As was already mentioned in the introductory section, the processing of VP-ellipsis constructions in real time requires a few steps that are necessary for the appropriate interpretation of the elided VP (second conjunct) including that of the elided pronoun. This process will be examined closely with the help of the example in (10).
(10) a. [The man touches his dog] and [the boy does <e> too]. b. man touches man’s dog & boy touches boy’s dog – semantic –
bound variable interpretation
c. man touches man’s dog & boy touches man’s dog – discourse –
coreference interpretation
It is natural to assume that in order to be able to process the second conjunct, the hearer must construct the syntactic structure (T1) and
assign reference to the pronoun in the first conjunct (T2 and T3) in order
to copy this information onto the second conjunct (Tc) where the VP is
elided. The hearer first constructs the VP in the first conjunct, the man
touches his dog, and assigns reference to the pronoun his. At the same
time the second conjunct is being processed. As we have seen in section 4.2.2, the hierarchy of referential dependencies assumes that the semantic (bound variable) interpretation is less costly than the discourse (coreference) interpretation. The bound variable interpretation in the first conjunct will be available sooner in time than the coreference interpretation because it is more economical. Hence, the bound variable representation will also be copied faster into the second conjunct. Nevertheless, it will not block a possible coreference interpretation, which will also become available, but at a later point. The coreference interpretation will be available in the first conjunct but will not be the preferred interpretation in the second conjunct simply because it is the more costly option for the unimpaired adults (see Figure 4.5).
Figure 4.5
So, why do agrammatic patients perform worse than the unimpaired subjects? The VP-ellipsis structures with pronouns represent cases where particular elements must be kept in store in order to be integrated with other elements that they form a dependency with. As the comprehension of these structures unfolds, the previously heard and not yet integrated material, such as the antecedent for the pronoun, needs to be connected to the pronoun once it is encountered. As was argued in the previous chapters I assume the syntactic structure building is delayed, which in turn affects the comprehension process in agrammatism. Syntactic structure building is a dynamic process that takes place in real time, therefore, the availability of particular bits of information that come from different sources is crucial for language processing. In the elided VP constructions first the syntactic structure of the initial VP phrase the man
touches his dog has to be built and a referent needs to be assigned to the
pronoun his. At the same time the second conjunct is being constructed as a copy of the first one.
In agrammatism, on the other hand, in an off-line task such as the one used in this experiment the picture becomes different (see Figure 4.6). Because the syntactic structure is built slower (T1 and T2 are later in time
than the same points in the neurologically intact individuals) the assignment of reference is also adversely affected. While the agrammatic patients are trying to construct the first conjunct and assign reference to the pronoun in the first conjunct, they are confronted with the second conjunct the boy does too. In order to interpret the second conjunct, they will have to have built the structure and assigned reference to the pronoun in the first conjunct. Because they ‘so-to-say’ take longer to build the syntactic structure, which needs to be consulted and copied (Tc) into the empty slot of the elided VP, economy considerations will
become more important to them than to the unimpaired speakers simply because they will be running out of time. Just like the unimpaired interpretive system, the agrammatic system has to assign some sort of an interpretation to the elided VP and the pronoun in it. The semantic dependency is cheaper and established faster (T2) than the coreference
dependency in the first conjunct. The pronoun in the second conjunct is more likely to be interpreted as a bound variable before the system is out of resources, specifically, out of time. The process of assigning a fixed discourse referent to the pronoun and then copying this referent into the elided VP, on the other hand, is a more costly and a more time consuming procedure. As illustrated in Figure 4.5, the point at which the discourse interpretation of the pronoun (T3) is completed in the first
conjunct will be delayed relative to the same point (T3) in the unimpaired
adults. In the agrammatic patients the coreference interpretation will not be ready on time to be copied into the second conjunct. This is why in Figure 4.6 the interpretation that is copied in the second conjunct is the semantic bound variable interpretation, which is what we see reflected in the results of the experiment. The agrammatic patients perform above chance (80% correct) on the condition where bound variable is the only correct response. When they are given a choice between the two possible interpretations, they prefer the bound variable interpretation (90% bound variable chosen) to coreference. When the agrammatic patients are confronted with the condition where the only correct response is coreference, they resort to guessing between the correct picture and the erroneous related distractor (53% correct). This indicates that the
coreference interpretation of the pronoun in the first conjunct is not ready on time for it to be copied into the second conjunct.5
5 One could argue that the distance between the site of the reconstruction of the
elided VP and the local DP2 (bound variable interpretation) is shorter than the
distance to the non-local DP11 subject of the first conjunct (coreference
interpretation).
(i) [DP1 The man] touches his dog and [DP2 the boy] does too.
As such, the more local DP2 could be more active in memory and accessed more
readily. However, the more active DP at the point of reconstruction is actually the non-local DP1 because at this point the whole VP (the event) of the first conjunct is
being reconstructed and copied. This would actually predict that the coreference interpretation should be easier for the agrammatic patients, which is obviously not the case. Another way of interpreting the data would be to assume that frequency plays a role and that the more frequent interpretation (bound variable) is less difficult. Online results obtained by Shapiro & Hestvik (1995) and Shapiro et al. (2003) show that this cannot be the case. Both interpretations are activated at the point of the reconstruction of the elided VP.
Figure 4.6
So, the agrammatic patients fail to obtain the coreference interpretation because their system runs out of time and resorts to the cheaper and faster available option. It could be the case that if these sentences, first conjunct in particular, are presented to the agrammatic patients in a slower-than-normal speech-rate, the coreference interpretation could
become equally available as the bound variable interpretation.6This is
only a suggestion, which would clearly need further investigation.
4.5 Conclusion
To conclude, agrammatic aphasic speakers are capable of obtaining the semantic (bound variable) interpretation for the possessive pronoun in VP-ellipsis constructions. However, they are incapable of establishing a discourse (coreference) dependency in the normal time course for the pronoun in the same constructions. The agrammatic patients also exhibit a preference for the bound variable interpretation that is much stronger than what the controls show. This preference reflects the processing hierarchy, which is related to complexity of these constructions. I propose that their grammatical knowledge of these constructions is not impaired and that their errors are not the result of a lack of knowledge but a consequence of a slower-than-normal syntactic processing. The results presented here support the approach that treats agrammatism as a reflection of an insufficiency of resources that are necessary to carry out linguistic operations in real time.
6 Swinney et al., (in preparation) (see also Zurif, 2003) conducted a priming study
where they presented agrammatic patients with non-canonical sentences involving filler-gap dependencies at a slower-than-normal speech rate (input rate reduced from six syllables per second to one syllable per second). Broca’s patients did reactivate the displaced constituents at the gap positions. In other words, they could establish the dependency between the gap and the antecedent.