A Pragmatic-Phonological Interface Model for Prosodic Meaning Construction

Thesis Abstract

Prosodic meaning construction remains inadequately explained by existing theories that treat pragmatics and phonology as separate modules; this study addresses the gap by proposing a Pragmatic-Phonological Interface Model (PPIM) that integrates contextual meaning-making with phonological encoding to predict prosodic patterns across communicative functions. The aim is to operationalize a model that accounts for how discourse-pragmatic factors (illocutionary force, topic status, and speaker stance) constrain prosodic realization (accent, intonation, tempo) in real-time interpretation and production. Specific objectives are to (i) delineate a theoretical mapping between pragmatic categories and prosodic cues within a probabilistic framework; (ii) develop a computational prototype that simulates PPIM predictions for a range of discourse contexts; (iii) validate the model against perceptual and production data from spontaneous speech corpora; and (iv) compare PPIM’s predictive power with two baseline theories Relevance Theory and the Autosegmental-Metrical (AM) framework. The study employs a mixed-methods design combining corpus-based analytics and experimental elicitation. The population comprises adult native speakers of English (N=60 for production tasks; N=120 for perception tasks) recruited from a metropolitan university, with stratified sampling by age group and gender to ensure demographic representation. Data collection uses (a) a discourse-production task yielding 240 short utterances embedded in varying pragmatic contexts, recorded at 44.1 kHz with high-fidelity equipment; (b) a matched perception task in which 120 listeners perform binary and rating judgments on perceived prosody while listening to the same stimuli. Instrumentation includes a multidimensional prosodic annotation scheme integrating ACs (accented syllables), tonal alignment, and tempo perturbations, combined with pragmatic coding of illocutionary force, stance, and topic-focus. Validity and reliability are ensured through double-coding of prosodic features (Cohen’s kappa > .80) and inter-rater agreement on pragmatic labels. Data analysis proceeds in three stages (i) Bayesian hierarchical modeling to estimate the influence of pragmatic factors on prosodic variables (pitch height, contour type, duration) across discourse contexts; (ii) regression analyses to test the predictive alignment of PPIM with observed production and perceptual judgments; and (iii) a cross-validated machine-learning component employing a conditional random field to simulate sequence-level prosodic realization under PPIM constraints. The model specification incorporates a pragmatic-prior module derived from Gricean implicature and Relevance Theory, and a phonological-grammar module grounded in AM theory, with interface rules operationalized as probabilistic guards that modulate prosodic realizations. Anticipated findings include robust effects of illocutionary force and discourse topic on pitch excursion and boundary marking, with PPIM outperforming Relevance Theory and AM-based accounts in predicting both production patterns (R2 around .42) and perceptual salience (AUC ~ .78). The study contributes to knowledge by offering a unified, testable framework that traverses linguistic subfields, provides a replicable methodology for interface research, and yields corpus-informed priors for computational dialog systems and language-learning applications. Theoretical implications include a formalized account of how pragmatic interpretation constrains phonological planning, while practical implications concern improved automatic speech recognition and text-to-speech systems that better reflect human prosodic inference and production. The conclusion posits that the Pragmatic-Phonological Interface Model captures essential cross-domain interactions missing from modular theories, and recommends extending the model to cross-linguistic datasets, incorporating multimodal cues, and exploring developmental trajectories in children to assess the universality of interface mechanisms.