Ultrasound analysis

One ultrasound frame was selected per recording depicting the maximal constriction of the anterior lingual gesture for/r/prior to any obvious movement into the following vowel. This was achieved by holistically examining the raw ultrasound images one by one in a sequence.

For each selected image, a spline was fitted to the visible surface of the midsagittal tongue using the edge-detection algorithm in AAA. A preliminary step to the edge detection process requires selecting the upper and lower limits between which the algorithm may detect a bright surface (Lawson et al., 2013). The upper and lower limits are set to remove traces of the hard palate and bright areas resulting from muscle structures inside the tongue (Lawson et al., 2013), which can be observed in the left image of Figure 4.6. After tracing the upper and lower limits (indicated by the green lines inFigure 4.6), a spline was roughly traced by hand around the midsagittal tongue contour (presented in pink in the middle image inFigure 4.6).

AAA’s edge-detection algorithm was then implemented by pressing the ‘Snap-to-fit’ button.

In the parts of the contour that have a good edge, the spline appears as a solid line. As the right image in Figure 4.6indicates, automatic edge detection removes parts of the spline at the extreme right and left of the image where no clear tongue surface exists in the original ultrasound image. Occasionally, the algorithm may miss certain areas of the tongue contour, particularly around the tongue tip due to shadowing from the jaw. In these cases, splines were manually corrected. Corrections were often achieved through a holistic examination of the ultrasound frames occurring before and after the selected frame, which generally allowed for more accurate tracking of the tongue tip, rather than relying on the one static ultrasound image selected.

Figure 4.6:Automatic detection of the midsagittal tongue contour in ultrasound data. (left) Raw ultrasound image depicting the visible surface of the midsagittal tongue represented by the lower edge of the bright white line; (middle) Upper and lower limits (green lines) and tongue surface spline (pink line) hand-traced for automatic edge detection; (right) after automatic edge-detection

spline fitting.

Imaging the occlusal plane

A reference spline was fitted to each speaker’s occlusal (i.e., bite) plane, which was imaged using a bite plate.⁴ Imaging a speaker’s occlusal plane improves interpretation of tongue position and inter-speaker comparison (Lawson et al., 2019). Bite plates are made from 2mm thick medical grade plastic and vacuum-moulded around a standard template (95x40 mm) (Lawson, Stuart-Smith, Scobbie, & Nakai, 2018), an image of which is presented in the top left image in Figure 4.7. When the bite plate is placed in the mouth (top right image ofFigure 4.7), a vertical ridge located near the middle of the bite plate rests against the front of the upper incisors.

Participants were recorded biting on the bite plate and were asked to press their tongue against the underside. The resulting flat surface of the tongue pressed against the bite plate allows for the identification of a flat plane in the ultrasound video image (Lawson, Stuart-Smith, Scobbie,

& Nakai, 2018), which is visible in the bottom left ultrasound image presented inFigure 4.7. A reference spline was fitted to this plane (as presented in the bottom right image ofFigure 4.7), which was used to rotate all subsequent splines to a quasi-horizontal position.Figure 4.8depicts

4Bite plates were kindly provided by the Clinical Audiology, Speech and Language Research Centre at Queen Margaret University, Edinburgh.

4.2. Methodology 119

our rotation technique: all contours are rotated so that the occlusal plane (green line) tracing is horizontal. An alternative technique is to adjust the probe-to-chin angle using the stabilising headset before recording so that the image of the occlusal plane is parallel to the upper and lower edges of the video pane, as described in Lawson et al. (2019). We decided to rotate the tongue surface splines at the post-processing stage because we found that in some speakers, adjusting the probe angle reduced the visibility of the tongue tip, which was of particular importance in the current study.

Figure 4.7:Imaging and detecting the occlusal plane with a bite plate. (top left) Plastic bite plate;

(top right) Subject biting on bite plate; (bottom left) Resulting ultrasound image depicting flat surface of tongue against the underside of the bite plate; (bottom right) Reference spline tracing

used to rotate all subsequent splines to a quasi-horizontal position.

Figure 4.8:Example of rotation of splines to the occlusal plane. The tongue tip is on the right.

The hard palate is traced in the top curve. All contours are rotated so that the occlusal plane (bottom line) is horizontal.

Identifying tongue shapes

Both the raw ultrasound images and the tongue splines rotated to the occlusal plane were used to classify tongue configurations for /r/on a continuum largely inspired by the one presented in Lawson et al. (2013) for Scottish English, which depicts four distinct shapes:

Mid Bunched, Front Bunched, Front Up and Tip Up (pp. 199–200), as presented inSection 2.4 (p. 41). Our classification differs in that it includes a fifth configuration: an ‘extreme’sublaminal retroflexinvolving curling up of the tongue tip, which has previously been associated with Anglo-English(as discussed inSection 2.4). The classification originally proposed by Lawson et al. (2013) grouped the curled up and the non-curled up tip up/r/together. Ultrasound images give some indication of the curling up of the tongue tip, which we describe below. However, we do not know to what extent the identification of these articulations is constrained by speaker anatomy. In some cases, it is possible that the jaw shadow obscures the tongue tip, which would make visualisingsublaminal retroflexionchallenging. It is therefore possible that the number of curled up articulations is underestimated in our analysis. The articulations of each

4.2. Methodology 121

configuration in our classification are described below,⁵andFigure 4.9presents raw ultrasound images of typical examples of each configuration from our dataset.

Mid Bunched (MB): the middle of the tongue is raised towards the hard palate, while the front, blade and tip are low.

Front Bunched (FB): the front of the tongue has a distinctlybunched configuration which results in a dip in the tongue’s surface behind thebunchedsection. The tip and blade remain lower than the rest of the tongue front.

Front Up (FU): the front, blade and tip are raised and the tongue surface forms a smooth convex curve.

Tip Up (TU): the tongue tip is pointing up resulting in a straight and steep tongue surface.

Curled Up (CU): the overall tongue shape is concave and the tip is curled up. Curling up of the tongue tip results in a near-parallel orientation of the tongue surface to the ultrasound scanlines, producing artefacts in the ultrasound image (Scobbie, Punnoose, &

Khattab, 2013). We tend to observe a bright white region above where the tongue tip is expected (Mielke et al., 2016) and a discontinuity in the tongue contour where the tongue tip is curled up (Bakst, 2016).

In order to facilitate the task of classifying tongue configurations, the decision tree presented inFigure 4.10was produced and used throughout the classification process. Tongue shapes were classified three times throughout the course of one year to ensure accuracy. Although discrepancies in the three classifications were rare, such cases were reexamined and the most common configuration of the three was selected.

If we employ the traditionalretroflex-bunchedclassification, the Mid Bunched and Front Bunched configurations have a low tongue tip and the primary constriction is located between the front to mid tongue body (Lawson et al., 2011), so we can consider them to bebunched.

5The first four configurations (MB, FB, FU, TU) are identical to those described in Lawson et al. (2011).

Figure 4.9:Raw ultrasound frames presenting typical examples of each of the five tongue configurations observed in Anglo-English/r/. The tongue tip is on the right. The top two images are bunched, while the bottom three are retroflex. The final retroflex configuration exhibits curling

up of the tongue which is evident from the bright white line where the tongue tip is expected towards the palate, and a discontinuity in the tongue contour where the tip starts to curl up

(indicated with an arrow).

If we adopt Hamann (2003)’s definition, any sound articulated with the tongue tip behind the alveolar region and involving a displacement of the tongue back towards the pharynx or velum may be consideredretroflex. Asbunched/r/has also been shown to include tongue root retraction (Delattre & Freeman, 1968; Proctor et al., 2019) and the drawing inwards of the tongue body away from the lips (Alwan et al., 1997), the main criterion we considered to define retroflexionfor/r/is the raising of the tongue tip, which results in the addition of asublingual space. The tongue tip and/or tongue front are raised towards the post-alveolar region in the last three configurations of our classification (FU, TU, CU), and so, we therefore consider them to beretroflex. As discussed inSection 2.4, the status of bladal configurations such as the one described in our Front Up category, has been disputed with some researchers who consider them to bebunchedrather thanretroflex. Although in some raw ultrasound images in our dataset the primary constriction (i.e., the highest point of the tongue) in some Front Up configurations may appear to be the tongue dorsum (as in the Front Up image presented in Figure 4.9), when the corresponding spline is rotated to the occlusal plane, the tongue tip does generally appear to be the primary constriction, or at least pointing up, an example of which

4.2. Methodology 123

Figure 4.10:Decision tree used to classify tongue configurations for/r/into five distinct categories from ultrasound data.

can be observed inFigure 4.8. The position of the tongue tip is a further indication that the Front Up configuration exhibits apicality, which is suggestive ofretroflexion.

Our classification would place the variant with the highest, most curled up tongue tip, the Curled Up configuration, at one end of the continuum. Curled Up is followed by the Tip Up and Front Up variants respectively. Deciphering which tongue shape is the most bunched category between Mid Bunched and Front Bunched is less evident. Although visualising the tongue contour tracings in speakers who present both configurations revealed that the tongue tip is generally lower in the Mid Bunched than the Front Bunched configuration, the Front Bunched category presents the most obviousbunching of the tongue i.e., with a dip in the tongue surface, orsulcalization(as can be seen in Figure 4.9). Furthermore, the very tip of the tongue is not always visible from ultrasound images and so we err on the side of caution regarding the accuracy of tongue tip tracings. It is hoped that results from this study may provide further insights into whichbunchedconfiguration is the most extreme of the two.