{"id":{"repo_id":"msu","oai_identifier":"oai:d.lib.msu.edu:etd_52643"},"canonical_url":"https://search.ndltd.org/etd/msu/oai:d.lib.msu.edu:etd_52643","repository":{"repo_id":"msu","name":"Michigan State University","base_url":"https://d.lib.msu.edu/oai"},"display":{"title":"Understanding Subsegmental Speech Perception with Bayesian Modeling","abstract":"Thesis (Ph.D.)--Michigan State University. Linguistics - Doctor of Philosophy, 2025","abstract_html":"Thesis (Ph.D.)--Michigan State University. Linguistics - Doctor of Philosophy, 2025","abstract_has_math":false,"creators":["Ye, Yongqing"],"institution":null,"degree_name":null,"degree_level":null,"degree_discipline":null,"degree_department":null,"school":null,"contributors":["Durvasula, Karthik","Wagner, Suzanne","Sneller, Betsy","Bongiovanni, Silvina","Lin, Yen-Hwei"],"advisors":[],"committee_chairs":[],"committee_members":[],"year":2025,"date_issued":"2025","date_published":"2025","updated_at":"2026-11-18T18:59:09Z","subjects":["Linguistics","Cognitive psychology"],"languages":["eng"],"rights":["Attribution-NonCommercial-NoDerivatives 4.0 International"],"rights_urls":[],"identifier_entries":[{"key":"dc:identifier","label":"Identifier","values":["etd:52643","local:Ye_grad.msu_0128D_21609"],"render_values":[{"text":"etd:52643","href":null,"code":true},{"text":"local:Ye_grad.msu_0128D_21609","href":null,"code":true}]}],"person_identifier_entries":[]},"links":{"outbound_url":"https://doi.org/doi:10.25335/recz-1p13","outbound_label":"DOI","outbound_source":"dc:identifier","record_url":"https://doi.org/doi:10.25335/recz-1p13","record_label":"Repository record","record_kind":"doi","record_source":"dc:identifier","download_url":null,"download_label":null,"download_source":null,"access_status":null,"embargo_until":null,"access_note":null},"source_record":{"url":"https://d.lib.msu.edu/oai?verb=GetRecord&metadataPrefix=oai_dc&identifier=oai%3Ad.lib.msu.edu%3Aetd_52643","prefix":"oai_dc"},"metadata_groups":[{"id":"people","label":"People","entries":[{"key":"dc:contributor","label":"Contributor","values":["Durvasula, Karthik","Wagner, Suzanne","Sneller, Betsy","Bongiovanni, Silvina","Lin, Yen-Hwei"]},{"key":"dc:creator","label":"Author","values":["Ye, Yongqing"]}]},{"id":"academic_context","label":"Academic Context","entries":[{"key":"dc:date","label":"Dc Date","values":["2025"]},{"key":"dc:relation","label":"Dc Relation","values":["Electronic Theses & Dissertations"]},{"key":"dc:type","label":"Dc Type","values":["text","Electronic dissertations"]}]},{"id":"subjects_keywords","label":"Subjects and Keywords","entries":[{"key":"dc:subject","label":"Dc Subject","values":["Linguistics","Cognitive psychology"]}]},{"id":"language_rights","label":"Language and Rights","entries":[{"key":"dc:language","label":"Dc Language","values":["eng"]},{"key":"dc:rights","label":"Dc Rights","values":["Attribution-NonCommercial-NoDerivatives 4.0 International"]}]},{"id":"identifiers","label":"Identifiers","entries":[{"key":"dc:identifier","label":"Identifier","values":["etd:52643","local:Ye_grad.msu_0128D_21609","https://doi.org/doi:10.25335/recz-1p13"]}]},{"id":"additional","label":"Additional Metadata","entries":[{"key":"dc:description","label":"Description","values":["Thesis (Ph.D.)--Michigan State University. Linguistics - Doctor of Philosophy, 2025","Perception presents a fundamental cognitive challenge: how do systems extract stable and meaningful interpretations from sensory input that is inherently continuous, ambiguous, and uncertain? Across perceptual domains, studies have identified three core properties that characterize this process: competition among alternative interpretations, probabilistic integration of sensory evidence, and categorical judgments derived from gradient input. Speech perception exemplifies this complexity, as it requires the integration of linguistic knowledge with performance factors that influence how phonological representations are processed and expressed in real time.This dissertation investigates speech perception within Marr's three-level framework for analyzing cognitive systems, focusing specifically on the level of computational theory -- the functional goals that the perceptual system must accomplish. It centers on the sub-segmental time course of perception, a dimension often unexplored in previous speech perception models, and uses vowel nasalization as a test case for examining how perceptual decisions unfold within a segment. Bayesian mathematical tools are employed as a formal modeling framework -- not as a theoretical commitment to Bayesian computation per se, but as a flexible and interpretable way to evaluate alternative hypotheses. To this end, a set of perception and production experiments was conducted in Mainstream American English and Standard Hindi respectively, two languages that exhibit distinct patterns of vowel nasalization. Listener perception was assessed using gated stimuli that incrementally revealed segments of the vowel, and these behavioral responses served as the empirical basis for evaluating the predictions generated by a set of formal perceptual models. Production data from the same participants were used to build timestep-conditioned acoustic models that generate likelihood distributions for each vowel category: /CVC/ words for oral vowels, /CVN/ for phonologically nasalized vowels, and /C~VC/ for phonemic nasal vowels. Perceptual models were then trained on this data and varied systematically along four dimensions: inference updating, time sensitivity, specificity of phonological contexts, and temporal normalization. Perceptual models were tested using the same gated stimuli employed in the perception experiments, and model performance was evaluated against human responses from gated perception experiments, with the best-fitting model identified based on its correspondence to listener behavior. The resulting framework proposes that any successful theory of speech perception must account for four key properties: incremental inference updating, sensitivity to time-series information, incorporation of non-underspecified distributions, and reliance on normalized rather than absolute time.Overall, the formal models outperformed human listeners in generating more accurate predictions of the test stimuli. Among the models tested, the updating, time-sensitive, context-specific model most closely captured listener behavior across both languages. Importantly, the results also show that even categorical, time-insensitive models can exhibit gradient perceptual patterns, highlighting the need for precise formal modeling in identifying the source of observed gradience. Gradience can emerge from the computational process of perception itself and does not necessarily imply gradient underlying representations. Finally, the findings indicate that perception operates over normalized rather than absolute time, posing challenges for any theoretical framework that assumes listeners directly track detailed temporal information without normalization.The modeling approach adopted in this dissertation, through training models on vowel production data independently of listeners' responses and then testing them against perceptual behavior, generates testable predictions rather than post hoc interpretation fitting to the data. This strategy mitigates the risk of data-dependent decision-making in interpretation, as highlighted by Gelman and Loken (2013) and Simmons et al. (2011). This strategy also constrains the degrees of freedom available for model adjustment, thereby reducing the risk of overfitting -- an important methodological consideration emphasized in recent critical assessments of modeling practices (Persson et al. 2025). Furthermore, by developing and evaluating formal models grounded in empirical data, this dissertation addresses two major gaps in the field: (1) it advances our understanding of the computational goals of speech perception, particularly how listeners interpret time-series acoustic input -- an aspect underexplored in prior perception models; and (2) it offers a formal framework for testing the assumptions and predictions of perceptual theories by isolating and examining theoretical subcomponents such as acoustic mapping, phonological underspecification, and temporal normalization.","Description based on online resource. Title from PDF t.p. (Michigan State University Fedora Repository, viewed ).","Includes bibliographical references."]},{"key":"dc:format","label":"Dc Format","values":["application/pdf","1 online resource (213 p.) : ill."]},{"key":"dc:title","label":"Title","values":["Understanding Subsegmental Speech Perception with Bayesian Modeling"]}]}],"canonical_facts":{"dc:contributor":["Durvasula, Karthik","Wagner, Suzanne","Sneller, Betsy","Bongiovanni, Silvina","Lin, Yen-Hwei"],"dc:creator":["Ye, Yongqing"],"dc:date":["2025"],"dc:description":["Thesis (Ph.D.)--Michigan State University. Linguistics - Doctor of Philosophy, 2025","Perception presents a fundamental cognitive challenge: how do systems extract stable and meaningful interpretations from sensory input that is inherently continuous, ambiguous, and uncertain? Across perceptual domains, studies have identified three core properties that characterize this process: competition among alternative interpretations, probabilistic integration of sensory evidence, and categorical judgments derived from gradient input. Speech perception exemplifies this complexity, as it requires the integration of linguistic knowledge with performance factors that influence how phonological representations are processed and expressed in real time.This dissertation investigates speech perception within Marr's three-level framework for analyzing cognitive systems, focusing specifically on the level of computational theory -- the functional goals that the perceptual system must accomplish. It centers on the sub-segmental time course of perception, a dimension often unexplored in previous speech perception models, and uses vowel nasalization as a test case for examining how perceptual decisions unfold within a segment. Bayesian mathematical tools are employed as a formal modeling framework -- not as a theoretical commitment to Bayesian computation per se, but as a flexible and interpretable way to evaluate alternative hypotheses. To this end, a set of perception and production experiments was conducted in Mainstream American English and Standard Hindi respectively, two languages that exhibit distinct patterns of vowel nasalization. Listener perception was assessed using gated stimuli that incrementally revealed segments of the vowel, and these behavioral responses served as the empirical basis for evaluating the predictions generated by a set of formal perceptual models. Production data from the same participants were used to build timestep-conditioned acoustic models that generate likelihood distributions for each vowel category: /CVC/ words for oral vowels, /CVN/ for phonologically nasalized vowels, and /C~VC/ for phonemic nasal vowels. Perceptual models were then trained on this data and varied systematically along four dimensions: inference updating, time sensitivity, specificity of phonological contexts, and temporal normalization. Perceptual models were tested using the same gated stimuli employed in the perception experiments, and model performance was evaluated against human responses from gated perception experiments, with the best-fitting model identified based on its correspondence to listener behavior. The resulting framework proposes that any successful theory of speech perception must account for four key properties: incremental inference updating, sensitivity to time-series information, incorporation of non-underspecified distributions, and reliance on normalized rather than absolute time.Overall, the formal models outperformed human listeners in generating more accurate predictions of the test stimuli. Among the models tested, the updating, time-sensitive, context-specific model most closely captured listener behavior across both languages. Importantly, the results also show that even categorical, time-insensitive models can exhibit gradient perceptual patterns, highlighting the need for precise formal modeling in identifying the source of observed gradience. Gradience can emerge from the computational process of perception itself and does not necessarily imply gradient underlying representations. Finally, the findings indicate that perception operates over normalized rather than absolute time, posing challenges for any theoretical framework that assumes listeners directly track detailed temporal information without normalization.The modeling approach adopted in this dissertation, through training models on vowel production data independently of listeners' responses and then testing them against perceptual behavior, generates testable predictions rather than post hoc interpretation fitting to the data. This strategy mitigates the risk of data-dependent decision-making in interpretation, as highlighted by Gelman and Loken (2013) and Simmons et al. (2011). This strategy also constrains the degrees of freedom available for model adjustment, thereby reducing the risk of overfitting -- an important methodological consideration emphasized in recent critical assessments of modeling practices (Persson et al. 2025). Furthermore, by developing and evaluating formal models grounded in empirical data, this dissertation addresses two major gaps in the field: (1) it advances our understanding of the computational goals of speech perception, particularly how listeners interpret time-series acoustic input -- an aspect underexplored in prior perception models; and (2) it offers a formal framework for testing the assumptions and predictions of perceptual theories by isolating and examining theoretical subcomponents such as acoustic mapping, phonological underspecification, and temporal normalization.","Description based on online resource. Title from PDF t.p. (Michigan State University Fedora Repository, viewed ).","Includes bibliographical references."],"dc:format":["application/pdf","1 online resource (213 p.) : ill."],"dc:identifier":["etd:52643","local:Ye_grad.msu_0128D_21609","https://doi.org/doi:10.25335/recz-1p13"],"dc:language":["eng"],"dc:relation":["Electronic Theses & Dissertations"],"dc:rights":["Attribution-NonCommercial-NoDerivatives 4.0 International"],"dc:subject":["Linguistics","Cognitive psychology"],"dc:title":["Understanding Subsegmental Speech Perception with Bayesian Modeling"],"dc:type":["text","Electronic dissertations"]},"updated_at":"2026-11-18T18:59:09Z"}