The velic traverse: An independent oral articulator?

The velum has long been considered as having two main functions in speech production: to open and close the velopharyngeal port, and to provide a passive or static surface in the oral tract against which the tongue can contract to produce oral constrictions (Catford, 1977; Laver, 1994). Depending on the size of the oropharyngeal opening, thought to be modulated primarily by tongue posture, this passive surface may sometimes vibrate against the tongue, as in uvular trills (Abercrombie, 1967, p. 49; Catford, 1977, pp. 127–128). However, at least part of the velum, the uvula, has been claimed to interact with the tongue in a more active way, bringing into question the velum's passive role in oral constrictions. In an x-ray study of French uvular fricative /ʁ/, Delattre (1969, p. 15) observes that, for at least some speakers “La luette se rapproche de la langue et tourne sa pointe en avant (The uvula approaches the tongue and turns its point forward).”

If part of the velum does participate actively in forming oral constrictions, it is important to consider exactly what structure is moving, and how is it able to move toward the tongue without lowering the entire velum and thus compromising velopharyngeal port closure. The uvula alone is unlikely to be responsible for a substantial movement toward the tongue, as the speech function of its only intrinsic muscle, the musculus uvulae, is to shorten and stiffen the uvula (Kuehn et al., 1988).

These observations lead us to hypothesize that a uvular oral constriction may engage a physical substructure of the velum that subsumes the uvula and has sufficient musculature to act as an autonomous oral articulator, but that is functionally independent of the parts of the velum used to close the velopharyngeal port. Specifically, we propose that the lower border of the velum, the fleshy top of the archway connecting the faucial (also sometimes called “faucal”) pillars, functions in this way in producing uvular constrictions. This structure, which we shall call the “velic traverse” as it forms the top of the velic archway traversing the faucial pillars, can easily be mistaken for the uvula in x-ray projections or magnetic resonance images; however, the uvula is too small to appear in most x-rays (see Baken and Orlikoff, 2007, p. 519), and visibly hangs posteriorly in magnetic resonance imagings (which are taken in a supine position). Figure 1 shows the location of this structure from the posterior and the midsagittal view.

If the velic traverse functions as an independent oral articulator, its gross position must be controlled mainly by its extrinsic muscles, palatoglossus and palatopharyngeus, which course anteriorly and posteriorly, respectively, downward from the velic traverse (forming the anterior and posterior faucial pillars), allowing them to pull the traverse forward or backward. Electromyographic studies have noted activity in these extrinsic muscles during some velar or uvular constrictions, but have generally concluded that their speech function is unclear (e.g., Fritzell, 1969; Lubker et al., 1970; Bell-Berti and Hirose, 1973; Seaver and Kuehn, 1980; Dixit et al., 1987), at least in part because of their inconsistent relationship with respect to tongue body raising. As for the turning or bending forward of the traverse, the extrinsic muscles of the velum, palatoglossus, palatopharyngeus, tensor veli palatini, and levator veli palatini, all have intrinsic portions with overlapping layers of fibers reaching deep into the traverse (perhaps most clearly illustrated in Drake et al., 2010). The specific functions of these intrinsic layers are poorly understood, but together they are able to shape the velum in complex ways, as may be seen in imaging and modeling studies of the posterior velum (Serrurier and Badin, 2008).

The present study measures the contribution of the velic traverse to uvular constrictions by examining x-ray films of native-speaker productions of French uvular /ʁ/. Our “velic traverse” hypothesis generates the prediction that the lower margin of the velum will contribute substantially to the oral constriction for the French /ʁ/, while the posterior velum, the part responsible for maintaining velopharyngeal port closure, will remain largely unaffected.

X-ray images for this experiment were obtained from the Laval X-Ray Database (Munhall et al., 1995), a database of cineradiography videos of speech collected at 50 frames/s. Six tokens of /ʁ/ from each of the nine native speakers of Québécois French in the data set were used, with each token consisting of a video clip containing the /ʁ/ plus the preceding vowel or consonant. Tokens where the /ʁ/ was in phrase-final position or in close proximity to a velar consonant, a nasal, or a non-low back vowel were omitted so as to prevent interference from unrelated velum movement; thus all tokens took one of the following forms: V-/ʁ/-V, C-/ʁ/-V, V-/ʁ/-C. Using ELAN (Sloetjes and Wittenburg, 2008), still frames were extracted at the point of maximum constriction between the tongue and the velum for the /ʁ/, as well as at the point of maximum constriction or opening for the immediately preceding consonant or vowel. The frame extracted from the /ʁ/ constriction is referred to as the “during” image, while the frame extracted from the preceding consonant or vowel is referred to as the “before” image. Once the frames were extracted, tracings were made of the velum, the pharyngeal wall, and the tongue using ImageJ (Rasband, 1997). At least two experimenters agreed on each tracing to ensure inter-rater reliability; as more usable tokens were available for some speakers than others, images found to be of lesser quality were omitted from analysis in favor of better quality images.

The “before” and “during” tracings for each token were overlaid on a single frame and measured using ImageJ (Rasband, 1997), as shown in Fig. 2. Vertical displacement of the superior and inferior surfaces and change in vertical thickness of the velum were measured from fixed anatomical landmarks above the posterior (A) and anterior (B) portions of the velum; horizontal displacement of the posterior and anterior walls, and change in horizontal thickness, of the velic traverse (C) were measured from fixed anatomical landmarks behind the traverse. As all measurements in this study were used only to compare differences between articulator positions “before” and “during” the /ʁ/ (rather than absolute positions), and because anatomical landmarks vary widely across subjects, similar results would be obtained using any landmark along a line perpendicular to the surfaces being measured. Finally, to determine the relative contributions of the velic traverse and the tongue to the uvular constriction, measurements of the positions of the velic traverse and the tongue were taken along a diagonal line (D) normal to the curve of the “during” tongue and running through the point of greatest tongue-velum constriction (excluding the uvula, which was seldom visible). Horizontal displacement of the posterior velum at the spatial point of velopharyngeal port closure was not measured, as it remained in constant contact with the rear pharyngeal wall throughout all tokens. In addition to the consensus method described above for tracings, a test-retest method was used to ensure measurement consistency; 96 measurements (11% of the data set) were taken from new tracings of the same raw images, with tracings and measurements made by different experimenters. All correlations between the two rounds of measurements were significant (all Pearson r-values > 0.70; all p-values < 0.01).

The measurements were normalized and converted to millimeters using a speaker-specific multiplier derived by calculating the ratio of the length of the line connecting the left gonion and the menton of each speaker's jaw to population average orthopantomogram measurements of that same line as reported by Ongkosuwito et al. (2009). SPSS version 20 (IBM Corp., Armonk, NY) was used to run statistics on the resulting normalized measurements.

Paired and one-sample t-tests were computed on data pooled across speakers to test whether the means of the before and during measurements differed. A Bonferroni correction was applied for multiple comparisons, resulting in an alpha level of 0.0045.

To determine the contribution of the velic traverse at the location of narrowest oral constriction, a one-sample t-test was computed comparing the position of the velic traverse along line D before vs during /ʁ/ against a hypothesized value of zero [see Fig. 2(b); M = 9.97, SD = 6.11], which showed a significant contribution to the uvular constriction, at t(53) = 12.00, p < 0.0045. A one-sample t-test was also computed comparing the position of the tongue along line D before vs during /ʁ/ against a hypothesized value of zero (M = 13.88, SD = 7.26) which also showed a significant contribution to the uvular constriction, at t(53) = 14.06, p < 0.0045. Figure 3 plots the subject-by-subject mean percentage contributions of the tongue and the velic traverse to the uvular constriction [averaged distances calculated along line D in Fig. 2(b)]. Overall, the tongue was responsible for an average of about 58% of the total uvular constriction, with the movement of the velic traverse making up the remaining 42%.

Horizontal measures [along line C in Fig. 2(b)] were used to determine the anteroposterior movement and thickness of the velic traverse. Paired-sample t-tests were significant for the before and during measurements of horizontal velic traverse position. The difference between before (M = 35.19, SD = 11.51) and during (M = 39.07, SD = 12.31) measurements of the posterior surface of the velic traverse showed significant forward movement, with t(53) = −7.89, p < 0.0045. Similarly, the difference between before (M = 49.41, SD = 10.52) and during (M = 60.17, SD = 11.87) measurements of the anterior surface of the velic traverse showed a significant forward movement of about 1 cm, with t(53) = −11.69, p < 0.0045. Last, a significant difference of t(53) = −9.36, p < 0.0045 was found in the calculated difference between the anterior and posterior surfaces of the traverse (i.e., traverse thickness) comparing before (M = 14.22, SD = 7.14) vs during (M = 21.10, SD = 6.47) /ʁ/. These measures indicate that the traverse is both moving forward and thickening in producing /ʁ/.

Measures along lines A and B in Fig. 2(b) were used to determine the vertical movement and thickness of the velic traverse. The difference between the before (M = 71.54, SD = 12.44) and during (M = 73.15, SD = 12.66) positions of the superior surface of the anterior velum showed significant lowering, at t(53) = −5.55, p < 0.0045. A similar result was obtained for before (M = 16.32, SD = 3.30) and during (M = 18.43, SD = 4.37) measurements of the superior surface of the posterior velum, which also showed significant lowering, at t(53) = −3.66, p = 0.001. No other comparisons yielded significant results.

The velic traverse contributes substantially to uvular constrictions, challenging long-standing notions of the velum as a passive player in forming oral constrictions. The mechanism by which this occurs is a flexing forward and thickening of the velic traverse, covering approximately 1 cm of distance toward the tongue. At no point was velopharyngeal port closure compromised, and no significant lowering of any inferior velic surface was observed, though there was a very small (2 mm) but consistent lowering of the superior surface, suggesting that portions of the velum not involved in the oral constriction may undergo some hydrostatic volume reduction, possibly in response to the thickening of the traverse.

Functioning as part of the oropharyngeal isthmus rather than the velopharyngeal port, the velic traverse may be considered as part of a broader functional distinction between the “orovelum” and the “nasovelum,” analogous with the oropharynx and nasopharynx, which operate at parallel levels in the vocal tract. Future production and modeling research should help to distinguish between those portions of the velum whose functions are primarily oral and those whose functions are primarily nasal.

An active role for the velic traverse challenges views maintaining that measuring and modeling tongue movements alone is sufficient to accurately characterize lingual constrictions. A more appropriate characterization of oral constrictions involving the velum might be one modeled on the lips, with sphincterlike actions of functionally defined structures in a three-dimensional space (Gick et al., 2011). The present results converge with evidence from other regions of the vocal tract such as the pharynx, where supposed “opposing surfaces” play an active role in articulation (e.g., Tiede, 1996; Edmondson and Esling, 2006). Lingual measurement techniques such as ultrasound imaging (e.g., Gick, 2002) and modeling approaches involving static vocal tract surfaces (e.g., Iskarous et al., 2003) should thus be used with full awareness of this limitation.

As with any methodology, using x-ray videos is not without its drawbacks. Because of ethical and safety concerns, researchers seeking to work with x-ray are limited to using existing databases, limiting researchers' abilities to control their data sets; the tokens produced in the Laval database (Munhall et al., 1995), however, did cover a wide range of phonetic contexts, indicating that the observed action of the velic traverse is sufficiently robust to overcome local variations. Further, as x-ray shows movement only along the mid-line, it provides an incomplete picture of actions of the surrounding structures. Additional work will be needed to describe actions off mid-line, which will help to elucidate the previously impenetrable actions of the palatoglossus and palatopharyngeus during oral constrictions. This will in turn lead to a better understanding of the actions of the velic traverse as part of a whole oral constrictor, including all of the structures surrounding the oropharyngeal isthmus.

The authors thank audiences at the Acoustical Society of America and elsewhere, and especially Fredericka Bell-Berti, for helpful comments. This work was supported by a Discovery Grant to B.G. from the Natural Sciences and Engineering Research Council of Canada.