| |
Abstract:
Many theories of language production assume that it involves
three processes: conceptualization (semantic encoding), formulation
(syntactic and phonological encoding), and articulation. One aspect
of formulation concerns the construction of syntactic structure. In
this paper, our main concern is whether one or more than one level
of syntactic encoding occurs. Bock et al. (1992) argue against the
existence of two levels, one concerned with the production of a
level akin to D(eep)-structure, the other concerned with
S(urface)-structure. However, an alternative possibility is that
there are two levels of structure, one concerned with dominance
relations alone, the other concerned with both dominance and
precedence relations (cf. Gazdar et al., 1985). Such an account
fits with recent psycholinguistic proposals (Harsuiker et al.,
1999; Vigliocco & Nicol, 1998). For instance, Hartuiker et al.
propose that language production involves a linearization stage
during which an unordered syntactic representation is converted
into a linearized form; whereas Vigliocco and Nicol suggest that
agreement is encoded at a syntactic level that is unspecified for
word order.
To investigate this, we report four experiments that employ
syntactic priming in written production to investigate aspects of
syntactic encoding (Pickering & Branigan, 1998; Branigan et
al., in press). Participants completed target sentence fragments
(e.g., the farmer gave ...) that were compatible with either a
*prepositional object (PO)* completion (e.g., ... some straw to a
horse) or a *double object (DO)* completion (e.g., ... a horse some
straw), immediately after completing prime fragments. Priming
occurred if participants tended to complete target fragments in the
same way as they completed a prime fragment. Experiment 1
demonstrated that priming is a two-way process: The proportion of
PO target responses was largest after a PO prime response,
intermediate after an intransitive prime response (the baseline
condition), and smallest after a DO prime response. Experiment 2
found no evidence of priming from either a simple transitive prime
or a prime consisting of a single PP argument.
Experiments 3 and 4 investigated whether heavy shifted
structures primed prepositional object completions, double object
completions, or neither. The results showed that heavy shifted
structures behaved like baselines, with respect to the proportion
of PO vs. DO responses. If the processor computed a syntactic
representation that was unordered, one would expect priming from
the heavy-shifted form to the PO form.
Experiment 3 also demonstrated that participants produced many
more Other (i.e., non PO or DO) completions with baselines than
with heavy-shifted constructions. We interpret this result as
showing a priming effect based on the number of arguments involved
in the prime and the target: A three-argument prime primed a the
production of a three-argument target. The results therefore
indicated the existence of two levels of priming: a priming of the
number of arguments produced, and a priming that appears restricted
to instances of the same constituents appearing in the same order.
This pattern (together with the findings of Bock et al., 1992)
strongly suggest that priming of different abstract levels of
linguistic structure occurs, and therefore that the absence of a
priming effect at an order-free level of syntactic structure
indicates that the language production system does not construct
such a representation.
|
