| |
Abstract:
Syntactic priming, i.e., the phenomenon of reusing syntactic
form over consecutive trials in a written or spoken production
task, bears important implications on representational as well as
architectural aspects of sentence generation. We present an
internet-based experiment (see
http://surf.to/experiments
) which was designed to explore the priming of two major
determinants of syntactic structure, namely subcategorisation
information and linear precedence (cf. Pollard & Sag, 1994).
While the former determines the "minimum legal sentence" by
specifying the number and type of complements required to saturate
a verb, the latter determines the ordering of constituents at the
sentence surface. German has scrambled constituent order and overt
case marking, allowing the direct manipulation of these factors.
Priming effects were investigated using sentence completion of
both prime and target sentences. The target constructions were
German VP-head-final sentence fragments like (i) and (ii) (see
example a) which, in principle, can be completed in two different
ways: Firstly, by appending a main verb which subcategorises for a
single object-NP (e.g., (i) could be completed by
angetroffen
["met"] which takes an accusative, and (ii) by
geholfen
["helped"] which takes a dative). Secondly, in each case an
additional object-NP followed by a ditransitive verb can be
appended (e.g.,
vorgestellt
["introduced"] subcategorises for an accusative and a dative
object). If the ditransitive option is chosen, (i) enforces an
accacc.
Each target was preceded by one of five different (semantically
unrelated) priming constructions which were completed by filling in
a missing object (indicated by "___"):
In (1) and (2), the final verb strictly subcategorises for two
object-NPs (NP[acc] and NP[dat]), where accacc orders are specified
respectively by the given object NPs. (3) and (4) are
monotransitive constructions where the final verb subcategorises
for an accusative (3) or dative (4) object. A baseline prime
condition (5) was strictly unrelated to the target sentences, as
were additional fillers.
The results indicate
no
priming of subcategorisation information
per se
(such as verb valency and case of complements). Instead, sentence
generation in German appears highly sensitive to linear precedence
information, highlighted by the fact that there is evidence for
priming and inhibition (!) of ditransitive structures, dependent on
whether a ditransitive prime implies the same word order as the
target or not. Our results suggest that positional processing must
at the least interact with "functional processing" (retrieval of
verb-frames and syntactic function assignment), contrary to certain
models of grammatical encoding in language production (e.g., Bock
& Levelt, 1994).
|
