Referential
processing: Evidence from eye tracking and ERPs
Peter C. Gordon
University of North Carolina at Chapel Hill
Tamara Y. Swaab and C. Christine Camblin
Duke University
Language comprehension involves the creation and the maintenance of
a discourse model that embodies the meaning of a sentence (or a series
of sentences) and which serves as a basis for the continued interpretation
of linguistic input. Processing referential expressions can in some
instances lead to the creation of new entities in the discourse model
that must be stored in working memory while in other instances existing
entities in working memory must be accessed in order to achieve a coreferential
interpretation. A coordinated pair of studies, one tracking eye movements
during normal reading and the other measuring ERPs during RSVP reading,
was conducted to examine the impact of the memory requirements imposed
by the number of entities in the discourse model and to examine how
working memory is accessed during the comprehension of coreferential
expressions. The experimental sentences manipulated two factors: (1)
whether the first clause of the sentences had one name (singular condition)
or two names (conjoined condition); (2) whether the referential expression
in the second clause was a repeated name or a pronoun (see example).
John (and Mary) went to the store so that he/John could buy some candy.
Both eye tracking and ERP measures showed effects of the number of names
on the comprehension of the words following the matrix subject but before
the coreferential expression. Gaze durations on those words were longer
and ERPs showed a stronger negativity in the N400 region in the two-name
condition than in the one-name condition. In
addition, both measures showed that the type of coreferential expression
(pronoun versus repeated name) interacted with the number of names in
the first clause. Of particular interest, gaze durations to repeated
names were longer in the one-name (singular) condition than in the two-name
(conjoined) condition, a pattern that was matched by a larger N400 to
the repeated name in the one-name than in the two-name condition. This
pattern is consistent with previous findings showing that repeated names
are processed more readily when referring to non-prominent referents
than to prominent referents.
The results show that eye tracking and ERPs provide consistent evidence
about referential processing in discourse, with respect both to the
memory requirements imposed by representing multiple discourse entities
and with respect to how those representations are accessed during the
establishment of coreferential relations.