Dissertation Defense: Alix Fetch
Candidate Name: Alix Fetch
Advisor: Elissa Newport, Ph.D.
Title: Does Learnability Predict Syntactic Universals? An Investigation Using Artificial Languages
Universals in natural language have been a focus of the generative syntactic literature. However, the source of these universals is not clear. Within Chomskyan generative syntactic literature, it was originally assumed that children were endowed with innate knowledge of natural language structure (see for example, Lightfoot 1999). However, research on language acquisition has shown that learning biases may partially explain why these universals arose (Newport 1981; Morgan, Meier & Newport, 1987; Culbertson et al., 2012; Culbertson & Newport, 2015; Culbertson & Newport, 2017). In a series of artificial language experiments, we have attempted to add further evidence to the hypothesis that biases in learning mechanisms give rise to language universals.
In our first study, we ask whether children can use distributional information to acquire the syntactic categories, phrases, and ultimately the sentence structure of a language. Our results show that children are able to use statistical learning mechanisms to acquire syntax from distributional cues to phrase structure. We then ask whether the addition of local asymmetry in a phrase structure grammar results in enhanced learning of sentence structure. Our results show that participants who were exposed to the locally asymmetric grammar learned the sentence structure better than those who were exposed to the control grammar, providing evidence for our hypothesis that local asymmetries bolster learning of sentence-length sequences.
We follow up on this finding by asking whether other types of local asymmetry are equally learnable. Our results suggest that only head-initial patterning was of additional learning benefit. We argue this may be due to effects of the head-initial native language (English) of our participants, in line with previous research by Onnis & Thiessen (2013).
Finally, we attempt to address the question of whether the universal Final-Over-Final Constraint (Holmberg, 2000) (FOFC) is a result of a bias in learning mechanisms. Our results showed that while there was no significant difference between the four grammars, when examined by FOFC category (allowed vs. disallowed) there was a significant difference between the two groups.
We conclude by discussing our findings and propose future work that will build upon the foundation we present here.