Doctor of Philosophy (Ph.D.)
Degree Granting Department
Communication Sciences and Disorders
Stefan A. Frisch, Ph.D.
Jean C. Krause, Ph.D.
Ruth H. Bahr, Ph.D.
Catherine L. Rogers, Ph.D.
categorical perception, Ganong effect, lexical identification shift, noise-vocoded speech, phonetic contrasts
In speech perception tasks with ambiguous bottom-up information, lexical processes have been shown to influence listener responses, such as in phoneme categorization tasks (Ganong, 1980). Proponents of interactive theories of speech perception and spoken word recognition assert this influence is a top-down feedback mechanism that can affect bottom-up perceptual processes (e.g., McClelland & Elman, 1986). While robust influences on phoneme perception have been reported in multiple studies (Connine & Clifton, 1987; Ganong, 1980; Gow, Segawa, Ahlfors, & Lin, 2008; Pitt & Samuel, 1993; among others), some phonetic contrasts, particularly those that distinguish place of articulation, have been tested in very limited circumstances. Furthermore, there is little research testing this phenomenon with impoverished stimuli or under poor listening conditions, where interaction from lexical processes in phoneme perception may increase, provided that bottom-up input is sufficient to activate higher-level processes.
To determine the extent of lexical influences in perception, or the “lexical identification shift” (Pitt & Samuel, 1993) with previously untested stimuli in non-degraded and spectrally-degraded conditions, a series of experiments analogous to Ganong’s (1980) paradigm were conducted. The first study investigated top-down influences in phoneme perception, including the contribution of coarticulation cues, in normal hearing adults. In the second study, the lexical identification shift was examined in multiple levels of spectral degradation with noise-vocoded (NV) speech. These experiments illustrate the interplay between top-down and bottom-up information in speech perception; specifically, the influence of top-down knowledge in identification of place-of-articulation contrasts, and the impact of spectral degradation on perception in these tasks. Results in the first study revealed a significant lexical identification shift with place-of-articulation phonetic contrasts across three separate experiments, providing further support for lexical interaction in speech perception. These results also established the contribution of coarticulatory information in relation to the lexical shift, which increased substantially with both a full continuum of stimuli and stimuli that are substantially more categorical. Mixed results were obtained in the second study, with a significant lexical identification shift in the 20-channel NV condition, and no significant differences in phoneme categorization with lexically-biased continua in undistorted, 10-channel and 16-channel NV conditions. Furthermore, the lexical shift was smaller in the 20-channel NV condition, compared to shifts observed in the first study that did not include spectrally-degraded stimuli. Overall, outcomes indicate that lexical influence in the perception of stop consonant place-of-articulation phonetic contrasts is restrained by spectrally-degraded input, with evidence of this restraint apparent in conditions with relatively low-levels of distortion.
Scholar Commons Citation
Smart, Jane Bradley, "Lexical Influence in Phoneme Perception with Non-degraded and Spectrally-degraded Speech" (2019). USF Tampa Graduate Theses and Dissertations.