| |
Abstract:
Research on subject-verb number agreement has revealed that
speakers occasionally make errors (1) (Bock & Miller, 1991),
and readers encounter processing difficulty (2) (Nicol et. al.,
1997), when a singular subject is followed by a plural "distractor"
noun.
(1) The key to the cabinets... WERE missing.
(2) The key to the cabinets was missing.
Eberhard (1997) has characterized these errors within an activation
framework (Note 1). The agreement mechanism checks for the
existence of an activated plural feature on the head noun: if no
feature is found, the verb is specified as singular by default. If
one is found, the verb is specified as plural. Errors (or
processing difficulties) occur because an activated plural feature
on the distractor noun creates noise in the system, allowing the
mechanism to possibly misinterpret the head noun's number. Within
this account then, the amount of noise created should be related to
the activation level of the interfering plural feature. And
activation levels should be related to the relative frequency with
which a noun appears in its plural form. For example, 'senator' is
four times more likely to appear as a singular than as a plural
while 'banker' is three times more likely to occur in its plural
form (Francis & Kucera, 1982). 'Banker' would then have a
strong association with its plural feature, while 'senator', would
have a relatively weak one. In short, 'bankers' should prove more
disruptive than 'senators'.
We investigated this by constructing two sentence sets: one in
which the distractor nouns were on average 3 times More Common as
Plurals (MCP) and one in which they were 3 times More Common as
Singulars (MCS). The head nouns used were all equi-probable as
singulars or plurals. A third set of sentences contained distractor
nouns that were not strongly biased in either direction (Control).
A sample set is displayed in Table 1.
Forty-four participants were tested on a whole sentence reading
task modeled on Nicol et. al. (1997). Following Nicol et. al.
(1997), a slow-down (reflecting processing difficulty) is expected
in each SP condition relative to its corresponding SS condition.
Regarding frequency effects, we predicted that the magnitude of
slowdowns would be MCP > Control > MCS.
Table 1 illustrates that the results did not fit our
predictions, as a significant effect was found only for the Control
items. However, an interesting post-hoc explanation presents
itself. For the MCP items, the distractor noun's association with
its plural feature could be so strong that interference occurred
even in the SS condition. Conversely, for the MCS items the
association was so weak that little interference occurred in the SP
condition. This possibility is strengthened by the fact that the
MCP sentences were significantly slower than the MCS sentences
overall. This explanation is plausible within an activation
framework, particularly one utilizing a connectionist architecture
(see Elman, 1993), and suggests that further work on the role of
word frequency in number agreement is warranted.
Note 1: Eberhard's (1993) model relates specifically to
production only, however, research on comprehension has indicated
that the mechanisms involved appear to be similar (see Nicol et.
al., 1997).
|
