Assessing mid-sagittal tongue contours in polar coordinates using generalised additive (mixed) models

Abstract

Statistical modelling of whole tongue contours has been mostly dominated by the use of Smoothing Splines Analysis of Variance (SSANOVA), although the quantitative analysis of UTI data remains a challenge. Recently, a variety of research disciplines witnessed an increased use of Generalised Additive Models (GAMs) and their mixed-effects counterpart. This family of models is a highly flexible solution which extends standard generalised linear mixed regressions to model non-linear effects. This paper offers a review of GAMs fitted to tongue contours in polar coordinates, as an alternative to polar SSANOVA, given the increasing popularity of these models among linguists. Polar GAMs fitting, significance testing, and model plotting are illustrated by means of an example study that compares tongue contours of voiceless and voiced stops of 12 speakers of Italian and Polish. A brief tutorial illustrates fitting and plotting of polar GAMs with the R package rticulate. The series of polar GAMs indicates a high degree of idiosyncrasy in tongue root position in voiceless and voiced stops, within and across speakers. Limitations of the current implementation of polar GAMs (such as across-speaker normalisation) and future directions are also briefly discussed.

Northern Tosk Albanian

“…Furthermore, Figure 13 shows midsagittal tongue contours from ultrasound imaging extracted from around the mid-point of the sustained tokens of each vowel (the front of the oral cavity is on the right side). In this figure, the tongue contours have been smoothed using a generalised additive model (Hastie & Tibshirani 1986, Wood 2006) with polar coordinates (Coretta 2019).…”

“…Early descriptive studies have suggested that declarative sentences have a falling intonational contour, as seen in our corpus in sentences (a) and (b), while questions have a rising one, as in sentences (c) and (d) (BoriCi 1987, Memushaj 2015. More recent work has proposed that different informational-structure constructs, such as topic, informational focus and contrastive focus associate with different tunes in Standard Albanian (Kapia & Brugos 2016, 2019Kapia, Kleber & Harrington 2020, Kapia, Harrington & Kleber forthcoming). The narrow focus pattern can be seen in the word murin in the polar question (c).…”

Northern Tosk Albanian

“…Early descriptive studies have suggested that declarative sentences have a falling intonational contour, as seen in our corpus in sentences (a) and (b), while questions have a rising one, as in sentences (c) and (d)(Boriçi 1987;Memushaj 2015). More recent work has proposed that different informational structure constructs, such as topic, informational focus and contrastive focus associate with different tunes in Standard Albanian(Kapia & Brugos 2016, 2019 Kapia et al 2020, forthcoming). The narrow focus pattern can be seen in the word murin in the polar question (c).…”