|
Abstract:
Construction grammar (CG) is a unification grammar in which
lexical items, idioms, and grammatical constructions are all
represented in the mental lexicon. Besides treating lexical and
syntactic processing equivalently, CG suggests the existence of
complex, non-local constructions with pragmatic constraints on
their well-formedness. For example, the distribution of the
let-alone construction (as in (1))
(1) Tony can't run 3 miles, let alone 5.
requires an account of its pragmatic and rhetorical purposes
(Fillmore, Kay, and O'Connor, 1988). This construction conjoins
two propositions, both of which can be interpreted on the same
scalar model,
a set of propositions ordered along some dimension such that
elements are related by material implication. So, if Tony can't
run 3 miles, a fortiori he can't run 5.
The let-alone construction can be violated in two obvious ways
as suggested in (2b) and (2c). The Polarity violation occurs
because "let alone," a negative polarity item, requires a
negative trigger. The Scalarity violation occurs because the
reordering of the elements fails to preserve the entailments of
the scalar model. Similarly, the positive polarity item "if not"
can also coordinate propositions on a scalar model, and is
subject to analogous violations of Polarity and Scalarity.
(2a) (Acceptable) They weren't friends, let alone lovers.
(2b) (Polarity) *They were friends, let alone lovers.
(2c) (Scalarity) *They weren't lovers, let alone friends.
(2d) (Acceptable) They were friends, if not lovers.
(2e) (Polarity) *They weren't friends, if not lovers.
(2f) (Scalarity) *They were lovers, if not friends.
We tested readers' (n=24) sensitivity to these two aspects of
constructional acceptability in a word-by-word self-paced reading
task. Stimuli in our 2x3 within-participants design included
sentences structured like those in (2). The overall analysis on
RTs for sentence-final words indicate reliable effects of
Construction (with longer RTs for if-not sentences) and
Acceptability. Separate analyses of Polarity and Scalarity
suggest both violations induced increased reading times relative
to Acceptable sentences: a 604 millisecond effect of Polarity,
and a 194 millisecond effect of Scalarity. Increased RTs for
Scalarity Violations are not attributable to lexical differences
of the sentence-final items in (2c/f) and (2a/d) since an
equivalent comparison at the last word of the first clause was
null (Acceptability (F(1,23) < 1)). Results suggest readers
were sensitive to both kinds of violations, although the response
to categorical Polarity violations was clearly more pronounced.
We suggest that the Polarity violations (2b/e) are more
disruptive because they prevent constructional access, while
Scalarity violations (2c/f) allow access but prompt readers to
conceptualize sentence content in an untenable manner. We discuss
the adequacy of Jurafsky's (1996) probabilistic parsing algorithm
to explain these results.
Fillmore, C., Kay, P., and O'Connor, M.C. (1988). "Regularity
and idiomaticity in grammatical constructions: The case of let
alone."
Language
64, 501-538.
Jurafsky, D. (1996). "A probabilistic model of lexical and
syntactic access and disambiguation."
Cognitive Science
20, 137-194.
|
