Paper

Reassessing a model of syntactic island acquisition

---

Abstract

This paper examines the limits of the learning model for syntactic islands from Pearl and Sprouse (2013), which challenges linguistic nativist perspectives by suggesting that island effects can be learned from language input and domain-general or learned abilities. Our investigations focus on sentences that would be ambiguous if there were no island constraints, where one conceivable interpretation violates an island constraint. A learner without any knowledge of islands could incorrectly treat the island-violating parses of such sentences as grammatical. We conducted simulations introducing these sentences in the model’s input and also analyzed their frequency in the childdirected speech corpora used as the model’s input. The results show that a small number
of potentially island-violating sentences in the model’s input impairs its ability to exhibit island effects, and potential island violations occur frequently enough in children’s input to degrade the model’s performance.

Keywords: syntactic islands

How to Cite:

Gulrajani, A. & Lidz, J., (2024) “Reassessing a model of syntactic island acquisition”, Society for Computation in Linguistics 7(1), 43–51. doi: https://doi.org/10.7275/scil.2128