• Keine Ergebnisse gefunden

5.6 The processing of Wh-Questions

5.6.3 Wh-questions in aphasia

Garraffa & Grillo (2008) investigate the production of Wh-question in an aphasic speaker of Italian. Their patient’s speech is characterized by agrammatism of the Broca’s type, as a consequence of a focal lesion in the frontal and parietal areas of the left hemisphere.

The study covers both comprehension and production tests on relative clauses, cleft structures and Wh-questions. The performance on the production of Wh-questions is interesting for the present discussion. Authors use an elicitation test based on the model of the studies described above for children (Guasti, 1996b; Thornton, 1990), with the difference that the puppet is substituted by an imaginary person, not represented in the experimental setting. Two factors are introduced in the study: subject versus object extraction and animacy of the Wh-element in use. Bare interrogative pronouns of the Who/What type are in use, so that four experimental conditions are created: Who subject questions, Who Object questions, What subject questions and What object questions.

Results clearly reveal a specific impairment on Who object questions (0%), despite a good production of What object questions (75%), What subject questions (83.3%) and Who subject questions (75%). In other words, Who object questions is the only condition in which no target output is produced.

In order to explain the results, the authors focus on the feature sets of the involved elements. In particular, they assume that the [+wh] feature is disregarded in the computation because of the reduced processing resources agrammatic speakers can count on. In their view, impaired speakers find particularly challenging to activate and maintain the activation of discourse-features, which, therefore, do not activate at all or fade away quickly. The assumption is known as Generalized Minimality and it predicts the

174

application of the Principle of Relativized Minimality (Rizzi, 1990) on a reduced (and therefore generalized) set of features. Once [+wh], the feature with a crucial role in the attraction of the moved element to its target position is omitted from the computation, the possibility for the subject to function as an intervener in the expected movement increases. The enhanced risk for intervention is indeed due to the fact that the feature arrays of the target, the intervener and the moved element are now very similar and the subject results in a suitable candidate for the target position. This condition corresponds to Who object questions and is particularly problematic for agrammatic speakers in Garraffa and Grillo’s (2008) view. Nonetheless, intervention is avoided with more ease whenever at least one feature (not of the discourse type) contributes in differentiating the target position and the moved constituent (the object Wh-element) from the potential intervener (the subject). This last described condition is precisely the one corresponding to the What object condition: in this, a major role is played by animacy. The [-animate]

feature on What suffices in making the Wh-element different enough from the subject.

The asymmetry between the inanimate subject and the animate object therefore allows for the movement of the object across the subject, and intervention is neutralized.

Summing up, Garraffa and Grillo (2008; see also Grillo, 2008) claim that reduced processing resources are the primary cause of the deficit in agrammatic speakers (i.e.

failed production of Who object questions). Indeed, reduced resources do not allow for the activation of all features needed in the derivation and syntactic underspecification follows. This underspecification corresponds in their view to the omission of discourse features (i.e. [+wh]).

This proposal is particularly tempting in the case of PADs, who also suffer from a generalized impairment of their cognitive abilities and are therefore likely to have difficulties at maintaining the activation of complex feature arrays over an extended timespan. The comprehension task on Wh-questions presented in Chapter 6 will help in verifying whether the proposal of a Generalized Minimality at work in sentence processing is adequate also for the description of comprehension abilities of speakers characterized by Alzheimer’s disease.

However, a criticism to Generalized Minimality can already be formulated at the theoretical level. The proposal builds on the idea that discourse features are the specific target of the impairment as they are omitted from syntactic computation. The difficulty

175

on object questions, object relatives and object clefts arises from the absence of this crucial feature, which, in principle, should play the role of distinguishing the element to be moved, in this case the object, from the subject. Actually, the basic function of this feature is not the creation of a mismatch between the elements involved; rather, it triggers A-bar movement of the constituent endowed with the same feature. If this feature is omitted, the derivation of a Wh-question or of a relative clause should be equally blocked, independently of the target element (the subject or the object). The case of What object questions can exemplify the observation. In Garraffa and Grillo’s (2008) view, agrammatic patients succeed in producing What object questions thanks to the animacy features which is represented in a configuration of mismatch in the relevant positions. It is hard to imagine though, how the object phrase could move towards an A-bar position although it does not entail any discourse feature, only on the basis of the [-animate]

feature it includes. The same reasoning could be applied to subject questions too:

following the same line proposed by the authors, it follows that subject questions are realized despite the absence of the relevant discourse feature. The authors fail to explain how such structure can be derived in absence of scope-discourse features.

In order to consider alternative explanations to the deficit in aphasic patients, it is therefore necessary to open the overview to studies concerning languages different from Italian. This is precisely the topic of the next section.

5.6.3.1 Wh-questions in aphasia: a cross-linguistic perspective

There exists few studies that analyse the agrammatic speech of patients affected by aphasia of the Broca’s type and find results that resemble data from language acquisition.

For instance, Hickok & Avrutin (1996) and Tait, Thompson & Balard (1995) find a specific impairment on WhichNP object questions, despite a rather well-preserved processing of the remaining three experimental conditions, namely WhichNP subject, Who subject and Who object questions. Hickok & Avrutin (1996) first interpret the data by building on the Trace Deletion Hypothesis (Grodzinsky 1986, 1990), a structural model of agrammatic speech: according to this, aphasic speakers fail at maintaining the activation of traces at the deep structure level and therefore, in case of moved elements as in interrogatives or in relative clauses, they are unable to form the syntactic chain that allows for the realization of the proper sentence structure. In other words, according to

176

this account, patients cannot retrieve the gap from where the fronted element was moved.

The account also predicts that patients resort to oversimplified strategies and, for instance, assign theta-roles to arguments according to a canonical linear order. The strategy accounts for the subject/object asymmetry, as theta-role assignment works fine in subject-first structures (i.e. subject questions), but it crucially fails in object-subject-first structures (i.e.

object questions). Moreover, the specific impairment on WhichNP questions is initially interpreted as a specific deficit at building binding chains (WhichNP), contra relations of the antecedent-government type (Who), which are parsed with higher accuracy. After focusing on the syntactic properties that distinguish WhichNP and Who questions, Avrutin also revises his interpretation by taking into account the semantic differences brought up by the two Wh-elements.

In a subsequent work, the author compares the performance of English-speaking children to the one of six aphasic patients (Avrutin, 2000); he observes that both groups produce a significant higher rate of mistakes on WhichNP object questions and concludes that there must be a common account for the difficulties suffered both by young children and by agrammatic speakers. The deficit, according to Avrutin’s (2000) hypothesis, is of the cognitive type: children and patients count on reduced processing resources with respect to healthy adult speakers. The shortage of processing resources becomes problematic in the case of WhichNP questions because the processing of these specific structures requires the integration of information coming both from the syntactic and from the semantic systems, while for all other conditions (Who questions and WhichNP subject questions), it suffices to deal with the syntactic structures. The integration of semantic information in the case of WhichNP questions derives from the fact that the Wh-element is referential in nature and therefore brings up discourse presuppositions in the processing. This factor is particularly relevant in the case of WhichNP object questions, while WhichNP subject questions can simply be analysed by proceeding with a canonical assignment of theta-roles.

However, in the light of previous discussion, it is quite evident that the results discussed so far are perfectly compatible with a processing strategy based on a stricter version of Relativized Minimality (Rizzi, 1990), which blocks the derivation of WhichNP object questions but allows the one of Who object questions on the basis of a mismatch in features (in this case, a mismatch in lexical restriction) between the intervening subject

177

(lexically restricted) and the bare Wh-element extracted from the object position. In this view, the results presented in Tait et al. (1995), Hickok & Avrutin (1996) and in Avrutin (2000) are perfectly compatible with those collected by Garraffa and Grillo (2008) with respect to the effects of aphasia of the Broca’s type in Italian. Despite this, the discussion on aphasic speakers is never exempted from the issue of variability (Caplan et al. 2001;

Caramazza et al. 2001; Drai et al. 2001); indeed, other studies fail at replicating the same results. For instance, Thompson, Tait, Ballard and Fix (1999) run similar tests on four aphasic patients, who actually perform very differently. Only one patient performs as predicted by Hickock and Avrutin (1996), while the other three do not. The answers provided are so different altogether that the individuation of a common alternative pattern is not possible, thus redirecting the discussion towards the impossibility of drawing clear conclusions with respect to forms of agrammatism that arise from similar (but still different) lesions to the left hemisphere.

There is one further hypothesis on the modalities of impairment of agrammatic speakers that has received attention and could also be discussed with respect to the performance of PADs. The proposal is known as the Tree Pruning Hypothesis and was advanced by Friedmann (2002), while discussing data on the production of questions in Hebrew- and Palestinian Arabic-speaking agrammatics. First in spontaneous speech and afterwards trough an elicitation task, Friedmann observes that agrammatic speakers deal well with Yes/no questions, while they are badly impaired on Wh-questions. In particular, within this second type, patients show a slight advantage for adjunct questions (especially Why-questions) in comparison to argument questions. Moreover, interestingly enough, no asymmetry between subject and object is detected, as both are impaired to the same extent. Friedmann (2002) interprets the data building on a specific syntactic characteristic of the Semitic languages at stake. Both Hebrew and Palestinian Arabic require movement of the Wh-element and of the verb to the CP-layer in case of Wh-questions, as many other languages do (for instance, English and Italian). In contrast, Yes/No questions do not require such movement and the Verb stays in TP. Therefore, the author hypothesizes that agrammatic speakers manage to project a reduced syntactic tree, which only projects as far as TP. Everything above that node can neither be projected nor be targeted by movement (from here the reference to tree pruning). With respect to Italian-speaking PADs, the hypothesis therefore predicts very poor performance of all kinds of Relative

178

Clauses and Wh-questions, considered that all sampled structures (see Chapter 6 and 7) entail phrase movement to the CP layer in Italian. The hypothesis will be further discussed in Chapters 6 and 7, in light of the data from the experimental tasks with PADs.

5.7 Conclusions

In the present chapter, I reported an analysis of Italian Wh-questions according the following lines: the Wh-operator is extracted from its argument position and raised to the upper portion of the sentence (Chomsky, 1977) in order to satisfy the Wh-criterion (Rizzi, 1996). The operation leaves a gap in the sentence, which corresponds to the point in which the element is interpreted. The satisfaction of the Wh-criterion requires a further step, namely the movement of the inflected verb from I to C. Once the two elements endowed with [+wh], the wh-operator and the verb, are both in CP, they can enter the Spec-Head relation required by the Wh-criterion. The operation blocks the subject DP from raising, so that this is left in a post-verbal position or realized as a phonetically null pro.

Empirical studies on the processing of Wh-questions have addressed mainly two questions: the role of the argument position from where the Wh-element is extracted and the type of Wh-element in use. The first question deals with the asymmetry between subject and object Wh-questions, while the second one concerns the differences between bare and lexically-restricted quantifiers, i.e. between operators of the Who and of the WhichNP type. Italian studies on comprehension in adult speakers (De Vincenzi, 1991a, 1991b, 1996), in L1 acquisition (De Vincenzi et al., 1999; Guasti, 1996b; Guasti et al., 2012) and in aphasia (Garraffa & Grillo, 2008; Grillo, 2008) all found a specific low performance on WhichNP object questions. The robust result is in line with the Principle of Relativized Minimality, which predicts an intervention effect of the subject on the movement of the object towards its target position in CP. This is particularly true in the case of object questions in which both the subject and the object share relevant features, i.e. both phrases are lexically restricted and take animate referents.

Alternative accounts have also been proposed in order to describe the performance of aphasic patients: the Trace Deletion Hypothesis (Grodzinsky, 1986, 1990), the Tree Pruning Hypothesis (Friedmann, 2002) and Generalized Minimality (Grillo, 2008).

According to Grodzinsky, aphasic speakers experience specific difficulties at maintaining active traces and therefore fail at individuating the gap from which the Wh-element was

179

extracted. Grillo (2008) hypothesizes that aphasics are unable to maintain the activation of complex feature sets over an extended time span and therefore tend to omit features of the scope-discourse type from the computation. The omission has decisive consequences on the computation of object Wh-questions whenever the object and the subject only distinguish for the (omitted) feature. Finally, Friedmann (2002) suggests that aphasic speakers might count only on a reduced portion of the syntactic structure, which does not comprehend the sentence left periphery. The hypotheses discussed here will be further analysed in Chapter 6, in light of the collected data, in order to see how they can account for the performance of PADs.

In the present Chapter I have also dealt with the syntactic structure of RCs. In particular, I have reviewed studies that speak in favour of an analysis that sees RCs as CPs that occupy a functional projection within the DP corresponding to the relative head.

The position that hosts the relative CP is either the complement of the DP (as proposed by Kayne, 1994) or a functional projection high in the DP (as claimed by Cinque, 2008, 2014). In both views, the internal argument raises to its target position in the CP-layer of the relative clause. In case the movement starts from the internal argument of the RC, locality effects arise. Indeed, a RM approach to relative clauses (Friedmann et al., 2009;

Grillo, 2009) accounts for the asymmetry between subject and object RCs with a precision that memory- and processing-based accounts lack (De Vincenzi, 1991a; Frazier & Fodor, 1978; Kidd et al., 2007; Mc Whinney & Pléh, 1988; amo). The well-attested phenomenon is indeed ascribed to the fact that the object DP moves across the subject DP and shares with it part of its feature array. The intervening subject makes the object movement difficult.

With respect to this, I have offered and overview of the factors that improve the comprehension and/or production of ORs. Of course, they are not all equal. As pointed out by Sanfelici et al. (2014), it might well be the case that some of the strategies adopted by children in order to avoid the production of the elicited ORs are universal (i.e. roles reversing), while other strategies might be language specific. I focused my attention on factors that, in my opinion, are relevant for Italian, either because they have already been proved to play a role, or because there are signs that indicate that this could be the case.

Summarizing, number mismatch, and passive voice (PORs) improve the comprehension of ORs. Both factors enhance the differences between the moved constituent and the

180

intervener, thus favouring extraction. As for resumption, the way it operates is not completely clear yet, but Contemori & Belletti (2014) proposed that it might be the case that the presence of a stranded clitic is relevant for the evaluation of feature arrays.

Two distinct studies (Adani et al., 2010; Belletti et al., 2012) robustly confirm that gender does not play any role in Italian at enhancing the difference between the feature arrays of the involved position, while an effect is visible in a language like Hebrew, which has gender agreement on the verb. This asymmetry is interesting in a cross-linguistic perspective, because it makes clear that not all features always enter the locality computation, rather only those that contribute to triggering movement do (Belletti et al., 2012).

Finally, data concerning the role of animacy in Italian is still very poor, as the feature was included only in one elicitation study with adults (Belletti & Chesi, 2011) and actually did not produce any remarkable effect.

The body of information now available on the derivation and the processing of Wh-questions and Relative Clauses is quite wide and rich. The information provided and summarized here are indeed the premises for the design of the experimental studies I am going to present in chapters 6 and 7.

181

6 Wh-QUESTION COMPREHENSION

IN PATIENTS WITH ALZHEIMER’S DISEASE

6.0 Introduction

In the present Chapter, I will present data from a Wh-question comprehension task performed by a group of Italian PADs, whose participants are characterized by different levels of dementia. The collected data will represent the starting point for the discussion on the syntactic deficit in PADs.

A number of reasons motivate the decision to include Wh-questions in the study.

First, the selected structures entail an operation of extraction, which moves a constituent out of its argument position and takes it to the clause left periphery through Wh-movement. Previous studies on L1 acquisition and aphasia (see Chapter 5 for an overview) used Wh-questions in order to sample speakers’ competence. Overall, they revealed that the operation of extraction increases the level of complexity in computation with respect to unmarked declarative clauses. The claim stems from the observation of increased difficulties in children and aphasic speakers in the comprehension and production of different kinds of Wh-questions, according to patterns that depend on the asymmetry between subject and object extraction, the kind of Wh-element in use (Who vs WhichNP), as well as the experimental task. Therefore, previous research shows that interesting and meaningful data can be collected through studies that include Wh-questions among their experimental material. Wh-Wh-questions have already proved to be a

First, the selected structures entail an operation of extraction, which moves a constituent out of its argument position and takes it to the clause left periphery through Wh-movement. Previous studies on L1 acquisition and aphasia (see Chapter 5 for an overview) used Wh-questions in order to sample speakers’ competence. Overall, they revealed that the operation of extraction increases the level of complexity in computation with respect to unmarked declarative clauses. The claim stems from the observation of increased difficulties in children and aphasic speakers in the comprehension and production of different kinds of Wh-questions, according to patterns that depend on the asymmetry between subject and object extraction, the kind of Wh-element in use (Who vs WhichNP), as well as the experimental task. Therefore, previous research shows that interesting and meaningful data can be collected through studies that include Wh-questions among their experimental material. Wh-Wh-questions have already proved to be a