Constraints on phonological statistical learning in adults

Auditory statistical learning studies typically explore whether participants can compute a given statistical pattern from an artificial language (e.g., classic transitional probability studies). Even when studies include several statistical cues, these cues often coincide (e.g., exploring how prosody impacts transitional probability learning). However, outside of controlled experiments, statistical learners face an overwhelming number of cues and need to determine what statistics to compute over others. The aim of this thesis was to identify the statistical information adults computed from artificial languages with concurrent, competing cues. Participants heard artificial languages containing sequences of CVC words. The voicing (voiced, voiceless), manner of articulation (plosive, nasal, fricative) and non-adjacent pairings of consonants were controlled so each language contained three concurrent types of statistical information: the single-feature bias (most codas were voiced), the multiple-feature bias (most codas for one manner of articulation - e.g., fricatives - were voiceless), and the within-word consonant relationships (onsets and codas co-occurred). Statistical learning was measured using a 2AFC task. Sixteen experiments contributed to this thesis (guide in Appendix 10). Convergent evidence suggested adults computed the single-feature bias (Experiment 1, Experiment 2, Experiment 3, Experiment 4, Experiment 9, Experiment 10, Experiment 11, Experiment 12). Participants computed this bias when carried by plosive and nasals (Experiment 1, Experiment 2, Experiment 9, Experiment 10, Experiment 11, Experiment 12) and fricative and nasals (Experiment 3, Experiment 4). Remarkably, adults computed this bias after 14 seconds of exposure (Experiment 12); generalising to new words after 7 minutes (Experiment 10). Adults also computed the multiple-feature bias (Experiment 5, Experiment 6, Experiment 7), but only when it was strong (≥90%), and only generalised this bias when it was presented with a weak concurrent single-feature bias (Experiment 7). There was a lack of evidence to suggest adults computed the within-word consonant relationships. A novel word search task provided supplementary evidence (Experiments 13a and 13b, Experiments 2, 5-7, Experiments 9-12). Adults readily compute statistics over one phonological feature (voicing) and can generalise to new words. Computations over two phonological features (voicing and manner of articulation) are possible given the right context. There are constraints on our auditory statistical learning mechanism to compute certain cues when faced with concurrent statistics. However, flexibility in our mechanism means we can circumvent these constraints. The nature of these constraints, the factors driving statistical learning and the theoretical implications are discussed.

Grantor dc:publisher

Oxford Brookes University

Year dc:date

2023

Author dc:creator

• Bond, Samuel

Contributors dc:contributor

• Pilling, Michael
• Gonzalez-Gomez, Nayeli
• Afonso, Olivia

Statement dc:rights

• All rights reserved

Language dc:language

en

DOI dc:identifier

https://doi.org/10.24384/hjd9-av19

OAI identifier oai:identifier

tle:464439b3-191b-455d-9b33-3da976e55b35:d6bd9758-527a-46cd-bfe2-c433766e8fca:1