| |
Abstract:
Three reading-time experiments investigated the
interpretation of discourse-linked fillers in various
constructions. A consistent preference was found to assign the
filler to a relatively later gap. We propose an account in terms of
discourse-licensing and interpretive preferences that reconciles
these findings with principles such as the Active Filler Strategy
(Frazier, 1987).
Experiment 1 examined filler-gap assignment preferences in NP/S
ambiguity structures (1). Sentences with
who
or
what
were easier to process when the
wh-
phrase received matrix object interpretation than when it was
construed as the embedded subject. With
which-
N the preference was reversed, favoring the embedded subject
interpretation. It is argued that
which-
N, being discourse-linked (thus salient), receives partial
licensing through discourse, thus there is less pressure to assign
it to the first available gap (cf. also DeVincenzi, 1996).
Moreover, subject interpretation of
which-
N is preferred as salient elements typically occur as subjects in
English.
Who/what
questions can also be interpreted as discourse-linked, given
appropriate context (Pesetsky, 1987). When they are, they should
behave just like
which-
N, preferring the later (embedded subject) gap (1b). This was
tested in Experiment 2, where items similar to (1) appeared after a
short context, which provided potential antecedents for the
wh-
phrase.Both types of
wh-
phrases were read faster in the embedded subject condition,
contrary to standard findings that
wh-
phrases prefer the first (here: matrix object) gap. The results
follow, however, assuming that discourse-linked fillers are
partially licensed and may be assigned to a later gap, especially
when that interpretation is preferred for semantic/discourse
reasons.
The third experiment investigated Hungarian sentences containing
two fillers, a topic and a
wh-
phrase, with the same case-marking (2). The topic is typically an
"active filler" in Hungarian, preferring an early gap. In this
construction, however, there was overwhelming preference to
interpret the
wh-
phrase as binding the first gap. This follows under the current
proposal: There is less pressure to assign discourse-linked fillers
(topics) to early gaps, since they are partially licensed through
discourse. As the
wh-
phrase is not discourse-licensed, the parser chooses to assign it
to the earlier gap.
Under this proposal, the parser tries to assign all fillers to
the earliest possible gaps. For discourse-linked fillers, however,
this might be overridden by other preferences, e.g., interpreting
salient elements as subjects in English, or the need to assign
another filler first (Hungarian).
|
