In order to study the nature of cognitive resources required in language
production, we elicited errors of subject-verb agreement in sentence
completion. We tested the effects of the conceptual number of the
to-be-uttered subject noun phrase (i.e., distributivity) in conditions
with and without a working memory load. We also determined a measure of
verbal working memory span for each participant (speaking span).
Conceptual number was manipulated by contrasting sentence fragments that
are likely to receive a distributive interpretation (e.g., the date
on the
coins) with items that are not (e.g., the author of the articles).
Previous studies with normal speakers found many more agreement
errors in
the former condition than in the latter condition (e.g., Hartsuiker et
al., 1999; Vigliocco et al., 1995; 1996a-b). However, there was no such
effect in speakers with aphasia (Hartsuiker et al., 1999). This suggests
that the production of verb agreement is at least partially
constrained by
verbal working memory resources, on the broadly supported
hypothesis that
aphasics suffer from a pathological reduction in these resources. The
present study therefore aimed at elucidating the role of working memory
resources in normal production of agreement, by manipulating the
demand for resources (presenting an extrinsic memory load) or by
considering the availability of resources (comparing speakers with
low and
high working memory capacities). We found significantly more agreement
errors when the sentence fragment was distributive; when the participant
had to maintain an extrinsic memory load; and when the participant had a
low speaking span. These variables did not interact, with the
exception of
a marginally significant interaction between speaking span and
distributivity (i.e., a reduced effect of distributivity for high-span
speakers). These results show that the production of agreement involves
processing steps that make a demand on working memory resources. We
interpret these findings as evidence for relatively automatic processes
that mark number on the subject-phrase (and possibly integrate it
with the
number specifications on morphemes, as recently suggested by Bock
et al.,
in press). These processes however are followed by a resource-demanding
step in which number is transmitted from subject-phrase to the verb.