Comprehension of wh-questions and declarative sentences in agrammatic aphasia: The set partition hypothesis

ARTICLE IN PRESS

Journal of Neurolinguistics 21 (2008) 375–399 www.elsevier.com/locate/jneuroling

Comprehension of wh-questions and declarative sentences in agrammatic aphasia: The set partition hypothesis Christos Salis, Susan Edwards School of Psychology and Clinical Language Sciences, University of Reading, Harry Pitt Building, Earley Gate Whiteknights, Reading RG6 6AL, UK Received 10 October 2006; received in revised form 26 October 2007; accepted 5 November 2007

Abstract Problematic trace-antecedent relations between deep and surface structure have been a dominant theme in sentence comprehension in agrammatism. We challenge this view and propose that the comprehension in agrammatism in declarative sentences and wh-questions stems from impaired processing in logical form. We present new data from wh-questions and declarative sentences and advance a new hypothesis which we call the set partition hypothesis. We argue that elements that signal set partition operations influence sentence comprehension while trace-antecedent relations remain intact. r 2007 Elsevier Ltd. All rights reserved. Keywords: Agrammatism; Aphasia; Comprehension; Wh-questions; Declaratives; Set partition

1. Introduction Agrammatic aphasia is an interesting and controversial condition. Its hallmarks are the fragmented and often ungrammatical sentences and difficulties understanding complex declarative sentences (Burchert, De Bleser, & Sonntag, 2003; Dickey & Thompson, 2004; Grodzinsky, 2000). In this paper we focus solely on the comprehension deficit. We start with a broad overview which serves as a background to our study. It has repeatedly been shown that reversible declarative sentences with canonical thematic role order Corresponding author. Tel.: +44 1183787467; fax: +44 1183784696.

E-mail address: [email protected] (C. Salis). 0911-6044/$ - see front matter r 2007 Elsevier Ltd. All rights reserved. doi:10.1016/j.jneuroling.2007.11.001

ARTICLE IN PRESS 376

C. Salis, S. Edwards / Journal of Neurolinguistics 21 (2008) 375–399

(actives, subject clefts and relatives) are understood better than non-canonical declarative sentences (passives, object clefts and relatives; Bastiaanse & Edwards, 2004; Edwards, 2005; Grodzinsky, 2000; Lee & Thompson, 2004). Nonetheless, non-reversible declarative sentences are understood well despite order of thematic roles (Ansell & Flowers, 1982; Brookshire & Nicholas, 1980; Caramazza & Zurif, 1976). The frequently reported pattern of comprehension differences between canonical and non-canonical reversible sentences is sometimes referred to as the ‘‘standard’’ profile of agrammatic comprehension (Grodzinsky, 1998, p. 179). Other less studied phenomena include comprehension of sentences with psychological predicates (Balogh & Grodzinsky, 1996; Nickels, Byng, & Black, 1991; Pin˜ango, 2000, 2006), sentences with prepositions (Kolk & Friederici, 1985; O’ Grady & Lee, 2005), aspectual and complement coercion (Pin˜ango & Zurif, 2001), sentences with dative alternations (Pin˜ango, 2006), sentences with quantifiers (Balogh & Grodzinsky, 2000; Saddy, 1995) and wh-questions that begin with who, what and which, the focus of this paper. In comparison with declarative sentences little is known about comprehension of wh-questions in aphasia, Gallagher and Guilford (1977) and Riley (1988) highlighted deficits in understanding wh-questions in aphasia. More recently, comprehension of wh-questions with canonical and non-canonical thematic role order has attracted attention in agrammatism (Hickok & Avrutin, 1996; Neuhaus & Penke, in press; Thompson, Tait, Ballard, & Fix, 1999; Van der Meulen, Bastiaanse, & Rooryck, 2002). The wh-questions under study are syntactically similar to canonical and non-canonical declarative sentences in that the order of thematic roles is a shared characteristic. In canonical sentences (declarative sentences and wh-questions alike) the order of thematic roles is agent– patient whereas in non-canonical the order of thematic roles is patient– agent. So, these sentences are ideal to test theoretical accounts in order to gain a more detailed insight into the underlying nature of aphasia and the architecture of the grammatical system. In this paper our aims are: To critically evaluate claims about comprehension of wh-questions in agrammatism; to present new data from five agrammatic speakers on different types of wh-questions; to explore, for the first time, the influence of raising and sentence length on wh-questions; to consider a range of possible explanations for our and other published data. Finally, we advance a new explanatory hypothesis about the comprehension deficit of declarative sentences and wh-questions in agrammatism. We are calling this the set partition hypothesis. 1.1. Accounts of agrammatic comprehension The aphasia field is crowded with competing accounts of agrammatic comprehension. These can be divided broadly into two main categories which reflect key theoretical orientations and debates. First, the government and binding framework (Chomsky, 1981, 1986; Haegeman, 1994) has motivated the trace deletion hypothesis (Grodzinsky, 1986, 2000) which explains the agrammatic comprehension deficit in terms of problematic traceantecedent representations. Second, reduction in processing capacity has been interpreted in a number of ways. These include: i. An inability to map thematic roles onto sentence constituents (Linebarger, 1995; O’ Grady & Lee, 2005; Saffran & Schwartz, 1988).

ARTICLE IN PRESS C. Salis, S. Edwards / Journal of Neurolinguistics 21 (2008) 375–399

377

ii. An inability to process closed-class vocabulary (Friederici & Gorrell, 1998; Pulvermu¨ller, 1995). Although Friederici and Gorrell attribute the underlying deficit to poor processing of closed-class vocabulary, in their view thematic role prominence helps comprehension. The agent is more prominent than the theme or patient (Grimshaw, 1990) and therefore will be assigned to the first NP. iii. Slow processing of lexical information and syntax (Burkhardt, Pin˜ango, & Wong, 2003; Dickey & Thompson, 2004; Friederici & Kilborn, 1989; Haarman & Kolk, 1991; Park, McNeil & Doyle, 2002; Pin˜ango, 2006; Swinney, Zurif, Prather, & Love, 1996; Zurif, Swinney, Prather, Solomon, & Bushell, 1993). iv. Fast decay of syntactic information (Haarman & Kolk, 1994). v. Loss of automatic access of information from the lexicon (Blumstein, Milberg, Dworetzky, Rosen, & Gershberg, 1991; Myers & Blumstein, 2005). We will return to some of these accounts when reviewing possible explanations for our data (Section 4). 1.2. Patterns and accounts of comprehension of wh-questions Hickok and Avrutin (1996) present data from two agrammatic speakers who showed the standard agrammatic profile, that is, better performance on canonical and poorer performance on non-canonical declarative sentences. These speakers were better at canonical who and which questions (see Table 2 for examples). However, performance on non-canonical who questions was also good while the only question type that yielded poor performance was non-canonical which questions. Hickok and Avrutin’s study was replicated by Thompson et al. (1999) on four speakers with the standard comprehension profile for declarative sentences. For comprehension of wh-questions only one speaker showed the pattern observed by Hickok and Avrutin. The other three speakers showed mixed profiles. Thompson and colleagues proposed a semantic explanation to account for the pattern observed by Hickok and Avrutin in one of their participants. It was claimed that a double conjoint computation is needed to solve the identity of a variable x in the case of which phrases whereas only one computation is required for who/what phrases. However, an additional syntactic computation, not proposed by Thompson et al., would need to be postulated for dealing with the noncanonical word order in which questions. In essence, Thompson and colleagues argued that the comprehension deficit regarding wh-questions stems from the interface between syntax and semantics. They also proposed a second account motivated by minimalism (Chomsky, 1995). Who is regarded a minimal projection while which a specifier. We believe their views regarding the nature of syntactico-semantic processing at the sentence level are correct and we build on their argument in formulating our hypothesis later on. In a retrospective study Avrutin (2000) reanalysed the six speakers (Hickok & Avrutin, 1996; Thompson et al., 1999) and confirmed the original pattern reported by Hickok and Avrutin (1996). This prompted Avrutin (2000) to develop an explanation, partly based on Pesetsky’s (1987) account of wh-questions and partly on reduced processing capacity. Pesetsky (1987) divides wh-questions into two categories. He claims that who/what are non-discourse-linked whereas which are discourse-linked. In which questions interlocutors have a set of referents in mind which are drawn from previously established discourse. Which questions are therefore discourse linked. No such requirement is imposed

ARTICLE IN PRESS 378

C. Salis, S. Edwards / Journal of Neurolinguistics 21 (2008) 375–399

on who/what questions and hence they are non-discourse linked. Based on this dichotomy, Avrutin claimed that integration from two cognitive modules, (a) syntax, in order to deal with the non-canonical order of thematic roles, and (b) discourse, to deal with linking which to previous discourse makes non-canonical which questions more difficult. In noncanonical who questions discourse is not required, only syntax and therefore comprehension would be better. We take issue with this conclusion as the same speakers had already been shown to have a deficit with non-canonical declarative sentences. Avrutin’s explanation deals with the poor results on which questions and fails to account for the relative ease agrammatic speakers have in understanding non-canonical who questions. In an attempt to explain Hickok and Avrutin’s (1996) data set, Grodzinsky (1995) and Balogh and Grodzinsky (2000) appealed to an explanation similar to that of Avrutin (2000). For Grodzinsky (1995) and Balogh and Grodzinsky (2000) who phrases are nonreferential and non-discourse linked while which phrases are referential and discourse linked (cf. Pesetsky, 1987). Based on the trace deletion hypothesis which posits that traces are deleted from the agrammatic sentence representation, it was claimed that a strategy helps assign agent only to referential phrases (which) but not to non-referential phrases (who). The examples in (1) show how the trace deletion hypothesis works and the predictions for which questions. The asterisks denote the positions of deleted traces.1 In (1a) the giraffe (patient) receives its correct thematic role directly from the verb because it has not moved. which hippoðagentÞ  kicked the giraffeðpatientÞ which hippoðagentÞ did the giraffeðagentÞ  kick 

above chance chance

(1a) (1b)

Performance in canonical which questions (1a) is above chance because the strategy compensates for the deleted traces, yet in (1b) there are two conflicting agents and hence the chance performance. Canonical and non-canonical who questions are exempt from the strategy and performance is above chance, even in non-canonical who questions. In these structures intact lexical semantic abilities correctly mediate thematic role assignment. It is unclear as to whether lexical semantics alone can support thematic role assignment in agrammatism (cf. Ostrin & Tyler, 1995). It is not clear why lexical semantics support sentence comprehension in who and not in which questions. As it stands the explanation appears selective in nature. With reference to the sentences types under discussion, the trace deletion hypothesis (Grodzinsky, 2000), the mapping hypotheses (Linebarger, 1995; O’ Grady & Lee, 2005) and the structural prominence hypothesis (Friederici & Gorrell, 1998) predict that canonical wh-questions (who/what and which), and generally canonical declarative sentences, would be understood better than non-canonical wh-questions and declarative sentences. If there is an inability to process closed-class vocabulary as suggested by Pulvermu¨ller (1995), then, there should be no difference between subject and object cleft sentences as both structures have the same number and type of closed-class words. Syntactic complexity has been shown to influence comprehension and processing of sentences in agrammatism (Dickey & Thompson, 2004; Goodglass et al., 1979; Haarman & Kolk, 1994), yet it is not clear whether it is due to slow activation or fast decay 1

In this example we assume the VP-internal subject hypothesis (e.g. Koopman & Sportiche, 1991).

ARTICLE IN PRESS C. Salis, S. Edwards / Journal of Neurolinguistics 21 (2008) 375–399

379

(Haarman & Kolk, 1994). The automaticity hypothesis also entails slowing down of processing according to Myers and Blumstein (2005). 1.3. The set partition hypothesis In this paper we suggest an additional factor that may impact on comprehension of sentences in people with agrammatism. We postulate that agrammatic speakers are unable to establish semantic representations in logical form. The main premise in the set partition hypothesis is that sentence meaning is compositional and agrammatic speakers have difficulties in identifying certain components of sentence meaning. This hypothesis is motivated primarily by work in formal-theoretical semantics especially from work by Katz and Fodor (1963), Bartsch (1973), Montague (1974) and Hervey (1979). The originality of our hypothesis lies in the application of key principles of this early research in formal-theoretical semantics to the domain of neurolinguistics. Lyons (1995) remarks that the sense of a word is the product of its component parts. So, the sense of a lexeme has a compositional function which means that its meaning is determined by the meaning of its components and the way they are combined (Lyons, 1995). We apply this principle to sentences rather than words. We follow a recent definition of compositionality proposed by Werning, Machery, and Schurz (2005, p. 10): ‘‘An interpreted representational system R is compositional if and only if for every complex representation r of R, the meaning of r is determined by the structure of r and the meaning of its constituents of r.’’ The concept of compositionality is not new and has been controversial in cognitive psychology (cf. Harley, 2001) and theoretical semantics (Bolinger, 1965; Fodor & Lepore, 2002; Katz & Fodor, 1963). It is a concept with different meanings in cognitive science as Werning et al. (2005) point out. To our knowledge the principle of compositionality has never been discussed with reference to the sentence comprehension deficit in agrammatic aphasia. We argue that it may be particularly relevant to who/what and which phrases and other sentences, especially those containing quantifiers. We posit that which phrases are more compositional and, in turn, more difficult in a processing sense. On the other hand, who and what phrases, being less compositional, facilitate processing. Since which modifies the noun we claim that which phrases instantly signal partition of a subset of a set. A set is a collection of features called members or elements (Heim & Kratzer, 1998). On the other hand, who/what phrases do not instantly signal such a partition. In a sentence, different sets combine together in larger or smaller sets to make up the meaning of the sentence. It is the inherent semantic difference between who/what and which phrases that determines these properties rather than the physical presence of individual set members, that is, the array of animal figures in our experiment. The latter is a non-linguistic factor. Our hypothesis and the semantic account by Thompson et al. (1999) are similar. In both accounts who/what phrases are thought to be, in some sense, less demanding and which phrases more demanding. In both accounts the distinction is attributed to the semantic processing of the wh-phrase. However, in Thompson et al.’s account it is not clear whether the additional demand is introduced by which or by the noun. In the set partition hypothesis the increment in processing load is introduced by which. Furthermore, the set partition hypothesis attempts to account for and unify other comprehension profiles in agrammatism as we discuss later in Section 4.3.1.

ARTICLE IN PRESS 380

C. Salis, S. Edwards / Journal of Neurolinguistics 21 (2008) 375–399

An aspect of grammatical ability that has received little attention in the literature on agrammatism is the level of logical form (LF), the semantic component of the grammar (Cappa, Moro, Perani, & Piattelli-Palmarini, 2000; Pin˜ango, 2006). Logical form ‘‘encodes logico-semantic properties’’ (Haegeman, 1994, p. 491) and is the third component of government and binding framework. Similarly, according to May (1985), logical form interprets aspects of semantic structure that are expressed syntactically. A more radical approach to logical form and government and binding was initially expressed by Van Riemsdijk and Williams (1981) and refined by Williams (1987) who argued that there are only two levels (as opposed to three in other versions of government and binding). The properties of logical form of government and binding are amalgamated by Williams. Chomsky (1979) distinguished logical form, which is part of syntax, from semantic representation which is purely a conceptual semantic level outside grammar. According to Jackendoff (1997) semantic representation does not encode the internal meaning of lexical items. However, in Chomsky (1986) the difference between syntax and the conceptual semantic level is less clear cut and logical form becomes an interface between language and other cognitive systems that mediate meaning. Jackendoff (1997, p. 48) reconciles these controversial definitions about the role of logical form under, what he calls, ‘‘syntactically transparent semantic composition.’’ According to this hypothesis, ‘‘all elements of content in y a sentence are found in the lexical conceptual structures (LCSs) of the lexical items composing the sentences’’ and ‘‘the internal structure of individual LCSs plays no role in determining how the LCSs are combined’’ (Jackendoff, 1997, p. 48). It is evident from this brief review that the function of logical form has been controversial and dynamic. It has also changed in the different incarnations of universal grammar (cf. Chomsky, 1995; Hornstein, 1995). Pin˜ango (2006) offers an account of agrammatic sentence comprehension which utilises LF and Jackendoff’s LCS. She appeals to the notion of slowing down of processing (Zurif et al., 1993) and postulates an interface level which shares some of the characteristics of the syntactic component and some characteristics of LCS. Grammatical relations are rules which connect structural properties of items to discourse prominent relations. The subject position is a prominent discourse position in English. An interesting feature of Pin˜ango’s account is that it is indifferent to canonicity of thematic roles. Pin˜ango’s account and the set partition hypothesis both highlight a new avenue of research. 1.4. Motivation for the present study Our review shows that the underlying nature of the sentence comprehension deficit is still not clear. Great emphasis has been placed on comprehension of declarative sentences and research has focused on trace-antecedent relations between deep and surface structures. Comprehension of wh-questions has received far less attention. The question that emerges is which theoretical account can best explain the comprehension deficit of declarative sentences and wh-questions in agrammatic aphasia? We address this key question in this paper. Further, our investigation includes two new sentence types, namely, wh-questions in raising structures and lexically padded wh-questions. Raising wh-questions provide an interesting insight to the syntactic abilities of aphasic speakers as they are complex in terms of syntax, while the order of thematic roles, which has been a crucial factor in agrammatic comprehension is preserved. Although the trace deletion hypothesis (Grodzinsky, 2000) does not predict a different pattern of performance

ARTICLE IN PRESS C. Salis, S. Edwards / Journal of Neurolinguistics 21 (2008) 375–399

381

between raising and non-raising sentences, processing accounts (Haarman & Kolk, 1994; Pulvermu¨ller, 1995) implicitly predict poorer performance with raising sentences. Raising verbs, e.g. seem, appear, are unaccusative verbs that fail to assign case and an external argument (Burzio, 1986). The syntax of a raising wh-question is shown in (2a) and is contrasted with an equivalent short, non-raising question (2b). A pragmatic or felicity violation did not occur in the context we report (Section 2.3). Raising sentences in BritishEnglish are used in politeness register even when there is no question as to whether an event is unclear. ½CP whoi ½IP ti ½VP seemed½IP ti to½VP ti ½kick the cow

(2a)

½CP whoi ½IP ti ½VP ti ½kicked the cow

(2b)

However, an obvious difference between short and raising questions is their length (in number of words). A difference in comprehension of short and raising questions could be attributed to this factor rather than syntactic complexity. To date, sentence length has not been found to play a role in agrammatism but only in relation to declarative structures (Berndt, Mitchum, & Wayland, 1997; Friedmann & Gvion, 2003; Love & Oster, 2002; Miera & Cuetos, 1998). We explore this dimension in this study. 2. Method 2.1. Participants Five chronically agrammatic aphasic speakers were recruited. All speakers were monolingual and native speakers of Scottish-English who had lived in Scotland all their lives. Mean age was 62 years and 2 months (range: 40–75 years) and mean level of education was 10 years 2 months (range: 9–11 years). Mean time post-onset of aphasia was 5 years 6 months (range: 1 year 7 months–17 years). Table 1 summarises the participants’ biographical information and neurological profiles. Diagnosis of agrammatic aphasia was based on the short Boston Diagnostic Aphasia Examination (BDAE; Goodglass, Kaplan, & Barresi, 2001) and a grammatical analysis of spontaneous language using the Northwestern Narrative Language Sample Analysis (Thompson, unpublished a). Around 150 words were elicited during picture description (cookie theft) and interview. The BDAE profiles and speech and language characteristics of the group and individual participants are shown in Appendices A and B, respectively. We also calculated speech rate for each agrammatic speaker as this is a useful diagnostic marker of non-fluent aphasia. Non-fluent aphasia comprises less than 50 words per minute (Kerschensteiner, Poeck, & Brunner, 1972). Further, we collected speech and language data from five neurologically control participants (control group, Appendix B) using the same elicitation method and analyses. It should be noted that HM and HT were classified as mixed non-fluent rather than Broca’s aphasic speakers because of the relatively low comprehension scores but had the agrammatic speech production characteristics. However, their single word comprehension and comprehension of canonical sentences were good as was their overall comprehension ability in conversation. Ethical approval from the relevant health authorities and the University of Reading was given. Written informed consent was obtained from the participants before the study.

382

Site of lesion

Classification

TPOb

Hemiplegia

Handedness

9

Infarct

Broca’s

3

None

R

Barmaid

10

Infarct

Mixed non-fluent

1; 7

R

R

71

Builder

10

Infarct

Mixed non-fluent

2

L

R

M

70

11

Infarct

Broca’s

17

R

R

F

40

Shop assistant Cleaner

11

Infarct

Left middle cerebral artery Left middle cerebral artery Left frontal/ parietal Left temporal/ parietal Left middle cerebral artery

Broca’s

1; 9

R

R

Gender

Age

Occupation

WA

M

75

Shepherd

HM

F

55

HT

M

NJ SW a

Years. TPO: time post-onset (years; months).

b

Educationa

ARTICLE IN PRESS

Aetiology

Participants

C. Salis, S. Edwards / Journal of Neurolinguistics 21 (2008) 375–399

Table 1 Biographical information and neurological profiles of participants

ARTICLE IN PRESS C. Salis, S. Edwards / Journal of Neurolinguistics 21 (2008) 375–399

383

2.2. Sentence stimuli Comprehension of canonical (actives, subject clefts) and non-canonical (passives and object clefts) declarative sentences was tested on two comprehension sentence–picture matching tests: The Verb and Sentence Test (VAST, Bastiaanse, Edwards, & Rispens, 2002) and the Northwestern Assessment of Verbs and Sentences (NAVS, Thompson, unpublished b). Mean lexical frequency of nouns that made up the wh-questions was 7 citations per million words (range: 1–20) and mean verb frequency was 5.25 citations (range: 1–10) per million words (Johansson & Hofland, 1989). The nouns denoting the animals were as follows: Cow, donkey, elephant, giraffe, goat, gorilla, hippo, kangaroo, penguin, pig, rhino, sheep and zebra. Five lexical verbs were also used: bump, pat, kick, scratch and shove. All animals were non-aggressive in order to neutralise responses (Caplan, Baker, & Dehaut, 1985). All participants were 100% successful at identifying the animals and actions. Overall, 240 questions were used per participant. There were three main types of questions in canonical and non-canonical word order. Table 2 shows the question types and the number of trials for each type. 2.3. Materials and procedure We used an act-out task to test comprehension of wh-questions (cf. Hickok & Avrutin, 1996; Thompson et al., 1999). Three animal figures (two of the same kind and a third of another kind) were placed in front of each participant and a scenario was acted out by the experimenter. For example, a small giraffe scratched the hippo and then the hippo scratched a large giraffe. In Hickok and Avrutin (1996) and Thompson et al. (1999) the Table 2 Examples and types of wh-questions

Short

Raising

Padded just now

Padded slowly

Canonical

Trials

Non-canonical

Trials

Who/what kicked the cow?

12/12

12/12

Which hippo kicked the cow?

12

Who/what did the cow kick? Which hippo did the cow kick?

Who/what seemed to kick the cow? Which hippo seem to kick the cow?

12/12

Who/what did the cow seem to kick? Which hippo did the cow seem to kick?

12/12

Who/what kicked the cow just now? Which hippo kicked the cow just now?

6/6

Who/what did the cow kick just now? Which hippo did the cow kick just now?

6/6

Who/what slowly kicked the cow? Which hippo slowly kicked the cow?

6/6

Who/what did the cow slowly kick? Which hippo did the cow slowly kick?

6/6

Total

120

Total

120

12

12

12

12s

12

12

12

ARTICLE IN PRESS 384

C. Salis, S. Edwards / Journal of Neurolinguistics 21 (2008) 375–399

correct response was always the first animal (small giraffe) that initiated the action in all canonical questions whereas in non-canonical questions it was always the third animal (large giraffe). The middle animal was never the target animal. We modified their task. So, in half of each type of who and what questions the target animal corresponded to the animal of which there was one in the scenario, i.e. the hippo in the above example. For the other half it corresponded to the animal of which there were two in the scenario, either the small or large giraffe. So, the same scenario, although not presented successively, was used with two different who and what questions and the required responses were different. A different scenario was used for each which question. Devising the scenarios in this way ensured that all three animals in the array were targets. All wh-questions were pseudorandomised. After acting out each scenario the animals remained in front of the participant. The participants had been told that they could ask for a repetition of the scenario and/or the stimulus question. However, very few repetitions were requested. The direction of the manipulation was left-to-right as well as right-to-left. This was in response to a finding by Deloche and Seron (1981) who reported that the left-to-right arrangement of pictures influenced responses. Although we considered this possibility in designing the task we did not examine if there was a spatial bias. Statistical data for the influence of task (probing information with respect to first vs. third figure, etc.) can be found in Salis (2006). 2.4. Analyses Firstly, in what we call the new data set group (N ¼ 5), we compared the following: (i) thematic role order, i.e. canonical vs. non-canonical in combined who, what and which questions, (ii) the structure of wh-phrase, (iii) short vs. raising questions and (iv) short vs. padded questions. Secondly, we analysed our data as a case-series. Thirdly, in order to work with a larger data set we combined our data and those reported by Hickok and Avrutin (1996) and Thompson et al. (1999). This yielded 11 speakers for the comparison between who and which questions and 9 speakers for the comparison between who and what. 3. Results In this section we report our findings of the new data set group and of individual profiles (case-series). We also report the results from the combined data sets. All results are reported as percentage correct performance. The statistical comparisons were based on raw data and not on percentages apart from the comparisons in the combined data sets. All statistical tests were two-tailed. 3.1. New data set group (N ¼ 5) Comprehension of canonical declarative sentences was 79% and 86% in the VAST and NAVS, respectively, while comprehension of non-canonical declarative sentences was 54% (VAST) and 45% (NAVS). There were differences between canonical and non-canonical declarative sentences in both tests, VAST: w2(1) ¼ 14.028, po.01, and NAVS: w2(1) ¼ 37.194, po.01. These findings conform to the standard profile of agrammatic comprehension. As far as comprehension of wh-questions is concerned in terms of thematic role order there was a difference between all canonical (75%) and non-canonical

ARTICLE IN PRESS C. Salis, S. Edwards / Journal of Neurolinguistics 21 (2008) 375–399

385

questions (54%, w2(1) ¼ 64.109, po.01), irrespective of the nature of wh-phrase. Table 3 shows the statistical comparisons according to canonicity and wh-phrase. Table 4 shows performance by question type in raising and padded questions. The comparison between all short and all raising questions was not significant w2(1) ¼ 4.126, NS. Considered individually, no comparison between short and raising questions in either canonical or non-canonical conditions was significant despite some variation in performance, although performance in which questions was more variable than who/what questions. The seemingly marked difference between short canonical which (67%) and short raising which (80%) did not reach significance w2(1) ¼ 2.727, NS. The two comparisons between all short and the two types of padded questions were also not significant (padded with just now w2(1) ¼ 3.347, NS, and padded with slowly w2(1) ¼ 4.126, NS). 3.2. Case-series Each participant’s performance on declarative sentences in the VAST and NAVS is shown in Tables 5 and 6 together with the statistical comparisons based on Fisher exact tests. All tests were two tailed. Table 8 shows the statistical results from each participant sorted by canonicity and wh-phrase. Table 7 shows the results for each agrammatic speaker in short, raising and padded conditions. We would like to note that HT was only tested on who padded questions and not what. The statistical comparisons Table 3 Performance and statistical comparisons on canonicity and wh-phrase Sentence comparisons and performance

Statistical comparisons

Who canonical (83%) vs. non-canonical (63%) What canonical (82%) vs. non-canonical (71%) Canonical who (83%) vs. what (82%) Non-canonical who (63%) vs. what (71%) Which canonical (67%) vs. non-canonical (38%) Canonical who/what (83%) vs. which (67%) Non-canonical who/what (65%) vs. which (38%)

w2(1) ¼ 18.171, po.01 w2(1) ¼ 5.974, po.01 NSa NS w2(1) ¼ 33.242, po.01 w2(1) ¼ 20.690, po.01 w2(1) ¼ 41.270, po.01

a

NS denotes non-significant results.

Table 4 Performance on short, raising and padded questions Wh-question types Short

Who/what Which

Raising

Padded just now

Padded slowly

Canonical (%)

Noncanonical (%)

Canonical (%)

Noncanonical (%)

Canonical (%)

Noncanonical (%)

Canonical (%)

Noncanonical (%)

83 67

74 33

83 80

61 33

82 52

70 43

83 68

60 52

ARTICLE IN PRESS C. Salis, S. Edwards / Journal of Neurolinguistics 21 (2008) 375–399

386

Table 5 Case-series: results (% correct) on declarative sentences Participants

WA HM HT NJ SW

VAST

NAVS

Canonical (%)

Non-canonical (%)

p-Values

Canonical (%)

Non-canonical (%)

p-Values

90 85 70 85 65

50 80 60 30 50

o.05 NS NS o.01 NS

90 90 80 95 75

40 65 30 30 50

o.01 NS o.01 o.01 NS

Table 6 Case-series: results on canonicity and wh-phrase Participants

Wh-questions

Canonical (%)

Non-canonical (%)

p-Values

WA

Who/what Which Who/what Which Who/what Which Who/what Which Who/what Which

88 88 89 50 88 81 75 46 78 69

71 23 74 25 28 6 68 79 85 69

w2(1) ¼ 6.063, po.05 w2(1) ¼ 40.481, po.01 w2(1) ¼ 5.516, po.05 w2(1) ¼ 6.400, po.01 w2(1) ¼ 52.589, po.01 w2(1) ¼ 54.857, po.01 w2(1) ¼ 0.852, NS w2(1) ¼ 11.378, po.01 w2(1) ¼ 1.140, NS w2(1) ¼ 0.000, NS

HM HT NJ SW

(based on Fisher exact tests) among different types of short questions (baselines) are shown in Table 8. Comparisons between baseline scores on short wh-questions and (i) raising and (ii) padded questions are based on cumulative results in each question type, irrespective of canonicity and wh-phrase (shown in Table 9). 3.3. Combined data sets (i. N ¼ 11, ii. N ¼ 9) i. Table 10 shows results from who, what and which questions from 11 agrammatic speakers (three data sets). Wilcoxon signed ranks tests showed no difference between canonical and non-canonical who questions and no difference between canonical who and canonical which questions. However, there were significant differences between canonical and non-canonical which questions (T+ ¼ 42, po.05) and between noncanonical who and non-canonical which questions (T+ ¼ 66, po.05). We undertook a further analysis of group performance without HM and HT. These speakers, although they had agrammatic speech output, were classified as having mixed nonfluent aphasia. Mean group performance without HM and HT was as follows: Canonical who 83% (sd 11.31), non-canonical who 79% (sd 9.24), canonical which 75% (sd 18.76), non-canonical which 51% (sd 13.81). There was no difference between canonical and non-canonical who questions but there was a difference

Participants

Types of wh-questions

WA HM HT NJ SW

Who/what Which Who/what Which Who Which Who/what Which Who/what Which

Padded just now

Raising

Padded slowly

Canonical (%)

Non-canonical Canonical (%) (%)

Non-canonical Canonical (%) Non-canonical Canonical (%) (%) (%)

Non-canonical (%)

83 92 92 33 75 92 79 33 83 83

83 33 88 33 46 8 71 58 83 33

71 0 50 25 17 0 63 75 79 67

42 42 83 17 8 17 67 92 100 83

88 92 92 58 100 100 67 50 67 100

83 83 83 33 92 58 75 42 75 42

75 17 83 17 33 0 75 92 83 83

83 83 83 75 83 75 75 58 92 50

ARTICLE IN PRESS

Short

C. Salis, S. Edwards / Journal of Neurolinguistics 21 (2008) 375–399

Table 7 Case-series: summary on short, raising and padded questions

387

ARTICLE IN PRESS C. Salis, S. Edwards / Journal of Neurolinguistics 21 (2008) 375–399

388

Table 8 Case-series: statistical comparisons on short wh-questions Comparisons

Group

WA

HM

HT

NJ

SW

Who/what canonical vs. non-canonical Which canonical vs. non-canonical Canonical who/what vs. which Non-canonical who/what vs. which

NS o.01 o.05 o.001

NS o.05 NS o.05

NS NS o.01 o.01

NS o.01 NS o.05

NS NS o.05 NS

NS o.05 NS o.01

Table 9 Case-series: comparisons on raising and padded vs. short wh-questions Participants

Short vs. raising

Short vs. just now

Short vs. slowly

WA HM HT NJ SW

w2(1) ¼ 1.246, w2(1) ¼ 1.516, w2(1) ¼ 0.028, w2(1) ¼ 0.030, w2(1) ¼ 0.038,

w2(1) ¼ 0.156, w2(1) ¼ 3.471, w2(1) ¼ 1.426, w2(1) ¼ 0.405, w2(1) ¼ 0.256,

w2(1) ¼ 0.395, w2(1) ¼ 0.520, w2(1) ¼ 1.426, w2(1) ¼ 0.777, w2(1) ¼ 0.645,

NS NS NS NS NS

NS NS NS NS NS

NS NS NS NS NS

Table 10 Performance (% correct) on who, what and which from three data sets Participants Who canonical (%)

Who noncanonical (%)

What canonical (%)

What noncanonical (%)

Which canonical (%)

Which noncanonical (%)

RDa FCa MDb CHb DLb FPb WA HM HT NJ SW Mean (sd)

87 87 84 80 90 60 75 75 50 75 75 76 (12)

– – 60 40 80 80 75 92 75 83 83 74 (15)

– – 70 60 85 60 92 100 42 66 92 74 (19)

87 87 76 56 80 80 92 33 92 33 83 73 (22)

47 47 76 64 50 50 33 33 8 58 33 45 (18)

a

80 93 76 60 95 90 92 92 75 75 83 83 (11)

Taken from Hickok and Avrutin (1996), no data on what questions. Taken from Tait, Thompson, and Ballard (1995).

b

between canonical and non-canonical which questions (T+ ¼ 33, po.05). Furthermore, there was no difference between canonical who and canonical which questions but there was a difference between non-canonical who and non-canonical which questions (T+ ¼ 45, po.05). So, even without HM and HT the group pattern remains. ii. The performance of nine agrammatic speakers on who and what questions is also shown in Table 10. Pairwise comparisons between canonical and non-canonical who and what questions were not significant.

ARTICLE IN PRESS C. Salis, S. Edwards / Journal of Neurolinguistics 21 (2008) 375–399

389

4. Discussion This section is organised into 3 parts. The first part discusses the results from the new data set. In the second part we discuss individual profiles. In the third part we discuss the combined data set group and the set partition hypothesis. 4.1. Discussion of the new data set results (N ¼ 5) The new data set results come from the five agrammatic speakers who participated in this study. As a group these speakers exhibited the classic fragmented speech output, they were unable to repeat sentences and had relatively high comprehension levels. They were better with canonical and poorer with non-canonical declarative sentences as both VAST and NAVS revealed. This profile conforms to the standard profile of agrammatic comprehension. This group’s performance with wh-questions also showed a statistical difference between all canonical (short, raising and padded considered together) and all non-canonical questions. Irrespective of the structure of the wh-phrase, the thematic role order determines comprehension in all types of wh-questions. This pattern is different from the pattern of the six agrammatic speakers Avrutin (2000) reported. In that data set thematic role order influenced only non-canonical which and not non-canonical who questions. However, if one considers the differences in the new data set between who/what and which questions alone (Table 3) a different pattern emerges. The difference between who/what and which was significant in both canonical and non-canonical conditions. Furthermore, there is also a difference between non-canonical who/what questions (65%) and canonical who/what (83%) questions. Similarly, there is a marked difference between canonical which (67%) and non-canonical which questions (38%). This pattern is compatible with Avrutin’s explanation for non-canonical which questions. Linking which to a referent appears to require more resources. However, integration from syntax and discourse requires even more resources than syntax or discourse alone. The different results between Avrutin’s group and our group show how small samples can generate contradictory data. We now focus on raising and padded questions. We hypothesised that: (a) comprehension of raising questions is not similar to short, non-raising questions and (b) length of wh-questions does not influence sentence comprehension in agrammatic aphasia. Table 4 shows performance by question type. The comparison between all short and all raising questions was not significant. So, comprehension of raising questions is similar to short questions. Although performance on raising questions was slightly higher than short questions, especially between short canonical which (67%) and short raising which (80%) (Table 4), none of the differences reached significance. We have no explanation for this finding. The two comparisons between all short and each type of padded questions were not significant. So, the length of wh-questions used in this study does not influence sentence comprehension in agrammatic aphasia. What do these findings reveal about our participants’ comprehension abilities? We hypothesise that although the thematic role order is the same in raising and short questions, there would be a difference between the two structures because of the different syntactic derivations and greater complexity of raising questions (cf. Goodglass et al., 1979; Haarman & Kolk, 1994; Pulvermu¨ller, 1995). The ability to represent traceantecedent relations is a crucial grammatical ability that has played a key role in research

ARTICLE IN PRESS 390

C. Salis, S. Edwards / Journal of Neurolinguistics 21 (2008) 375–399

in the field (Grodzinsky, 2000; Zurif et al., 1993). It has also played a key role in normal language processing (Shapiro, 2000). Our data show that the non-canonical thematic role order is a more important factor than the presence of intermediate traces in raising questions. Zurif et al. (1993), using an auditory-visual, cross-modal priming task showed that their four agrammatic speakers were unable to link a trace to its antecedent. However, Blumstein et al. (1998) found the opposite; their eight agrammatic speakers were able to link a trace to its antecedent. Blumstein and colleagues discuss the reasons for these discrepant findings which lie in the adopted methodology. The presence of dyslexia may have confounded the finding in Zurif et al. as in the cross-modal task the participants had to identify the words visually whereas in Blumstein et al. only the auditory modality was used. Furthermore, no baseline data on semantic facilitation were obtained by Zurif et al., which may have given rise to the absence of priming within sentences (Blumstein et al., 1998). More recent studies have shown that trace-antecedent relations are resolved, albeit in a delayed manner (Burkhardt et al., 2003; Love, Swinney, & Zurif, 2001). In particular, Burkhardt and colleagues showed that antecedent reactivation occurred 650 and 800 ms after the verb in canonical declarative sentences. Our participants were able to deal with trace-antecedent relations in who and what structures better than in which structures. Although we have used an off-line paradigm which can offer little insight into the time course of processing, the pattern we observe in wh-questions appears compatible with the findings by Love and colleagues and Burkhardt and colleagues. 4.2. Discussion of case-series results This section focuses on the performance of each of the five participants. Different patterns emerge when the results are scrutinised discretely. Beginning with performance on declarative sentences (Table 5) a clear syntactic deficit is evident in three of the five speakers. WA and NJ were poorer at understanding non-canonical sentences in both comprehension tests while for HT there was a statistical difference only on the NAVS. In the VAST, there was trend of poorer performance on non-canonical sentences which did not reach significance. HM’s comprehension was relatively high for both canonical and non-canonical sentences in the VAST although she was diagnosed as presenting with mixed non-fluent aphasia on the BDAE and having a mild to moderate comprehension deficit. However, her 65% accuracy on non-canonical sentences on the NAVS reveals difficulties with these structures. Finally, SW was roughly equally impaired with canonical and non-canonical sentences in both tests. Table 6 shows the participants’ performance by question type (canonicity and wh-phrase) in all question subtypes (short, raising and padded pooled together). First we focus on who/what questions. The difference between canonical and non-canonical who/what questions is significant in WA, HM and HT but not in NJ or SW. Interestingly, the same comparisons in short questions were not significant in any of the speakers. We now focus on which questions. The difference between canonical and non-canonical which questions was significant in WA, HM, HT and NJ but not in SW. Interestingly also, for WA and HM the percentage differences in these comparisons were wider than the differences between canonical and non-canonical who/what questions. This is also reflected in the level of significance, which was o.01 in which questions and o.05 in who/what questions. These trends suggest a more severe impairment with non-canonical which than

ARTICLE IN PRESS C. Salis, S. Edwards / Journal of Neurolinguistics 21 (2008) 375–399

391

non-canonical who/what questions. HT was equally impaired in understanding noncanonical who/what and non-canonical which questions. In NJ we see a different pattern. Although there is a non-significant difference between canonical (75%) and non-canonical (68%) who/what questions, there is a difference between canonical (46%) and noncanonical (79%) which questions. Unlike WA, HM and HT, NJ has a more severe deficit with canonical which questions and a less severe deficit with non-canonical which questions. Finally, for SW there was no significant difference between canonical and noncanonical who/what questions and between canonical and non-canonical which questions. No speaker found raising wh-questions more difficult than short questions. Similarly, padded questions in both conditions (slowly, just now) were understood at similar levels to short questions by all participants. 4.3. Discussion of combined data sets (N ¼ 11, N ¼ 9) We undertook a further analysis between, what we call, short wh-questions and other published data since we followed similar procedures in previous studies. Table 10 shows results from who, what and which questions from 11 agrammatic speakers (three data sets). In this data set there was no difference between canonical and non-canonical who questions and no difference between canonical who and canonical which questions. However, there were significant differences between canonical and non-canonical which questions and between non-canonical who and non-canonical which questions. None of the comparisons between who and what questions was significant. The crucial question that emerges from these data is why comprehension of wh-questions does not parallel the pattern found in declarative sentences. Consequently, are trace-antecedent relations a common determinant of comprehension between declarative sentences and wh-questions? In the combined data set the agrammatic speakers showed the standard agrammatic profile as they were better at canonical declarative sentences and poorer at non-canonical declarative sentences as in Hickok and Avrutin (1996) and Thompson et al. (1999). However, this difference was only evident in which questions but not in who questions. So, trace-antecedent relations appear to predict performance in which but not in who questions. We appeal to the set partition hypothesis which we introduced in Section 1.3 to account for the pattern found in the combined data sets. This hypothesis has not been triggered by the presence of new patterns in the data. The data discussed in this section are similar to the pattern presented in Avrutin (2000). Consequently, they do not disconfirm previous accounts of wh-questions. The set partition hypothesis provides a more specific account of processes involved in the comprehension of wh-questions and can be viewed as an extension of previous accounts. We believe that the sentence comprehension deficit in agrammatism does not stem from impaired representation or processing from the deep to surface structure as has been argued (Grodzinsky, 2000). We propose that it stems from a diminished processing ability that affects processing in logical form. Edwards and Lightfoot (2000) argued that grammatical knowledge remains intact in agrammatic aphasia, yet access to the grammar is intermittent. Here we build on their ideas. The term ‘‘grammar’’ in Edwards and Lightfoot appears to refer to trace-antecedent relations as they responded to Grodzinsky (2000). In this sense the grammar is intact. Here we suggest that what may not be intact is the ability to perform set partitioning

ARTICLE IN PRESS 392

C. Salis, S. Edwards / Journal of Neurolinguistics 21 (2008) 375–399

operations which are carried out by the logical form, the semantic component of the grammatical system. In government and binding theory (Chomsky, 1981; Haegeman, 1994) interpretation of sentences is largely mediated by successful representation of trace-antecedent relations between deep and surface structures. Logical form is not implicated. This aspect of the theory has driven research into agrammatism. Here, we deviate from this position and ask: What would be the consequences if one accepts that the ability to represent traceantecedent relations is intact (cf. Edwards & Lightfoot, 2000) and one postulates that agrammatic speakers have difficulties in representing structural aspects of meaning at logical form? Trace-antecedent relations influence comprehension in some, but not all, declarative sentences (cf. Balogh & Grodzinsky, 2000; Saddy, 1995) and some, but not all, wh-questions. A crucial similarity in all sentences under consideration is that they are reversible and thus have two animate thematic roles. We postulate that the thematic roles in reversible sentences belong to an overarching set (or superset) represented in logical form and this superset encompasses all other sets where animacy is a shared feature. Within this superset, there are semantic features that identify the agent and features that identify the patient. A set is defined in terms of animacy features. As to whether it can also be defined further in terms of +or human remains to be seen in further experiments. Because of diminished processing abilities the agrammatic speakers have difficulties in successfully partitioning the animacy superset in order to match the semantic features that identify the agent and the features that identify the patient. In canonical declarative sentences these demands are fewer as thematic role order follows the canonical and therefore expected pattern of agent– patient and does not impose additional computations. Agrammatic speakers are therefore more likely to identify and match the correct features to the two thematic roles and therefore interpret canonical declarative sentences. In noncanonical declarative sentences this operation can be hindered by the non-canonical, patient– agent, order of thematic roles. The probability of identifying the patient is greatly reduced as speakers expect the canonical and more frequent agent– patient order (cf. Bever, 1970). Due to diminished processing ability for set partitioning operations semantic feature identification is incorrect as the features of the agent are matched to the patient. Our data support this explanation. It may appear at first that the set partitioning demands in an animacy superset are qualitatively different from the notion of selecting a member of a set in the case of which questions. In the set partition hypothesis there two main constructs: The animacy superset in reversible sentences and the extent to which a phrase is modified by an element that either signals partition, for example which, or not, for example, every (Section 4.3.1 below). So, these elements either increase or decrease set partitioning demands in the overarching superset and consequently determine success rates in sentence comprehension. Although it could be argued that this is qualitatively different from the animacy superset it does not influence the way the hypothesis works. We maintain the animacy aspect in our hypothesis because it is crucial in explaining comprehension patterns in declarartive sentences and whquestions. Which questions do not only select an animacy entity but they restrict the animacy features that identify that entity. Which questions do formulate a more restricted set (i.e. partition of a larger set). Agrammatic speakers are unable to make the partition and identify the features in that restricted/partitioned set. Table 11 summarises the set partitioning demands that different types of sentences introduce. Below we discuss other data patterns that strengthen our hypothesis.

ARTICLE IN PRESS C. Salis, S. Edwards / Journal of Neurolinguistics 21 (2008) 375–399

393

Table 11 Summary of set partitioning demands Sentence types Declarative sentences Who/what questions Which questions

Interpretive demands Canonical Non-canonical Canonical Non-canonical Canonical Non-canonical

Simple syntax, animacy Complex syntax, animacy+/partitioning demand Simple syntax, animacypartitioning demand Complex syntax, animacypartitioning demand Simple syntax, animacy+partitioning demand Complex syntax, animacy+partitioning demand

4.3.1. The set partition hypothesis and other data Agentive passive sentences, e.g. a man was photographed by a child, are difficult for agrammatic speakers, yet, quantified passives, e.g. every man was photographed by a child, are understood well (Balogh & Grodzinsky, 2000; Saddy, 1995). This finding shows that every facilitates sentence comprehension despite the inability of these speakers to deal with non-canonical thematic role order and trace-antecedent relations in sentences without every. However, this finding is not consistent (Edwards & Varlokosta, 2007; Edwards, Varlocosta, & Payne, 2003). Their agrammatic speakers did not understand quantified active sentences with pronouns. Using a truth-value judgement task, they compared only reversible active sentences in phrases with a universal quantifier every in agent position and a pronoun as a theme, e.g. is every mother looking at her? Saddy (1995) and Balogh and Grodzinsky (2000) compared agentive and quantified passive sentences without pronouns. So, sentences with universal quantifier with and without pronouns may be processed differently. These studies highlight the need to recruit new or expanded explanations to account for new findings. As briefly discussed earlier, sentence length does not influence sentence comprehension in people with agrammatic aphasia. Inserting additional lexical items (lexical padding) in a sentence contributes very little to sentence meaning, at least as required in the experimental tasks. Items that do not signal a contrastive difference between padded sentences and unpadded sentence do not influence sentence comprehension (Berndt et al., 1997; Friedmann & Gvion, 2003; Love & Oster, 2002; Miera & Cuetos, 1998). Intuitively, the presence of additional lexical items would seem to increase the ‘‘compositionality’’ of the sentence. Arguably the padded sentences are more compositional than unpadded sentences. However, the additional lexical items were not making a significant contribution to sentence meaning. For instance, in the study by Friedmann and Gvion (2003) the three Hebrew-speaking agrammatic speakers had to match a sentence to pictures. An example of a padded sentence was: This is the king with the long gown, the golden crown and nice moustache that the boy combs. The unpadded version was this is the king that the boy combs. The two pictures were similar in terms of the depicted characteristics of the king and the boy. Only the thematic role was reversed so there was little difference in the identifying features of the king that would have influenced comprehension. Consequently, the contrastive meaning of the sentence in terms of the additional prepositional phrases did not change. If however, there were contrastive differences, for example, kings with different coloured gowns and rusty crowns, the set partition hypothesis would predict that those sentences would be more compositional than unpadded sentences. In this case higher error rates would be expected.

ARTICLE IN PRESS 394

C. Salis, S. Edwards / Journal of Neurolinguistics 21 (2008) 375–399

Frawley (1992) draws a distinction on English quantifiers between total and partial quantities. All, every and each are total or collective quantifiers whereas partial quantities are some, a few, most, many, and several. Although which is not a quantifier there is a similarity in meaning between which phrases and partial quantifiers as they all call for a partition from a set, unlike who/what and every that do not call for this operation. The partial quantifiers discussed by Frawley denote partial quantity and therefore partition from a set in terms of quantity while which denotes partition from a set in terms of identity. The shared property between partial quantifiers and which is on set partition. If one accepts this similarity in meaning, then, clear predictions can be formulated for phrases modified by an element that signals partition from a set. So, phrases with partial quantifiers and which phrases increase processing load, while phrases that do not signal partition from a set, that is, who/what and all, every, each do not. Data from comprehension of quantified passives is limited and no speaker has been tested on all structures discussed here. The set partition hypothesis predicts that noncanonical sentences with who, what, all, every and each would be understood better than noncanonical sentences without these elements. In the combined data set, only the difference between canonical and non-canonical which questions was significant. So, a more compositional phrase or sentence would be more complex or in lay terms ‘‘harder’’ to process. In turn, identifying the component parts, that is, partitioning demands that make up this complexity would increase the probability of errors. Frawley (1992, p. 468) classifies the indefinite article a together with partial quantifiers because it means ‘‘one or more,’’ so it makes essential reference to the rest of a collection. Prima facie the prediction for NPs headed by a according to the set partition hypothesis would be the same as for partial quantifiers and would therefore require further processing resources. Consequently, error rates would be predicted to increase in such sentences. However, Heim (1982, p. 127) argued that indefinite expressions have no quantificational force of their own but are bound by nearby quantifiers. She gives the following example to illustrate this point: ‘‘If a table has lasted for 50 years it will last for another 50’’. The paraphrase of this sentence is: ‘‘Every table that has lasted 50 years will last for another 50’’. In this case a is similar to every so no additional processing cost would arise.2 However, this may not be the case because in the paraphrase there is, in our view, a crucial set partitioning operation introduced not by every but the relative clause which modifies and restricts/partitions the set that every refers to; every table that has lasted 50 years could be interpreted as partitioning a large set of tables. The examples in (3) clarify this difference. Assuming a large set of tables of different ages we could propose the following: Every table that has lasted 50 years will last for another 50

(3a)

Every table that has lasted 10 years will last for another 30

(3b)

If we accept this argument then Heim’s example is not an accurate paraphrase and consequently the indefinite a should be treated in a different way from every. Nonetheless, this issue highlights the complexity and diversity of set partitioning demands that can be imposed by quantifiers and the way they interact with other elements in a sentence. What are, then, the predictions for the definite article the? The has a more restrictive meaning, similar to those of partial quantifiers (cf. Heim & Kratzer, 1998). So, the prediction for NPs headed by the would be the same as per partial quantifiers and the 2

We thank one of the reviewers who brought this issue to our attention.

ARTICLE IN PRESS C. Salis, S. Edwards / Journal of Neurolinguistics 21 (2008) 375–399

395

would signal set partitioning demands and error rates would be higher. Following this line of argument we see that although a and the belong semantically to two distinct classes of indefinites and definites, respectively, the predictions formed by the set partition hypothesis are the same. The limited research that has been undertaken on the comprehension of quantified expressions in agrammatism suggests that there is no processing difference between NPs headed by a and the. The experimental sentences in Saddy (1995) contained indefinite NPs, e.g. every man was photographed by a child, and the sentences in Balogh and Grodzinsky (2000) contained definite NPs, e.g. every man was photographed by the child. There is no evidence in the data that this difference influenced comprehension of sentences although the absence of difference could be due to task effects. Finally, we consider another comprehension pattern which strengthens our hypothesis. Non-reversible sentences, that is, sentences with one inanimate and one animate thematic role, facilitate comprehension in agrammatic speakers (Burchert et al., 2003; Caramazza & Zurif, 1976). We put forward a different explanation. The two thematic roles in nonreversible sentences belong to two different sets, animate and inanimate, where semantic features are already segregated and there is no animacy superset. This may be a factor additional to lexical knowledge (cf. Caramazza & Zurif, 1976) that contributes to the good comprehension on non-reversible non-canonical sentences. 5. Summary and conclusions In this paper we investigated the comprehension of wh-questions in an experimental group (N ¼ 5) made up from our participants and six other participants taken from published studies. We compared comprehension of wh-questions and declarative sentences exploring the effect of syntactic complexity, structure of wh-phrases, raising and sentence length. We have argued that the underlying comprehension deficit in agrammatic aphasia is common between declarative sentences and wh-questions and stems from an inability to reliably implement semantic aspects of grammatical knowledge. We proposed the set partition hypothesis to unify diverse sentence comprehension patterns. According to this hypothesis a component of processing load is the set partitioning semantic operations. We are aware that this deviates from the accepted views regarding the function and purpose of the logical form as this offers an explanatory account for sentence comprehension patterns. Grodzinsky (2000) pointed out that no experiments have investigated the status of logical form in people with agrammatic aphasia. We believe we have made a healthy start in unravelling the potential advantage of considering the implications of logical form in agrammatism and we have offered testable predictions that advance our understanding of the underlying mechanism in agrammatic aphasia. Acknowledgements We would like to thanks the participants and clinicians in the following NHS (UK) trusts: Scottish Borders, Tayside and Berkshire West. We would also like to thank Drs. Miseon Lee, Vesna Stojanovik, Spyridoula Varlokosta and professors Jane Marshall and Stephanie Stokes for their helpful comments and guidance. We also thank the two anonymous reviewers who made useful comments and suggestions. An earlier version of this paper was presented at the Academy of Aphasia 2005 meeting (Amsterdam). This project was funded by the Economic and Social Research Council (UK), PTA 030-2003-01072.

ARTICLE IN PRESS C. Salis, S. Edwards / Journal of Neurolinguistics 21 (2008) 375–399

396

Appendix A BDAE group and individual percentiles are shown in Table A1. Table A1 BDAE subtests Severity rating Fluency

Conversation Auditory comprehension

Articulation Recitation Repetition Naming

Phrase length Melodic line Grammatical form Simple social responses Basic word discrimination Commands Complex ideational material Articulatory agility Automatised sequences Words Sentences Responsive naming Boston naming test Special categories

Mean percentiles

WA

HM

HT

NJ

SW

60 18 34 26 68 56 54 50 26 80 50 34 70 60 64

60 20 60 30 90 60 70 20 10 90 40 10 90 60 90

40 10 30 10 20 40 20 70 10 90 50 0 70 70 80

70 20 30 30 90 60 40 20 30 100 70 60 90 50 90

60 20 30 30 90 60 70 90 40 90 60 50 70 90 30

70 20 20 30 50 60 70 50 40 30 30 50 30 30 30

Appendix B Speech and language characteristics of participants are shown in Table B1. Table B1 Speech & language characteristics a

Speech rate MLU (words) Grammatical sentences (%) Simple sentences (%) Complex sentences (%) a

Control group, mean

87 7.2 84.4 47.8 52.2

Experimental group, mean

34 3.5 19.86 95.3 4.7

Individual participants WA

HM

HT

NJ

SW

34 3.4 18 80 20

30 3.5 30 100 0

42 3.2 10 96.5 3.5

29 3.8 26.3 100 0

32 3.6 15 100 0

Words per minute.

References Ansell, B. J., & Flowers, C. R. (1982). Aphasic adults’ use of heuristic and structural linguistic cues for sentence analysis. Brain and Language, 16, 61–72. Avrutin, S. (2000). Comprehension of discourse-linked and non-discourse-linked questions by children and Broca’s aphasics. In Y. Grodzinsky, L. Shapiro, & D. Swinney (Eds.), Language and the brain: Representation and processing (pp. 295–313). San Diego, CA: Academic Press. Balogh, J., & Grodzinsky, Y. (1996). Varieties of passives in agrammatic Broca’s aphasia: y-grids, arguments and referentiality. Brain and Language, 55, 54–56.

ARTICLE IN PRESS C. Salis, S. Edwards / Journal of Neurolinguistics 21 (2008) 375–399

397

Balogh, J. E., & Grodzinsky, Y. (2000). Levels of linguistic representation in Broca’s aphasia: Implicitness and referentiality of arguments. In R. Bastiaanse, & Y. Grodzinsky (Eds.), Grammatical disorders in aphasia: A neurolinguistic perspective (pp. 88–104). London: Whurr. Bartsch, R. (1973). The semantics and syntax of number and numbers. In J. P. Kimball (Ed.), Syntax and semantics, Vol. 2 (pp. 51–93). New York: Seminar Press. Bastiaanse, R., & Edwards, S. (2004). Word order and finiteness in Dutch and English Broca’s and Wernicke’s aphasia. Brain and Language, 89, 91–107. Bastiaanse, R., Edwards, S., & Rispens, J. (2002). Verb and sentence test. Bury St. Edmunds, UK: Thames Valley Test Company. Berndt, R. S., Mitchum, C. C., & Wayland, S. (1997). Patterns of sentence comprehension in aphasia: A consideration of three hypotheses. Brain and Language, 60, 197–221. Bever, T. G. (1970). The cognitive basis for linguistic structures. In J. R. Hayes (Ed.), Cognition and the development of language (pp. 279–362). New York: Wiley. Blumstein, S. E., Milberg, W. P., Dworetzky, B., Rosen, A., & Gershberg, F. (1991). Syntactic priming in aphasia: An investigation of local syntactic dependencies. Brain and Language, 40, 393–421. Blumstein, S. E., Byma, G., Kurowski, K., Hourihan, J., Brown, T., & Hutchison, A. (1998). On-line processing of filler-gap constructions in aphasia. Brain and Language, 61, 149–168. Bolinger, D. (1965). The atomization of meaning. Language, 41, 555–573. Brookshire, R. H., & Nicholas, L. E. (1980). Verification of active and passive sentences by aphasic and nonaphasic subjects. Journal of Speech and Hearing Research, 23, 878–893. Burchert, F., De Bleser, R., & Sonntag, K. (2003). Does morphology make the difference? Agrammatic sentence comprehension in German. Brain and Language, 87, 323–342. Burkhardt, P., Pin˜ango, M. M., & Wong, K. (2003). The role of the anterior left hemisphere in real-time sentence comprehension: Evidence from split intransitivity. Brain and Language, 86, 9–22. Burzio, L. (1986). Italian syntax. Dordrecht, the Netherlands: Reidel. Caplan, D., Baker, C., & Dehaut, F. (1985). Syntactic determinants of sentence comprehension. Cognition, 21, 117–175. Cappa, S. F., Moro, A., Perani, D., & Piattelli-Palmarini, M. (2000). Broca’s aphasia, Broca’s area, and syntax: A complex relationship. Brain and Behavioral Sciences, 23, 27–28. Caramazza, A., & Zurif, E. B. (1976). Dissociation of algorithmic and heuristic processes in language comprehension: Evidence from aphasia. Brain and Language, 5, 572–582. Chomsky, N. (1979). Language and responsibility. New York: Pantheon. Chomsky, N. (1981). Lectures on government and binding. Dordrecht, the Netherlands: Foris. Chomsky, N. (1986). Knowledge of language, its nature, origin, and use. New York: Praeger. Chomsky, N. (1995). The minimalist program. Cambridge, MA: MIT Press. Deloche, G., & Seron, X. (1981). Sentence understanding and knowledge of the world: Evidence from a sentence–picture matching task performed by aphasic patients. Brain and Language, 14, 57–69. Dickey, M. W., & Thompson, C. K. (2004). The resolution and recovery of filler-gap dependencies in aphasia: Evidence from on-line anomaly detection. Brain and Language, 88, 108–127. Edwards, S. (2005). Fluent aphasia. Cambridge, UK: Cambridge University Press. Edwards, S., & Lightfoot, D. (2000). Intact grammar but intermittent access. Brain and Behavioral Sciences, 23, 31–32. Edwards, S., & Varlokosta, S. (2007). Pronominal and anaphoric reference in agrammatism. Journal of Neurolinguistics, 20, 423–444. Edwards, S., Varlocosta, S., & Payne, E. (2003). Pronominal reference and agrammatic comprehension. Brain and Language, 87, 21–22. Fodor, J. A., & Lepore, E. (2002). The compositionality papers. Oxford, UK: Clarendon Press. Frawley, W. (1992). Linguistic semantics. Hove, UK: Laurence Erlbaum Associates. Friederici, A. D., & Gorrell, P. (1998). Structural prominence and agrammatic theta-role assignment: A reconsideration of linear strategies. Brain and Language, 65, 253–275. Friederici, A. D., & Kilborn, K. (1989). Temporal constraints on language processing: Syntactic priming in Broca’s aphasia. Journal of Cognitive Neuroscience, 1, 262–272. Friedmann, N., & Gvion, A. (2003). Sentence comprehension and working memory limitation in aphasia: A dissociation between semantic–syntactic and phonological reactivation. Brain and Language, 86, 23–39. Gallagher, T. M., & Guilford, A. M. (1977). Wh-questions: Responses by aphasic patients. Cortex, 13, 44–54.

ARTICLE IN PRESS 398

C. Salis, S. Edwards / Journal of Neurolinguistics 21 (2008) 375–399

Goodglass, H., Kaplan, E., & Barresi, B. (2001). The Boston diagnostic aphasia examination (3rd ed.). Baltimore: Lippincott, Williams & Wilkins. Goodglass, H., Blumstein, S. E., Berko Gleason, J., Hyde, M. R., Green, E., & Statlander, S. (1979). The effect of syntactic encoding on sentence comprehension in aphasia. Brain and Language, 7, 201–209. Grimshaw, J. (1990). Argument structure. Cambridge, MA: MIT Press. Grodzinsky, Y. (1986). Language deficits and the theory of syntax. Brain and Language, 27, 135–159. Grodzinsky, Y. (1995). A restrictive theory of agrammatic comprehension. Brain and Language, 50, 27–51. Grodzinsky, Y. (1998). Comparative aphasiology: Some preliminary notes. In E. Visch-Brink, & R. Bastiaanse (Eds.), Linguistic levels in aphasia (pp. 175–192). San Diego, CA: Singular. Grodzinsky, Y. (2000). The neurology of syntax: Language use without Broca’s area. Behavioral and Brain Sciences, 23, 1–71. Haarman, H. J., & Kolk, H. H. J. (1991). Syntactic priming in Broca’s aphasics: Evidence for slow activation. Aphasiology, 5, 247–263. Haarman, H. J., & Kolk, H. H. J. (1994). On-line sensitivity to subject–verb agreement violations in Broca’s aphasics: The role of syntactic complexity and time. Brain and Language, 46, 493–516. Haegeman, L. (1994). Introduction to government and binding (2nd ed.). Oxford, UK: Blackwell. Harley, T. A. (2001). The psychology of language: From data to theory (2nd ed.). Hove, UK: Psychology Press. Heim, I. (1982). The semantics of definite and indefinite noun phrases. PhD dissertation, University of Massachusetts, Amherst, MA, unpublished. Heim, I., & Kratzer, A. (1998). Semantics in generative grammar. Oxford, UK: Blackwell. Hervey, S. G. (1979). Axiomatic Semantics: A theory of linguistic semantics. Edinburgh: Scottish Academic Press. Hickok, G., & Avrutin, S. (1996). Comprehension of wh-questions in two Broca’s aphasics. Brain and Language, 52, 314–327. Hornstein, N. (1995). Logical form: From GB to minimalism. Oxford, UK: Blackwell. Jackendoff, R. (1997). The architecture of the language faculty. Cambridge, MA: MIT Press. Johansson, S., & Hofland, K. (1989). Frequency analysis of English vocabulary and grammar, Vol. 1. Oxford, UK: Clarendon Press. Katz, J. J., & Fodor, J. A. (1963). The structure of a semantic theory. Language, 39, 170–210. Kerschensteiner, M., Poeck, K., & Brunner, E. (1972). The fluency–non-fluency dimension in classification of aphasic speech. Cortex, 8, 233–247. Kolk, H. H. J., & Friederici, A. D. (1985). Strategy and impairment in sentence understanding by Broca’s and Wernicke’s aphasics. Cortex, 21, 47–67. Koopman, H., & Sportiche, D. (1991). The position of subjects. Lingua, 85, 211–258. Lee, M., & Thompson, C. K. (2004). Agrammatic aphasic production and comprehension of unaccusative verbs in sentence contexts. Journal of Neurolinguistics, 17, 315–330. Linebarger, M. (1995). Agrammatism as evidence about grammar. Brain and Language, 50, 52–91. Love, T., & Oster, E. (2002). On the categorization of aphasic typologies: The SOAP (a test of syntactic complexity). Journal of Psycholinguistic Research, 31, 503–529. Love, T., Swinney, D., & Zurif, E. (2001). Aphasia and the time-course of processing long distance dependencies. Brain and Language, 79, 169–170. Lyons, J. (1995). Linguistic semantics: An introduction. Cambridge, UK: Cambridge University Press. May, R. (1985). Logical form: Its structure and derivation. Cambridge, MA: MIT Press. Miera, G., & Cuetos, F. (1998). Understanding disorders in agrammatic patients: Capacity or structural deficits. Brain and Language, 64, 328–338. Montague, R. (1974). Formal philosophy: Selected papers of Richard Montague. New Haven: Yale University Press. Myers, E. B., & Blumstein, S. E. (2005). Selectional restriction and semantic priming effects in normals and Broca’s aphasics. Journal of Neurolinguistics, 18, 277–296. Neuhaus, E., & Penke, M. (in press). Production and comprehension of wh-questions in German Broca’s aphasia. Journal of Neurolinguistics, doi:10.1016/j.neuroling.2007.05.001. Nickels, L., Byng, S., & Black, M. (1991). Sentence processing deficits: A replication therapy. British Journal of Disorders of Communication, 26, 175–199. O’ Grady, W., & Lee, M. (2005). A mapping theory of agrammatic comprehension deficits. Brain and Language, 92, 91–100. Ostrin, R. K., & Tyler, L. K. (1995). Dissociations of lexical function: Semantics, syntax and morphology. Cognitive Neuropsychology, 12, 345–389.

ARTICLE IN PRESS C. Salis, S. Edwards / Journal of Neurolinguistics 21 (2008) 375–399

399

Park, G. H., McNeil, M. R., & Doyle, P. J. (2002). Lexical access rate of closed-class elements during auditory sentence comprehension in adults with aphasia. Aphasiology, 16, 801–814. Pesetsky, D. (1987). Wh-in-situ: Movement and unselective binding. In E. J. Reuland, & A. G. B. ter Meulen (Eds.), The representation of (in)definiteness (pp. 98–129). Cambridge, MA: MIT Press. Pin˜ango, M. M. (2000). Canonicity in Broca’s sentence comprehension: The case of psychological verbs. In Y. Grodzinsky, L. Shapiro, & D. Swinney (Eds.), Language and the Brain: Representation and processing (pp. 327–350). San Diego, CA: Academic Press. Pin˜ango, M. M. (2006). Thematic roles as event structure relations. In I. Bornkessel, M. Schlesewsky, B. Comrie, & A. D. Friederici (Eds.), Semantic role universals and argument linking: Theoretical, typological and psycholinguistic perspectives (pp. 303–326). Berlin: De Gruyter. Pin˜ango, M. M., & Zurif, E. (2001). Semantic combinatorial operations in aphasic comprehension: Implications for the cortical organization of language. Brain and Language, 79, 297–308. Pulvermu¨ller, F. (1995). Agrammatism: Behavioural description and neurobiological explanation. Journal of Cognitive Neuroscience, 7, 165–181. Riley, L. (1988). Aphasic patients’ responses to wh-questions. Doctoral dissertation, University of Missouri, Columbia, unpublished. Saddy, D. (1995). Variables and events in the syntax of agrammatic speech. Brain and Language, 50, 135–150. Saffran, E. M., & Schwartz, M. F. (1988). ‘‘Agrammatic’’ comprehension it’s not: Alternatives and implications. Aphasiology, 2, 389–394. Salis, C. (2006). Comprehension of wh-questions and declarative sentences in agrammatic aphasia. PhD dissertation, University of Reading, unpublished. Shapiro, L. P. (2000). The processing of long distance dependencies in normal listeners: Evidence for form-driven activation. In R. Bastiaanse, & Y. Grodzinsky (Eds.), Grammatical disorders in aphasia: A neurolinguistic perspective (pp. 35–50). London: Whurr. Swinney, D., Zurif, E., Prather, P., & Love, T. (1996). Neurological distribution of processing resources underlying language comprehension. Journal of Cognitive Neuroscience, 8, 174–184. Tait, M. E., Thompson, C. K., & Ballard, K. (1995). Subject-object asymmetries in agrammatic comprehension. Brain and Language, 51, 77–79. Thompson, C. K. (unpublished a). The Northwestern narrative language sample analysis: Theory and methodology. Northwestern University, IL. Thompson, C. K. (unpublished b). The Northwestern noun, verb and sentence test. Northwestern University, IL. Thompson, C. K., Tait, M. E., Ballard, K. J., & Fix, S. C. (1999). Agrammatic aphasic subjects’ comprehension of subject and object extracted wh questions. Brain and Language, 67, 169–187. Van der Meulen, I., Bastiaanse, R., & Rooryck, J. (2002). Wh-movement in French agrammatism. Brain and Language, 83, 184–187. Van Riemsdijk, H., & Williams, E. (1981). NP-structure. The Linguistic Review, 1, 171–218. Werning, M., Machery, E., & Schurz, G. (Eds.). (2005). The compositionality of meaning and content: Vol I. Foundational issues. Frankfurt: Ontos Verlag. Williams, E. (1987). A reassignement of the functions of LF. Linguistic Inquiry, 17, 265–299. Zurif, E., Swinney, D., Prather, P., Solomon, J., & Bushell, C. (1993). An on-line analysis of syntactic processing in Broca’s and Wernicke’s aphasia. Brain and Language, 45, 448–464.