| |
Abstract:
In two sentence continuation tasks and a corpus analysis the
production of relative clauses was investigated in sentences of the
NP1-of-NP2-RC structure like "Someone shot the servant of the
actress who was on the balcony". The crucial feature of this type
of sencences is that the relative clause (RC) can be attached to
either of two noun phrases ("the servant" or "the actress", NP1 or
NP2). The present research (in Dutch) is centered around two main
questions. First, it examines the correspondence between text
corpora and the sentence continuation task (two modes of sentence
production). The second research question concerns the suggestion
made by Gibson and Schütze (1999) that the production of
sentences is only governed by recency and that no factor favouring
high attachment (i.e., to NP1) is involved.
Concerning the correspondence between the sentence continuation
task and text writing our results reveal that the modification of a
complex head containing two noun phrases seems to be distinct for
sentences with a human NP1 and sentences with a non-human NP1.
According to the corpus analysis, the latter show a very strong
preference for NP2 attachment of the relative clause, while the
former show an NP1 attachment bias. Submitting sentences that were
selected from the corpus to the sentence continuation task revealed
a pattern that was very similar to the one obtained by the corpus
study (see Table 1). For three of the four sentence types the
similarity in NP2 attachment percentages was very striking.
Although the similarity was less striking for the sentences with a
human NP1 and a human NP2 (h/h), it should be noted that both the
corpus analysis and the sentence continuation task showed an NP1
attachment bias and that this result was only based upon 15
sentences. Given the important differences between text writing and
the continuation of a presented sentence beginning (as in the
sentence continuation task) this correspondence is a noteworthy
finding in itself, with some major implications for present
accounts of human parsing. Moreover, this finding has some
important methodological implications. This means, for instance,
that the sentence continuation task is a very useful task for
verifying whether carefully constructed experimental sentences are
still comparable to the majority of sentences readers encounter in
their language.
Second, our results falsify the suggestion made by Gibson and
Schütze (1999), who state that production is only governed by
locality considerations. In Dutch, sentences containing complex
heads with a human NP1 seem to be consistently produced with an
attachment to this human site. This means that factors favouring
high attachment, such as predicate proximity (Gibson, Pearlmutter,
Canseco-Gonzales, & Hickok, 1996) or anaphoric binding
(Hemforth, Konieczny, & Scheepers, in press) can influence
sentence production and are not only involved in sentence
comprehension.
Finally, our results show that the humanness of the first noun
phrase plays an important role in the production of sentences. Our
corpus study shows that h/h sentences are only a minority of the
NP1-of-NP2-RC sentences that readers encounter when reading texts.
It is rather problematic that researchers have nonetheless mainly
used this type of sentences to investigate modifier attachment.
This may even lead to the conclusion that the results obtained by
the majority of studies at this time can not be generalised to all
NP1-of-NP2-RC sentences.
|
