|
Abstract:
Mauner and Koenig (1999) showed that agent-dependent
rationale clauses were difficult to process following intransitive
but not short passive clauses that both entailed an agent. This was
interpreted as evidence that agents are derived from information
strongly associated with lexical representations of verbs. A
challenge to this lexical interpretation is that _all_ participant
information is instead derived from syntactic constructions (e.g.,
Goldberg, 1995). In three experiments, using a self- paced reading
task with a make-sense judgment, we show the processing of
participant information associated with lexical sources can be
distinguished from both constructions and from participant
information less strongly associated with verbs (i.e. adjuncts).
Experiment 1 addressed the constructional challenge by comparing
the processing of recipient-dependent purpose clauses (e.g., 1b)
following sentences whose verbs were hypothesized to introduce or
not introduce a recipient (e.g., "provide" vs. "update"
respectively in 1a). Crucially, matrix clauses were created to be
constructionally identical by not including optional dative
complements. At the post-verbal gap where purpose and rationale
clauses become disambiguated and a recipient is required to
interpret PRO, purpose clauses following "update" sentences
elicited significant anomaly effects in judgments and reading times
compared to control sentences with explicit subjects (e.g., "for
students") instead of PRO. No anomaly effects were observed
following "provide" sentences relative to their explicit controls.
These results demonstrate implicit recipients arguments associated
with optional dative complements of "provide" verbs are lexically
and not constructionally derived.
1a. Several college libraries provide/update on-line
databases(i)
1b. ...for students/PRO to complete class assignments with ___(i)
for law school.
Experiments 2-3 explored the distinction between participant
information that is strongly associated with the lexical
representations of verbs (i.e., arguments) vs. that which is not
(i.e. adjuncts). We focused on optional constituents for which
linguistic argumenthood criteria typically yield ambiguous results
(Schutze & Gibson, 1999). Motivating Experiment 2 was the idea
that if the argument-adjunct distinction is psychologically real,
there should be clear differences in how participant information
associated with arguments versus adjuncts is processed on-line. We
explored whether the processing of filler-gap sentences, which have
been used to examine the influence on parsing of participant
information associated with _required_ constituents (e.g., Boland,
et al., 1995), could be used to identify the argument status of
optional wh-dative complements. We compared the processing of
filler-gap sentences with "provide" and "update" verbs (e.g., 2a)
whose wh-fillers were implausible as direct objects but plausible
as dative complements. We obtained significant anomaly effects in
judgments and reading times at the verbs in "update" but not
"provide" sentences relative to controls that explicitly marked
wh-fillers as dative complements (e.g., 2b). Experiment 3
replicated these results with wh-fillers equated for plausibility
in their respective sentences via completions and ratings obtained
in separate studies.
2a. Which students does the library provide/update on-line
databases to/for ___ regularly?
2b. To/For which students does the library provide/update on-line
databases ___ regularly?
Conclusions:Experiments 1-3 provide evidence that
participant/argument information cannot be completely reduced to
either constructions or plausibility. It appears that the lexical
representations of verbs independently contributes participant
information during processing. Experiments 2-3 demonstrate that the
processing of filler-gap sentences is sensitive to the argument
status of optionally- expressed constituents.
|
