• Keine Ergebnisse gefunden

2.5 The syntactic competence in spontaneous speech

2.5.4 Relative clause comprehension

In order to investigate sentence processing in PADs, linguists take into consideration syntactic complexity as a relevant factor. In particular, they aim at verifying whether the performance of PADs decreases in correlation with factors of syntactic complexity. Two structures have often be implied in order to test the effects of syntactic complexity: passive voice and relative clauses. I considered the former in the previous section; the latter will be the object of the present one.

Background literature provides at least four studies that include relative clauses among their experimental conditions: Bickel et al. (2000); Kempler et al. (1998); Small, Kemper & Lyons (2000); and Waters, Rochon & Caplan (1998). Those works have

50

already been reviewed in the present study because of their contribution to the investigation of the computation of passive. In what follows, I will therefore review the papers again with a specific focus on relative clauses, in order to summarize what is known so far about the computation of relative clauses in PADs, and what the results can tell us about the syntactic competence in Alzheimer’s disease.

Waters, Rochon & Caplan (1998) deny the presence of a syntactic deficit in PADs.

However, in my opinion, the reasons for this strong claim stem from authors’ definition of syntactic complexity and the design of the experimental conditions in use. Waters, Rochon & Caplan (1998) engage participants in a sentence-to-picture matching task, which includes nine different conditions: active sentences (9a), active sentences with a conjoined DP in the object position (9b), active sentences with a direct object and a prepositional phrase (9c), passive sentences (9d), truncated passive sentences (9e), objet clefts (9f), right-branching subject relative clauses (9g), two conjoined sentences (9h) and left-branching object relatives (9i)14:

(9) a. The lion kicked the elephant

b. The pig chased the lion and the cow

c. The elephant pulled the dog to the horse

d. The elephant was pushed by the cow

e. The pig was touched

f. It was the dog that the horse passed

g. The horse kicked the elephant that touched the dog

h. The elephant followed the lion and pulled the dog

14 Examples are reported from Waters et al. (1998: 369).

51

i. The dog that the pig followed touched the horse

Patients repeat the test several times in slightly different conditions, which vary according to the supplied visual material. They see either two or three pictures and must point to the one that correctly represents the action described by the experimental sentence. In two different versions of the test, patients see either one picture or one short video and must decide whether these correctly depict the sentence meaning. All versions of the test provide the same results: with respect to the baseline condition (9a), patients are significantly less accurate in the comprehension of sentences that include two conjoined clauses (9h) and in sentences with a relative clause (in (9g) and (9i)).

Authors consider three factors for the interpretation of the results. These are canonicity of theta-role assignment, number of arguments in the clause and number of predicates. The first two factors represent syntactic complexity and their manipulation allow the distinction between active and passive sentences, and between sentences with two arguments (subject and object as in (9a)) or with three arguments (one subject and two objects as in (9b)), or subject, object and indirect object as in (9c)). The third factor, namely the number of predicates in the clause, is considered only as a factor for sentence length, but not for syntactic complexity. Authors intend to use it as a test for working memory, thus assuming that two predicates imply longer sentences and therefore higher processing workload for working memory. Based on these premises, Waters et al. (1998) conclude that PADs deal well with syntactic complexity because they show a good comprehension both of active and passive sentences, both with two or three arguments.

However, patients’ poor comprehension on relative clauses is interpreted as the reflex of sentence length and it is ascribed to poor working memory. In their discussion, authors completely disregard the syntactic derivation that characterizes relative clauses and the asymmetry between matrix and embedded clauses. They only evaluate sentence length as a predictor for sentence comprehension: in their view, sentences including relative clauses are more difficult to comprehend because they are usually longer than simple sentences.

Contra to authors’ definition of syntactic complexity, I will claim that relativization is precisely a factor of syntactic complexity. In my view, the comprehension of relative clauses posits some difficulties, which stem from the syntactic derivation per se, rather

52

than from the number of words in the sentence (see Chapter 5 on the derivation of Relative Clauses).

Unfortunately, the only two relatives in use are a right-branching subject relative and a left-branching object relative. The two conditions in use differ in two factors, namely relativization site (subject vs object relative clauses) and direction of branching (left- vs right-branching clauses). The study therefore lacks minimal pairs for the evaluation of the single components, such that it is not possible to compare subject and object relatives and left- and right-branching as separate factors. Based on the experimental material and the results at hand, the only information we can infer from Waters et al. (1998) is that PADs are more impaired at comprehending relative clauses than at comprehending simple sentences15.

Kempler et al. (1998) also investigate patients’ ability with a sentence to picture matching task. Authors aim at evaluating two factors by including four kinds of sentences.

The four experimental conditions comprehend active sentences (10a), active sentences with two conjoined DPs in the object position (10b), passive sentences (10c) and relative clauses (10d)16:

(10) a. The boy pushes the girl

b. The boy scratches the dog and the cat

c. The boy is kissed by the girl

d. The dog chases the girl that chases the boy

Kempler et al. (1998) design the experimental conditions above in order to observe two factors: syntactic complexity (simple active vs passive, and simple active vs relative clauses) and the number of arguments (two in (10a) and (10b) vs three in (10b) and (10d)).

15 The study also reveals that the number of arguments in the experimental sentences and the number of pictures displayed in the visual stimuli have an impact on patients’ successful comprehension. However, the two issues are not of primary interest for the present discussion and therefore will not be further addressed in this section.

16 Examples are reported from Kempler et al. (1998: 301).

53

Results report the highest level of accuracy on simple active sentences and the lowest one on relative clauses. Intermediate levels of successful comprehension correspond to the two remaining conditions, with no significant differences between them. According to the described pattern of performance, authors conclude that both considered factors are relevant and influence patients’ rate of comprehension. Relative clauses are more difficult than simple active sentences because they are complex from the syntactic point of view; moreover, sentences with three arguments are more difficult than sentences with two arguments. Unfortunately, the experimental material does not allow the evaluation of fine-grained syntactic factors within relative clauses. For instance, authors report to have included subject relatives both of the centre-embedded and of the right-branching kind in their experimental material, but then they collapse the two kinds into one single condition for data analysis.

Bickel et al. (2000) attempt to evaluate different factors internal to relative clauses in their sentence to picture matching task performed by German-speaking PADs. Their study includes fourteen different sentences, among which there are four kinds of relative clauses: centre-embedded subject relatives (11a), centre-embedded object relatives (11b), centre-embedded subject relatives with intransitive verbs (11c), and right-branching subject relatives (11d)17:

(11) a. Das Abendkleid, das auf dem Anzug liegt, ist aus blauer Seide 'The robe that lies on the suit is made up of blue silk.'

b. Das Abendkleid, auf dem der Anzug liegt, ist aus blauer Seide 'The robe, on which the suit is lying, is made up of blue silk. '

c. Der Junge, der im Auto sitzt, unterhält sich mit dem Mann 'The boy, who is sitting in the car, is talking with the man.'

d. Der Junge unterhält sich mit dem Mann, der im Auto sitzt 'The boy is talking with the man sitting in the car.'

17 Examples are adapted from Bickel et al. (2000: 435).

54

Bickel et al. (2000) subdivide their patients into two groups according to their level of dementia. Patients who score above 20/30 in the Mini Mental-State Examination Test (Folstein et al., 1975)18 are classified as mildly impaired. Participants with scores lower than 20 on the same test form the group of patients with moderate impairment. Both groups show a preference for subject relatives over object relatives. In particular, centre-embedded object relatives correspond to the condition with the lowest level of accuracy (at-chance performance). In contrast, comprehension of centre-embedded subject relatives is good: moderately impaired PADs perform above chance level and mildly impaired PADs do not differ from controls. What actually surprises in the results is the fact that within the subject relative condition, PADs perform better on centre-embedded than on right-branching clauses. The result is rather unattended from the syntactic point of view because it seems to indicate that the processing of matrix clauses benefits from the interposition of a centre-embedded clause (thus leaving the initial NP stored in the working memory without an assigned role in the matrix clause). Bickel et al. (2000) interpret the result as an effect of task performance. In the case of centre-embedded relative clauses, the task can be completed in a shorter time because the information necessary for the picture selection is provided by the embedded clauses. Participants can therefore perform the task without completing the processing of the matrix clauses. In contrast, right-branching relative clauses require the complete processing of both the matrix and the subordinate clause prior to the selection of the target image. Based on this observation, Bickel et al. (2000) also resort to limited working memory in order to explain the pattern of performance.

Overall, the main result of the reviewed study consists in the differences in performance between mild and moderate patients. The first group performs similarly to controls, while the second one performs poorly. I will try to examine in depth the issue in Chapter 6 and 7, in order to achieve a better understanding of how patients’ syntactic competence changes along with the worsening of the disease.

The fourth study I am going to review in this section represents a further attempt to disentangle two factors of syntactic manipulation in relative clauses, in order to comprehend their role in sentence processing. Small, Kemper & Lyons (2000) enrol their

18 The Mini Mental State Examination Test (Folstein et al., 1975) is a test for the evaluation of cognitive impairment. Section 4.2.2 contains more information on it.

55

patients in a sentence repetition task. The factors at stake for relatives are canonicity of theta-role assignment (subject vs object relatives) and branching direction (centre-embedded vs right-branching relatives). Both factors are crossed twice in order to obtain four conditions: centre-embedded subject relatives (12c), centre-embedded object relatives (12d), right-branching subject relatives (12e) and right-branching object relatives (12f). The task also includes simple active sentences (12a) and passive structures (12b)19:

(12) a. The circus at the convention centre attracted thousands of children

b. The motorist was stopped by the highway patrol

c. The tornado that swept through the town destroyed several homes

d. The job that the woman wanted required a college education

e. The operator assisted the merchant who made a long distance call

f. The angry parent disciplined the student who the teacher sent home

Small et al. (2000) calculate percentages of accurate repetition for each condition and find an unattended pattern of performance. PADs repeat passive sentences and right-branching subject relatives better than simple active sentences. In turn, simple active sentences are repeated more accurately than the three other conditions with a relative clause. Such a poor repetition of simple active sentences is most probably due to the prepositional phrase, which modifies the subject DP and delays verb processing. Moreover, the PP increases the mean length of simple active sentences, a rather relevant factor for a sentence repetition task.

As for relative clauses, authors observe that subject relatives are usually better processed than object relatives. However, the facilitating effect disappears in case of centre-embedding. PADs usually prefer right-branching over centre-embedding, but the

19 Examples are reported from Small et al. (2000:236).

56

asymmetry is nullified in case of object relatives. The interpretation of results is made difficult by the crossing of the two factors, which seem to interact but not in an additive way. If the two factors played an additive effect, results would report the lowest accuracy on centre-embedded object relatives, but this is not the case (the poorest performance is actually registered on centre-embedded subject relatives). Overall, Small et al. (2000) fail at clearly disentangling the effect brought in by the two factors. They observe recency effects in patients’ performance as these accurately repeat the first part of the sentence, while they fail at repeating the final one. The discussion therefore brings into play once again working memory effects, but lacks the analysis of strictly syntactic factors. Authors ascribe the pattern of performance to limited resource capacity. When the request for information activation exceeds the available pool of resources, the task can be executed only partially.

Altogether, the reviewed studies point out a specific impairment in the processing of relative clauses in comparison to simple active sentences. The observation immediately turns relative clauses into an interesting point of observation to gain insights into the syntactic competence of PADs. However, the studies mentioned above fail at offering the opportunity to analyse the impairment in details. This is partially due to characteristics of the experimental designs in use. Conditions often lack minimal pairs for the evaluation of specific factor manipulations (Kempler et al., 1998; Waters et al., 1998). For instance, in Kempler et al. (1998) centre-embedded and right-branching relatives are collapsed into one condition although they are also characterized by different extraction sites. In the latter case, right-branching SRs are compared to centre-embedded ORs, but the analysis is made difficult by the impossibility of disentangling the two factors (extraction site and branching direction). Moreover, it is not always clear whether authors control for potential confounding factors like, for instance, animacy and word frequency. Animacy (and therefore sentence reversibility) is crucial for the computation of passive sentences and of ORs (see Chapter 5 on this). Word frequency cannot be overlooked in studies with patients that suffer from severe anomia, because low-frequency words might compromise sentence comprehension and the results. For instance, Small et al. (2000) report the full list of sentences they use: each sentence contains different lexical items, such that the differences between active sentences and relatives might be due to lexical items rather than different syntactic derivations. Finally, the mentioned studies do not provide a

57

syntactic analysis for the experimental sentences in use. The lack of a theoretical framework impoverishes both the task design and the result interpretation. For all these reasons, further investigation into the processing of relative clauses in PADs is needed.