| |
Abstract:
The idea that the preferences observed in human sentence
parsing reflect the application of the principles of grammar in a
transparent way has been brought forward a number of times
(Pritchett 1992, Gorrell 1995, Phillips 1996). Such approaches seem
to suffer from both theoretical and empirical problems and have
therefore not played a dominant role in the psycholinguistic
literature. In this paper, we argue that the theoretical and
empirical problems disappear when a suitable theory of grammar,
viz. Optimality theory, is applied.
When a sentence is parsed incrementally, a structure is built up
that integrates the words perceived so far. We follow Tesar (1994)
in the assumption that additional structure may be added to this
immediately (the phonetic string is "overparsed") as long as doing
so improves the constraint violation profile associated with the
syntactic object. That this additional structure is predicted
because a syntactic object with such a structure violates less
principles than a candidate without it implies parsing preferences:
the parser's expectation is that incoming material can be
integrated smoothly into this structure. We will consider a number
of syntactic principles that have been motivated independently in
syntactic theory (e.g. the Extended Projection Principle (Chomsky
1993), the principle "Obligatory Head" (Grimshaw 1997) and the
principle Stay) and show that they are able to predict major
parsing preferences (e.g. Active Filler and late closure effects).
One particular prediction made by such a model of sentence
processing is that fairly unimportant grammatical principles may
have a major influence on parsing. Recall that Optimality Theory
assumes that all principles are universal but differ in ranking
only. If such a grammar is applied in a transparent fashion,
parsing should be affected by principles that have nearly no effect
on the grammatical system (because they have such a low rank). We
have identified at least one case in point. Consider the following
structure
(1) NP _ [relative pronoun _. verb ]
There is a principle of Universal Grammar that implies that the
head noun of a relative clause construction and the relative
pronoun should agree in case. Effects of this principle can be
observed for example in Ancient Greek. In most languages, however,
this agreement principle is dominated by a principle that requires
that the case selection properties of the verb of the relative
clause must be respected. Consequently, from a grammatical point of
view, NP cannot transmit its case to the relative pronoun in (1).
Note, however, that the relative pronoun is encountered before the
verb during incremental processing of a relative clause That is,
the information that rules out the case agreement between the NP
and the relative pronoun in the case of a conflict, viz. when the
case requirements can be read off the verb, is not present when the
relative pronoun is parsed and when its case may need to be derived
from grammatical principles because it is morphologically
ambiguous. In other words, we predict an on-line preference for
case agreement in (1) when the relative pronoun is case ambiguous.
We present empirical evidence from a number of reading time
experiments that show that this prediction is borne out.
|
