Use this URL to cite or link to this record in EThOS: https://ethos.bl.uk/OrderDetails.do?uin=uk.bl.ethos.804282
Title: Assessing the impact of co-occurrence frequency and diversity in statistical learning accounts of language processing
Author: Turk, Russell
Awarding Body: Nottingham Trent University
Current Institution: Nottingham Trent University
Date of Award: 2019
Availability of Full Text:
Access from EThOS:
Access from Institution:
Abstract:
Language is heralded as one of the foremost human achievements and is vital in scaffolding the successful development of many other skills. Yet, the mechanism by which language is acquired is still poorly understood. One possible account is Statistical Learning Theory, an explanation of language acquisition that has grown in popularity over the past three decades. The central tenet of Statistical Learning Theory is that learners are guided by statistical regularities in their environment and can utilise these to develop an implicit understanding of their natural language. Current theory holds that transitional probabilities are the best predictor of learner performance in statistical learning tasks. However, little has been done to investigate alternative statistical measures. This thesis presents two such metrics: Bigram frequency and bigram diversity and contrasts them with transitional probability in predicting task performance. Through the repurposing of primed lexical decision and sequence learning tasks, I present a novel approach to examining the impact of statistical priming on task performance in a naturalistic dataset. Model comparison using Bayesian multilevel modelling suggests that transitional probability is not as reliable a predictor as was previously believed. Moreover, I demonstrate that bigram frequency may represent a better metric for predicting task performance in these tasks. The current work highlights the importance of considering alternative metrics of statistical regularity when describing the underlying mechanisms of language acquisition and showcases alternative methods of examining statistical learning performance.
Supervisor: Not available Sponsor: Not available
Qualification Name: Thesis (Ph.D.) Qualification Level: Doctoral
EThOS ID: uk.bl.ethos.804282  DOI: Not available
Share: