|
Abstract:
In this paper we present results from a Japanese
questionnaire evaluating the predictions of two classes of
syntactic complexity theories: locality-based theories (e.g., the
syntactic prediction locality theory (SPLT) Gibson, 1998) and
interference-based theories (e.g., the case-based theories of
Lewis, 1993, and Stabler, 1994). Babyonyshev & Gibson (1999)
presented results suggesting that the complexity of a nested
structure depends on the number of NPs between dependent phrases.
However, there were methodological problems with this study which
make the results open to alternative interpretations. The study
reported here avoided these problems while addressing the same
issues.
48 participants took part in the questionnaire experiment,
rating the complexities of sentences on a scale from 1 to 5. The
questionnaires contained 32 items and 80 fillers in a balanced
Latin Square design. In the first four conditions we compared
nested structures containing intransitive verbs in the most deeply
embedded clauses (as in 1a and 1b) to nested structures containing
transitive verbs in the most deeply embedded clauses (as in 1c and
1d). In (1a), the most deeply embedded clause contains a nominative
subject and a verb. In (1b)-(1d), the most deeply embedded clause
contains a nominative subject and a verb and one additional
constituent: an adverb in (1b); an accusative direct object in
(1c); and a dative object in (1d).
The SPLT predicts that the transitive conditions (1c) and (1d)
should be more complex than the intransitive conditions because an
extra NP (introducing a new discourse referent) separates a number
of subject-verb dependencies. The SPLT predicts no difference
between the two intransitive conditions (1a) and (1b) because the
distance metric is based on new discourse referents, and the
intervening adverb does not introduce a new discourse referent. The
SPLT also predicts no complexity difference between the accusative
and dative transitive conditions. The interference-based theories
predict no differences among any of the four conditions, because
the intervening element in (1b)-(1d) is distinct from all preceding
constituents in both its case-marking and its X-bar phrase
structure position.
The results of the experiment bore out the predictions of the
SPLT: there was no difference between the ratings for (1a) and (1b)
or for (1c) and (1d), but there were significant differences for
each comparison between an intransitive condition and a transitive
condition (Fs>5; ps<.05). The observed differences between
the transitive and intransitive conditions were not predicted by
the interference-based theories.
Two additional conditions tested the complexities of structures
with five sentence-initial NPs but only one level of nesting, as in
(2). (2b) is identical to (2a) with an initial modifying clause
containing a subject and an intransitive verb. Thus (2b) contains
the same number of clauses as the sentences in (1), so complexity
differences among the structures in (1) and (2b) are not due to
differing numbers of clauses. If syntactic complexity is determined
in part by the number of NPs lacking thematic roles at a given
parse state (Hakuta, 1981; Gibson, 1991) then the structures in (2)
should both be more difficult than the structures in (1), which
include at most four NPs lacking thematic roles at any parse state.
Both the locality-based and the interference-based theories predict
that structures like those in (2) should be easier than structures
like those in (1).
Intuitions have suggested that sentences like (2a) are easier to
process than any of the examples in (1) (Lewis, 1993). The results
of the questionnaire confirm the intuition: examples like (2a) and
(2b) were both easier to process than any of the examples in (1)
(Fs>5; ps<.05). Taken together, the evidence reported here
supports a locality-based complexity metric like the SPLT.
|
