When the restrictor of a quantifier is not overt, but must be inferred, multiple possibilities are the rule rather than the exception. How is this type of semantic ambiguity resolved? We hypothesize that (1) semantic processing is immediate and incremental and (2) is guided by a principle of 'forward directionality' (Hendriks & De Hoop 2000): 'select the smallest restrictor allowed by the preceding discourse'.
Experiment 1: Completion of interrupted discourses illustrated in (1)below. Each of the three versions of an item allows two discourse-determined (anaphoric) restrictors for the bare quantifier, in (1) the set of boys examined/passed as fit (small restrictor), or the set of unspecified boys (large restrictor). Subjects generally preferred anaphoric interpretations, except for the (c) items. The (a) items pre-dominantly elicited large restrictor responses, the (b) items small restrictor readings. This effect is due to a predominance of predicates in the (a) items which readily evoke a contrast set (e.g. passed as fit > rejected).
Experiment 2: Difficulty rating of items similar to (1), except that the
second sentence was completed with a predicate which was either compatible
(b) or incompatible with (a, c) the small restrictor, e.g. "... were
declared unfit on account of a bad back." In the (c) items however, the
cardinality of the bare quantifier itself already blocks a small restrictor.
Subjects found the (c) items more difficult than the (b) items. This
suggests that they attempted to select a small restrictor, but had to
re-analyze due to the ensuing cardinality conflict. The (a) items were
expected to lead to re-analysis at the predicate, also to be reflected in
the rated difficulty. It appeared, however, that only (a) items containing
non-contrast-evoking predicates were more difficult than the (b) items.
Apparently, contrast-evoking predicates immediately cue a large restrictor
reading. The (c) items were rated as more difficult than the (a) items,
despite identical introductory sentences. This can be related to the earlier
observation that the (c) items tend to evoke non-anaphoric readings.
Conceivably, switching from a non-anaphoric interpretation to an anaphoric
one is more costly than re-analyzing from one anaphoric (small restrictor)
interpretation to another (large restrictor).
The results are in line with the forward directionality principle, but also
show that 'contrastiveness' has an interfering effect. Moreover, if forward
directionality leads to a cardinality clash, anaphoric interpretation tends
to be abandoned right away. The results also suggest that restrictor
selection is immediate, and that reanalysis may ensue if a contradiction
arises.