| |
Abstract:
According to Fodor & Inoue (1994), reanalysis operations
triggered by a syntactic cue, as in (1a), are easier to carry out
than reanalysis operations triggered by plausibility information,
as in (1b). Experiment 1 consisted of a self-paced word-by-word
moving window experiment conducted on examples like these along
with controls in which the disambiguating PP in (1) ("in/into the
ice cream") is attached as an argument of "put", so that reanalysis
is not required. The results of the experiment confirmed the
intuition: The reanalysis effect in examples like (1b) was
significantly larger than that in examples like (1a).
However, these results are compatible with two alternative
explanations. First, (1a) contains two cues for reanalysis: the
syntactic cue (e.g., "into") and the plausibility cue (e.g., the
unlikelihood of the bowl being in the ice cream). In contrast, (1b)
contains only the plausibility cue. The faster reading times in
(1a) could therefore be due to the presence of an additional cue to
reanalysis. Second, the reanalysis cue in (1a) ("into") is
available two words earlier than the reanalysis cue in (1b) ("ice
cream"). If the difficulty of reanalysis depends on how long the
parser retains its initial analysis, then reanalysis may be more
difficult in (1b) than (1a).Experiment 2 was designed to address
the same issue while avoiding these confounds. Within the two
conditions illustrated in (2a) and (2b), the structures contain a
single unambiguous cue (in the form of phrase structure or temporal
information) which tells the parser that the preferred local
attachment of the phrase is impossible and the dispreferred
non-local attachment must be pursued. In (2a), the goal PP "into
the suitcase" is incompatible with the phrase structure
subcategorization of the local VP "borrow", but it is compatible
with the non-local VP headed by "packed". In (2b), the temporal
adverb "yesterday" is incompatible with the tense of the local VP,
but it is compatible with the tense of the non-local VP. Note that
the cues become available to the parser at the same point within
the two sentences, i.e., immediately after the embedded verb.
The results of Experiment 2 are compatible with Fodor &
Inoue's hypothesis distinguishing syntactic and "semantic" cues to
reanalysis. The reading times for two additional conditions
containing local attachments of "into..." and "yesterday" were not
significantly different either at the point of disambiguation or at
the clause boundary, so that these cues appeared to be equally
effective in local attachments. However, in the two conditions that
contain non-local attachments (2a) and (2b), the semantic
(temporal) disambiguation towards the dispreferred structure was
less effective than the syntactic (subcategorization-based)
disambiguation. Sentences like (2b) exhibited a significant
slowdown at the clause boundary ("before the weatherman...").
In the paper we discuss two additional features of the
experiment that go beyond Fodor & Inoue's hypothesis. First,
the "semantic" cues utilized in sentences like (2b) differ from the
sort of "semantic" cues that have been studied previously in that
they do not rely on world knowledge or judgements about the
plausibility of certain situations. Rather, the cue in (2b) is
based on temporalinformation which is represented within a sentence
by morphosyntactic means. This raises the possibility that the
parser distinguishes between cues based on a specific type of
syntactic information (i.e., subcategorization information) and all
other types of cues (semantic, word-knowledge, and other
syntactic). Second, given the structure of sentences in (2), it is
possible that the experiment is tapping into the differences within
first analysis mechanisms, rather than the differences within
reanalysis mechanisms.
|
