| |
Abstract:
One of the more active debates in the parsing literature
concerns the time course over which lexical information exerts its
influence during on-line sentence comprehension. Two distinct
hypotheses can be identified: (1) the processor constructs an
initial parse based only on structural information (Ferreira &
Clifton, 1986; Frazier, 1987); (2) the processor accesses and uses
lexical information very rapidly (MacDonald, Pearlmutter, &
Seidenberg, 1994; Trueswell & Tanenhaus, 1994). In the work
reported here we examined the influence of verb control information
in on-line sentence processing. Lexically-specified control
information determines, for a particular verb, which of its
arguments (subject or object) will be the antecedent of, i.e.
controls, the implicit subject of its infinitival complement. The
processor, in order to correctly coindex the implicit subject with
its antecedent, must access and use this verb-specific information.
Event-related brain potentials (ERPs) were recorded as
Spanish-speaking subjects listened to sentences presented as
continuous, natural speech. The critical sentences contained an
NP-adjective gender agreement violation in the complement, and
these were directly compared with grammatical version. Spanish
grammar requires that an adjective agree in number and gender with
the NP it modifies. We studied the effects of the lack of such
agreement in clauses where the adjective is predicated of the null
subject of a controlled infinitival complement. The gender of the
phonologically null subject must be inherited from its antecedent.
Two classes of verbs were used: (1) subject control verbs and (2)
object control verbs. In each item, the main clause was followed by
an infinitival complement, identical in the grammatical (a) and
ungrammatical (b) versions. By alternating the gender of the
appropriate antecedent for the null subject, a gender agreement
violation was introduced in half of the sentences.
1 Subject Control Verb
a. Pedro ha prometido a Maria ser estricto con los alumnos
b. *Maria ha prometido a Pedro ser estricto con los alumnos
(Pedro(masc)/Maria(fem) has promised Maria(fem)/*Pedro(mas) to be
strict(masc) with the students)
2 Object Control Verb
a. Maria ha aconsejado a Pedro ser educado con los trabajadores
b. *Pedro ha aconsejado a Maria ser educado con los trabajadores
(Maria(fem)/Pedro(masc) advised Pedro(masc)/ *Maria(fem) to be
polite(masc) with the employees)
The detection of such a grammatical violation would imply that the
processor had established the coreference relation between the null
subject and its antecedent. We hypothesized that violating
sentences would evoke a positive-going wave, similar to that found
in previous studies of agreement violations (Osterhout \&
Mobley, 1995). The results showed that ungrammatical sentences
elicit a biphasic ERP pattern with a negativity followed by a
positivity, indicating the early influence of verb control
information in sentence processing. The results will be discussed
in relation to current models of sentence processing.
|
