| |
Abstract:
According to Hankamer and Sag (1977), there is a distinction
between surface anaphora (SA) and deep anaphora (DA). SA are
exemplified by reflexive pronouns, traces, and VP-ellipsis, whereas
DA are exemplified by expressions such as "do it", "do so", "the
same thing". Antecedents of SA are determined by and highly
contrained by the syntactic form of the expressions containing them
and anteceding them, whereas DA find their antecedents in
non-syntactic, propositional discourse representations, and are
therefore less constrained by grammatical form. In terms of
sentence processing, this leads to the expectation that SA should
show evidence of highly automatic and reflexive processing (i.e
show immediate antecedent reactivation), whereas DA resolution
should be more sensitive to context effects and have a "drawn out"
time course for anaphora resolution, a characteristic of slower and
less automatic higher cognitive processing. We examined these
predictions with several experiments using behavioral and
electrophysiological measures of the time course of antecedent
reaccess. An all-visual lexical decision RT experiment was run to
replicate the finding by Nicol and Swinney (1989) that reflexive
pronouns immediately reactivate their antecedents. Subjects were
presented with Norwegian sentences illustrated in (i) on a
word-chunk by word-chunck basis, and the probe word appeared
displaced on the screen in either the temporal position 1 (before
the anaphor) or 2 (immediately after the anaphor). Subjects had to
determine whether the probe was a word or a non-word by pressing a
button as fast as possible. An ANOVA with probe position as group
factor and probe relation as within-subject factor revealed a
significant main effect of probe relation, F(1,34)=7.82, p <
.011 (mean RT to related probes: 820ms; to unrelated probes:
865ms). Furthermore, RTs to related probes was significantly
shorter than RTs to unrelated probes in position 2, and RTs to
related probes was significantly shorter in position 2 than in
position 1, as verified with t-tests (p < .005 and p < .045).
Thus, Norwegian reflexives immediately reactivate their
antecedents.
Another experiment similar in design was run but with the
Norwegian deep anaphor "gjoere det" ('do so/do it') instead of a
reflexive pronoun, and with probe positions before the anaphor,
immediately after the anaphor, and in a "down-stream" position 1000
ms after the onset of the anaphor to test for late reactivation
(see (ii)). We here found no effect of probe relation immediately
after the anaphor or in the downstream position. Thus, the
Norwegian deep anaphor shows no immediate antecedent reactivation
effect.
In addition, an ERP experiment was conducted with stimuli from
the surface anaphora reflexive experiment, partly in order to
explore the suitability of this measure for reaccess studies, and
partly to provide a possibly more finegrained data about the time
course of reactivation. In this experiment, subject were watching
the sentence being presented on a screen, and the probe was flashed
in a displaced position as in the RT-experiments. Subjects were
however instructed to delay their word/non-word decision until the
end of the sentence. We expected that if a probe was unrelated to
the information activated by the reflexive, a semantic incongruity
effect should be generated, and an N400 should occur. If the probe
was related, however, no semantic inconguity effect was expected.
Data were analyzed in difference waveforms (ERPs to unrelated minus
related probes) as mean amplitudes in time ranges. Amplitudes at Cz
in the time ranges 400-500, 500-600 and 600-700 ms were
significantly negative from zero in position 2, only (one-tailed
t-tests, ps < .033, .022 and .013, respectively). These effects
evidence that ERPs are sensitive to reaccess of the antecedent to
reflexives.
|
