Syntactic Accuracy in Sentence Production: The Case ... - Springer Link

6 downloads 89 Views 2MB Size Report
Wisconsin, 1202 W. Johnson St., Madison Wisconsin 53703. 623 ... culine or feminine, such as: fiurne/fiumi (river/rivers-M); nave/navi ..... The (fem.) eagle (fem.) ...
Journal of Psycholinguistic Research, Vol. 28, No. 6, 1999

Syntactic Accuracy in Sentence Production: The Case of Gender Disagreement in Italian Language-Impaired and Unimpaired Speakers Gabriella Vigliocco1,3 and Tiziana Zilli2 We report three experiments with language-impaired and unimpaired speakers of Italian, assessing: (1) whether nonsyntactic (both conceptual and morphophonological) information is used in encoding the syntactic structure of a sentence; and (2) whether the integration of syntactic and nonsyntactic information can be differentially impaired in Broca's aphasics. In all the experiments, gender agreement errors between a noun, subject of the sentence, and a predicative adjective were induced by presenting participants with sentence fragments to complete. The first experiment assessed the role of conceptual information. The second experiment investigated whether agreement is disrupted by the presence of another noun with different gender in the subject noun phrase. In the last experiment, we assessed whether morphophonological cues are used. We found that both populations used nonsyntactic information (both conceptual and morphophonological). However, patients were disrupted to a greater extent than normals by the presence of a gender mismatching noun in the subject noun phrase. The results are discussed in terms of how information integration during production is achieved and how it can be disrupted in aphasia.

The research reported here was supported by NSF grant (SBR 9729118) to the first author and by a NATO Collaborative Grant (CRG 941138) to Brian Butterworth, Bernard Comrie, and Carlo Seraenza. Portions of the study reported here were presented to fulfill a senior thesis requirement by Tiziana Zilli and were also presented to the 38th meeting of the Psychonomic Society (1997). A number of ideas reported in the paper come from conversations with Rob Hartsuiker. We would also like to thank Dave Vinson and Carlo Semenza for their helpful comments on previous versions of the manuscript; patients C. G. and C. D. for their enormous patience and willingness to complete this never ending study; and the staff of the Laboratorio di Neuropsicoogia, Ospedale Civile di Brescia directed by Prof. Stefano Cappa, in particular Marina Frugoni and Patrizia Pasquali for their help in recruiting patients and for allowing Tiziana Zilli to use space and equipment to conduct the experimental sessions. 1 University of Wisconsin-Madison. 2 University of Padova. 3 Address all correspondence to Gabriella Vigliocco, Department of Psychology, University of Wisconsin, 1202 W. Johnson St., Madison Wisconsin 53703.

623 0090-6905/99/1100-0623$16.00/0 © 1999 Plenum Publishing Corporation

624

Vigliocco and Zilli

To fulfill its communicative goal, the sentence production system must succeed at multiple levels. These include meaning, syntax, morphology, phonology, and motor commands. There are ample opportunities for sentences to be ill formed at any of these levels. However, as estimated by Bock (1991), errors in production do not exceed 1 in 1,000 words. This level of accuracy is remarkable for a system that produces two to three words per second (Levelt, 1989). In this paper we will focus on accuracy in realizing a syntactic representation for a sentence; more specifically, on accuracy in realizing agreement between two elements in a sentence. The questions we address are: first, whether nonsyntactic information, both conceptual and morphophonological, is used in production to ensure accuracy during grammatical encoding (i.e., those processes responsible for building syntactic representations for to-beuttered sentences). Second, whether the integration of syntactic information across constituents and the integration of nonsyntactic and syntactic information during encoding can be differentially disrupted in Broca's aphasia. Agreement relations between sentential elements are particularly well suited for addressing these questions. First, agreement is a syntactic operation that reflects conceptual features (e.g., the sex of a referent). Second, it is a syntactic relation that is realized in languages through morphophonological markers. Furthermore, questions about the use of agreement are directly relevant for many linguistic theories (e.g., Bresnan, 1982; Pollard & Sag, 1994) because the mechanisms underlying agreement are also used for other constraints (e.g., to determine the permissibility and placement of arguments). They are also relevant for many computational and psycholinguistic theories of both parsing and production (McDonald, Pearlmutter, & Seidenberg, 1994; Kempen & Vosse, 1989; Kempen, 1997; Vigliocco, Butterworth, & Garrett, 1996). Gender agreement between a noun, subject of a sentence, and a predicative adjective in Romance languages seems to be a perfect example. In Italian, nouns are always marked for gender (either masculine or feminine). For most nouns referring to humans, the gender marking is determined by the sex of the referent. The reference to male or female entities is achieved by changing the word ending; for example, ragazzo (boy) and ragazza (girl). Nouns of this type are referred to as having conceptual gender. For nouns referring to objects and abstract entities, gender is largely arbitrary (so that, for example, pietra (rock) is feminine while sasso(stone) is masculine.) Nouns in this category are referred to as having grammatical gender. For a large number of animal names, a single form (either feminine or masculine) is used to refer to both sexes [e.g., scimmia (monkey-F); gufo (owl-M)]. A few nouns referring to humans also belong to this latter class [e.g., vittima (victim-F)].

Gender Disagreement in Italian

625

Most of the time, the gender of a noun can be inferred from the word ending: masculine nouns end in -o and -i (for the singular and plural, respectively), while feminine nouns end in -a and -e. These nouns are referred here as morphophonologically marked. However, there are exceptions. Relevant here are nouns that end in -e, and -i that can be either masculine or feminine, such as: fiurne/fiumi (river/rivers-M); nave/navi (ship/ships-F). These nouns will be referred to as morphophonologically ambiguous. Agreement with a predicate containing an adjective follows the same rule regardless of whether the noun has conceptual or grammatical gender and whether it is morphophonologically marked or ambiguous; if the noun is masculine, then the adjective must be masculine; if the noun is feminine, then the adjective must be feminine (Vincent, 1990). Therefore Italian provides us with an excellent vehicle to assess whether conceptual and morphophonological information is used in agreement computation. Vigliocco and Franck (1999) have recently used gender agreement errors in Italian and French to assess whether information regarding the sex of a referent is used in grammatical encoding during production. They tested the predictions deriving from two different views concerning the role of nonsyntactic information, which they labeled the minimalist and the maximalist views. In the minimalist view, accuracy in syntactic operations is ensured by keeping nonsyntactic information outside the domain of grammatical encoding (thereby protecting the encoder from the possibility of interference). In the maximalist view, accuracy is ensured by using the additional nonsyntactic information. The availability to the encoder of this information would protect against the risks of information loss. Using a constrained version of a sentence completion task (Bock & Miller, 1991; Vigliocco, Butterworth, & Semenza, 1995), Vigliocco and Franck (1999) presented speakers of Italian and French with sentence fragments such as: (1) La ragazza nel parco [The (fern.) girl in the (masc.) park (masc.)] (2) La panchina nel parco [The (fem.) bench (fem.) in the (masc.) park] Participants were required to complete the fragments with an adjective (that was provided). The subject noun in (1) has conceptual gender; in (2) it has grammatical gender. It is assumed that nouns with conceptual gender receive their gender specification from conceptual structures (in a manner analogous to number specification), while gender specification for nouns with grammatical gender is a lexical property (Levelt, 1989; Vigliocco, Antonini, & Garrett, 1997). If the grammatical encoder uses only the minimal information from conceptual structures, then nouns with conceptual and

626

Vigliocco and Zilli

grammatical gender should induce similar numbers of errors. For nouns with conceptual gender, the conceptual information regarding the sex of the referent would be used to establish the gender of the subject noun but agreement is then computed solely on the basis of this syntactic information. For nouns with grammatical gender, only the syntactic gender feature can be used for agreement computation. Therefore, while (1) and (2) differ in terms of how the subject noun is specified for gender (conceptually vs. lexically), they do not differ insofar as agreement is concerned, hence no difference in error rates is predicted.4 In contrast, if the grammatical encoder uses maximal information from conceptual structures, fewer errors for nouns with conceptual gender than for nouns with grammatical gender should be observed. For nouns with conceptual gender, there is redundant congruent information at the conceptual and syntactic level. If conceptual information is allowed to influence grammatical encoding, being redundant and congruent, it should help correct agreement computation. For nouns with grammatical gender, instead, the only available information is the syntactic specification of the noun. If, during the encoding, the gender information is lost, the likelihood of an error should be higher when the only source of information is lexical, as in (2), than when there is also supporting conceptual information, as in (1). Vigliocco and Franck (1999) found that errors were more common for fragments in which the subject noun had grammatical gender than for fragments in which the subject noun had conceptual gender, as predicted by the maximalist view. The first objective of this paper is to further assess the maximalist view. We extend the investigation of conceptual effects on agreement with Broca's aphasics and we assess whether the grammatical encoder also uses information concerning morphophonological regularities of gender marking. Consider the two examples below: (3) La barca nel porto [The (fern.) boat (fem.) in the (masc.) harbor (masc.)] (4) La nave nel porto [The (fern.) ship ?? in the (masc.) harbor (masc.)] If morphophonlogical marking is taken into account during grammatical encoding, we predict that errors should be more common for (4) than for (3).5 4

5

See Vigliocco and Franck (1999) for a more detailed description of the minimal and maximal input hypothesis. Note that an effect of morphophonological regularity could come about in different ways: either as feedback in the system from phonological encoding to grammatical encoding (e.g., Dell, 1988) or as a result of monitoring (e.g., Levelt, 1989). Regardless of how such an effect could occur, it would indicate that morphophonological information contributes to accuracy at the level of syntactic structures.

Gender Disagreement in Italian

627

The integration of nonsyntactic information with syntactic information is not the only integration process. The integration of syntactic information across constituents is clearly also a fundamental process for grammatical encoding in general and agreement in particular. Either or both of these integration processes can go wrong in Broca aphasia. The second objective of this study is to investigate these two integration processes in Broca aphasia. In the comprehension domain, a number of proposals have been put forward. We contrast here two broad views: one in which the integration of nonsyntactic information would be most problematic and one in which the integration of syntactic information across constituents would be most problematic. The first hypothesis is the mapping hypothesis (Linebarger, Schwartz, & Saffran, 1983; Schwartz, Linebarger, Saffran, & Pate, 1987). It assumes that patients' difficulties with syntactic processing are related to a deficit in mapping syntactic information onto conceptual information. According to this account, integrating nonsyntactic conceptual information during encoding requires processing resources that can be reduced by brain damage. The second view implies that patients may suffer limitation of resources devoted to the building of syntactic structures. Under this category, we consider a number of quite different approaches to resource limitations, all of which have in common the idea that processing limitations can affect the computation of a syntactic structure. Different proposals differ in terms of whether they assume a single resource pool that would be affected by brain damage [e.g., Carpenter, Miyake, & Just (1994) vs. Caplan & Waters, (1999) and Martin (1995)]; whether the processing limitation can be conceived in terms of limited capacity or limited efficiency [e.g., Martin & Romani (1994) vs. Kolk (1995)]; whether or not processing limitations can arise in different manners depending on the task demands [e.g., Caplan and Waters (1999), Friederici (1995), and Martin (1995) vs. Kolk and Weijts (1996) and Linebarger et al. (1983)].6 In the production domain, and for the task at hand, the mapping hypothesis would predict that patients with Broca's aphasia do not use addi6

Note that here we are considering two different approaches in the same category. The hypothesis that patients are selectively impaired in the processes responsible for building syntactic structures (Caplan & Waters, 1999, Friederici, 1995; Kolk, 1995; Martin & Romani, 1994), and the hypothesis that patients can allocate a reduced pool of resources to different integration processes. In this latter view, resources can be allocated to the building of a syntactic structure or to the use of nonsyntactic information; patients would experience a tradeoff between allocating resources to syntactic and nonsyntactic information (Hartsuiker & Kolk, 1999; Kolk & Weijts, 1996; Linebarger et al., 1983). The experiments we describe below do not allow us to tease apart the hypothesis that patients suffer from a reduction of resources specifically dedicated to syntactic processing and the hypothesis that the patients experience a trade-off.

628

Vigliocco and Zilli

tional conceptual information. More generally, while their ability to integrate syntactic information across constituents may be within normal limits, they should behave differently than normals in integrating nonsyntactic with syntactic information. The alternative hypothesis predicts that while patients use nonsyntactic information as normals do, they should be impaired in their ability to integrate syntactic information across constituents. To summarize, the present paper attempts to answer two main questions. First, whether nonsyntactic information, either conceptual or morphophonological, is used in production to ensure accuracy during grammatical encoding, as predicted by the maximalist view. Second, whether the integration of syntactic information across constituents and the integration of nonsyntactic with syntactic information during encoding can be differentially disrupted in Broca's aphasia. STUDY PLAN We report a series of three experiments conducted with two Broca's aphasic speakers (patients C. G. and C. D.), and language-unimpaired speakers of Italian. All the experiments use the same methodology, consisting of presenting speakers with sentence fragments (preambles), and adjectives. The speakers' task is to use the adjective to complete the preamble. The dependent measure of interest is the numbers of errors in gender agreement between the subject head noun and the predicative adjective.7 In experiment 1, we assess whether the patients use conceptual information in agreement computation in the same manner as normal speakers. In experiment 2, we assess whether patients are more disrupted than normal speakers are by a "syntactic complexity" manipulation (i.e., the presence of 7

The two patients described in this study exhibit a characteristic agrammatic spontaneous speech. Therefore a question that needs to be addressed before we introduce the experiments is whether we expect to observe errors [which could be classified as paragrammatisms (Butterworth & Howard, 1987)] or whether we expect to observe omission of closed class morphemes (both free and bound). Following Kolk (1995), we expect that patients will tend to produce more paragrarnmatic responses than agrammatic omissions in a constrained task like the one employed in this study (Caplan & Hanna, 1998; Haarmann & Kolk, 1994; Kolk & Heeschen, 1992). In other words, we expect both normals and patients to produce agreement errors in the task at hand (Butterworth & Howard, 1987; Kolk, 1995). Whether errors produced by the patients are distributed across different experimental conditions than errors produced by normal subjects is informative with respect to the hypotheses discussed above. Evidence that Broca's aphasics tend to produce agreement errors (instead of omitting the closed class morphemes) in the sentence completion task can be found in Hartsuiker & Kolk (1999) and in Vigliocco, Butterworth, Semenza, and Fossella, (1994), who investigated Broca's aphasics' performance in computing subject-verb agreement in number.

Gender Disagreement in Italian

629

a mismatching modifier in the subject NP). Experiment 3 assesses whether patients and normals use morpho-phonological information in agreement computation. EXPERIMENT 1 The first experiment replicates experiment 1 reported in Vigliocco and Franck (1999). Normal and language-impaired speakers were presented with sentence fragments in which the subject head noun had either conceptual or grammatical gender. On the basis of the results reported by Vigliocco and Franck (1999) in the language-unimpaired population, we expect to find fewer gender agreement errors for nouns with conceptual than for nouns with grammatical gender. The performance of the aphasic patients will provide us with information concerning whether redundant conceptual information is used when processing resources are reduced. Method Participants. Two nonfluent aphasic individuals and two groups of 18 age- and education-matched nonaphasic individuals (control 1: mean age: 49; range: 41-60, mean education: 7.50, range: 5-13; control 2: mean age: 70, range:61-80; mean education: 6, range: 5-13). They were all native Italian speakers. The same participants took part in all three experiments, which were conducted throughout a year. Each experiment was separated from the other by a minimum of three months. The order in which the experiments were conducted, both with the patients and the normal speakers, is the same as reported here. Participants were debriefed about the nature of the experiments only upon completion of the study. The patients participating in the study were selected on the basis of the following criteria. They received a diagnosis of Broca aphasia, their neurological condition was stable, and their speech was free of articulatory problems. Patient C. G. C. G. is a 73-year-old male farmer, with 5 years of formal education. In March 1996, he suffered a stroke. A CT scan showed a hypodense area in the cortical-subcortical junction in the left frontal-central region. On admission, the neurological examination revealed right hemiplegia and severe aphasia. At the time of the present study, the patient presented mild hemiplegia and Broca's aphasia with comprehension and production disorders. In the Italian version of the Aachen Aphasia Test (AAT) (Luzzatti, Willmes, & de Bleser, 1991), C. G. received a score of 3 for spontaneous speech. He received a score of 24 in the token test (24/36). He could repeat a maximun of three words, and only simple sentences.

Vigliocco and Zilli

630

Table I reports his comprehension scores (errors) in different tests and the results of a grammaticality judgment task. Patient C. D. C. D. is a 45-year-old male hotelkeeper, with 8 years of formal education. In July 1995, he suffered a stroke and was admitted to the hospital in a coma state. The patient woke up three days later with rightside hemiplegia and severe aphasia. The CT scan revealed a large hypodense area in the region irrorated by the left medial cerebral artery, including the temporal-insular area, the omolateral basal ganglia, and extending to the mediofrontal gyrus. At the time of the experiments he showed right hemiplegia, more severe for the upper limb, and Broca's aphasia, with comprehension relatively spared but with agrammatic production. Patient C. D. received a score of 3 for spontaneous speech in the AAT. He received a score of 35 in the token test (35/36). He could repeat a maximum of six words and somewhat complex sentences. Grammaticality judgments and comprehension scores are reported in Table I. Table I. Grammaticality Judgments and Comprehension Scores of Patients C. G. and C. D.a

Grammaticality judgments Acoustical presentation Errors total Grammatical sentences Ungrammatical Sentences Visual presentation Errors total Grammatical Sentences Ungrammatical Sentences Comprehension Acoustical presentation Reversible sentences Active voice Passive voice Morphological errors Active voice Passive voice Semantic errors Active voice Passive voice Visual Presentation Reversible Sentences Morphological Semantic From Miceli, Laudanna, and Burani (1989). The numbers refer to errors in the different conditions.

a

C.G.

C.D.

11/48 0/24 11/24

2/48 0/24 2/24

10/24 8/12 2/12

6/24 5/12 1/12

11/20 4/10 7/10 6/20 2/10 4/10 8/20 4/10 4/10

2/20 1/10 1/10 0/20

7/15 5/15 3/15

1/15 0/15 1/15

1/20 0/10 1/10

Gender Disagreement in Italian

631

Table II. Examples of Experimental Sentence Preambles for Experiment 1 Conceptual gender Masculine head noun, feminine local noun Lo sposo in chiesa The (masc.) groom (masc.) in church (fern.)

Feminine head noun, masculine local noun La zingara nel pacheggio The (fem.) gypsy (fem.) in the (masc.) parking lot (masc.) L'inquilino della casa La bidella nell' atrio The tenant (masc.) of the (fem.) house (fem.) The (fem.) janitor (fem.) in the (masc.) hall (masc.) Grammatical gender Masculine head noun, feminine Local Noun II cero in chiesaa The (masc.) candle (masc.) in church (fem.) Lo sgabuzzino della casaa The (masc.) closet (masc.) of the (fem.) house (fem.) a

Feminine head noun, masculine local noun La macchina nel parcheggio The (fem.) car (fem.) in the (masc.) parking lot (masc.) La lampada nell' atrio The (fem.) lamp (fem.) in the (masc.) hall (masc.)

Note that in Italian, both the articles il and lo are marked as masculine (their use depends on the word onset of the noun).

Materials. Materials for this experiment were the same as used in Vigliocco and Franck (1999), experiment 1. Stimuli consisted of sentence preambles composed of a subject head noun and a local noun embedded in a prepositional noun phrase that modified the subject noun phrase, and adjectives that could plausibly be used to complete the preambles. The variables experimentally manipulated were (1) the gender type of the subject head noun (conceptual vs. grammatical) and (2) the gender of the subject head noun (masculine vs. feminine). In the experimental items, the local noun always mismatched the head noun in gender, leading to two combinations: masculine head noun with feminine local noun and feminine head noun with masculine local noun. All nouns used in the experimental materials were singular. All the local nouns had grammatical gender. Examples of experimental preambles are provided in Table II. For each preamble, the two forms of the adjective (masculine and feminine) were presented one above the other. The position (above or below) of each form of the adjective was counterbalanced in the two lists. All the nouns used in the experimental materials were transparently marked for gender.8 Two experimental lists were generated. Each list was composed of 64 experimental preambles and 64 fillers. There were 16 items per condition. 8

A full listing of the experimental items used in experiments 1-3 is available at the following URL: http://psych.wisc.edu/faculty/pages/gvigliocco/public98.html

632

Vigliocco and Zilli

The difference between the two lists was whether the masculine form of the adjective for each item was presented above or below the feminine form. Filler preambles also consisted of a noun phrase and a prepositional phrase. In 48 of the filler preambles, the gender of the head noun and of the local noun matched, while it mismatched in the remaining 16. Thirty-two of the filler preambles had a plural head and a plural local noun, and the remaining 32 had a plural head and a singular local noun. In 32 filler preambles the head noun referred to an animate entity, while it referred to an inanimate entity in the remaining 32 items. Procedure. Trials consisted of the visual presentation of the adjectives (masculine and feminine form) followed by the auditory presentation of the preamble. The participants' task was to repeat the preamble and complete it using the adjective. For the normal speakers, the two forms of the adjective, one above the other, were printed on a card and presented to the participants for approximately 2 s. As soon as the experimenter took away the card, the preamble was played using an analog tape player. For the patients, the card with the two forms of the adjective was left in view during the acoustical presentation of the preamble. Eight practice trials performed before the experimental trials ensured that the participants understood the instructions. The experimental sessions were tape-recorded using an analog recording system (Marantz 201) and then transcribed. Filler and experimental stimuli were presented in a pseudorandom order with the constraints that each experimental list started with at least four filler items, and no more than three experimental items were presented in succession. Patients received each list three times. Each presentation was separated from the other by at least three weeks. Normals were presented with only one list. Scoring Criteria. Produced utterances were transcribed and scored according to the following criteria: (1) Correct responses were scored when the participant correctly repeated the preamble and completed the sentence using the correctly inflected form of the adjective. For the patients, we also scored as correct response cases in which freestanding closed-class elements such as determiners and prepositions were omitted. (2) Agreement errors were scored when the completion met the criteria above but the adjective form failed to agree in gender with the subject of the sentence. (3) Miscellaneous responses were scored when the participant failed to repeat the preamble or the adjective (or parts of it), or when he/she failed to repeat some words in the preamble. If two different utterances were produced in succession, only the first was scored. Design and Data Analysis. Statistical tests were conducted on agreement errors. The design of the experiment included group (control 1 vs.

Gender Disagreement in Italian

633

control 2) as between subject and within item factors. Gender type of the subject noun (conceptual vs. grammatical) and gender of the subject noun (masculine vs. feminine) as within subjects and between items factors. Analyses of variance (with both subjects and items as random factors) were conducted on the nonaphasics data; analyses by items only were conducted separately on the patients' data. Paired t tests were used to compare the patients' performance with the controls (patient C. G.'s performance was compared with control 2 and patient C. D.'s performance was compared with control 1). All comparisons used proportion of errors as dependent variable. Results Table III reports the general distribution of responses in the different scoring categories. Figure 1 reports the percentage of errors in the different experimental conditions. For the normal speakers, errors were more common among the older speakers than the younger [85 vs. 36; significant main effect of group [F1 (1, 34) = 15.45, p < .0001; F2 (1, 60) = 34.53, p < .0001]; when the subject head noun had grammatical gender than when it had conceptual gender [93 vs. 28; significant main effect of gender type F 1(1, 34) = 49.3, p < .0001; F2 (1, 60) = 23.1, p < .0001], while no difference was found between masculine and feminine head nouns (63 vs. 58, respectively, Fs < 1). No other main effects or interactions were significant (all Fs < 1.5). A similar distribution of responses was found for patients C. G. and C. D. Errors were more common for subject head nouns with grammatical than conceptual gender (57 vs. 14 for patient C. G. and 43 vs. 4 for patient C. D.). No difference was found for nouns with masculine and feminine gender (35 vs. 36 for patient C. G. and 25 vs. 22 for patient C. D.). The effect of gender type was significant for both patient C. G. [F2 (1, 60) = 31.34, p < .0001] and patient C. D. [F2 (1, 60) = 22.09 p < .0001]. No other main effect or interaction was significant (all Fs < 1). Pairwise comparisons between the patients and the normal speakers revealed that patients Table III. Distribution of Responses in Different Scoring Categories: Experiment 1 Patient

Correct responses Agreement errors Miscellaneous responses

Controls

C.G.

C.D.

2137 (92.7%) 121 (5.2%) 46 (2.3%)

275 (71.6%) 71 (18.5%) 38 (9.9%)

300 (78.1%) 47 (12.2%) 37 (9.7%)

634

Vigliocco and Zilli

Fig. 1. Percent errors in the different experimental conditions for patients C. D. and C, G. and the controls (percents are computed as the number of errors in a condition divided the total number of observations in the condition).

made significantly more errors than normals [C. G. vs. control 2: t(63) = 4.25, p < .0001; C. D. vs. control 1: t(63) = 3.97, p < .0001]. Patients made more errors than normals when the subject noun had grammatical gender [C. G. vs. control 2: t(31) = 4.81, p < .0001; C. D. vs. control 1: t(31) = 4.12, p < .0001], When the subject noun had conceptual gender, however, patients' and normals' performance did not differ [C. G. vs. control 2: t(31) = 1.60, p = .12; C. D. vs. control 1: t(31) = -.86, p = .40]. Discussion We found that the gender type of the subject noun had an impact on performance for both language-impaired and unimpaired speakers. This result replicates what was previously reported by Vigliocco and Franck (1999) with college students. As predicted by the maximalist view, information concerning the sex of the referent is taken into account in computing agreement. Since conceptual information is congruent with the syntactic information, fewer errors are produced for nouns with conceptual gender than for nouns with grammatical gender. In the latter case, there is no conceptual support, and the only relevant information for agreement computation is the lexical specification of gender of the noun. Patients C. G. and C. D. made a substantially higher proportion of errors than normal speakers did; crucially, their performance differed from normals' only when the subject noun had grammatical gender. This result indicates that patients (like normals) benefit from congruent conceptual information and therefore they can integrate conceptual and syntactic information. This result conflicts

Gender Disagreement in Italian

635

with the predictions made by the mapping hypothesis introduced above, according to which Broca's aphasics suffer a processing limitation in their ability to map conceptual information onto syntactic information. Will the patients show a deficit in integrating syntactic information across constituents? This question is addressed in experiment 2. EXPERIMENT 2 This experiment aims first to assess the role of animacy unconfounded with the type of gender (conceptual, grammatical) of the subject head noun. Vigliocco and Franck (1999) reported that animacy did not influence performance; furthermore, animacy does not seem to be a linguistic cue on which Italian speakers rely (MacWhinney, Bates, & Kliegl, 1984). However, it is possible that it becomes more important for Broca's aphasics. Nouns referring to animate entities may be more salient and therefore processed more efficiently by the patients (Kilborn, 1991; Saffran, Schwartz, & Marin, 1980). Crucially, this experiment also allows us to assess the process of integrating syntactic information across constituents. Previous studies on agreement have shown that agreement errors (both number as well as gender agreement) are more common when the local noun mismatches in number or gender with the head noun (e.g., Bock & Miller, 1991; Vigliocco et al., 1995, Vigliocco & Franck, 1999). This effect seems to be a genuine effect of syntactic complexity (see, e.g., Bock, 1995; Bock & Eberhard, 1993; Vigliocco & Nicol, 1998) arising because the syntactic properties of the local noun interfere with the properties of the head noun. Method Participants. They are the same as in Experiment 1. Materials. Materials were the same as used in Vigliocco and Franck (1999), experiment 3. Stimuli consisted of preambles composed of a subject head noun and a local noun embedded in a prepositional phrase and adjectives to be used in the completions. The manipulated variables were: (1) the animacy of the head noun (animate vs. inanimate), (2) the gender of the head noun (masculine vs. feminine), and (3) the gender of the local noun (masculine vs. feminine). All the nouns in the experimental items had grammatical gender, and the animate head nouns referred to animate entities that have just one gender form in Italian [such as animal name, e.g., scimmia (monkey-fem.), gufo (owl-masc.)]. All nouns used in the experimental materials were morphophonologically marked and singular. Each experi-

636

Vigliocco and Zilli

mental condition was represented by 16 items. As in experiment 1, two lists were created (to balance the position of the adjective). Each list was composed of 128 experimental preambles and 64 fillers. Examples of experimental items are reported in Table IV. Filler items were the same as used in experiment 1. Procedure. The same procedure was used as in experiment 1. Each patient received each list twice (four lists total). Scoring. The scoring was the same as in experiment 1. Design and Data Analysis. Statistical tests were the same as in experiment 1. The design of the experiment included four main factors for the normals: group, animacy of the subject noun, gender of the subject noun, and gender of the local noun and three for the patients (same as above except group). All manipulations were realized between items. Results Table V reports the general distribution of responses in the different scoring categories. Figure 2 reports the percentage of errors in the different experimental conditions. For normal speakers, errors were more common for the elderly group [69 vs. 44; significant main effect of group, F1 (1, 34) = 7.67, p = .009, F2 (1, 120) = 4.5, p = .04]. Errors were more common when the head and local noun mismatched in gender [103 vs. 10, significant interaction between the gender of the head and gender of the local noun F1 (1, 34) = 185.41, p < .0001; F2 (1, 120) = 49.43, p < 0001], while they were equally common for animate and inanimate head nouns (50 vs. 63, Fs < 2) and for masculine and feminine head nouns (53 vs. 60, Fs < 1). The distribution of errors for patients C. G. and C. D. is similar. Errors were more common for preambles in which the head and local noun mismatched [45 vs. 6 for patient C. G., significant interaction F2 (1, 120) = 37.82, p < .0001; and 36 vs. 2 for patient C. D., significant interaction F2 (1, 120) = 40.85, p < .0001]; no difference was found for animate and inanimate head nouns (23 vs. 28 for patient C. G. and 17 vs. 21 for patient C. D) and for masculine and feminine head nouns (30 vs. 21 for patient C. G., 21 vs. 17 for patient C. D.). Pairwise comparisons between the patients and the normal speakers revealed that patients overall made more errors than normals [C. G. vs. control 2: t(127) = 5.6, p < .0001; C. D. vs. control 1: t(127) = 5.16, p < .0001]. Patients made significantly more errors than normals when the subject noun and the local noun mismatched in gender [C. G. vs. control 2: t(63) = 5.64, p < .0001; C. D. vs. control 1: t(63) = 5.37, p < .0001], but not when the subject noun and the local noun matched in gender [C. G. vs. control 2: t(63) = 1.83, p = .08; C. D. vs. control 1: t(63) = 1.46, p = .148].

Table IV. Examples of Experimental Sentence Preambles for Experiment 2 Animate Masculine head noun, masculine local noun II ghepardo nel villaggio The (masc.) cheetah (masc.) in-the (masc.) village (masc.) Masculine head noun, feminine-local noun II delfino nella vasca The (masc.) dolphin (masc.) in-the (fem.) tub (fem.) Feminine head noun, feminine local noun L'aquila sulla cima The (fem.) eagle (fem.) on-the (fem.) peak (fem.) Feminine head noun, masculine local noun La talpa nell'orto The (fem.) mole (fem.) in-the garden (masc.) Masculine head noun, feminine local noun II ghepardo nella foresta The (masc.) cheetah (masc.) in-the (fem.) forest (fem.) Masculine head noun, masculine local noun II delfino nell'acquiario The (masc.) dolphin (masc.) in-the aquarium (masc.) Feminine head noun, masculine local noun L'aquila nel cielo The (fem.) eagle (fem.) in-the (masc.) sky (masc.) Feminine head noun, feminine local noun La talpa nella buca The (fem.) mole (fem.) in-the (fem.) hole (fem.) Inanimate Masculine head noun, masculine local noun II capanno nel villaggio The (masc.) hut (masc.) in-the (masc.) village (masc.) Masculine head noun, feminine local noun II ciottolo nella vasca The (masc.) rock (masc.) in-the (fem.) tub (fem.) Feminine head noun, feminine local noun La nuvola sulla cima The (fem.) cloud (fem.) on-the (fem.) peak (fem.) Feminine head noun, masculine local noun La zappa nell'orto The (fem.) hoe (fem.) in-the garden (masc.) Masculine head noun, feminine local noun II capanno nella foresta The (masc.) hut (masc.) in-the (fem.) forest (fem.) Masculine head noun, masculine local noun II ciottolo nell'acquario The (masc.) rock (masc.) in-the aquarium (masc.) Feminine head noun, masculine local noun La nuvola nel cielo The (fem.) cloud (fem.) in-the (masc.) sky (masc.) Feminine head noun, feminine local noun La zappa nella buca The (fem.) hoe (fem.) in-the (fem.) hole (fem.)

638

Vigliocco and Zilli

Table V. Distribution of Responses in the Different Scoring Categories, Experiment 2 Patient Controls Correct responses Agreement errors Miscellaneous responses

4,435 (96.2%) 113(2.5%) 60 (1.3%)

C.G.

C. D.

461 (90%) 51 (10%) 0 (0%)

474 (92.6%) 38 (7.4%) 0 (0%)

Discussion Animacy did not influence performance either for normal or languageimpaired speakers. These results exclude the possibility that the difference in amounts of errors we observed in the previous study is due to the fact that speakers were paying more attention to animate nouns in general. Our results are in line with other studies that showed that while animacy affects the process of grammatical function assignment (Bock & Warren, 1985), it does not affect the processes involved in building a syntactic skeleton for the utterance (Bock & Levelt, 1994; Vigliocco & Nicol, 1998), hence the process of agreement computation (Bock & Miller, 1991). The presence of a local noun mismatching in gender with the subject head noun was an important determinant of agreement errors. As discussed elsewhere (e.g., Vigliocco & Nicol, 1998), this effect can be interpreted as resulting from interference by the gender feature of the local noun. The performance of the Broca's aphasics was significantly more disrupted by a mismatching local noun than the performance of nonaphasic individuals. This is consistent with the hypothesis that aphasic speakers suffer from a reduction of processing resources devoted to syntactic integration.

Fig. 2. Percent errors for animate and inanimate subject nouns matching and mismatching with a local modifier for the two patients and their control groups.

Gender Disagreement in Italian

639

EXPERIMENT 3 The last experiment assesses whether the morphophonological realization of gender has an impact on agreement. If syntactic accuracy is achieved by using all the information available, then morphophonological information should also contribute. Conceptual and morphophonological effects come about in different manners. The effect of conceptual information indicates that the conceptual system exerts a fine-grained control over the operations of the grammatical encoder (in a top-down manner), an effect of morphophonological regularity would indicate that information at a subsequent level (phonological encoding) influences operations at a preceding level (grammatical encoding). In this experiment, we also manipulated whether the local noun matched or mismatched with the subject head noun in gender. Method Participants. They are the same as in the previous experiments. Materials. Stimuli consisted of sentence preambles and adjectives. The manipulated variables were: (1) morphophonological marking of the subject head noun (marked vs. ambiguous), (2) gender of the subject head noun (masculine vs. feminine), and (3) gender of the local noun (masculine vs. feminine). Thirty-two experimental items were created. Each condition was represented by four items. In all experimental noun phrases (determiner + noun), the gender of the noun was unambiguously marked on the determiner (i.e., there were no nouns starting with a vowel). The manipulation of morphophonological marking, therefore, strictly refers to whether the head noun ended in "-a" or "-o" (transparently marked for feminine and masculine) or whether it ended in "-e" (ambiguously marked). All local nouns were morphophonologically marked for gender. Items in the different conditions were matched for number of syllables and frequency. Sixteen singular and 16 plural forms of the head noun were used. Local nouns were also singular and plural, always matching in number with the head noun. Since no difference with respect to the number manipulation was observed for singular and plural forms, this difference is not discussed further. Two lists were created. Each list had the same experimental items, the only difference between the two lists was the position of the masculine and feminine form of the adjective (above/below). Each list was composed of the 32 experimental preambles and 32 fillers. Filler preambles consisted of a noun phrase and a prepositional phrase. In all fillers the gender of the head noun and of the local noun matched. Sixteen of the filler preambles had a plural head and a plural local

640

Vigliocco and Zilli

noun, and the remaining 16 had a singular head and a singular local noun. Examples of experimental preambles are reported in Table VI. Procedure. This was as in the previous experiments. Each patient received three repetitions of the same list (six lists total). Scoring Categories. These are the same as in the previous experiments. Design and Data Analysis. The design of the experiment included the three main factors described above (and group for the nonaphasics. Data analysis was conducted as in the previous experiments. Results Table VII reports the general distribution of responses in the different scoring categories. Figure 3 reports the percentage errors in the different experimental conditions. For normal speakers, errors were more common for the elder speakers than the younger [44 vs. 16; significant main effect of group F1 (1, 34) = 17.54, p < .0001; F2 (1, 24) = 13.67, p - .001]; they were more common when the head noun was ambiguously marked for gender than when it was unambiguously marked [main effect of morphophonological marking 45 vs. 15; F1 (1, 34) = 11.99, p = .001; F2 (1, 24) = 7.58, p = .11], and when the Table VI. Examples of Experimental Sentence Preambles for Experiment 3 (Gender Mismatching Condition Only) Morphophonologically transparent Masculine head noun, feminine local noun II viaggio verso I'isola The (masc.) trip (masc.) toward the island (fern.) Masculine head noun, masculine local noun II quadro sul muro The (masc.) painting (masc.) on the (masc.) wall (masc.)

Feminine head noun, masculine local noun La trappola per il topo The (fem.) trap (fem.) for the (masc.) mouse (masc.) Feminine head, noun feminine local noun La casa sulla collina The (fem.) house (fem.) on the (fem.) hill (fem.)

Morphophonologically ambiguous Masculine head noun, feminine local noun II lampione nella strada The (masc.) light-O in-the (fem.) road (fern.) Masculine head noun, masculine local noun II televisore sul ripiano The (masc.) television-O on the (masc.) table (masc.)

Feminine head noun, masculine local noun La canzone del gruppo The (fem.) song-O of-the (masc.) band (masc.) Feminine head noun, feminine local noun La dichiarazione della cameriera The (fem.) speech-O of the (fem.) waitress (fem.)

641

Gender Disagreement in Italian

Table VII. Distribution of Responses in Different Scoring Categories in Experiment 3 Patient

Correct responses Agreement errors Miscellaneous responses

Controls

C.G.

C. D.

1,076 (93.4%) 60 (5.2%) 16 (1.4%)

134 (70.3%) 50 (26%) 8 (4.7%)

159 (82.8%) 24 (12.5%) 9 (4.7%)

head and local noun mismatched in gender than when they matched [51 vs. 9; interaction between gender of the head and of the local noun: F1 (1, 34) = 37.67, p < .0001; F2 (1, 24) = 14.87, p = .001]. The effect of morphophonological marking was stronger for preambles in which the head and local noun mismatched in gender [three-way interaction between morphophonological marking, gender of the head noun and gender of the local noun F1 (1, 34) = 7.56, p = .01, F2 (1, 24) = 4.8, p = .03]. No other main effects or interactions reached significance (all Fs < 1). The effect of morphophonological marking was significant in the items analysis for patient C. G. [F2 (1, 24) = 4.8, p = .03] and marginally significant for patient C. D. [F2 (1, 24) = 2.9, p = .10]. The interaction between the gender of the head and of the local noun (indicating a gender mismatch effect) was significant for both patients [C. G.: F2 (1, 24) = 9.38, p = .005; C. D.: F2 (1, 24) = 9.39, p = .005]. No other main effects of interactions were significant (Fs < 2). Pairwise comparisons revealed that, overall, patients made significantly more errors than normals [C.G. vs. control 2: t(31)= 3.89, p