Natural languages instantiate one-to

advertisement
Learning Non-Adjacent Dependencies: A Mechanism for Language Acquisition?
Natural languages instantiate one-to-one correspondences between non-adjacent
elements that mark relevant grammatical relations. In a sentence like Chomsky is
always driving Tomasello crazy, the dependency between the auxiliary be and the
suffix –ing on the verb marks an important morpho-syntactic relationship between the
functional domain of auxiliaries and the lexical domain of the VP. Infants as young as
18 months are already sensitive to such non-adjacent dependencies (NADs) in their
native language, and a common claim in the literature is that NADs are detected based
on a distributional learning mechanism which keeps track of co-occurrence statistics
(i.e. probability or frequency of co-occurrence) between non-adjacent elements
(NAD-learning mechanism).
However, NADs in natural languages are usually instantiated between functional
elements, or functors, which have been shown to differ in their perceptual properties
from lexical elements (cf. Shi, Morgan & Allopena, 1998). Namely, functors are
perceptually ‘reduced’, and thus less salient than lexical words, a trait which has been
shown to affect their acquisition (cf. Shi, Werker & Cutler, 2006, among others).
The question addressed in this talk is whether and how the perceptual properties of
functors affect the ability to detect and learn discontinuous dependencies between
them. I propose (and discuss the technicalities of) an experiment that wishes to
directly answer two questions: 1) is NAD-learning facilitated or inhibited when the
dependent elements have the acoustic make-up of functors? and 2) can we thus show
that NAD-learning, as exhibited in artificial grammar learning studies, is a potentially
useful tool for L1 acquisition?
Download