Title: The dynamics of specificity-baed phonetic beliefs in multi-speaker speech perception.
Supervisor: Meredith Tamminga
Committee member: David Embick, Jianjing Kuang
In the recent decades, a substantial (and still increasing) number of studies have reported facilitation on speech processing from specific phonetic information, such as idiosyncratic phonetic properties or incidental surface details that are specific to particular tokens, speakers or contexts (Mullennix et al., 1989; Church and Schacter, 1994; Goldinger, 1996, 1998; Nygaard and Pisoni, 1998; among others). These studies lend support to involvement of surface phonetic details in the encoding of linguistic representations, and point to a necessity for speech processing models to cope with the trade-off between structure and variability in more serious ways. The proposed dissertation is motivated to bridge the gap between abundant observations on the facilitation of phonetic specificity in various aspects of speech processing, and the lack of a relative precise understanding about the scope and limitation within which specificity plays its role. With three experiments, the dissertation probes the dynamics of speaker-specific phonetic beliefs in the context of perceptual learning (Norris et al., 2003; Samuel and Kraljic, 2005) -- a process by which listeners adjust the boundaries of phonemic categories to become better aligned with the phonetic input from particular speakers – in multi-speaker listening scenes. Experiment 1 investigates whether an acquired phonetic distribution of a particular phoneme from speaker A is still retained after exposure to interfering allophones with different distributional properties from speaker B with a different gender. Experiment 2 investigates how the same question applies to individual speakers with the same gender, and whether the observed pattern changes after top-down information about idiosyncratic speakers (i.e., photos) is included. Two pairs of contrasted phonemes, i.e., t/d and s/ʃ are investigated in Experiment 1 and 2, to help examine the possibility of asymmetric behaviors with phonetic beliefs between phoneme types. Different from the previous two experiments, Experiment 3 investigates the potential impediment of phonetic-belief retention by multi-speaker speech materials that do not contain any acoustic instances of the phonemes in question. The last experiment is intended to investigate whether the interference from multi-speaker speech is really caused by an additional amount of allophonic variability, or rather, that a perceived speaker change by itself is sufficient to cue a switch in listeners’ attentional or strategic listening behaviors, with or without containing allophonic inputs of the critical phoneme. Results of these experiments are interpreted towards a more accurate depict of the mental associations listeners established between idiosyncratic speakers or speaker groups and specific acoustic properties of linguistic and nonlinguistic attributes.