Priebalsių balsingumo įtaka formantinei bendrinės lietuvių kalbos balsių struktūrai: faktorinė analizė

Acta Linguistica Lithuanica 2025, vol. 92, pp. 1–24 Contents link / Turinys nuoroda ISSN 2669-218X (online / elektroninis) THE INFLUENCE OF CONSONANT VOICING ON THE FORMANT STRUCTURE OF STANDARD LITHUANIAN VOWELS: FACTORIAL ANALYSIS Effects of Consonant Voicing on Vowel Formant Distribution in Standard Lithuanian: A Factorial Analysis EVALDAS ŠVAGERIS Institute of the Lithuanian Language E-mail: [email protected] ORCID ID: https://orcid.org/0009-0002-5624-7275 Research areas: experimental phonetics, Baltic language prosody and vocalism. https://doi.org/10.35321/all92-10 ABSTRACT This article continues experimental research on the dynamics of phenomena of Standard Lithuanian vocalism (its positional variation). Attention is focused on the interrelations and hierarchy of factors directly affecting the formant structure of vowels. After the empirical data were supplemented with vowel sequences containing voiced plosive consonants, the power of the voicing factor to affect the distribution of F1 and F2 in the spectrum was analysed from the perspective of the whole. For this purpose, a factorial analysis of the data (ANOVA) was conducted to determine the hierarchical positions of the consonant voicing factor and its potential for coordinating its operation with other factors. The results revealed that the weak positions of the factor under discussion in hierarchical chains indicate a rather low level of autonomy. Its relative power appears to correlate with the articulatory vowel series: the farther back the vowel, the more pronounced the formant differences determined by the voicing factor. Moreover, although this factor does not possess greater differential power, it expands the articulatory–spectral space, enabling competing factors to act with greater force (for example, highlighting the fronting of high back vowels in the position between dental plosive consonants and preceding vowels). KEYWORDS: vowel formants, Standard Lithuanian, factorial analysis, factor hierarchy, positional vowel variants. ABSTRACT This study aims to continue experimental research on the dynamics of vowel formant distribution in Standard Lithuanian. The primary focus is on the interaction and hierarchy of factors that directly influence Copyright © 2025 Evaldas Švageris. Published by the Institute of the Lithuanian Language. This is an Open Access article distributed under the terms of the Creative Commons Attribution Licence, which permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited. // Išleido Lietuvių kalbos institutas. Šis straipsnis yra atviros prieigos, platinamas pagal „Creative Commons“ priskyrimo licencijos sąlygas, leidžiančias neribotai naudoti, platinti ir atkurti turinį bet kokioje laikmenoje, nurodant autorių ir šaltinį. Received: / Received: 2025-04-11. Accepted: / Accepted: 2025-06-02. 1 formant placement in the spectra of Lithuanian vowels. After CVC combinations with voiced plosive consonants were added to the dataset, the differential power of the voicing factor was analysed. A factorial analysis (two-way ANOVA) was performed to determine what position in the hierarchy the voicing factor can take. Additionally, the question of factor coordination in the field of formant distribution was raised. The results suggest that the voicing factor occupies weak positions in the hierarchy and exhibits low-level autonomy. To some extent, its relative power correlates with the backness of the vowel, as it shifts towards the back tongue position, differences implied by the consonant voicing factor on formant distribution increase. Although it has no significant impact, consonant voicing plays a role in expanding the spectral space, allowing other factors to exert greater influence (for example, it creates conditions for high back vowels to be more fronted in a dental stop environment). KEYWORDS: vowel formants, Standard Lithuanian, factorial analysis, factor hierarchy, vowel allophones. INTRODUCTION The first attempts to analyse in greater detail, by experimental means, the dynamics of variation in Standard Lithuanian vocalism—more precisely, the interrelations among the factors determining the phonetic structure of vowels—produced essentially positive results (Švageris 2024; Švageris, Urbanavičienė 2024). Although the limited amount of empirical material and the small number of factors included in the analysis have so far prevented far-reaching conclusions, it can nevertheless be observed that the dynamic dimension of quantitative and qualitative vowel features is not chaotic (for research based on a similar concept, see, for example, Lindblom, Moon 1994: 40–55; Roesler, Yung 2018: 535–540; from a more theoretical perspective, see Lindblom 1990: 65–100). It is evident that the factors are connected by hierarchical relations, and the spectral characteristics of positional vowel variants can be inferred from correlations established through statistical calculations. For example, the more open a vowel is, the more its frequency spectrum depends on the prosodic factor (its position in relation to phrase stress); the farther back it is, the more clearly its formant structure signals the articulatory type of the plosive consonant adjacent to it, and so on. In addition, the results revealed that research practice relying on the acoustic characteristics of speech sounds pronounced in isolation may distort tendencies regarded as ordinary. Under the conditions in question, the influence of the subjective speaker factor often increases (hyperarticulation is possible; on this factor, see Johnson et al. 1993: 701–714; Lee et al. 2019: 1568–1579; from the perspective of the Baltic languages, see Čeirane, Indričianė 2012: 43; Ambrazevičius, Leskauskaitė 2014: 168; Kazlauskienė 2018: 41), sometimes neutralising, for example, qualitative differences between long and short vowels. It therefore seems logical and justified to draw the broader conclusion that naturalness of speech is an important condition of its systematicity (Švageris, Urbanavičienė 2024). Looking ahead, research of this kind could be useful in two ways. First, the direction of analysis itself could eventually broaden and supplement the discourse of experimental research on Lithuanian vocalism and provide a basis for comparative work with analogous material from other languages (Hillenbrand et al. 2001: 754–755; Roesler, Yung 2018: 535–540). Second, if increasing the number of factors and the amount of empirical data yielded even clearer and more specific results, the possibilities for predicting and modelling the features of positional vowel variants would expand (an applied aspect). In other words, from the perspective of applied speech synthesis, it would be useful to refine a model that classifies the set of features of positional vowel variants from the perspective of the interrelations among the factors directly affecting them. An additional alternative for predicting 2 the acoustic structure of speech sounds could contribute to improving the naturalness of computer-generated human speech. All this seems to provide sufficient incentive to continue the research already begun. 1. Research Problem, Aim and Objectives The problem addressed by this experimental study arises from the need to increase the number of factors directly affecting the qualitative spectra of Standard Lithuanian vowels included in the analysis. Naturally, the more factors are incorporated into the model, the more comprehensive and universal the results should be. Since this is essentially a continuation of the work already begun (Švageris 2024; Švageris, Urbanavičienė 2024), the factor of plosive consonant voicing is added here to those included in earlier stages of the research (speaker, phonological quantity, articulatory type of plosive consonants, and phrase stress). In experimental research on Lithuanian, the possible effect of this factor on the distribution of vowel formants has apparently not yet been examined in detail. Indirect connections can be discerned only in isolated studies, and mainly in discussions of coarticulation among speech sounds (Ambrazevičius, Leskauskaitė 2014: 202–225; Jaroslavienė et al. 2019: 62–167; from the perspective of other languages, see Kewley-Port 1982). Foreign researchers indicate that, in comparative terms, vowels occurring between voiced consonants are characterised by lowering of the first formant. The formant distribution of front vowels is affected more strongly by this factor (F1 is lowered by approximately ~100 Hz on average), whereas its effect on mid and back vowels is supposedly much weaker (~20 Hz; see Stevens, House 1963: 111–128; Hillenbrand et al. 2001: 754–755; also Adriaans 2018: 21–25). It is true, however, that consonants in prevocalic rather than postvocalic positions leave a stronger trace in the spectrum. Researchers suppose that this correlation is probably a spectral reflection of variation in laryngeal position (when pronouncing voiced consonants, the larynx is lowered). More generally, these configurations are apparently related to the dynamics of subglottal and supraglottal pressure variation required to (de)activate vocal-fold phonation (Ambrazevičius, Leskauskaitė 2014: 171). In order for phonation to be maintained as long as possible during the explosion phase of voiced plosive consonants, supraglottal pressure must be reduced (the volume of the supraglottal vocal-tract cavity must be increased); therefore, lowering the larynx appears clearly motivated from this perspective (Stevens 1998: 465–485). Moreover, if the configuration of the vocal tract were detailed according to the logic of perturbation theory, it would have to be acknowledged that near the larynx the tract not only extends but also widens (which is another favourable condition for reducing supraglottal pressure). More precisely, the first formant can be lowered in two cases: when the vowel is labialised (because the antinode of the first resonant mode is located at the lips) or when the vocal tract widens near the larynx (where the node of this mode is located; see Chiba, Kajiyama 1941; Kent, Read 2002: 28–32). Thus, the aim of this article is to determine, by experimental means, the relative power of the voicing factor of plosive consonants to affect the formant distribution of vowels in Standard Lithuanian. Indeed, there is probably little doubt that, from the perspective under discussion, the influence of this factor would differ significantly from that established in analogous studies of other languages. Rather, this is a phonetic universal, determined exclusively by human anatomy and the physical causes of speech-sound generation (i.e. the conditions necessary for phonation in voiced plosives 3 of consonants), rather than an expression of linguistic specificity (Stevens 1998: 465–485). Therefore, as already mentioned, what is more relevant to this study is the expansion of the factorial analysis itself (the addition of another factor), after which it should become clear whether the hierarchical relations among the factors change as their number increases. For example, it would be useful to determine whether the voicelessness–voicing factor becomes a serious obstacle to differentiating vowel formant structure according to those factors whose influence, based on earlier data, was more strongly reflected in differences in the first formant. To construct a general dynamic picture of qualitative vowel features, not only the power of the factors themselves is important, but also the directions of their operation and their compatibility; therefore, all these aims determine somewhat broader research objectives: 1. To expand and supplement the empirical material of the study (Švageris 2024; Švageris, Urbanavičienė 2024) with data classified according to the consonant voicing factor (i.e. to record and include in the study all identical positions containing voiced plosive consonants). 2. To compare statistically the relative power of five factors—speakers, phonological quantity, phrase stress, consonant place of articulation, and consonant voicing—to affect vowel formant structure, and also to specify the directions and compatibility of their operation. 3. To supplement and, if necessary, clarify the hierarchical chains of factors established during the earlier stage of the study (i.e. to expand the model itself), and to determine the position of the consonant voicing factor within them. 2. Brief Review of Research, Empirical Material and Method Since this study continues work already begun, it follows essentially the same methodological guidelines (Švageris 2024; Švageris, Urbanavičienė 2024). Studies on closely related topics have already been discussed in greater detail, so here it is possible only to briefly report what emerged from the results of the latest factorial analysis (Švageris 2024). It is known that the quantitative and qualitative features of vowels are influenced by various linguistic and extralinguistic factors (for a general perspective on the effect of various factors on the spectral characteristics of Lithuanian vowels, see Vaitkevičiūtė 1961; Pakerys 1982: 132–134; 2003 [1986]: 24, 35–36, 190–191; Kaukėnienė 2002; Kudirka 2009; Jaroslavienė 2014; 2015; Urbanavičienė et al. 2019: 78; Ledichova 2020a: 1–33; 2020b; and cf. studies of other languages: Johnson et al. 1993: 701–714; Hillenbrand et al. 2001; Tsukada 2009). The distribution of a vowel’s formants in the frequency spectrum—how the vowel is articulated—depends on vowel height and backness, phonotactic and prosodic factors of speech, the individual articulation of the speaker, and so on (ibid.). Nevertheless, a more detailed acoustic picture of vowels can be formed when all these factors are viewed from an interactional perspective, when by objective criteria ¹ Let us remember that this factorial analysis (using statistical methods) aims to clarify the relations and hierarchy of the factors determining variation in speech sounds. 4 the comparative weight of each of them is assessed. The data analysed in this way show that, in terms of their effect on the distribution of F1 and F2, the highest hierarchical positions are generally occupied by phonological quantity (Švageris 2024)2. Nevertheless, this conclusion loses its evidential force when articulatory vowel subgroups are analysed separately. When the analysis focuses on back vowels, the plosive-consonant factor displaces phonological vowel length from the dominant position. This means that, without grouping the data, the general distribution of formants makes it most probable to distinguish the voiceless [p], [k] and [t] adjacent to vowels [ɑː / ɛ] and [uː / ʊ]. More specifically, the greatest qualitative change in the spectrum of the back low [ɑː / ɛ] is caused by articulation of the bilabial plosive (p[ɑː / ɛ]p, because of the lowering of F2 by an average of 200 Hz, is particularly clearly separated from the k[ɑː / ɛ]k and t[ɑː / ɛ]t variants). The qualitative spectrum of the back high [uː / ʊ] diverges most from other positional variants when surrounded by the front lingual dental [t] (F2 rises by the same average amount of 200 Hz; this phenomenon should also be regarded as a phonetic universal; see Hillenbrand et al. 2001). Several other correlations also have a strong statistical basis: vowel openness and phrase stress (the more open the vowel, the more its formant structure depends on its position in relation to phrase stress); vowel frontness/height and phonological quantity (the more front and higher the vowel, the more clearly the formants distinguish its short and long variants); and articulatory tenseness and the weight of the speaker factor (the tenser the articulation of the vowel, the farther it is from the centre of the acoustic spectrum, and the higher the speaker factor ranks in the hierarchical chain). The observed cases of unidirectional factor operation also retain considerable added value. Apparently, the labialisation of open vowels between labial consonants is organically compatible with the openness differences in [ɑː / ɛ] implied by phrase stress. The relation between the latter factor and phonological quantity is different in this respect, since, as the data show, their influence cannot increase in parallel: as the influence of the former increases, that of the latter simultaneously tends to decrease (meaning that their directions of operation are incompatible). To compare the latest data with the earlier data, strict empirical-data selection must be maintained. This applies first of all to the speakers. If other speakers were recorded at this stage of the study, the latter factor would naturally have to be eliminated from the analysis, since the data could not be compared directly. For this reason, the same men aged 38 (D1) and 20 (D2) were invited to record the additional material; they can be regarded as suitable representatives of the substandard of Standard Lithuanian (Miliūnaitė 2018: 53). Let us recall that the analysis used trisyllabic CVC-type sequences with so-called “corner corner vowels” [iː / ɪ], [uː / ʊ] and [ɑː / ɛ], corresponding to tongue positions farthest from the neutral position (Titze 2000: 162–163). In total, 36 sequences were formed with plosive consonants [p], [t], [k], [b], [d], [g] and short [ɪ], [ʊ] and [ɛ] and long [iː], [uː] and [ɑː] vowels; short three-word phrases were constructed with them (e.g. ‘say KAK so’; ‘say BŪB so’). By additionally varying them according to the other factors analysed (the position of the sequence in relation to phrase stress, the study’s 2 Here it should be emphasised that what is meant is the relative similarity of phonological quantity, based on statistical calculations, with respect to the remaining three factors (phrase stress, voiceless plosive consonants, and speakers). As the number of factors analysed increases, their hierarchical arrangement may naturally change. 5 position in relation to phrase stress), 108 slides were prepared for recording, which both speakers were asked to read twenty times each (five at a time). A total of 4320 tokens were recorded. The recordings were made in the phonetics laboratory of Vilnius University using professional equipment and following the usual requirements for experiments of this kind. All recorded sequences were segmented, annotated and processed using the sound-analysis program Praat. A special script for this program automatically extracted the data required for the study—the values, in hertz, of the first three formants (F1, F2, F3) in the central portions of the vowels—with which two-way analysis-of-variance (TWO-WAY ANOVA, Rietvel, van Hout 2005) calculations were then performed. First, this statistical model was used to assess the power of the consonant voicing factor to differentiate the data in general, i.e. when the empirical data were not grouped according to any other factors (these results are illustrated in three-dimensional graphs). Then, after conditions of direct competition were created (when competing factors were exchanged), hierarchical chains of factors were calculated for each articulatory type of vowel using the same model. On this occasion, the compatibility of the factors was also analysed using additional graphical methods. In this case, not only the number of times one factor surpassed another was calculated (as a percentage), but cases were also recorded in which, despite the superiority of one factor, the data were successfully separated according to both directly competing factors. If the critical threshold of p = 0.05 was not exceeded in either comparison, the effects of both factors on the frequency spectrum were considered compatible. The directions of factor operation were also illustrated by additional graphical methods (using the neutral schwa as the conditional reference point). These graphical data are presented to detail the dynamic shifts of the formants as fully as possible and to relate them more clearly to articulatory vowel changes (for all other details of the research methodology, see Švageris 2024; Švageris, Urbanavičienė 2024). 3. Results of the Study. The Effect of Consonant Voicing on the Formant Distribution of the Low Back [ɑː/ɑ] It is optimal to begin the factorial analysis with the question of what statistical probability exists, in the overall dataset (1440 tokens in total), of distinguishing qualitative [ɑː / ɑ] variants occurring between voiced and between voiceless plosive consonants. Naturally, without grouping the data in greater detail, the homogeneity (integrity) of the samples being compared suffers and the reliability of the statistical results decreases. Nevertheless, judging from the overall graphical distribution of the data (see Figure 1) and the calculated probability estimate (p = 1,72765E−12), there appears to be sufficient reason to believe that the consonant voicing factor makes a substantial contribution to differentiating the qualitative spectrum of low back vowels. Although differences in both formants could be discerned in three-dimensional graphical space, the mean values clearly indicate that positional vowel variants of this type are distinguished only by the first formant (among voiceless consonants—F1 = 714 Hz, F2 = 1298 Hz; among voiced consonants F1 = 634 Hz, F2 = 1283 Hz; difference ~ 80 Hz). Essentially, the phonetic universal indicated in the introductory part of the article is confirmed (additional lowering of the larynx to initiate phonation → lowering of F1). It should also be added that the results of analogous studies in other languages suggested a much smaller and essentially statistically insignificant effect of this factor on back vowels (differences in the first formant amounted to only 15–20 Hz; Hillenbrand et al. 2001). 6 and, in principle, statistically insignificant effect of this factor (differences in the first formant amounted to only 15–20 Hz; Hillenbrand et al. 2001). Figure 1. Distribution of the first two formants of low back [ɑː / ɛ] according to the consonant voicing factor (p = 1,72765E−12)³ Moreover, despite the tendency implied by the statistical calculations, another detail of the graphical data dispersion is striking. Unequal formant-variation errors are noticeable, and their mean values do not reflect them. The voicelessness of the consonants (red ellipse) appears to somewhat restrict the F2 shifts of [ɑː / ɛ] in the spectrum. It follows logically that the articulation of voiceless plosive consonants, traditionally regarded as tenser (Pickett 1999: 125), correlates in these cases with lower mobility of the horizontal tongue position, whereas the widened glottal region (the overall lowering of F1) in sequences with voiced consonants opens the way for somewhat greater changes in vowel F2. The direction of operation of the voicelessness–voicing factor and, more importantly, how it correlates with the corresponding vectors of the other factors can be seen in Figure 2. The points placed there (F1 and F2 values) illustrate the greatest recorded qualitative differences in the vowels of the group under discussion with respect to each factor. For example, according to the speaker factor, post-stress long [ɑː] in a sequence with voiced dental [d] had the highest statistical estimate (p = 1,6E−14). The formant means for this particular case are also plotted in the figure (D1 F1 = 614 Hz, F2 = 1414 Hz: D2 F1 = 672, F2 = 1173 Hz). Following the same principle, analogous points for the other factors were also arranged. This analytical approach to the data creates a visual impression of the limits of variation in spectral characteristics that positional [ɑː / ɛ] ³ Among voiceless consonants—the red ellipse; among voiced consonants—the blue one (the overall dataset). 7 variants can reach when the factors act with maximum force (yielding the greatest statistical probability of a difference). Figure 2. Potential directions of factor operation ([ɑ: / e] spectral space) Taking as the conditional reference point the first two formants of the neutral voice schwa (F1 = 500 Hz, F2 = 1500 Hz; the large black point in the figure), the effect of the factors on the qualitative vowel spectrum can be assessed through the prism of articulatory (un)tenseness (the shifts of F1 and F2 from or toward the centre of the acoustic spectrum). From this perspective, the consonantal-environment factor stands out more clearly, differentiating the spectrum to the greatest degree (p = 1E-28) essentially only along the ordinate, the y-axis (because of marked labialisation in sequences with labial consonants), whereas all the other factors operate almost in a single (de)centralisation direction. If the entire study were limited to this graphical material, it could be stated that in these pairs one member is simply articulated less strongly (is less tense). Naturally, short vowel variants shift toward centralisation more than long ones; what is unexpected, however, is that their greatest statistically significant difference (p = 2,1E-22) was established in the post-stress position in a sequence with voiced velar consonants (D2, means: g[ɑ:]g F1 = 591 Hz, F2 = 1332 Hz; g[ɑ:]g F1 = 667 Hz; F2 = 1166 Hz). Correspondingly, vowels occurring among voiced rather than voiceless consonants move farther from the centre of the spectrum (p = 1E-14) (D2; post-stress position; means: k[ɑ:]k F1 = 629 Hz, F2 = 1326 Hz; g[ɑ:]g F1 = 667 Hz, F2 = 1164 Hz). An acoustic difference of the same kind (p = 1,6E-14) occurs between the speakers (post-stress position, long d[ɑ:]d; means: D1 F1 = 614 Hz, F2 = 1414 Hz; D2 F1 = 672 Hz, F2 = 1173 Hz), and between positions in relation to phrase stress (p = 7E-18; D2, long k[aː]k; means: stressed F1 = 782 Hz, F2 = 1228 Hz; post-stress F1 = 629 Hz, F2 = 1326 Hz). It is true that the spectral expression and statistical weight of all these differences do not correlate. The statistical power of phonological quantity and the articulatory type of consonants is greater than that of all the remaining factors, but this advantage is not always accompanied by the greatest graphical distance between the points. Again, this graphical dimension merely illustrates the most pronounced differences. It must not be forgotten that in some cases the acoustic indicators differed only slightly, or not at all, according to all these factors; therefore, the informativeness of these data is not absolute. Each phonetic realisation of a vowel is one result or another of the interaction of ALL factors, so the added value of this diagram lies only in highlighting their potential directions of operation. The latter should clarify the possibilities for compatibility among the factors affecting the formant structure of vowels. The dominant position of consonants is also confirmed by an analysis of direct factor competition. The threefold increase in the data, considered as a whole, does not weaken the effect of labial [p] and [b] (lowering of F2) on the more open back low vowels. As the number of directly comparable pairs increased from 16 to 44, the percentage expression of this factor’s superiority over the others remained the highest (79 percent; see Figure 3). This means that in nearly four out of five cases, when analysing the mixed data, it is possible to determine with fairly high confidence between which plosive consonants (more precisely, labial or non-labial ones) the vowels of the [ɑː / ɛ] subgroup occur. This is also clearly demonstrated by the means of the overall data set ([p]/[b] F1 = 689 Hz, F2 = 1178 Hz, [t]/[d] F1 = 675 Hz, F2 = 1337; [k]/[g] F1 = 675 Hz, F2 = 1357 Hz; interaction probability, enormous, p = 1,8E−122). This entire configuration of indicators shows that, in the general case, the consonant factor clearly occupies the highest position in terms of its effect on the formant structure of low back vowels. The hierarchical positions of the remaining factors are considerably more problematic. It is obvious that the increased volume of data strengthened the power of the speaker factor. Apparently, vowel sequences with voiced consonants (compared with voiceless ones) provide a more favourable environment for individual articulation. Perhaps it is no coincidence that the directions of action of these two factors are highly similar (see Figure 2). When only four factors were compared (Švageris 2024), distinguishing one speaker from another on the basis of the distribution of the first two formants was rather difficult (only 23 percent), whereas now this probability is twice as high (55 percent). The overall means (without voiced consonants – D1 F1 = 736 Hz F2 = 1326 Hz; D2 F1 = 692 Hz F2 = 1269 Hz; with voiced consonants D1 F1 = 682 Hz F2 = 1340 Hz; D2 F1 = 672 F2 = 1239 Hz) suggest that the space for individual pronunciation expanded for two reasons: because of the stronger effect of voiced consonants on D1’s glottal region (lowering of F1), and probably because of the intensified D2 labialisation under these conditions (more pronounced lowering of F2). The somewhat weakened effects of phrase-stress position are probably a consequence of the increased power of the consonant-voicing and speaker factors. The phonological quantity criterion remains the weakest in this entire context. The probability of distinguishing short [ɛ] from long [ɑː] is the lowest (38 percent), although, for the sake of accuracy, it should be said that this factor’s lag behind the phrase-stress and voicing factors probably does not exceed the limits of standard error. 9 Figure 3. Relative factor power under conditions of direct competition ([αː / ɛ] F1 and F2 differentiation) It must therefore be stated that a more pronounced divide forms only between the articulatory type of consonants and all the remaining factors (the speaker distinction is somewhat more significant), and the results of the direct-competition analysis allow the following hierarchical chain to be constructed (from strongest to weakest): place of consonant articulation → speakers → phrase stress → voicing → phonological quantity. The results of the factor-coordination analysis make it possible to specify all the ideas expressed above somewhat further (see Figure 4). It should be recalled that, in this case, what is assessed is whether interchanging the factors does not become an obstacle to classifying the two opposing sides of the data at a sufficiently high statistical level (when p <0,05). If, for example, failure to distinguish long and short vowels (combining them into a common sample) did not prevent D1 from being reliably distinguished from D2 according to the F1 and F2 distribution, and if, at the same time, failure to distinguish the speakers did not prevent the vowels from being differentiated according to the phonological-quantity criterion, the operation of the factors is considered coordinated. Since there are many phonetic contexts in which comparisons are made in this respect, it is possible simply to count how many times the data are reliably distinguished according to both directly competing factors and how many times according to only one of them. The graphic space (see Figure 4) reflects the percentage expression of this ratio—the higher the percentage, the more cases in which the effects of different factors were statistically coordinated. All this reveals whether high / low hierarchical factor positions are in some way reflected in their tendency to coordinate their power with one another. Put simply, it is examined whether, under special experimental conditions, both factors are strong or only one of them is. If the former scenario is more frequent, the question arises of which factors have the potential to form these “coherent” pairs. In this way, the tendencies observed in the graphical analysis of the directions of factor action are supplemented, as it were. 10 Figure 4. Factor coordination ([ɑː / ɛ] spectral space) This type of graphical space is fairly extensively filled (the dark zone covers a large radar area). The higher the coordination percentage, the larger the area covered. This should suggest that the spectral space / articulatory mobility of the low back vowel is quite conducive to coordinating factor effects. It becomes evident that the highest hierarchical position of the consonant-environment factor is accompanied by the highest coordination score (the overall mean reaches 77 percent). This suggests that, at least in these cases, the influence of the linguistic factor on F1 and F2 distribution depends on its potential to interact (to operate in the same directions of differentiation) with other factors. The most pronounced opposition in this respect is formed by the speaker factor, since its coordination is at the lowest level (54 percent). This divide should not be surprising, since these two factors differ in nature: consonant articulation is directly influenced essentially by the linguistic system (an objective criterion), whereas individual pronunciation is determined by the speaker’s choice and by various linguistic circumstances (a subjective criterion). The latter graphical data somewhat rehabilitate the positions of phrase stress (64 percent) and phonological quantity (69 percent). Despite the fact that these factors cannot equal the articulatory type of consonants and the speakers in absolute power, their potential to adapt to other factors somewhat increases their comparative weight. The picture of consonant voicing (64 percent) can be interpreted very similarly. As a certain point of reference, it may be indicated that the least coordinated factors in this context are voicing and speakers (39 percent), while the most coordinated are voicing and consonants (83 percent). 11 4. The effect of consonant voicing on the distribution of the formants of the high front [iː / ɪ] A completely different picture emerges when the spectral realisation of high front vowels and the relationships among the factors directly affecting it are viewed from the same perspectives. Without carefully grouping the data, there is no possibility (p = 0,16) of determining whether the plosive consonants adjacent to these vowels are voiced or voiceless. Naturally, this is the overall data set, in which a lower-degree effect on F1–F2 distribution may not become visible (overall means: [iː / ɪ] among voiceless consonants—F1 = 360 Hz, F2 = 2133 Hz; [iː / ɪ] among voiced consonants F1 = 359 Hz, F2 = 2104 Hz). Nevertheless, the same standards were applied in the analysis of the low back vowel, and there the data differentiation was entirely different (see Figure 1). This is probably the first warning signal that, when vowels shift to the front row and reach high height, the voicing factor essentially loses its power. Such a conclusion would be somewhat unexpected, since the effect of this factor recorded by foreign researchers (lowering of F1) applies specifically to front vowels (Hillenbrand et al. 2001). Figure 5. Distribution of the first two formants of high front [iː / ɪ] according to the consonant-voicing factor (p = 0,16)⁴ The scales in the graphical plane intended to illustrate the potential directions of factor action suggest the same thing (see Figure 6). The phonological-quantity and speaker factors are clearly dominant, standing out sharply from the general context both through the highest statistical probability and through graphical distance. In terms of vector direction, the consonant-environment ⁴ Among voiceless consonants—the red ellipse; among voiced consonants—the blue one (overall data set). 12 factor again diverges somewhat from the others, this time producing differences only in the second formant because of (non-)velar articulation (D2, pre-stress position, short vowels, environment of voiced consonants—means: b[ɫ]b F1 = 415 Hz, F2 = 1654 Hz; d[ʲ]d F1 = 419 Hz, F2 = 1701 Hz; g[ʲ]g F1 = 398 Hz, F2 = 2041 Hz; p = 4E−38). The spectral variation implied by all the other factors essentially coincides with the (de)centralisation axis. The acoustic difference between long and short vowels raises the fewest doubts in this respect. Interestingly, the highest statistical indicator was again recorded in the weak phrase position (D2, pre-stress position—means: d[ʲ]d F1 = 419 Hz, F2 = 1701 Hz; d[ʲː]d 302 Hz, F2 = 2420 Hz; p = 1,03E−61). This factual circumstance repeats the previously established weak hierarchical positions of phrase stress in terms of its effect on the formant structure of the vowels in this group (Švageris 2024). The speaker factor is not far behind (post-stress position, short vowels, sequences with voiced labial consonants b[ɫ]b—means D1 F1 = 348 Hz, F2 = 2048 Hz; D2 F1 = 451 Hz, F2 = 1596 Hz; p = 5E−44). Articulatorily, vowels are pronounced more weakly among voiced (D2, stressed position, short vowels, g[ʲ]g—means: F1 = 463 Hz, F2 = 1731 Hz), rather than voiceless consonants (respectively—F1 = 448 Hz, F2 = 1831 Hz, p = 7E−08), and, naturally, vowels in the weak phrase position (D2, short vowels, p[ʲ]p—means: pre-stress F1 = 413 Hz, F2 = 1695 Hz, stressed F1 = 422 Hz, F2 = 1833 Hz, post-stress F1 = 446 Hz, F2 = 1558 Hz; p = 6E−22). Among the details observed, one may also point to point positions that have exchanged places relative to the neutral vowel according to the consonant-voicing criterion: in the [α / ɐ] spectral plane, the vowels closer to schwa were those occurring among voiceless consonants, whereas in this case the reverse is true. How regular this is can be assessed objectively in the next paragraph, but for now this change itself is indeed a sign of the voicing factor’s lack of independence. Figure 6. Potential directions of factor action (in the [iː / ɪ] spectral space) 13 The results of the direct factor-competition analysis provide even more support for all these ideas (see Figure 7). There is no doubt that, in the general case, the long variants of the vowels in this subgroup can be distinguished from the short ones, with the highest statistical probability, on the basis of the distribution of the first two formants in the spectrum. The highest hierarchical position of phonological quantity is more than evident (95 percent). The overall means change little either (without the voicing factor—long [iː] F1 = 325 Hz, F2 = 2318 Hz; short [ɪ] F1 = 393 Hz, F2 = 1958 Hz; with the voicing factor—long [iː] F1 = 320 Hz, F2 = 2307 Hz; short [ɪ] F1 = 400 Hz, F2 = 1935 Hz). Including the consonant-voicing factor in the analysis appears to function as a catalyst. Although it has no substantial power (11 percent), this factor creates additional space for the remaining factors to become stronger (place of consonant articulation 31 percent → 40 percent, speakers 47 percent → 65 percent, phrase stress 6 percent → 29 percent). Comparing the hierarchy shown in this diagram with that established in the previous stage of the study (naturally, without the voicing factor; Švageris 2024), it can be stated that the two replicate one another. The factors differentiating the qualitative spectrum of low back vowels formed two more pronounced blocks according to the power index: the articulatory type of plosive consonants and all the others, whereas here the hierarchy is clearer and more consistent. In other words, there is a sequence of gradually weakening factors. This is a strong argument for reiterating the correlations formulated earlier: as articulatory tension increases in the direction of fronting / raising, the statistical weight of qualitative differentiation according to phonological length and speakers increases (the more tense the vowel, the more favourable the conditions for individual articulation). It was also stated that high vowel height is a serious obstacle to reflecting the positions of the sounds under study with respect to phrase stress through spectral characteristics. As the data show, neither of these tendencies has essentially changed (the power of phrase stress under direct competition reaches only 29 percent). It is difficult to say why the position of the voicing factor is so weak. For now, it would suggest that the influence of consonant voicing on F1 and F2 distribution is inversely proportional to vowel frontness: as the vowel moves forward, it decreases. It is obvious that in many cases the statistical weight of the differences is determined by the second formant—the resonance of the oral cavity. When the F2 differences are so large, the much smaller F1 variations caused by the opposition of voicelessness and voicing in the glottal region may simply no longer be detectable by statistical means. On the other hand, this should be an objective sign that, from the perspective of the whole, they are insignificant. Thus, bearing all this in mind, the following hierarchical chain can be constructed: phonological quantity → speakers → place of consonant articulation → phrase stress → voicing. The diagram illustrating factor coordination also reveals important aspects. First of all, compared with the corresponding graphical material for the [ɑː / ɛ] vowel (see Figure 4), it can be seen that the grey zone covers a considerably smaller radar area. This means that the articulation of [iː / ɪ] is less favourable for coordinating factors. The voicing factor fits in with the others least successfully (its mean is 36 percent). If it was previously assumed that a low coordination percentage could express high factor power, this time the opposite conclusion must be drawn. Combining the results of the direct factor-competition analysis with the data in this diagram into a single logical chain, 14 it must once again be stated that the other factors practically leave no room for differentiated spectral realisation of [iː / i] according to the criterion of consonant voicelessness and voicing. Apparently, this can only be a matter of individual cases and positions, in no way determining the overall distribution of the spectral domains. Figure 7. Relative factor power under conditions of direct competition ([iː / i] F1 and F2 differentiation) Figure 8. Factor coordination ([iː / i] spectral space) 15 In addition, it should be concluded that factor coordination and its actual power nevertheless do not correlate directly. To understand this relationship better, it is necessary to examine the articulatory and spectral dynamics of vowels in still greater depth (in other words, to move to the level of microarticulation). Plosive consonants constitute the smallest obstacle to other factors producing F1 and F2 differences. The most natural combination in this respect is that of consonants and phrase stress (the coordination of this pair is 75 percent), whereas it is most difficult to distinguish velar plosives from non-velars without grouping the data according to speakers (55 percent). Indeed, the overall coordination indices of the phonological-quantity and speaker factors occupying the highest hierarchical positions are similar (46 percent and 40 percent, respectively). Phrase stress occupies an intermediate position according to this indicator (52 percent). 5. The effect of consonant voicing on the distribution of the formants of the high back [uː / ʊ] Finally, the spectral characteristics of high back [uː / ʊ], in terms of their relations with the voicing–voicelessness of plosive consonants, appear to occupy an intermediate position between the two vowels discussed earlier. The division that separates the overall data set into two groups according to this criterion has a tangible statistical basis (p = 1,95E−07), but the means (among voiceless consonants—F1 = 372, F2 = 885 Hz; among voiced consonants—F1 = 387, F2 = 1019 Hz) reveal no articulatory connection with the glottal region (F1 is essentially stable). The statistical weight of the difference is reflected in the second formant (a difference of ~ 100 Hz), implying configurational dynamics in the front part of the oral cavity. The overall errors clearly show that voiced consonants (as in the case of low back vowels) become an additional stimulus for freer tongue movements along the horizontal axis. Moreover, if the overlap of the ellipses in three-dimensional space is not merely accidental or a consequence of the inhomogeneity of the empirical data (although the large quantity of data should theoretically reduce this probability), the conclusion should be the opposite of that suggested by the theoretical discourse: the vowels in this group may more often reach lower F1 values next to voiceless (rather than voiced!) plosive consonants. Thus, although a result favourable to differentiated data realisation was recorded, its interpretation is currently rather problematic. The directions of factor action in the [uː / ʊ] spectral space are perhaps the most confused. It is no longer sufficient to rely solely on the (de)centralisation vector, and a much more multidimensional (more precisely, multidirectional) distribution of the data must be acknowledged. Since the scales of both axes have identical intervals (“steps”) (every 100 Hz), it is easy to see that the positional vowel variants differ more with respect to the ordinates (F2). In terms of vector magnitude and statistical weight, the factors of consonant articulation place and phonological quantity stand out from the others. Their directions of action, at least on the basis of this graphical material, clearly do not coincide. The vector of the former illustrates a phonetic universal, since it was observed by foreign researchers—a marked raising of F2 adjacent to plosive anterior dental consonants. The greatest contrast of this kind (p = 1E−60) was recorded in the front position among voiced consonants (D2, short vowels—means: g[ʊ]g F1 = 397 Hz, F2 = 1022 H; b[ʊ]b F1 = 413, F2 = 943 Hz; d[ʊ]d F1 = 403 Hz, F2 = 1394 Hz). 16 Figure 9. Distribution of the first two formants of high back [u / ɔ] according to the consonant-voicing factor (p = 1,95E-07)⁵ Figure 10. Potential directions of factor action (in the [u: / ɔ] spectral space) ⁵ Among voiceless consonants—the red ellipse; among voiced consonants—the blue one (overall data set). 17 Phonological quantity rather regularly shifts the short variants of these vowels closer to the epicentre of the spectrum (D2, pre-stress position, environment of voiced dental plosives—means: d[ʉ̯]d F1 = 403 Hz, F2 = 1394 Hz; d[u̟]d F1 = 329 Hz, F2 = 712 Hz; p = 4E–45). It is rather unusual that the phrase-stress factor does not follow the same vector and, apparently, yields to the influence of consonants (D2, short vowels, environment of voiced dental plosives—means: pre-stress F1 = 403 Hz, F2 = 1394 Hz; stressed F1 = 459 Hz, F2 = 1293 Hz; post-stress F1 = 431 Hz, F2 = 1290 Hz; p = 3E–20). The differences in individual articulation acquire features of a tendency. Comparing the corresponding vectors of all three vowels, one can observe that D1 generally tends to centralise low vowels somewhat, whereas D2 does so with high vowels. This latter regularity is also illustrated by the graph in this section (post-stress position, short vowels, environment of voiceless labial plosives—means: D1 F1 = 370 Hz, F2 = 671 Hz; D2 F1 = 399 Hz, F2 = 883 Hz; p = 5E–19). Meanwhile, the irregular spectral shift according to the voicing criterion probably indicates the factor’s lack of independence. The expected lowering of F1 was recorded in vowels adjacent to voiceless consonants, while the statistical significance of the difference itself was essentially determined by a marked raising of F2 (D1, post-stress position, short vowels—means: t[ʉ̯]t F1 = 345 Hz, F2 = 973 Hz; d[ʉ̯]d F1 = 372 Hz, F2 = 1268 Hz). It is worth noting that dental consonants figured in the most pronounced oppositions (i.e. those with the greatest statistical significance) according to the factors of phrase stress, voicing, and quantity; therefore, the influence of consonants on the vector direction itself cannot be ruled out. It should not be forgotten that all these directions of factor action are merely indicative—an illustration of the maximum effect of these factors on the formant structure of vowel [uː / ʊ]. The factors are constantly linked by hierarchical relationships, and therefore the vectors emerging in this graph are probably strongly adjusted by their interactional mechanism (in principle, they cannot operate in isolation, independently of one another). Figure 11. Relative factor power under conditions of direct competition ([uː / ʊ] F1 and F2 differentiation) 18 When directly competing with one another, all the factors under examination position themselves quite clearly in relation to one another (see Figure 11). At first glance, it might seem that the scenario observed in the corresponding [iː / i] section is repeating itself—that is, by adding the voicing factor to the set of components, it does not acquire independent status or tangible power, but expands the articulatory-spectral space for other factors to act with greater force. Nevertheless, its not being in the very lowest hierarchical position does not allow this conclusion. According to this criterion, the weakest factor is phrase stress, which contributed least to the distribution of F1 and F2 even when only four factors were analysed (the change is small, 18 percent → 20 percent; Švageris 2024). The power of the phonetic universal (dental-consonant environment → raising of F2 in [uː / ʊ]) remains unquestioned (75 percent), and it makes no difference whether consonant voicing appears in this dynamic plane (overall means—velar-consonant environment—F1 = 376 Hz, F2 = 803 Hz; labial—F1 = 383 Hz, F2 = 756 Hz; dental—F1 = 375 Hz, F2 = 1012 Hz; p = 2,63E-83) or not (respectively, velar-consonant environment—F1 = 373 Hz, F2 = 770 Hz; labial—F1 = 371 Hz, F2 = 729 Hz; dental—F1 = 363 Hz, F2 = 956 Hz; p = 5,05E-51). At least the mean values indicate that the new factor became an additional stimulus strengthening the fronting effect of high back vowels among dental consonants. Compared with the data from the previous study, the position of the quantity factor becomes somewhat stronger (57 percent → 71 percent). This change makes it possible to assert with still greater justification that the influence of phonological length on the distribution of the first two formants correlates with the shift of vowels towards the front row and high height. It may be thought that, as F2 rose even further among voiced dental consonants (in the general case 956 Hz → 1012 Hz), the space for the spectral differentiation of these vowels according to quantity expanded. The speakers’ position remains stable. Although, naturally, this factor cannot equal either vowel length or the articulatory type of plosive consonants in power of effect, their combined influence on the qualitative spectrum remains sufficiently significant (46 percent). The most complicated assessment in this context concerns the consonant-voicing index. All previous indications would suggest that this relatively high score (35 percent) has no more rational basis. Apparently, as already mentioned, resolving this question requires an even more detailed analysis of microarticulatory phenomena. At present, it can only be said approximately that, perhaps by acting in the same direction as the other factors (primarily the consonant factor and phonological quantity), this factor retains certain positions, and its influence correlates with the backness of vowels—the farther one moves from the front row, the more pronounced the effect of voicing–voicelessness on F1 and F2 distribution becomes. It remains to add that the hierarchical arrangement of all these factors is as follows: place of consonant articulation → phonological quantity → speakers → voicing → phrase stress. That the qualitative spectral dynamics of high back vowels are the most difficult to analyse is well illustrated by the factor coordination chart (see Chart 12). There is no doubt that the [uː / ʊ] spectrum and articulation constitute the most problematic environment for coordinating the factors. Only in very rare cases does the coordination percentage exceed the 50 percent threshold. The consonant and speaker factors (67 percent) and the consonant and quantity factors (58 percent) alone possess more noticeable potential for coordinating their power. All other pairs remain below the fifty-percent threshold. For this reason, the graphic space appears very empty, while the dark zone covers only a small part of the radar chart. If we were to rely on the overall averages, then according to the coordination criterion the factors are arranged in the following sequence: place of consonant articulation (51 percent) → speaker (36 percent) → voicing (30 percent) → quantity (28 percent) → phrase stress (24 percent). It is interesting that there were cases in which coordination was impossible in principle (the interaction between quantity and phrase stress is equal to zero in this respect). Apparently, the articulatory specificity of this type of vowel destabilises interactional relations, selectively strengthening the power of individual factors. It is true that, from both theoretical and practical perspectives, the recurring correlation between the relative power of the consonant factor and vowel backness is more relevant (the further back the vowel, the more distinctly the formant structure differentiates the articulatory type of the adjacent plosive consonants), and the latest data confirm it even more clearly. Nevertheless, in the future all the observed tendencies should be examined in greater depth. Chart 12. Factor coordination ([uː / ʊ] spectral space) CONCLUSIONS 1. The research material analysed according to the statistical weight of the impact exerted on the vowel formant structure makes it possible to establish the following hierarchical factor chains (from strongest to weakest): [ɑː / e] – place of consonant articulation → speaker → phrase stress → voicing → phonological quantity; [iː / ɪ] – phonological quantity → speaker → place of consonant articulation → phrase stress → voicing; [uː / ʊ] – place of consonant articulation → phonological quantity → speaker → voicing → phrase stress. 20 2. The inclusion of the new voicing factor in the analysis did not substantially alter the correlations established during the earlier stages of the study. The influence of plosive consonant type on F1 and F2 distribution remained correlated with vowel backness (the further back the vowel, the greater this influence), that of phonological quantity with vowel backness and height (the closer the articulation approaches the [i] type, the more pronounced the spectral differences between short and long vowels), and that of phrase stress with vowel openness (the more open the vowel, the greater the influence of the prosodic factor). The relative power of the consonant voicing factor itself seems to depend on the articulatory shift of vowels towards the back series (the more pronounced the shift, the more noticeable the formant differences determined by this factor). 3. The weak positions of the voicing factor in the hierarchical chains imply a low level of autonomy. This could be one of the reasons for the rather irregular differences in F1 and F2 distribution according to this factor. The material of Standard Lithuanian does not confirm the phonetic regularity attested by data from other languages—that F1 in front vowels lowers when they are surrounded by voiced consonants. This hypothesis is supported only in part by the measurements of low back vowels. There is more reason to believe that, by supplementing the data set without acquiring a more significant influence, the voicing factor expands the articulatory-spectral space in which competing factors can act with greater force, for example, highlighting the fronting of high back vowels in positions between dental plosive consonants. It is true that the possibility cannot be excluded that microarticulatory vowel differences associated with the factor under discussion exist but are obscured by the excessively strong influence of other factors on the qualitative spectrum. 4. Additional analysis of factor coordination revealed that the most favourable environment for coordinating the effects of all of them is the spectral space of the low back vowel [ɑ]. In other words, the more open the vowel, the more readily the effects of the factors shaping its formant structure can be coordinated. Among other things, this should mean that the space of F1 and F2 variation in open vowels is the broadest. LITERATŪRA Adriaans Frans 2018: Effects of consonantal context on the learnability of vowel categories from infant-directed speech. – The Journal of the Acoustical Society of America 144(1), 20–25. DOI: https://doi.org/10.1121/1.5045192. Ambrazevičius Rytis, Leskauskaitė Asta 2014: Priebalsių akustinės ypatybės: palatalizacija ir balsingumas [Acoustic Properties of Consonants: Palatalization and Voicing], Kaunas: Technologija. Chiba Tsutomu, Kajiyama Masato 1941: The Vowel: Its Nature and Structure, Tokyo: Kaiseikan. Čeirane Solveiga, Indričāne Inese 2012: Latviešu valodas troksneņu raksturojums pēc lokusa vienādojumiem [Characterization of Latvian Language Sounds Using Locus Equations]. – Baltistica 47(1), 37–50. DOI: http://dx.doi.org/10.15388/baltistica.47.1.2132. 21 Hillenbrand James, Clark Michael, Nearey Terrance 2001: Effects of consonant environment on vowel formant patterns. – The Journal of the Acoustical Society of America 109(2), 748–763. DOI: https://doi.org/10.1121/1.1337959. Jaroslavienė Jurgita 2014: Spectral Characteristics of the Lithuanian Vowels: Some Preliminary Results of a New Experimental Research. – Linguistica Lettica 22, 68–84. Jaroslavienė Jurgita 2015: Lietuvių kalbos trumpųjų ir ilgųjų balsių kiekybės ir kokybės etalonai [Standards of Quantity and Quality of Short and Long Vowels in the Lithuanian Language]. – Bendrinė kalba 88, [1–17]. Jaroslavienė Jurgita, Grigorjevs Juris, Urbanavičienė Jolita, Indričāne Inese 2019: Baltų kalbų garsynas XXI a. pradžioje: balsių ir garsų sąveikos instrumentinis tyrimas [Sounds of the Baltic Languages in the Early 21st Century: An Experimental Study of Vowels and Coarticulation], Vilnius: Lietuvių kalbos institutas. DOI: https://doi.org/10.35321/e-pub.1.baltu-garsynas. Johnson Keith, Ladefoged Peter, Lindau Mona 1993: Individual difference in vowel production. – The Journal of Acoustical Society of America 94(2), 701–714. Kaukėnienė Lidija 2002: Lietuvių bendrinės kalbos nekirčiuotų balsių spektrai [The Spectra of Unstressed Vowels in Standard Lithuanian]. – Kalbotyra 51(1), 41–59. Kazlauskienė Asta 2018: Bendrinės lietuvių kalbos fonetikos ir fonologijos pagrindai [Basics of General Phonetics and Phonology of the Lithuanian Language], Kaunas: Vytauto Didžiojo universitetas Kent Raymond, Read Charles 2002: The Acoustic analysis of speech, Albany: Thomson Learning. Kewley-Port Diane 1982: Measurement of formant transitions in naturally produced stop consonant–vowel syllables. – The Journal of the Acoustical Society of America 72(2), 379–389. Kudirka Robertas 2009: Lietuvių bendrinės kalbos kirčiuotų ilgųjų ir trumpųjų balsių formantinė struktūra [Formant Structure of the Accented Long and Short Vowels in Lithuanian Standard Language]. – Respectus Philologicus 16(21), 141–152. Ledichova Edita 2020a: Labializuotų bendrinės lietuvių kalbos balsių eksperimentinis tyrimas [The Experimental Research of Qualitative Characteristics of Labialised Vowels in Standard Lithuanian]. – Bendrinė kalba 93, [1–33]. Ledichova Edita 2020b: Lietuvių bendrinės kalbos vokalizmo variantai: eksperimentinis kokybės ypatybių tyrimas: daktaro disertacija, Kaunas: Vytauto Didžiojo universitetas, Lietuvių kalbos institutas. Lee Yoonjeong, Keating Patricia, Kreiman Jody 2019: Acoustic voice variation within and between speakers. – The Journal of Acoustical Society of America 146(3), 1568–1579. Lindblom Björn 1990: Models of phonetic variation and selection. – Perilus 11, 65–100. 22 Lindblom Björn, Moon Seung-Jae 1994: Interaction between duration, context, and speaking style in English stressed vowels. – The Journal of the Acoustical Society of America 96(1), 40–55. Miliūnaitė Rita 2018: Kalbos normos ir jų savireguliacija interneto bendruomenėje [Language Norms and Their Self-regulation in the Internet Community], Vilnius: Lietuvių kalbos institutas. Pakerys Antanas 1982: Lietuvių bendrinės kalbos prozodija [Prosody of Lithuanian Standard Language], Vilnius: Mokslas. Pakerys Antanas 2003 [1986]: Lietuvių bendrinės kalbos fonetika [Phonetics of the Lithuanian Common Language], 3-iasis patais. ir papild. leidimas, Vilnius: Enciklopedija. Pickett James M. 1999: The Acoustics of Speech Communication. Fundamentals, Speech Perception Theory, and Technology, Needham Heights: Allyn & Bacon. Rietveld Toni, van Hout Roland 2005: Statistics in Language Research: Analysis of Variance, Berlin, New York: Mouton de Gruyter. Roesler Lindsay, Yung Song Jae 2018: Acoustic characteristics of tense and lax vowels across sentence position in clear speech. – The Journal of the Acoustical Society of America 144(6), 535–540. Stevens Kenneth N. 1998: Acoustic Phonetics, Cambridge, MA: The MIT Press. Stevens Kenneth N., House Arthur 1963: Perturbation of vowel articulations by consonantal context: An acoustical study. – Journal of Speech and Hearing Research 6(2), 111–128. DOI: https://doi.org/10.1044/jshr.0602.111. Švageris Evaldas 2024: Formantinę balsių struktūrą lemiančių veiksnių ryšiai: žvalgomasis bendrinės lietuvių kalbos ilgųjų ir trumpųjų balsių tyrimas [Interaction Between Factors Determining the Formant Structure of Long and Short Vowels in Standard Lithuanian: A Pilot Study]. – Kalbotyra 77, 61–84. DOI: https://doi.org/10.15388/Kalbotyra.2024.77.3. Švageris Evaldas, Urbanavičienė Jolita 2024: Lietuvių kalbos balsių formantinę struktūrą ir trukmę lemiančių veiksnių sąveika bei hierarchija [Phonetic Distribution of Vowel Quality and Quantity in Standard Lithuanian: A Factorial Analysis], Vilnius: Lietuvių kalbos institutas. DOI: https://doi.org/10.35321/e-pub.109.balsiu-veiksniai-hierarchija. Titze Ingo R. 2000: Principles of Voice production, Iowa City: National Center for Voice and Speech. Tsukada Kimiko 2009: An acoustic comparison of vowel length contrasts in Arabic, Japanese and Thai: Durational and spectral data. – International Journal of Asian Language Processing 19(4), 127–138. Urbanavičienė Jolita, Indričianė Inese, Jaroslavienė Jurgita, Grigorjevs Juris 2019: Baltų kalbų garsynas XXI a. pradžioje 2: Priebalsių instrumentinis tyrimas [Sounds of the Baltic 23 Languages in the Early 21st Century 2: An Experimental Study of Consonants], Vilnius: Lietuvių kalbos institutas. DOI: https://doi.org/10.35321/e-pub.2.baltu-garsynas. Vaitkevičiūtė Valerija 1961: Lietuvių literatūrinės kalbos balsinės ir dvibalsinės fonemos [Vowel and Diphthong Phonemes of the Lithuanian Literary Language]. – Lietuvių kalbotyros klausimai 4, 19–46. Effects of Consonant Voicing on Vowel Formant Distribution in Standard Lithuanian: A Factorial Analysis SUMMARY It is well known that from the phonetic point of view, any speech sound is a matter of great dynamicity. The vowel system in a language is no exception. In more general terms, the spectral characteristics of the vowel depend on a great number of both linguistic and extralinguistic factors. This is the reason why a factorial analysis must be conducted to gain a better understanding of the dynamicity of the speech. The first attempt to analyse the dynamics of the phonetic vowel realisation in Standard Lithuanian has yielded positive results. Although for some reason, far-reaching conclusions cannot be drawn yet, it can nevertheless be observed that the dynamic plane of quantitative and qualitative vowel features is not fully chaotic. It can be seen that all related factors are linked by hierarchical relationships. This study aimed to determine what position in the hierarchy of factors (i.e. between consonantal environment, phonological length, prosodic focus and speaker) directly influences formant distribution in the spectra of Lithuanian vowels (so-called corner vowels [iː / ɪ], [uː / ʊ] and [ɑː / ɐ]) is occupied by the consonant voicing factor. A factorial analysis (two-way ANOVA) was performed for this purpose. Results revealed quite weak positions of the voicing factor in the hierarchy and its low-level autonomy. Its relative power on F1–F2 distribution correlates with the backness of the vowel, as it shifts towards the back position of the tongue, differences implied by the consonant voicing factor on formants can be noticed more reliably. Although having no significant impact, the consonant voicing plays some role in the expansion of the spectral space for other factors to act with greater power (for example, it creates conditions for high back vowels to be more fronted in a dental stop environment). There may indeed be microarticulatory differences related to the factor in question, but quite obviously, their detection is hindered by the excessive influence of co-acting factors. Further analysis is to be conducted to get this dynamic picture in more detail. 24