|
Abstract:
During sentence processing, words are continuously recognized
and integrated with the existing syntactic and semantic
representation of a sentence. This integration is often supposed to
be a rather 'passive' match of the lexical content of the incoming
word with the representation that has been built so far. However,
in some cases people are quite good at actively predicting the next
word of a given sentence, at least in off-line tasks. The question
is to what extent do readers use this predictive ability during
on-line sentence processing. To investigate this issue we conducted
an ERP-experiment, using sentences in which the syntactic
structure, together with the content of the preceding words, leads
the reader to expect a specific lexical item, such as 'gebakken'
(baked) in 1a (the sentences are in Dutch; the literal English
translation is given in parentheses).
1a. Het bruine brood werd door de bakkers GEBAKKEN.
1b. Het bruine brood werd door de bakkers BEDREIGD.
(lit: The brown bread was by the bakers BAKED / THREATENED.)
In 1b, this expectation is violated by the occurrence of
'bedreigd' (threatened), which makes the sentence implausible, and
will presumably lead to difficulty in semantic integration. This
difficulty will most likely lead to a N400-effect. However, the
word 'bedreigd' not only violates a lexical prediction, it is also
semantically unrelated to the words that make up the rest of the
sentence. Passive, bottom-up integration would thus also predict a
N400- effect. Therefore, we included two types of implausible
sentences where the final words are either related or unrelated,
but in which the syntactic structure does not support any
expectation for a specific lexical item, as in 2a and 2b.
2a. Het bruine brood heeft de bakkers GEBAKKEN.
2b. Het bruine brood heeft de bakkers BEDREIGD.
(lit: The brown bread has the bakers BAKED / THREATENED.)
We found a significant interaction between lexical prediction
(prediction vs. no prediction) and semantic relatedness (related
vs. unrelated) on the amplitude of the N400 elicited by the final
words. The N400-effect (relative to condition 1a) was smallest for
sentences such as 2a, where the implausible word is semantically
related but where there is no specific lexical prediction, and
largest for sentences as 1b, which end with an unrelated word which
does violate a specific prediction. The N400-effect for sentences
with an unrelated word and no specific prediction (2b) was exactly
in between. This suggests that syntactic structure and lexical
content together create specific predictions of a lexical item;
processing a word that does not fit this prediction then Becomes
quite difficult.
Unexpectedly, we also found an interaction of lexical prediction
and semantic relatedness on a P600-like effect which was present in
all implausible sentences, relative to plausible ones such as 1a.
The effect was largest for implausible sentences containing a
semantically related word (1a), which presumably indicates that the
largest effort to reanalyze an implausible sentence is made when
the incorrect word bears a semantic relation to the other words in
the sentence. This is compatible with the view that the P600
reflects a (syntactic) reanalysis process which can be initiated by
either semantic or syntactic violations.
|
