| |
Abstract:
Verbs like "begin", "finish", "enjoy", and "endure" have been
argued to be examples of classes of verbs that semantically select
for an activity or an event (e.g., Jackendoff, 1997; Pustejovsky,
1991, 1995). However, these verbs can occur with a complement that
need not directly denote either an activity or event ("The critic
began the book" or "The critic enjoyed the book"). We argue that
such complements result in a semantic version of the syntactic
garden path. When readers attempt to construct a semantic
representation of the string "began the book", the default reading
of "the book" as an object produces an unacceptable semantic
interpretation. Readers must revise their initial interpretation by
accessing either telic or agentive information in the lexical
representation for "book" in order to construe it as an event. The
telic information for "book" can be argued to be a two-argument
event structure headed by the verb "read" (X read the book);
agentive information can be viewed as a two-argument event
structure, headed by the verb "write (X wrote the book)". Prior to
collecting reading time data, we constructed a number of sentences
like (1):
1a. The author began the book before going on vacation.
(Coerced)
1b. The author wrote the book before going on vacation.
(Preferred)
1c. The author read the book before going on vacation.
(Dispreferred)
A completion pre-test established that for the string "The
author began the book," the preferred interpretation was "The
author wrote the book" (as in 1b) and "The author read the book"
was dispreferred. (In another sentence, "read" might be preferred,
as in "The student began the book.") The first self-paced reading
experiment established that both 1a and 1c were more difficult than
1b at the word "book." One word later (e.g., at "before"), 1a
produced slow reading times, and 1b and 1c were processed equally
quickly. The results at the word "book" are consistent with either
type coercion or the immediate adoption of a dispreferred semantic
interpretation of "began the book." The fact that 1c is as fast as
1b, but 1a is slow, one word after the critical word shows that
sentences like 1a and 1c are processed differently and suggests
that slow reaction times on 1a at "book" are the result of
additional computations necessary to recover an event reading for
"book" in 1a. A second self-paced reading experiment employed
unbounded dependency constructions like (2):
2a. The book that the author began to write ended up on the
best-seller list. (Preferred)
2b. The book that the author began to read ended up on the
best-seller list (Dispreferred)
2c. The book that the author wanted to write ended up on the
best-seller list. (Uncoerced, Preferred)
2d. That's the book that the author wanted to read before going on
vacation. (Uncoerced, Dispreferred)
In this experiment, readers had greater difficulty processing
"read," in the coerced case ("The book the author began to
read...") relative to the other three conditions, which did not
differ. This suggests that processing verbs that require coercion
is no more difficult than processing other types of verbs and that
coercion occurs in an incremental fashion. If subsequent text
confirms the coerced interpretation (as in the coerced, preferred
cases like 2a), no disruption of processing ensues.
The results across the self-paced reading experimetns are
consistent with recently published probe-task results (Pinango,
Zurif, & Jackendoff, 1999) and the results of as yet
unpublished speed-accuracy tradeoff experiments (McElree, in
prep).
We also plan to present some evidence from eye-tracking.
|
