Back to search

Digital Repository at the University of Maryland

Islands in a Sea of Statistics

Abstract

dc:description.abstract

Children learn the language of their environment, so of course that environment plays a crucial role in shaping the language they come to speak. However, while the language environment is in principle consistent with many possible grammars, humans overwhelmingly converge on consistent mental grammars, exploiting generalizations across disparate phenomena in language. In order to account for the generalizations that form human grammars, we must understand the particular inductive bias the human child brings to the task of language learning. In this dissertation, I explore what type of inductive biases might lead to human generalizations in A' movement. To do so, I consider two case studies. In the first case study, I consider cross construction generalization of island constraints. In a series of experimental studies, I show that humans are sensitive to islands formed by clefting for both cleft and relative clauses and that they are sensitive to islands formed by relative clauses for both clefts and relatives. I argue that this reciprocality is evidence of a shared generalization. In a computational study with large language models, I show that an alternative hypothesis, that the shared generalization can be learned through string-level distributional information in language environment is not born out. In the second case study, I consider crossover effects. I show that large language models do not come to human-like generalizations with crossover and argue that these models reveal that the language environment is consistent generalizations that nevertheless do not appear to be recognized by humans. I then turn to the question of how crossover might be learned by turning to methodologies for language acquisition studies. In this final series of behavioral experiments, I show that uncovering children's knowledge of crossover is not straightforward and previous claims of their robust knowledge might be overblown. I then discuss how our methodologies might be better refined.

Degree

thesis:*
Department dc:contributor.department
Linguistics
Grantor dc:publisher
Digital Repository at the University of Maryland
Year dc:date.issued
2026

Author and committee

dc:creator, dc:contributor.*
Author dc:creator
  • Howitt, Katherine Gandolfo
Advisors dc:contributor.advisor
  • Lidz, Jeffrey
  • Phillips, Colin

Rights

Language dc:language.iso
en

Identifiers

dc:identifier.*

Chain of custody

source
Harvested from
University of Maryland
Base URL
api.drum.lib.umd.edu/server/oai/request
Last updated
2026-11-19
Source record
OAI-PMH GetRecord
related terms
citation

Howitt, Katherine Gandolfo. Islands in a Sea of Statistics. Digital Repository at the University of Maryland, 2026. http://hdl.handle.net/1903/36331