not be independent of meaning; thus, the priming mechanism may encourage speak-
ers to categorize on the basis of form and meaning—exactly the sort of categorization
that is required by construction-based approaches. We briefly discuss a very general
processing advantage of CxG and other theories that assume a parallel representation
of linguistic knowledge (e.g., Jackendoff 2002; Sag and Wasow 1999).
Recognition of the Incrementality of Language Processing
Like other parallel constraint-based models of grammar (e.g., head-driven phrase
structure grammar), CxG assumes that syntactic, semantic, and phonological infor-
mation is represented in parallel. Parallelism has some attractive features from the
perspective of processing (Bencini 2002; Sag and Wasow 1999; Jackendoff 2002).
These features are evident when one contrasts parallel grammars with the main-
stream generative grammar architecture, in which syntax is the “engine” and seman-
tics and phonology are interpretative levels (e.g., Chomsky 1981). Although main-
stream generative grammar is a model of competence, it is possible to derive a certain
general processing implication from its main organizational features (see Bock,
Loebell, and Morey 1992 for a good example): Namely, it is presented as inherently
directional. Derivations are always computed from the syntax (through various inter-
mediate derivational steps) to meaning and sound.
This type of grammar clearly requires some sort of adjustment to be compatible
with the disparate processes of language comprehension and production. In language
production, the starting point is a meaning to be conveyed, and the goal is to convert
this meaning into a motor program for speech. Along the way, syntactic information
is used to produce a well-formed utterance (Bock 1995; Bock and Levelt 1994;
Garrett 1980). In comprehension, the starting point is a phonetic representation de-
rived from auditory or visual input, and the goal is to compute a meaning.
The parallel architecture, on the other hand, is not inherently directional and
therefore is compatible in a transparent way with both comprehension and produc-
tion. Parallelism allows for differential weighting and ordering of linguistic informa-
tion in processing (e.g., Bencini 2002). For example, in comprehension one need not
assume that a syntactic structure must be computed “first.” A more rapid and incre-
mental integration of different sources of information can be assumed to constrain
the interpretation as information becomes available. This analysis is consistent with
many findings from online comprehension studies (e.g., Just and Carpenter 1982;
Garnsey et al. 1997).
14
Do'stlaringiz bilan baham: