|
Abstract:
All other things being equal the parser favors attaching an
ambiguous modifier to the most recent possible site (Kimball, 1973;
Frazier, 1979; Gibson, Pearlmutter, Canseco-Gonzalez, & Hickok,
1996). A plausible explanation is that locality preferences arise
in the service of minimizing memory costs --more distant sentential
material is more difficult to reactivate than more recent material.
Note that processing *any* sentence requires linking each new
lexical item with material in the current parse. Often this
involves construction of long-distance dependencies. Under a
resource-based view, lengthy integrations should induce difficulty
even in unambiguous sentences (cf. Gibson, 1998; Stevenson, 1994).
To date there has been little direct quantitative evidence in
support of this perspective. This paper presents two self-paced
reading studies exploring the hypothesis that locality is a
fundamental determinant of reading complexity in unambiguous
constructions. Evidence suggests that difficulty associated with
integrating a new input item is heavily determined by length of the
material intervening between the input item and the site of its
target dependents.
Following Gibson (1998) distance can be measured in terms of
words or discourse referents.
In the first experiment, participants read sentences in which
the matrix subject was modified either by a subject or object
extracted relative clause (as in (1)). Integration distances by
region were significantly related to reading times accounting for
over 60% of the variance.
In the second experiment, participants read a variety of nested
and non-nested structures (e.g., (2)). Points of high integration
cost occurred primarily at the main and embedded verbs. Magnitude
of predicted integration cost was manipulated (1) by inserting or
omitting modifiers on the the matrix and embedded subjects and (2)
by varying the type of modification (relative clause or
prepositional phrase). Integration costs were highly correlated
with residual reading times accounting for roughly 45% of the
variance across regions. Looking only at points of high complexity
across conditions, integration costs accounted for 27% of the
variance. This work emphasizes the role of a resource-based
configurational component in sentence processing. The effect of
locality dwarfed contributions of non-configurational factors such
as word length, lexical frequency, and the plausibility of
intervening material. The role of locality is even more pronounced
if distance is fit to a psychologically plausible non-linear model
(e.g, a sigmoid). Patterns observed here are not straightforwardly
accounted for within an experience-based model of configurational
complexity (MacDonald & Christiansen, 1999). Such models posit
that the ease of processing a construction relates directly to how
frequently it is encountered. Without appealing to resource
limitations in comprehension or production, there is no a priori
reason to posit that longer integrations should be rarer for the
sentences given in our second experiment.
In sum, locality is not a specialized heuristic for choosing
between alternative attachments, but is a ubiquitous factor in
language processing.
(1) a. (The reporter) (who sent) (the photographer) (to the
editor) (hoped for) (a story).
b. (The reporter) (who the photographer) (sent to) (the editor)
(hoped for) (a story).
(2) a. (The administrator) (who the nurse) (from the clinic)
(supervised) (scolded the) (medic while) ...
b. (The administrator) (who the nurse) (supervised) (scolded the)
(medic while) ...
c. (The nurse) (from the clinic) (supervised the) (administrator
while) ...
d. (The nurse) (supervised) (the administrator) (while) ...
e. (The administrator) (who the nurse) (who was) (from the clinic)
(supervised the) (administrator while) ...
f. (The nurse) (who was) (from the clinic) (supervised the)
(administrator while)
|
