Speaker: Peter Graff (MIT)
Title: Constraints on Possible Meanings
Time: Tues 10/26, 12pm
Location: 46-3189
The meanings of natural language determiners can be understood as relations
between sets (Barwise and Cooper, 1981). It has since been noted that natural
language determiners are crucially constrained in the set relations they can
express. One influential formalization of this typological pattern is
Conservativity (Keenan and Stavi, 1986), a constraint taken to universally hold
for all denotations of natural language determiners. In this talk we propose a
second universal generalization over possible denotations in natural language,
which we term Myopia. We show that this constraint not only excludes many
denotations of natural language determiners previously excluded by
Conservativity, but also generalizes to predicates of all other lexical
categories, including nouns, verbs, adpositions and adjectives. We go on to
present evidence from an on-line artificial determiner learning study (N=217),
showing that both Conservativity and Myopia are active in determiner learning.
Conservative determiners are easier to acquire than non-conservative myopic
determiners, which in turn are easier to acquire than non-conservative
non-myopic determiners providing further evidence that absolute typological
universals may have different relative impact on language learning.