| |
Abstract:
A widely-held belief in psycholinguistics is that sentence
comprehension is an all-or-nothing process; readers either derive a
fully interpreted representation of a sentence, or, if the sentence
for example contains a difficult garden path, fail to arrive at any
coherent interpretation (e.g., MacDonald, Pearlmutter and
Seidenberg, 1994; Frazier and Clifton, 1997).However, a recent set
of experiments (Christianson, Hollingworth, Halliwell, and
Ferreira, submitted) used probe questions following ambiguous
sentences to cast doubt on this premise; participants were shown to
be likely to give incorrect answers, reflecting thematic role
assignments which should have been discarded when the ambiguity was
resolved, particularly in cases where the parser would have
retained an incorrect thematic interpretation for a certain amount
of time. Christianson et al. interpreted their findings as showing
that the human parsing process may simply result in an
interpretation that is "good enough" for the task in hand, even if
that interpretation contains unresolved inconsistencies.
The present study extends the exploration of "good enough"
interpretations, measuring reading times for sentences in which the
role played by an (unambiguous) postnominal relative clause is
contextually constrained to be pragmatically more or less important
to the understanding of the assertion. An example pair of materials
is given in (1). If such sentences are fully interpreted, we would
expect the reading time for (1a) to be shorter than that for (1b),
because the pronoun anaphor within the relative clause is congruent
with social expectations about the gender of the sentential subject
(in this case, "the engineer"; cf. Carreiras, Garnham, Oakhill, and
Cain, 1996).
(1) a. The engineer who had lost his hard hat argued with the
foreman.
b. The engineer who had lost her hard hat argued with the
foreman.
However, the pragmatic salience of the relative clause can be
manipulated by preceding it with a context mentioning either one or
two engineers.
Following Crain & Steedman (1985), we assume that in the
case where two engineers have entered the discourse, it is central
to the representation of the assertion made in (1) to know which of
the two engineers had an argument; that is, it is important to
fully comprehend the relative clause, which will be interpreted
restrictively. In the case where there is one engineer, on the
other hand, the noun phrase "the engineer" already refers uniquely
and it is more likely that the relative clause will be interpreted
nonrestrictively, as simply supplying additional information about
the engineer. In the latter case, a "good enough" representation of
the main assertion in (1) may be obtained without fully
interpreting the relative clause; the difference between (1a) and
(1b), which is predicted for a two-antecedent context, may
disappear in the one-antecedent case.
Evidence from the study directly supported this hypothesis. In
two-antecedent contexts, there was a clear difference in reading
times between (1a) and (1b), but there was *no* difference when the
target sentences were preceded by one-antecedent context sentences.
These findings, together with those of Christianson et al.,
strongly suggest that it is not the case that a fully-interpreted
representation of a sentence is the only alternative to failure to
interpret; indeed, it may be the case that the parser makes
pragmatic cost/benefit decisions about the importance of
interpreting, or reinterpreting, elements of a sentence. Moreover,
the present study demonstrates that one aspect of such decisions
may be determined by the contexts in which sentences occur, and the
pragmatic functions served by their constituents in relation to
those contexts. Two follow-up studies, including an eyetracking
experiment, are currently in progress.
|
