Roger Yu-Hsiang Lo
Doctor of Philosophy in Linguistics (PhD)
From annotation to assessment: Tools for sign language phonetic and phonological research
G+PS regularly provides virtual sessions that focus on admission requirements and procedures and tips how to improve your application.
This dissertation compares the lexical tone perception abilities of two populations with different bilingual configurations: Cantonese-dominant adults who grew up in Hong Kong (referred to as homeland speakers), and English-dominant adults who grew up in a Cantonese-speaking household in Canada (heritage speakers). From infancy both were exposed to Cantonese as a first language in terms of chronological order; however, after the onset of schooling, each became dominant in the majority language of their respective society. Given this background, this study investigates whether heritage speakers' perception of lexical tones of a non-dominant first language (Cantonese) exhibits cross-language effects from a dominant second language (English) that does not have a contrastive dimension of tone. A series of perception experiments was conducted using the word identification paradigm. Eight types of audio stimuli were presented to homeland and heritage speakers (N=34 per group), each of which represented a specific configuration of four variables: whether the acoustic signal contained segmental and tonal information, whether the target word was isolated or embedded in a carrier sentence with semantic context, and whether the meaning of the target word was congruous with the carrier sentence. In each trial, participants saw pictures of the target word and minimally contrastive tonal competitors, and were instructed to choose the picture that represented what they heard.Major findings of this study were: (1) among the eight stimulus types, the accuracy gap between the two groups was the biggest when the stimuli were low-pass-filtered monosyllables with no segmental information or semantic context, which suggests that homeland speakers have a significantly greater ability to identify tonally contrastive words by solely relying on tonal information. (2) Both groups showed confusion of overlapping subsets of tone pairs, but heritage speakers had a higher error percentage, which indicates a quantitative but not qualitative difference between the two groups. (3) When the target word was semantically incongruous with the carrier sentence, homeland speakers outperformed heritage speakers by attending to acoustic information, while heritage speakers relied on semantic information relatively more often. In other words, the two groups used different listening strategies in tone identification.
This thesis replicates work in Gender Stereotype Effects and Speech Processing (Strand, 2000), referred to as GSESP, which explored the influence of gender stereotypes on low-level speech processing; and includes two additional experiments to investigate the types of processing situations wherein the effects noted in GSESP and in this work are active.Strand found a significant effect of voice stereotype level upon speech processing time such that stereotypical voices are processed more quickly than non-stereotypical voices; a significant effect of voice gender upon processing time such that female voices are processed slightly more quickly than male voices; and a significant effect of face stereotype level upon processing time of female voices only, such that a stereotypical female voice primed with a stereotypical face is processed more quickly than when primed with a non-stereotypical face.To explore the replicability of these findings, two experiments are conceptually replicated to select stereotypical and non-stereotypical face and voice stimuli using speeded gender classification tasks; one experiment is conceptually replicated to explore the effect of voice stereotype level upon speech processing using an audio-only speeded shadowing task; and one experiment is conceptually replicated to explore the effect of stereotype priming upon speech perception using a face-primed speeded shadowing task. A version of the audio-only speeded shadowing task and a version of the face-primed speeded shadowing task with stimuli separated by stereotype level are also carried out to determine whether any effects noted are present in processing situations where there are 1-2 voices of a single stereotype level (single talker-type) and in those with more voices of varying stereotype levels (multi talker-type).Reaction time results normalized for word duration show effects of voice stereotype level and voice gender upon processing time as noted in GSESP, but no priming effect of stereotypical faces upon female voices, unlike in GSESP. These results are consistent across the two audio-only speeded shadowing experiments and the two face-primed speeded shadowing experiments.These results support the hypothesis that gender stereotypes influence speech processing, and also suggest that stereotypes are used to filter speech as a default, in multiple types of processing situation.