| |
Abstract:
We present results of two self-paced reading experiments
designed to investigate the effects of the discourse status of a
noun phrase and its implications for reading times in sentences.
These experiments build upon results presented in Warren and
Gibson, 1999. In that poster, we presented questionnaire evidence
showing that the discourse status of certain NPs in doubly center
embedded sentences had effects on the judged complexity of those
sentences. This evidence was interpreted as supporting a discourse
based distance metric in the Dependency Locality Theory (Gibson,
1998), a locality based theory of linguistic complexity. The
current experiments expand upon that work, showing that the DLT
makes correct predictions about the timing of discourse status
effects in self paced reading.
The two experiments investigate the role building and
identifying referents plays in the distance metric within the DLT.
Past studies (e.g. Haviland & Clark, 1974; Murphy, 1984)
indicate that there is a cost for building a new referent, but
these studies reported whole sentence reading times and did not
have predictions about location and time course of the cost. The
DLT predicts the cost of introducing a new referent will be seen
most strongly at points in the sentence where dependencies which
cross the new referent are resolved.Experiment 1 investigated the
effects of building a new discourse referent on reading times.
Experiment 1 had two conditions, which were equally plausible
according to the results of a norming questionnaire. In one
condition, a context sentence introduced a referent which was
referred back to with a definite NP in the test sentence. In the
other condition, the referent was not introduced in the context
sentence, and was introduced with an indefinite NP in the test
sentence. DLT predicts that reading times on the verbal regions
will be slower in the indefinite (new) condition than in the
definite (old) condition, because the cost of building a new
referent will interrupt the integrations that cross the new
referent. A paired t-test showed the predicted effect over the
verbal region.
Experiment 2 varied the referential availability of the
antecedent of a definite NP. Each experimental sentence was
preceded by one of three different context sentences. In the first
condition, the context sentence explicitly mentioned the referent
that was the subject of the embedded clause in the experimental
sentence. In the second condition, the antecedent could be
identified in the context based on general world knowledge (easy
bridge). In the third condition, the identification of the
antecedent was more difficult, based on the results of a norming
questionnaire (hard bridge). The condition with explicit reference
should have the fastest reading times, followed by the easy bridge
and then the hard bridge. DLT predicts that these differences will
be most evident on the verbal regions. Results of a preliminary
group of subjects show a trend in the predicted direction at the
embedded verb, the main verb and also at the next word. Results
provide evidence supporting the DLT's discourse based distance
metric for computing locality.
Experiment 1
1a. A suspect in a bank robbery was caught by a detective on
Friday. The suspect, who the detective had sighted on Wednesday,
struggled but was eventually subdued.
1b. A suspect in a bank robbery was caught on Friday. The suspect,
who a detective had sighted on Wednesday, struggled but was
eventually subdued.
Experiment 2
Context sentences:
(explicit)
2a. A physicist attended the cross-disciplinary conference at the
university. (easy bridge)
2b. Some scientists attended the cross-disciplinary conference at
the university.
(hard bridge)
2c. Some professors attended the cross-disciplinary conference at
the university.
Test sentence:
A speaker who the physicist questioned explained the results in
more detail.
|
