| |
Abstract:
We examine the use of subcategorization information
associated with syntactically alternating and non-alternating
ditransitive verbs. These verbs allow us to separate the
contribution of semantic and syntactic subcategorization
information to parsing decisions, and demonstrate a strong case of
immediate access to verb subcategorization information. However,
even full access to subcategorization information does not
guarantee accurate parsing: successful grammatical search is also
required.
Previous studies have demonstrated rapid use of verb
subcategorization information in parsing (e.g. Ford et al. 1982,
Marslen-Wilson et al. 1988, Boland et al. 1991, Trueswell et al.
1993 and many others), by comparing verb classes with different
numbers or types of thematic roles. Our experiments compare verbs
which allow the dative/double-object alternation (e.g. give, show)
and verbs which do not (e.g. donate, display). These classes are
matched in the number and type of internal arguments, but differ in
the available syntactic frames (1-2).
(1) The millionaire gave/donated the painting to the museum.
[dative]
(2) The millionaire gave/*donated the museum the painting. [double
object]
If syntactic subcategorization information is fully accessed and
exploited, then substitution of a non-alternating verb for an
alternating verb should render the classic garden path in (3)
unambiguous (4). Our self-paced reading experiments compare
sentences like (3) and (4) with unambiguous controls (with the
complementizer 'that'), and additional conditions containing
monotransitive verbs (e.g. visit, want) in place of the
ditransitives. 36 experimental items were interspersed with 84
fillers.
(3) The duke bought the museum the elaborate antique chandeliers
had gracefully adorned for his son.
(4) The duke purchased the museum the elaborate antique
chandeliers had gracefully adorned for his son.
In Experiment 1, alternator conditions showed a large cost of
ambiguity at the disambiguating VP [F(1,53)=27.5, p<.0001], but
not at the second post-verbal NP. The monotransitive conditions
showed the reverse: a large cost of ambiguity at the second
post-verbal NP [F(1,53)=16.4, p<.0001] but not at the embedded
VP. The crucial non-alternating conditions showed a large cost of
ambiguity at both the second post-verbal NP [F(1,53)=30.1,
p<.0001] and the disambiguating VP [F(1,53)=18.1, p<.0001].
The clear contrast between alternating and non-alternating verbs
implies full access to syntactic subcategorization information.
However, the difference between non-alternating verbs and the
monotransitive verbs at the disambiguating VP remains to be
explained. Results from a preliminary experiment and from a
split-half analysis of Experiment 1 provide an account of this
contrast. In a preliminary experiment, which lacked monotransitive
conditions and had many fewer relative clauses in the filler
conditions, we found alternating and non-alternating verbs showed
identical reading-time profiles. Furthermore, in Experiment 1, the
large cost of ambiguity at the disambiguating VP is only seen with
non-alternators in the first half of the experiment (whereas
alternating verbs show a garden-path effect throughout the
experiment).
Conclusion: we assume that full access to lexical-syntactic
information occurred throughout both experiments. What varied was
subjects' ability to use information about the impossibility of a
double object analysis to search their grammar for the appropriate
relative clause analysis of the V NP NP sequence. This search
succeeded only in the context of numerous other relative clause
parses. Experiment 2, currently underway, explicitly manipulates
the availability of support for grammatical search. We conclude
that it is necessary to distinguish full access to syntactic
information associated with verbs from the ability to use this
information to find a grammatical parse for a sentence.
|
