|
Abstract:
Considerable research has been conducted in recent years on
the production of subject-verb agreement errors. Typically,
participants in such experiments are presented with a complex NP
subject (containing a head NP and a nonhead NP) and asked to repeat
it and complete the sentence. Verb agreement errors (as in (1)) are
tallied.
(1) The key to the cabinets ARE missing.
A recurring hypothesis throughout this literature has been that
errors may occur as a result of a misidentification of the head
noun-phrase, so that the plural nonhead NP erroneously controls
agreement (Bock & Miller, 1991; Thornton & McDonald, 1999).
This issue has broader significance with regard to the possible
interactions between the stage of processing at which agreement is
computed and the stage at which grammatical roles are assigned, and
whether both processes can be influenced by probablistic or
semantic factors related to the likelihood of a particular argument
bearing a particular role.
We investigated four such factors: animacy, plausibility,
semantic overlap, and relative frequency. In our studies,
participants were presented (visually) first with a verb or
adjective, and then with a sentence preamble, e.g. time 1 tired
time 2 The boy near the dog
and were then to construct a complete sentence, as in "The boy near
the dog was very tired."
1. Test of Animacy: Animacy is highly correlated with
subjecthood, therefore one prediction is that sentence preambles
containing animate plural non-heads should be more prone to error.
We ran a fully counterbalanced set of items in which we used all
combinations of head and non-head animacy (II, IA, AA, AI) . We
found no evidence that an animate nonhead induces more errors than
an inanimate one (replicating Bock & Miller's (1991) results
for a subset of the animacy conditions; AI and IA).
2. Test of Plausibility: The relative plausibility with which
each NP can be associated with the sentence predicate could also be
a contributing factor to error rates: preambles in which only the
head noun bears a plausible relation to the predicate should yield
fewer errors than those in which both the head and non-head bear
plausible relations. In a post-hoc study, we had all of the items
from the animacy study rated for exactly this factor. Analyses
revealed a near zero correlation between error rates and the degree
to which the non-subject noun had a plausible relationship to the
predicate. This is in contrast to claims by Thornton & McDonald
(1999).
3. Test of Word Frequency: The relative frequency of the head
vs. non-head noun should also be predictive of errors. Sentences in
which the non-head is of significantly higher frequency than the
head should lead to more errors than the reverse case. We found a
statistically non-significant trend in the predicted direction,
indicating that any effect of frequency is likely to be weak.
4. Test of Semantic Relatedness: Sentences in which the head and
local nouns have significant semantic overlap should lead to more
errors than those in which the two nouns are semantically distinct.
We are currently conducting a study to investigate this
question.
To date, our findings suggest that head misidentification is not
a strong factor in the commission of agreement errors. Factors such
as animacy, which play a significant role in the assignment of
functional roles, appear to be largely irrelevant to the
computation of agreement. Implications for the architecture of the
production system will be discussed.
|
