| |
Abstract:
This paper presents two experiments comparing three
contemporary theories of sentence processing: the serial,
garden-path theory (Frazier & Clifton, 1996), a parallel
processing theory, and the SPLT model based on computable memory
cost only (Gibson, 1998). It is claimed that the results obtained
constitute evidence in favor of the serial theory and cannot be
accounted for in the other two models.
Experiment 1 focuses on the comparison between serial and
parallel models through the study of temporary ambiguity
resolution. In Catalan, the conjunction that introduces the second
term of a comparison is homonymic with a relative pronoun: they are
both
que.
We used sentences in which the ambiguity introduced by a first
que
was not disambiguated until the appearance of a second instance of
que.
The resulting structures are as follows:
1a. 1st term comparison -
que-
2nd term comparison -
que
relative clause
1b. 1st term comparison -
que-
relative clause -
que-
2nd term comparison
Theories of serial processing argue that when an ambiguous item is
encountered, the many alternative analyses are processed one at a
time, starting by a preferred analysis and reanalyzing the sentence
if it is the case that the first analysis is disconfirmed. Applied
to this case, the Construal principle (Frazier & Clifton, 1996)
predicts that when the context provides with a first term of a
comparative, the preferred analysis for que is that of introducer
of a second term of the comparative. If so, (1a) will be easier to
process than (1b), since (1b) requires an extra step of reanalysis.
Parallel processing theories argue instead that, faced with
ambiguity, the processor computes all possible analyses, and waits
for cues to eliminate all but the final right analysis. This theory
predicts that (1a) be as easy as (1b), since they are both equally
ambiguous. Our results suggest a preference for (1a) over (1b), as
predicted by the serial theory. This is unexpected under the
parallel alternative.
In Experiment 2, the garden path theory is compared with the
SPLT. Based on memory cost alone, the SPLT predicts with the garden
path theory that (1a) be preferred to (1b), since (1b) contains a
delayed instantiation of a predicted element (the second term of
the comparison) and hence a higher load of processing. However, the
two theories differ in still one prediction: given no ambiguity,
the garden path theory predicts no effect of preferring one
analysis over another, and therefore no difference between the
unambiguous counterparts of (1a) and (1b), since no reanalysis
takes place. The SPLT predicts rather that even when there is no
ambiguity, an unambiguous counterpart of (1a) will still be easier
than an unambiguous counterpart of (1b), since in (1a) there is a
greater memory cost. The results show no difference in the
processing of unambiguous counterparts of (1a) and (1b), hence
confirming once more the predictions of the garden path theory and
leaving the data unexplained under the SPLT model.
|
