of one or two arguments, and of the same vs. different degree of animacy of the grammatical subject and the grammatical object. Consider for instance the sentence completion task. In the case of [1agentive] contexts, CAN showed a comparable performance with sentences like Il tuo amico [ha bevuto] tre lattine di birra e si e' ubriacato, `Your friend [drank] three cans of beer and he got drunk' (one argument), and with sentences like Per giorni [ho chiamato] Giulia al telefono, lett. `For days, I [called] Julia on the phone' (two arguments, both animate). On the other hand, when he is faced with [2agentive] contexts, he has trouble with verbs having two arguments not differing in animacy (e.g. La cassa [contiene] antichi gioielli, `The box [contains] old pieces of jewellery) as well as with verbs having only one argument (e.g. Il ®ume Po [scorre] lungo la pianura Padana, `The Po river [¯ows] along the Padana plain'). Moreover, the hypothesis of a general dif®culty in assigning thematic roles would not predict differences in performance when verbs are presented in isolation. CAN still showed a selective impairment for [2agentive] verbs even in the two naming tasks. In this connection, CAN's performance on semantically reversible sentences should be emphasized. CAN performed ¯awlessly in comprehension and production of the semantically reversible sentences included in the ENPA. This result de®nitively rules out the hypotheses of impaired mapping procedures and dif®culties in assigning the verb its arguments as possible sources of the de®cit. Note that reversible sentences are typically agentive and involve two animate participants, both able in principle to perform the action denoted by the verb. The ¯awless performance of CAN on this type of sentences is perfectly in line with his spared abilities on [1agentive] verbs and [1agentive] sentences. One could argue then, that the difference in the type of thematic roles [1agentive] and [2agentive] verbs can take is responsible for CAN's pattern of performance. Thus, CAN's performance would be impaired only when the subject of the verb maps onto a thematic role different from that of agent. Another possibility is that CAN suffers from a very mild de®cit restricted to the grammatical class of verbs. His abilities would be suf®cient to allow the retrieval of the most prototypical items of the category (namely, verbs denoting actions performed by agents). However, the subject's performance on the peripheral items of the category (non-action or [2agentive] verbs) would be signi®cantly worse. According to this account, CAN would perform better on [1agentive] verbs simply because they are easier than [2agentive] verbs. This is unlikely, however. Finocchiaro and Miceli (in preparation) described a patient,
C. Finocchiaro / Journal of Neurolinguistics 15 (2002) 433±446
443
GSC, who showed a selective preservation of state verbs as opposed to other type of verbs. As it happens, state verbs are intrinsically [2agentive] (by de®nition, a state cannot be an action). This fact cannot de®nitely rule out the hypothesis of a greater dif®culty of [2agentive] as opposed to [1agentive] verbs. However, it strongly suggests that different dimensions may be at play so that the greater dif®culty argument can hardly explain all the effects observed. Thus, it does not seem unreasonable to believe that CAN suffers from a semantic de®cit restricted to [2agentive] verbs and [2agentive] sentences. His de®cit occurs in the context of spared abilities in production and comprehension of semantically reversible sentences and cannot be ascribed to a general impairment of the mapping procedure of grammatical functions onto thematic roles and vice versa. It is hard to say whether CAN's de®cit involves thematic roles different from that of agent, or the verb [2agentive] feature per se. The ®rst hypothesis, in order to account for the results, must be augmented by the assumption that CAN is sensitive to the type of thematic role of the grammatical subject, but he is insensitive to the type of thematic role of the grammatical object. Without this assumption, the ®rst hypothesis would erroneously predict an impaired performance also for transitive [1agentive] verbs (their grammatical objects can be animate but they cannot be agents). However, by assigning a predominant importance to the grammatical subject, the ®rst hypothesis becomes empirically indistinguishable from the second hypothesis. They are, de facto, the same. An additional question of interest concerns the relevance of the opposition [1living] vs. [2living] in the verbal domain. As said in the introduction, there are serious linguistic reasons to believe that verbs are not classi®ed with respect to the [^living] feature. Some neurolinguistic evidence seems to converge on this point. Moss, De Mornay Davies, Jeppeson, McLellan and Tyler (1998) described a subject, RC, whose de®cit for [1living] nouns was not mirrored in his verb knowledge. RC, even being signi®cantly better on [2living] as opposed to [1living] nouns, performed equally on [1living] and [2living] verbs. Two hypotheses are compatible with the data: (1) the [^living] feature is not cognitively relevant for the verbal domain; (2) the [^living] feature, although relevant for the verbal domain, may in¯uence the subjects' performance on nouns but not on verbs, or may have the reverse effect. Thus, sensitivity to the [^living] feature could be limited to one syntactic category (nouns vs. verbs). Hypothesis (2) would enforce the hardly plausible conclusion that two independent modules (one for nouns, one for verbs) are responsible for the same semantic information (i.e. of the [^living] feature). There are no theoretical or empirical reasons (in lack of positive evidence) in support of hypothesis (2). On the other hand, (1) provides a straightforward explanation of the results obtained, consistent with linguistic considerations. As to the correspondence between the [^living] feature in the nominal and the [^agentive] feature in the verbal domain, CAN may provide some evidence. This subject, submitted to a picture naming test, performed equally well on nouns (29/30) and on verbs paired to them (28/30). Most important, his performance was unaffected by the animacy value of both nouns and verbs. Thus, he correctly named 7/7 [1living] nouns (e.g. dog), 22/23 [2living] nouns (e.g. gun), 6/7 [1living] verbs (e.g. to bark), 22/23 [2living] verbs (e.g. to shoot).
444
C. Finocchiaro / Journal of Neurolinguistics 15 (2002) 433±446
Obviously, these data are not so striking as to allow de®nite hypotheses. However, they suggest that the [^living] and the [^agentive] features are governed by distinct neural mechanisms. That is to say, a subject may be sensitive to the [^living] feature of nouns, but not to the [^agentive] feature of verbs. For instance, s/he may perform signi®cantly better on [1living] as opposed to [2living] nouns, failing, at the same time, to perform signi®cantly better on [1agentive] as opposed to [2agentive] verbs. If these speculations are on the right track, it should be possible to ®nd even double dissociations in a single subject (for example, a signi®cantly better performance on [1living] with respect to [2living] nouns, and on [2agentive] with respect to [1agentive] verbs). This appears as a plausible hypothesis, in so far as animacy and agentivity are related but different concepts; furthermore, as several recent studies have shown, verbs and nouns may be selectively impaired (Caramazza & Hillis, 1990; De Renzi & Di Pellegrino, 1995; Hillis & Caramazza, 1991; Miceli, Silveri, Villa & Caramazza, 1984; Zingeser & Berndt, 1988). Thus, even the semantic principles according to which they are classi®ed, could be, to a large extent, separately represented in the brain. 6. Conclusions The main results discussed here provide substantial evidence to the claim that the [^agentive] feature is a relevant parameter for the classi®cation of verbs in the brain. Moreover, it seems that: 1. the [^living] feature, although relevant for nouns, does not seem to play any role for verb classi®cation; 2. the [^living] and the [^agentive] features are governed by independent neural mechanisms. By hypothesis, apart from generic semantic concepts, such as the concrete/abstract dimension, verbs and nouns appear to be organized according to quite different parameters. Hopefully, further research will shed light on verb semantics and on the relationship between animacy and agentivity. Acknowledgements This work is part of my graduation thesis, defended in Pisa on July 1999. I am sincerely grateful to Pier Marco Bertinetto and Gabriele Miceli for their constant support from the beginning of this study. I also thank Alfonso Caramazza who made helpful comments and suggestions to a previous version of this manuscript. References Bever, T. G. (1970). The cognitive basis for linguistic structures. In J. R. Hayes (Ed.), Cognition and the development of language. New York: John Wiley & Sons.
C. Finocchiaro / Journal of Neurolinguistics 15 (2002) 433±446
445
Bortolini, U., Tagliavini, C., & Zampolli, A. (1971). Lessico di frequenza della lingua italiana contemporanea, Milano: IBM Italia. Breeedin, S. D., Saffran, E. M., & Schwartz, M. F. (1988). Semantic factors in verb retrieval: an effect of complexity. Brain and Language, 63, 1±31. Capasso, R. & Miceli, G. (2001) Esame Neuropsicologico per Pazienti Afasici (ENPA), Milano: Springer-Verlag. (in press). Caplan, D., Baker, C., & Dehaut, F. (1985). Syntactic determinants of sentence comprehension in aphasia. Cognition, 21, 117±75. Caramazza, A., & Berndt, R. S. (1985). A multicomponent de®cit view of agrammatic Broca's aphasia. In M. L. Kean (Ed.), Agrammatism, London: Academic Press. Caramazza, A., & Hillis, A. E. (1990). Lexical organisation of nouns and verbs in the brain. Nature, 349, 788±90. Caramazza, A., & Miceli, G. (1991). Selective impairment of thematic role assignment in sentence processing. Brain and Language, 41, 402±36. Caramazza, A., & Shelton, J. R. (1998). Domain-speci®c knowledge systems in the brain: the Animate-Inanimate distinction. Journal of Cognitive Neuroscience, 10, 1±34. Cennamo, M. (1998) TransitivitaÁ e inaccusativitaÁ in testi antichi abruzzesi e napoletani. In P. Ramat and E. Romani (Eds.), SLI, Sintassi storica. Atti del xxx congresso internazionale della societaÁ di linguistica italiana. Roma: Bulzoni. Comrie, B. (1981). Language universals and linguistic typology, Oxford: Blackwell. Damasio, H., Grabowski, T. G., Tranel, D., Hichwa, R. D., & Damasio, A. R. (1996). A neural basis for lexical retrieval. Nature, 380, 499±505. De Mauro, T., Mancini, F., Vedovelli, M., & Voghera, M. (1993). Lessico di frequenza dell'italiano parlato (LIP), Milano: Etas s.r.l. De Renzi, E., & Di Pellegrino, G. (1995). Sparing of verbs and preserved but ineffectual reading in a patient with impaired word production. Cortex, 31, 619±36. De Renzi, E., & Lucchelli, F. (1994). Are semantic systems separately represented in the brain? The case of living category impairment. Cortex, 30, 3±25. Finocchiaro, C. & Miceli, G. (2002) Verb actionality in aphasia (in preparation). Hillis, A. E., & Caramazza, A. (1991). Category-speci®c naming and comprehension impairment: a double dissociation. Brain, 114, 2081±94. Hillis, A. E., & Caramazza, A. (1995). Representation of grammatical categories of words in the brain. Journal of Cognitive Neuroscience, 7, 396±407. Hoepelman, J. (1986). Action, comparison and change. A study in the semantics of verbs and adjectives, TuÈbingen: Gunter Narr Verlag. Hoepelman, J. (1981). Verb classi®cation and the Russian verbal aspect, TuÈbingen: Gunter Narr Verlag. Jones, E. V. (1984). Word order processing in aphasia: effect of verb semantics. In F. C. Rose (Ed.), Advances in neurology: Vol. 42. Progress in aphasiology, New York: Raven Press. Levin, B., & Rappaport Hovav, M. (1995). Unaccusativity, Cambridge, MA: The MIT Press. Martin, R. C., & Blossom-Stach, C. (1986). Evidence of syntactic de®cits in a ¯uent aphasic. Brain and Language, 28, 196±234. Miceli, G., Silveri, M. C., Villa, G., & Caramazza, A. (1984). On the basis for the agrammatic's dif®culty in producing main verbs. Cortex, 20, 207±20. Moss, H. E., De Mornay Davies, P., Jeppeson, C., McLellan, S., & Tyler, L. K. (1998). The relationship between knowledge of nouns and verbs in a category-speci®c de®cit for living things. Brain and Language, 65, 92±95. Perlmutter, D. (1978). Impersonal passives and the Unaccusative hypothesis. Proceedings of the Berkeley Linguistics Society, 4, 157±89. Saffran, E. M., Schwartz, M. F., & Marin, O. S. M. (1980). Evidence from aphasia: isolating the components of a production model. In B. Butterworth, Language production, New York: Academic Press. Saffran, E. M., Schwartz, M. F., & Marin, O. S. M. (1980). The word order problem in agrammatism. II. Production. Brain and Language, 10, 263±80. Samson, D., Pillon, A., & De Wilde, V. (1998). Impaired knowledge of visual and non-visual attributes in a patient with a semantic impairment for living entities: a case of a true category-speci®c de®cit. Neurocase, 4, 273±90.
446
C. Finocchiaro / Journal of Neurolinguistics 15 (2002) 433±446
Sartori, G., & Job, R. (1988). The oyster with four legs: a neuropsychological study on the interaction of visual and semantic information. Cognitive Neuropsychology, 5, 105±32. Schwartz, M. F., Saffran, E. M., & Marin, O. S. M. (1980). The word order problem in agrammatism. I. Comprehension. Brain and Language, 10, 249±62. Silveri, M. C., & Gainotti, G. (1988). Interaction between vision and language in category-speci®c impairment. Cognitive Neuropsychology, 5, 677±709. Sorace, A. (1995). Acquiring linking rules and argument structures in a second language: the unaccusative/ unergative distinction. In Eubank, Selinker & Sharwood Smith, The current state of Interlanguage. Studies in honour of William E. Rutherford, Amsterdam/Philadelphia: John Benjamins. Van Valin, R. D. (1990). Semantic parameters of split intransitivity. Language, 66, 221±60. Warrington, E. K., & Shallice, T. (1984). Category speci®c semantic impairments. Brain, 107, 829±54. Zingeser, L. B., & Berndt, R. S. (1988). Grammatical class and context. Effects in a case of pure anomia: implication for models of language production. Cognitive Neuropsychology, 5, 473±516.