|
Abstract:
In this study we use gender agreement to investigate the
insulation of syntactic operations from discourse information in
language production. In romance languages, every noun is marked
for gender, however even for nouns referring to humans the
correspondence between the gender of the noun and the sex of the
referent is not always transparent. For example, there exists a
class of animate nouns referring to humans which nevertheless
bear an arbitrary grammatical gender, also called surface
gender.
We exploited such nouns to manipulate the congruency between
grammatical and conceptual features of gender and to test the
effect of conguency on agreement computation. Nouns with surface
gender were put in a referential context such that their
grammatical gender could either match the biological sex of the
person they referred to or mismatch it. Participants first saw a
context sentence introducing either a man or a woman:
Les violences se sont déroulées sous les yeux de
Jean/Jeanne qui était sur son balcon.
(The violence occurred in front of Jean-M/Jeanne-F who was on
his/her balcony)
Following this, an adjective appeared on the screen presented
in both masculine and feminine forms:
suspect
(suspect-M)
suspecte
(suspect-F)
Immediately after the adjective, a sentence preamble appeared
which referred to the person introduced in the first
sentence:
Le témoin de la bagarre ...
(The witness-M of the fight ...)
Participants had to complete this preamble using the copula
'est' (is) and the adjective they had just been shown.The subject
head noun within the preamble had a grammatical surface gender
that either matched the conceptual gender of the person it
referred to (témoin-M, Jean-M) or mismatched it
(témoin-M, Jeanne-F). Note that the subject head noun
'témoin' is masculine regardless of whether it refers to a
male or female. We recorded the number of gender agreement errors
between the subject head and the adjective.
Agreement errors when grammatical and conceptual gender
matched were relatively rare (3.2). However, when they
mismatched, a high proportion of errors ensued (10.8), with the
adjective being made to agree with the sex of the referent
introduced in the contextual sentence (Jeanne-F) rather than with
the grammatical gender of the subject noun (témoin-M).
This result implies that conceptual information from the
discourse model is accessible to the process that computes
agreement between a subject noun and an attribute, and therefore
that discourse level representations can exert a fine control on
some syntactic operations.
|
