|
Abstract:
A considerable amount of work suggests that the comprehension
of anaphora is affected by the implicit causality information
associated with verbs (e.g. Caramazza, Grober, Garvey and Yates,
1977). However, there has been little consideration of how other
aspects of verb semantics can affect anaphoric processing. In
this talk, we define another type of verb semantic information
which we call implicit consequentiality, and investigate its
effects on anaphoric resolution.
In sentence-completion tasks, we found that for fragments such
as 'Because Harold dreaded Justin...', there was a strong
preference to write a continuation that provided information
about the consequences associated with the first mentioned
character, Harold, rather than with the second mentioned
character, Justin. Hence, the verb 'dread' is an NP1 biasing verb
for consequentiality. We also found a set of NP2 biasing verbs,
in which there was a strong preference to write a continuation
about the second mentioned character. In general, implicit
consequentiality biases were distinct from implicit causality
biases.
We report two reading experiments that demonstrate effects of
implicit consequentiality on comprehension. Subjects read
sentences like:
1. Because Harold dreaded Justin, he steadfastly refused to go
back to school.
2. Because Harold dreaded Justin, he was told to try acting
less aggressively.
The verb 'dreaded' is NP1 biasing. In (1), the continuation is
compatible with this bias, but in (2) the continuation is
incompatible with this bias. We found an implicit
consequentiality congruency effect, with the continuation in (1)
being read faster than the continuation in (2). This is similar
to the implicit causality congruency effect (e.g. Stewart et al.,
1996).
In conditions in which the pronominal anaphor was replaced
with a repeat name, we found a repeat name penalty (e.g. Gordon,
Grosz and Gillion, 1993) that was independent of verb bias. This
penalty simply corresponds to a reading difficulty associated
with repetition of the first mentioned character's name. We
suggest that implicit consequentiality does not affect the
initial interpretation of sentences containing repeat name
anaphors.
A second experiment found that the congruency effect was
larger if the pronouns were locally ambiguous (as in (1) and (2))
than if they were disambiguated by gender. This suggests that
readers can use implicit consequentiality information in the
initial processing of pronouns. This provides a clear contrast
with our findings on the processing of implicit causality
information.
It is possible that the processor can access implicit
consequentiality more rapidly than implicit causality. We argue
against this interpretation, and instead suggest that the
processor pays immediate attention to resolving pronouns in the
subject position of main clauses, as in (1) and (2).
|
