Dan Cameron Burgdorf and Dr. Sam Tilsen present paper at LSA 2021
Dan Cameron Burgdorf and Dr. Sam Tilsen presented a paper titled "Glides Prioritize Articulation, Vowels Prioritize Acoustics" at the 95th Annual Meeting of the Linguistic Society of America (LSA 2021), held virtually Jan 7-10, 2021
Little phonetic research has been done on the articulatory and acoustic differences between glides and vowels. These segments have been characterized in numerous ways phonologically, most commonly in terms of intrinsic differences of features or structural differences of syllable position, with respective predictions of phonetic differences in constriction or timing effects.
Articulatory and acoustic data from multiple experiments show that glides and vowels differ in another way entirely: glides are realized with greater articulatory precision than vowels, while vowels are realized with greater acoustic precision and greater sensitivity to acoustic feedback.
Experiment 1 collected articulatory (EMA) data and acoustic data from six native English speakers in an imitation task. Stimuli varied in duration and intensity from vowel-like to glide-like, for both [i]-[j] and [u]-[w]. The experiment was designed to test hypotheses concerning constriction, timing effects, and categoricity, but post hoc analysis of participants’ variances in measured parameters showed that spatial parameters, such as trajectory curvature and movement range, had more variance (less precision) for more vowel-like productions.
In contrast, acoustic parameters, formants and their transitions, had more variance (less precision) for more glide-like productions. Some parameters found to have substantial variance responses across the stimulus space had little or no response in the parameter itself, indicating that greater or lesser precision was not necessarily contingent on any shift in mean value.
Experiments 2 and 3 expanded on this finding with the use of an altered feedback paradigm (AFP). AFPs are studies in which the participant hears a real-time playback of their voice which is, unknown to them, altered in some fashion. This alteration elicits compensation, in which participants change their production so that the feedback they hear better resembles their intended target. AFPs have been applied to vowels previously (e.g. Houde 2011) by altering the formants of the (usually mid front) vowel to be perceived as a different vowel (e.g., [ɛ] to [i] or [æ]). The use of an AFP on high vowels and glides allowed for testing of the hypothesis that glides prioritize other forms of feedback (e.g. somatosensory) over acoustic feedback and would therefore show less compensation than vowels.
Experiment 2 presented 21 native English speakers with auditory models of /biə/ ‹bia› and /bjə/ ‹bya› and subsequently cued participants orthographically. Feedback alteration was achieved with Audapter, a software package for real-time manipulation of acoustic parameters of speech. Maximum alteration was an increase of F2 by 250Hz and a decrease of F1 by 120Hz, effectively making the input sound more high and more front. Alteration tapered off within each word to avoid altering the formants of the final vowel.
Experiment 3 (underway) applied a similar alteration to /duə/ ‹dua› and /dwə/ ‹dwa›. Some participants responded unexpectedly by lengthening their vowels to outlast the period of alteration. Of participants whose vowels were short enough to be fully altered as intended, most showed substantially more compensation for vowels than for glides, as predicted.
Additionally, the acoustic findings from Experiment 1 were replicated: both during baseline periods with no alteration and while alteration was applied, most participants had greater variance in their formants for glides than for vowels. These findings point to a kind of difference between glides and vowels that, despite being unpredicted by standard phonological approaches, is not unintuitive if the category of glides is generalized to consonants.
Glides, as consonants, prioritize the articulatory domain, where they’re realized with a degree of consistency regardless of acoustic results. Vowels prioritize the acoustic domain, and more readily adjust their articulation to meet acoustic targets.
This difference may underlie broad patterns in the phonological and phonetic behavior of consonants and vowels.
8th January 2021
Rachel Vogel presents poster at LSA 2021
Rachel Vogel presented a poster titled "Vowel devoicing is not one size fits all: the case of Cheyenne" at the 95th Annual Meeting of the Linguistic Society of America (LSA 2021), held Jan 7-10, 2021.
Phonological accounts of vowel devoicing cross-linguistically typically involve the spreading or insertion of a laryngeal feature, however, there is debate over which particular feature is involved.
This paper provides a phonological account of three vowel devoicing processes in Cheyenne (Algonquian) and demonstrates that the segmental environments in which the different processes occur in fact require both [spread glottis] and [-voice]. It thus contributes to cross-linguistic work arguing that VD is not a single unitary phenomenon (e.g., Cho 1993, Gordon 1998, Kilbourn-Ceron & Sonderegger 2018) and that moreover, several distinct phonological vowel devoicing phenomena may exist within a single language.
8th January 2021
Seung-Eun Kim to present paper at ISSP 2020
PhD candidate Seung-Eun Kim and Dr. Sam Tilsen will present a paper titled "Evidence for F0 preplanning with delayed stimuli" at the 2020 International Seminar on Speech Production Conference.
Many prosodic theories hold that different syntactic structures are mapped to distinct prosodic organizations; these theories predict that acoustic and articulatory correlates of these structures differ mainly at phrase boundaries, yet no studies have investigated whether such predictions are correct.
This study uses a novel neural network-based analysis method for temporally localizing prosodic information that is associated with syntactic contrast in acoustic and articulatory signals.
Specifically, we focus on the contrast between non-restrictive and restrictive relative clauses. Neural networks were trained on multi-dimensional acoustic and articulatory data to classify the two types of relative clauses, and the network accuracies on test data were analyzed.
The results found two different patterns syntactically conditioned prosodic information was either widely distributed around the boundaries or narrowly distributed at specific locations.
The findings suggest that prosodic expression of syntactic contrasts does not occur in the uniform way or at a fixed location, but rather it is accomplished with various strategies.
3rd January 2021
Five Cornell Phonetics Lab Researchers present papers at the 12th International Seminar on Speech Production (Providence)
Five Phonetics Lab researchers presented papers at the 12th International Seminar on Speech Production (Providence), which is held Monday, Dec. 14 to Friday, Dec. 18.
19th December 2020