Learning nonadjacent dependencies embedded in sentences of an artificial language: When learning breaks down

Area of Science:

• Cognitive Science
• Developmental Psychology
• Linguistics

Background:

• Natural language structure involves dependencies within word sequences.
• Infants and adults detect adjacent statistical properties, but nonadjacent dependency learning is limited.
• Existing theories propose different mechanisms for adjacent and nonadjacent dependency learning.

Purpose of the Study:

• To challenge the view that distinct mechanisms underlie adjacent and nonadjacent dependency learning.
• To investigate how sequence length and silences affect the learning of both dependency types.
• To propose a unified theory of dependency learning mechanisms.

Main Methods:

• Four experiments were conducted to test dependency learning.
• Participants processed sequences with adjacent and nonadjacent dependencies.
• Experimental conditions manipulated sequence length and the presence of silences.

Main Results:

• Learning of both adjacent and nonadjacent dependencies was hindered in longer sequences.
• Isolation by silences facilitated the learning of both dependency types.
• Findings suggest NAD learning is computationally more complex than adjacent dependency learning.

Conclusions:

• Similar learning mechanisms are deployed for adjacent and nonadjacent dependency detection.
• Nonadjacent dependency learning success is enhanced by constraining information, similar to adjacent dependency learning.
• The computational complexity of nonadjacent dependencies explains observed limitations in learning.