Back to search

University of British Columbia

Beyond noise : the role of speaker variability on statistical learning

Abstract

dc:description

Statistical learning describes a phenomenon where individuals can detect probabilistic patterns in their environment. In language acquisition, this mechanism has been shown to support learners’ induction of phonological categories (Cristia, 2018), words (Black & Bergmann, 2017), and even morphosyntactic rules (Finn & Hudson Kam, 2008). However, studies demonstrating this phenomenon are typically tightly constrained and artificial; when the laboratory conditions are slightly more complex, learning is surprisingly fragile. In this thesis, I explore the idea that these learning conditions are too artificial and fail because they are missing a key component of the natural language environment: variability. Speaker variability—the acoustic variability that occurs when different people produce the same sounds—has been shown to improve the acquisition of novel language structures (e.g., Graf Estes & Lew-Williams, 2015). I hypothesized that adding speaker variability to an otherwise artificial statistical learning task would enhance learners’ ability to detect underlying statistical patterns. Two experiments were designed to assess the extent of listeners’ sensitivity to underlying structure and word segmentation in a statistical learning paradigm. Participants heard one or multiple voices speaking with either English or non-English phonology when listening to the syllable stream. The syllable stream is constructed such that four trisyllabic pseudowords can be extracted based on the transitional probabilities between syllables. Previous work had shown that adult learners fail to extract words when the phonology is unfamiliar (Black, 2018). It was anticipated that speaker variability would enhance learning in this condition. The second of the two experiments was designed to address questions raised by Experiment 1; it implements adjustments to the stimuli and longer exposure to the syllable stream and limits the sample to only monolingual English speakers. Results from both experiments show that speaker variability may interfere with statistical learning, no matter if the phonology is familiar or unfamiliar. This evidence is consistent with the prediction that speaker variability introduces processing demands that compete with tracking underlying statistics. However, a lack of evidence that the control conditions meet the expected learning outcomes indicate the experimental paradigm likely produces unexpected confounds that warrant further testing.

Degree

thesis:*
Name thesis:degree_name
Master of Science - MSc
Level thesis:degree_level
master's
Discipline thesis:degree_discipline
Audiology and Speech Sciences
Grantor dc:publisher
University of British Columbia
Year dc:date
2024

Author and committee

dc:creator, dc:contributor.*
Author dc:creator
  • Melville, Larissa S.
Contributors dc:contributor
  • Black, Alexis K.

Rights

dc:rights
Statement dc:rights
  • Attribution-NonCommercial-NoDerivatives 4.0 International
Language dc:language
eng

Identifiers

dc:identifier.*
Handle dc:identifier
http://hdl.handle.net/2429/87801

Chain of custody

source
Harvested from
University of British Columbia
Base URL
circle.library.ubc.ca/oai/request
Last updated
2026-11-19
Source record
OAI-PMH GetRecord
related terms
citation

Melville, Larissa S.. Beyond noise : the role of speaker variability on statistical learning. master's thesis, University of British Columbia, 2024. http://hdl.handle.net/2429/87801