A syllable-based investigation of coarticulation - PhDData

Access database of worldwide thesis




A syllable-based investigation of coarticulation

The thesis was published by Liu, Zirui, in June 2023, UCL (University College London).

Abstract:

Coarticulation has been long investigated in Speech Sciences and Linguistics (Kühnert &
Nolan, 1999). This thesis explores coarticulation through a syllable based model (Y. Xu,
2020). First, it is hypothesised that consonant and vowel are synchronised at the syllable
onset for the sake of reducing temporal degrees of freedom, and such synchronisation
is the essence of coarticulation. Previous efforts in the examination of CV alignment
mainly report onset asynchrony (Gao, 2009; Shaw & Chen, 2019). The first study of this
thesis tested the synchrony hypothesis using articulatory and acoustic data in Mandarin.
Departing from conventional approaches, a minimal triplet paradigm was applied, in
which the CV onsets were determined through the consonant and vowel minimal pairs,
respectively. Both articulatory and acoustical results showed that CV articulation started
in close temporal proximity, supporting the synchrony hypothesis. The second study
extended the research to English and syllables with cluster onsets. By using acoustic data
in conjunction with Deep Learning, supporting evidence was found for co-onset, which
is in contrast to the widely reported c-center effect (Byrd, 1995). Secondly, the thesis
investigated the mechanism that can maximise synchrony – Dimension Specific Sequential
Target Approximation (DSSTA), which is highly relevant to what is commonly known
as coarticulation resistance (Recasens & Espinosa, 2009). Evidence from the first two studies show that, when conflicts arise due to articulation requirements between CV, the
CV gestures can be fulfilled by the same articulator on separate dimensions simultaneously.
Last but not least, the final study tested the hypothesis that resyllabification is the result of
coarticulation asymmetry between onset and coda consonants. It was found that neural
network based models could infer syllable affiliation of consonants, and those inferred
resyllabified codas had similar coarticulatory structure with canonical onset consonants. In
conclusion, this thesis found that many coarticulation related phenomena, including local
vowel to vowel anticipatory coarticulation, coarticulation resistance, and resyllabification,
stem from the articulatory mechanism of the syllable.



Read the last PhD tips