Palatalization in coronal consonants of Polish: A three-/four-dimensional ultrasound study Open Access

The consonant inventory of Polish includes a phonemically contrastive “hard” series of posterior sibilants, /ṣ, ẓ, $ts͜$̣, $dz͜$̣/, and a “soft” series, /ɕ, ʑ, $tɕ͜$⁠, $dɕ͜$/.¹ The hard posterior sibilants are variously called “post-alveolar” or “retroflex” [here the transcription proposed by Ladefoged and Disner (2016) is used]. The soft posterior sibilants are variously called “alveopalatal” or “prepalatal,” and are inherently palatalized. These soft posterior consonants occur as surface alternants of dental consonants /t, d, s, z, n, l^γ, r/ in morphophonological “coronal palatalization” contexts [cf. Rubach (1984)], as well as within morphemes that show no synchronic alternations.² Additionally, the hard posterior sibilants, like all other obstruents in Polish, become allophonically palatalized when preceding /i/ or /j/.

Polish phoneticians have interpreted the distinction between the hard and soft posterior series primarily in terms of the place of articulation. For example, Bąk (1977) describes the hard series as being articulated with the front of the tongue, and the soft series with the middle part of the tongue. Koneczna and Zawadowski (1951) argue that the soft series is palatal and the hard series is coronal and alveolar (Polish term: dziąsłowe). Wierzchowska (1971) describes the soft series as prepalatal and the hard series as alveolar. Both Koneczna and Zawadowski (1951) and Wierzchowska (1971) stress the exceptionally high level of tongue raising towards the hard palate in the soft series. Hamann (2003), Żygis (2003), Żygis et al. (2012), and a number of other recent studies describe the hard posteriors of Polish as retroflexes. An electromagnetic articulography (EMA) study of Lorenc and Świȩciński (2014/2015) refers to the concave shape of the tongue blade, while Bukmaier and Harrington (2016) distinguish between /s, ʂ, ɕ/ (/s, ṣ, ɕ/ in our notation) in terms of the tongue tip orientation. Lulich and Cavar (2019) point to a systematic difference in the position of the tongue root between the alveopalatals and hard posteriors. Phonological accounts of Polish treat both series /ṣ, ẓ, $ts͜$̣, $dz͜$̣/ and the soft /ɕ, ʑ, $tɕ͜$⁠, $dɕ͜$/ as posterior, differentiating between the two series in terms of palatalization features, e.g., [+/− back] (Rubach, 1984; Szpyra, 1995; Rydzewski, 2014) or feature [+/− ATR] [as a palatalization feature (Cavar, 2004)].

In this paper we build on the work Lulich and Cavar (2019) to directly examine the properties of Polish palatalized consonants in opposition to non-palatalized consonants. In particular, we identify the systematic articulatory characteristics of palatalized sounds as opposed to hard consonants.³ While we report on a subset of coronal consonants, preliminary results indicate that the findings extend to the allophonic palatalization in other places of articulation in Polish. The findings are also confirmed by results from other languages—Russian (Cavar and Lulich, 2018; Matsui and Kochetov, 2018) and Irish (Bennett et al., 2018). Thus, the current study contributes to our improved understanding of the mechanism of palatalization in general [cf. Bateman (2011), Bhat (1978), Cavar (2004), Cavar and Hamann (2003), Chen (1996), Kim (2001), Żygis (2003)]. We examine the effects of vowel context on phonemic palatalization, the effects of vocalic context, consonantal place and the differences (if any) between phonemically and allophonically palatalized consonants.

This study makes use of new, complementary data from the same speakers as described by Lulich and Cavar (2019), which reported primarily on the allophonic variation of vowels. Ten native speakers of standard Polish (Greenberg et al., 2017), age 35–58 years old, participated in this study (female: speakers 1, 3, 4, 6, 7; male: speakers 2, 5, 8, 9, 10). All speakers had at least a college degree with their education completed in Poland. There were no perceptible dialectal differences among the speakers. Seven speakers (speakers 1, 2, 3, 6, 7, 9, and 10) had lived outside of Poland for over 10 years, the remaining three participants were short-term visitors at Indiana University with their permanent residence in Poland. None of the participants showed a detectable foreign accent in their Polish, but they retained a Polish accent when speaking English. Only speaker 1 (author MEC) was aware of the purpose of the study. Speakers 1–5 and speakers 6–10 were recruited separately. Real words were elicited from speakers 1–5, while we switched to nonce words for speakers 6–10 to better control for the syllable- and word-position and to expand the vocalic environments. In both groups, all words were elicited in isolation, and presented in written (orthographic) form on a computer monitor. The words that were analyzed in this study are listed in Table S1 of the supplementary materials.⁴ Words were elicited individually within a block, and the block was repeated at least two times. This study compares the articulation of corresponding soft-hard pairs of coronal consonants articulated at the same place of articulation—anterior or posterior—that differ only in softness,⁵ i.e., phonemic pairs /ɕ/ vs /ṣ/ in different vocalic contexts, and allophonic pairs [t]-[t^j] and [ṣ] - [ṣ^j] (5 pairs of words for each of 10 speakers, repeated two times, for a total of 200 tokens). Twenty-four tokens were discarded due to technical problems. The complete set of data including the coding variables used in the analysis are provided in the supplementary materials.⁴

Ultrasound images of the tongue were recorded in the Speech Production Laboratory at Indiana University, using a Philips EPIQ 7G ultrasound system with a Philips xMatrix x6–1 digital 3D/4D transducer (Lulich et al., 2018) secured under the chin using an Articulate Instruments ultrasound stabilization headset (Scobbie et al., 2008). Ultrasound recordings were synchronized with simultaneous audio recordings as described elsewhere (Lulich et al., 2018; Lulich and Cavar, 2019; Lulich and Pearson, 2019). They were further analyzed using a custom toolbox for matlab called “WASL.” ⁶

Although 3D/4D ultrasound was used to record tongue articulation, this study focuses on the traditionally 2D midsagittal tongue contours. We selected the frames closest to the temporal mid point of the consonant. For each consonant, we identified two points along the tongue surface: the point of maximum raising of the tongue body, and the point opposite the tendon of the genioglossus to examine the tongue root position as in Fig. 1, cf. Lulich and Cavar (2019). For every pair of corresponding soft-hard sounds within each repetition, the Euclidean distance between the points of maximum raising in the soft vs hard consonant [representing tongue body advancement (TBA)] was calculated, as well as the horizontal distance [representing tongue body fronting (TBF)] and the vertical distance [representing tongue body raising (TBR)], where $TBA2=TBF2+TBR2$⁠. The Euclidean distance between the point opposite the tendon of the genioglossus in the soft and hard consonants was likewise calculated [representing tongue root advancement (TRA)].

Positive values for TBA, TBF, and TRA indicate advancement in the soft consonant rather than in the hard consonant. Positive values for TBR indicating raising in the soft consonant rather than the hard consonant. We also analyzed the ultrasound images qualitatively by tracing the tongue surface and comparing tongue surface shapes among the different consonants.

Normality of the variables of TBA, TBF, TBR, and TRA was tested using the Shapiro-Wilk statistic (with Bonferroni correction, α = 0.05/4 = 0.0125). All four variables were found to be non-normally distributed. Non-parametric tests were consequently used for all statistical analyses. One-sample Wilcoxon signed rank tests were used (with Bonferroni correction, α = 0.05/4 = 0.0125) to assess the hypotheses that fronting/raising/advancement are greater in soft consonants than hard consonants. Kruskal-Wallis post hoc tests were used to test for effects of gender (male, female), length of residence outside of Poland (short, long), word type (real, nonce), vowel context ([i, e, a]), consonant ([t, ṣ]), and palatalization type (phonemic vs allophonic), with Bonferroni correction $α/48=0.001$⁠. matlab was used to carry out statistical analyses and generate graphs.

Complete descriptive statistics (medians) and inferential statistics (Shapiro-Wilk, Wilcoxon signed rank, and Kruskal-Wallis tests) are given in Table S2 of the supplementary material.⁴ Four independent Shapiro-Wilk tests indicated that all four variables (TRA, TBF, TBR, and TBA) had distributions that were significantly different from normal after Bonferroni correction (α = 0.05/4 = 0.0125). Four independent one-tailed Wilcoxon signed rank tests indicated that all four variables had medians values that were significantly larger than 0 after Bonferroni correction. Median values for TRA, TBF, TBR, and TBA were 0.82, 0.79, 0.35, and 0.97 cm, respectively. TBF, TBR, and TBA all included negative values, indicating retraction and/or lowering of the tongue blade in the soft consonants, while TRA did not include any negative values (Fig. 2). Twenty-four independent Kruskal-Wallis tests of main effects were carried out on four variables over six factors (4 × 6 = 24). The factors were vowel context ([i, e, a]), consonant ([t, ṣ]), palatalization type (phonemic, allophonic), gender, duration of residence outside of Poland, and word type (real, nonce). The critical value α after Bonferroni correction was therefore α = 0.05/24 = 0.002. No significant main effects of gender, duration of residence outside Poland, or word type was found on any of the four variables (p > 0.005). No significant main effect of vowel context was found on any of the four variables within the phonemic type of palatalization (p > 0.07). Similarly, no effect of consonant type was found on any of the four variables within the allophonic type of palatalization (p > 0.1). A main effect of palatalization type was found only for TBF (p ≪ 0.001) and TBA (p ≪ 0.001), with effect sizes (difference in medians) of 0.46 and 0.41 cm, respectively. Box plots in Fig. 2 show the distribution of values for TRA, TBF, TBR, and TBA. The distribution by palatalization type for TBF and TBA, which were found to be significantly affected by palatalization type, is shown in Fig. 3.

From the statistical analyses, we conclude that soft consonants exhibit advancement/fronting/raising of the tongue relative to hard consonants, without regard to vowel context or consonant, but that TBR and TBA are sensitive to the type of palatalization. Moreover, although TBF and TBA have median values comparable to TRA, they also exhibit larger overall variability, including some negative values (indicating greater fronting/advancement in the hard consonant). The variable that is most consistently associated with palatalization in this study is therefore TRA.

For a qualitative examination of tongue shapes in phonemically hard [ṣ] and [t], and soft [ɕ] consonants, representative tongue contours for each speaker are shown in Fig. S1 in the supplementary material.⁴ Tongue blade and tip position are nearly identical in [t] and [ṣ] for all 10 speakers. The tongue body is raised and fronted in [ɕ] for most speakers, but the difference is minimal for speakers 3, 8, and 10. For most speakers, tongue root advancement is greater for the soft consonants than for the hard anterior consonants, and greater for the hard anteriors than for the hard posteriors. This difference between hard anteriors and posteriors is minimal for speakers 2, 8, and 10, while the difference between soft and hard anterior consonants is minimal for speaker 4.

For a qualitative examination of tongue shapes in allophonically soft consonants, [ṣ^j] and [t^j], in contrast to their hard [ṣ] or [t] variants, representative tongue contours for each speaker are shown in Fig. S2 in the supplementary material.⁴

For palatalized consonants, the tongue body is raised or fronted in most speakers, although raising is minimal for at least one of the consonant pairs in speakers 1, 2, 4, 5, and 10, and fronting is minimal for at least one of the consonant pairs in speakers 1, 2, 3, 5, and 8. The tongue root is advanced in the palatalized consonants in all speakers, although it is minimal for [ṣ^j] vs [ṣ] in speakers 2 and 5, for [t^j] vs [t] in speakers 8 and 10, and for both pairs of consonants in speaker 3.

In this study, we analyzed 3D/4D ultrasound images of phonemically and allophonically palatalized voiceless consonants in Polish. The results have shown that for both phonemic and allophonic palatalization in Polish, the most consistent correlate of palatalization is advancement of the tongue root, with an effect size of 0.82 cm. This is similar to the effect size for soft vs hard consonants reported by Lulich and Cavar (2019) for a different subset of data from the same speakers (0.79 cm). There is a comparable effect size for tongue body fronting and advancement (which, incidentally, are strongly correlated with each other), but tongue body fronting and advancement are more variable and sometimes indicated a reversal, in which the hard consonant was more fronted/advanced than the soft consonant. Tongue body raising was associated with a significantly smaller effect size.

Based on an analysis of soft-hard vowel allophony, Lulich and Cavar (2019) concluded that Polish vowel allophony can be accounted for by the tongue root advancement. The present findings support and extend these conclusions. We conclude that the tongue root advancement is the unique most stable correlate of palatalization in Polish, driven by tongue root advancement in both phonemic and allophonic palatalization, and that the observed associations of tongue body fronting and advancement are by-products of tongue root advancement via the muscular hydrostat properties of the tongue (Gilbert et al., 2007). Furthermore, we propose that palatalization cross-linguistically is in general an effect of tongue root advancement (except in cases where historical sound change has rendered the involvement of the tongue root opaque). Future studies are needed to investigate the role of the tongue root in palatalization in other languages.

We would like to thank Max Nelson and Olivia Foley for their help with data collection for this project.