3.1 Research questions and hypotheses

The aim of the current research is to conduct an acoustic study of rhythmic stress in Ukrainian. As very little is known about Ukrainian stress, the present research is exploratory and the precursor of a much broader study of the Ukrainian metrical system. To answer the current research questions (formulated in (5) below), we look at stress patterns of words with the main stress removed from the left edge of the word by three, four and five syllables: $\unicode[STIX]{x1D70E}\unicode[STIX]{x1D70E}\unicode[STIX]{x1D70E}$ ˈ $\unicode[STIX]{x1D70E}$ ( $\unicode[STIX]{x1D70E}_{\text{n}}$ ), $\unicode[STIX]{x1D70E}\unicode[STIX]{x1D70E}\unicode[STIX]{x1D70E}\unicode[STIX]{x1D70E}$ ˈ $\unicode[STIX]{x1D70E}$ ( $\unicode[STIX]{x1D70E}_{\text{n}}$ ), $\unicode[STIX]{x1D70E}\unicode[STIX]{x1D70E}\unicode[STIX]{x1D70E}\unicode[STIX]{x1D70E}\unicode[STIX]{x1D70E}$ ˈ $\unicode[STIX]{x1D70E}$ ( $\unicode[STIX]{x1D70E}_{\text{n}}$ ).Footnote ^[5]

As mentioned above, standard descriptions of secondary/tertiary prominence have not been confirmed by acoustic evidence. Moreover, Brovčenko (Reference Brovčenko1969: 14–16) explicitly denies its presence in Ukrainian. Therefore, our main objective is to find out whether there is secondary/tertiary stress in the first place. If the standard descriptions of Ukrainian stress are correct, secondary stress is expected to surface on the initial syllable in all the data under investigation, e.g. ˌ $\boldsymbol{\unicode[STIX]{x1D70E}}\unicode[STIX]{x1D70E}\unicode[STIX]{x1D70E}$ ˈ $\unicode[STIX]{x1D70E}$ ( $\unicode[STIX]{x1D70E}_{\text{n}}$ ). Another conceivable scenario would be to stress the second syllable in even-parity words: $\unicode[STIX]{x1D70E}$ ˌ $\boldsymbol{\unicode[STIX]{x1D70E}}\unicode[STIX]{x1D70E}$ ˈ $\unicode[STIX]{x1D70E}$ ( $\unicode[STIX]{x1D70E}_{\text{n}}$ ).Footnote ^[6]

An additional (tertiary) stress intervening between the main stress and the initial stress should appear in words with the main stress removed from the left edge of the word by four syllables: ˌ $\unicode[STIX]{x1D70E}\unicode[STIX]{x1D70E}$ ˌ $\boldsymbol{\unicode[STIX]{x1D70E}}\unicode[STIX]{x1D70E}$ ˈ $\unicode[STIX]{x1D70E}$ ( $\unicode[STIX]{x1D70E}_{\text{n}}$ ). Standard grammars do not specify the position of tertiary stress in words with the main stress removed from the left edge of the word by five syllables: $\unicode[STIX]{x1D70E}\unicode[STIX]{x1D70E}\unicode[STIX]{x1D70E}\unicode[STIX]{x1D70E}\unicode[STIX]{x1D70E}$ ˈ $\unicode[STIX]{x1D70E}$ ( $\unicode[STIX]{x1D70E}_{\text{n}}$ ). Assuming that secondary stress falls on the initial syllable, either the third or the fourth syllable counting from the left can potentially bear tertiary stress: ˌ $\unicode[STIX]{x1D70E}\unicode[STIX]{x1D70E}$ ˌ $\boldsymbol{\unicode[STIX]{x1D70E}}\unicode[STIX]{x1D70E}\unicode[STIX]{x1D70E}$ ˈ $\unicode[STIX]{x1D70E}$ ( $\unicode[STIX]{x1D70E}_{\text{n}}$ ) or ˌ $\unicode[STIX]{x1D70E}\unicode[STIX]{x1D70E}\unicode[STIX]{x1D70E}$ ˌ $\boldsymbol{\unicode[STIX]{x1D70E}}\unicode[STIX]{x1D70E}$ ˈ $\unicode[STIX]{x1D70E}$ ( $\unicode[STIX]{x1D70E}_{\text{n}}$ ). Interestingly, only the former pattern has been attested cross-linguistically, and languages with a lapse not adjacent to the primary stress are predicted not to exist (Kager Reference Kager2001, Alber Reference Alber2005, among others).

In sum, the current research aims to provide answers to the following questions:

3.2 Participants

The data were collected from 16 monolingual native speakers of Ukrainian (four male, 12 female). All of the participants were born and had continuously lived in the Drohobyč region, in western Ukraine. Given the dialectal diversity of modern Ukrainian, the choice of a single region enhances the chance of detecting a uniform pattern. Although most of the participants declared some knowledge of foreign languages, predominantly Russian, Polish, and English, Ukrainian was the language of their everyday use. Therefore, we do not expect any significant interference from other languages. The age of the participants ranged from 18 to 67 years (M = 33.4 years, SD = 13.6). The participants had different educational backgrounds – nine had a Masters degree, four had a high-school degree, and three were college students. All of them were naı̈ve about the purpose of the study.

3.3 Stimuli

The data set consisted of 28 single-root words with primary stress either on the odd-numbered syllable or on the even-numbered syllable; see Appendix A. The lexical stress in the words chosen for the experiment fell on the antepenultimate, penultimate or final syllable. It was separated from the left edge by minimally three syllables and maximally five syllables. Overall, three prosodic types were distinguished, as shown in Table 1.Footnote ^[7]

Table 1

Division of tokens.

In all the three types, secondary stress is expected to appear on the first syllable. Type 1 and type 3 are predicted to have a lapse on the pretonic syllable, and types 2 and 3, to have an additional secondary stress on the third syllable. Only words with the rightward iteration of the rhythmic stress are included in the present study.Footnote ^[8]

Care was taken to select words in accordance with criteria minimising the risk of biased results. First, words with consonantal clusters in the second or third syllable were discarded. Second, the distribution of high and non-high vowels in the second versus the third syllable did not show bias in favour of the iterative stress hypothesis.Footnote ^[9]

3.4 Experimental procedure

Recordings were produced in Ukraine, using an H4 Zoom portable digital audio recorder, set at a sampling rate of 44.1 kHz, and an AT897 microphone. Participants read the list of 28 words, consisting of single-root nouns, adjectives and verbs. The words were presented as a list of isolated words printed in Arial typeface, 13 pt letter size. Participants were asked to embed the tokens within a carrier sentence Vin skazav …druhyj raz ‘He said …for the second time’, which was printed at the top of the page. The critical item appeared in a position where it did not carry sentence stress and was not subject to sentence-final lengthening. Primary stress was marked orthographically to avoid hesitations on the speakers’ part.Footnote ^[10] The participants were asked to read the sentences at a natural pace, taking breaks when needed.

3.5 Data analysis

The data were analysed using Sound Forge and Praat (Boersma & Weenink Reference Boersma and Weenink1992–2017). Segmentation was done manually by the authors based on the inspection of the waveforms and the spectrograms as well as auditory perception. The syllable boundaries were marked at vowel–consonant junctures. In the tokens containing sonorant–obstruent clusters, the sonorant was syllabified into the coda of the first syllable. There were five such words in the data set: ver.sy.fi.ka.ci.ja ‘versification (noun)’, ver.sy.fi.ka.cij.nyj ‘versification (adj)’, kon.fe.de.ra.ci.ja ‘confederation’, kon.fe.de.ra.cij.nyj ‘confederate’, and or.ha.ni.zu.va.ty ‘organise’. Syllable boundaries were aligned with the beginning of the closure phase in stops and the onset of noise in fricatives. Sonorant consonants were identified according to the formant transitions between consonants and preceding vowels. Initial onsetless syllables were extracted on the basis of the formant structure and the dynamics of glottalisation patterns. Out of 448 recorded tokens, 94 tokens were rejected due to mispronunciations and hesitations on the speakers’ part. Rejected tokens were mostly long unfamiliar words of six or seven syllables. Additionally, 170 vowel-initial syllables were excluded from the statistical analyses in order not to underestimate the duration score of the initial syllable. We also discarded a handful of syllables containing clusters or codas. In total, 1410 syllables in six positions were included in the analysis.

The parameters which usually manifest both primary and secondary stress include duration, intensity, and pitch change (Fry Reference Fry1955, Lehiste & Peterson Reference Lehiste and Peterson1959, Peterson & Lehiste Reference Peterson and Lehiste1960, Morton & Jassem Reference Morton and Jassem1965, Lehiste Reference Lehiste1970, Beckman Reference Beckman1986, Cutler Reference Cutler, Pisoni and Remez2005, Plag, Kunter & Schramm Reference Plag, Kunter and Schramm2011). However, a preliminary inspection of the collected data as well as the previous work on primary stress (Brovčenko Reference Brovčenko1969) have indicated that the main cue of stress in Ukrainian is duration. The results reported in an accompanying pilot study by Łukaszewicz & Mołczanow (Reference Łukaszewicz and Mołczanow2016) point to no correlation between increased intensity or f0 values and lexical or subsidiary stress in Ukrainian. The contours of the two parameters obtained for the iterative subset, i.e. ˌ $\unicode[STIX]{x1D70E}\unicode[STIX]{x1D70E}$ ˌ $\unicode[STIX]{x1D70E}\unicode[STIX]{x1D70E}$ ˈ $\unicode[STIX]{x1D70E}$ ( $\unicode[STIX]{x1D70E}_{\text{n}}$ ) words, are presented in Figures 1 and 2. Clearly, the tonic vowel is not characterised by the highest level of intensity or f0. Also, there is no discernible pattern of alternating stresses in the syllables preceding main stress.

Figure 1

Mean relative intensity in ˌ $\unicode[STIX]{x1D70E}\unicode[STIX]{x1D70E}$ ˌ $\unicode[STIX]{x1D70E}\unicode[STIX]{x1D70E}$ ˈ $\unicode[STIX]{x1D70E}$ ( $\unicode[STIX]{x1D70E}_{\text{n}}$ ) words. Relative intensity stands for amplitude standardised in dB (with average amplitude in the word as the reference level).

Figure 2

Mean relative f0 in $\unicode[STIX]{x1D70E}\unicode[STIX]{x1D70E}$ ˌ $\unicode[STIX]{x1D70E}\unicode[STIX]{x1D70E}$ ˈ $\unicode[STIX]{x1D70E}$ ( $\unicode[STIX]{x1D70E}_{\text{n}}$ ) words. Relative f0 stands for f0 standardised in semitones (with average f0 in the word as the reference level).

A recent acoustic study of secondary stress in Polish, which is geographically and historically related to Ukrainian, has revealed that secondary stress is expressed by an increased length of consonants in the onsets of stressed syllables (Łukaszewicz Reference Łukaszewicz2015, Reference Łukaszewiczto appear). As for Ukrainian, there are indications in the literature that vowels have different durations in unstressed syllables depending on their distance from lexical stress (Toc’ka Reference Toc’ka1973). Since it is possible that both vocalic and consonantal length may be affected in positions of secondary/tertiary stress, a decision was made to measure the duration of the whole syllable. Overall, we expect both primary and secondary/tertiary stress to be manifested by a substantial difference in duration of adjacent syllables.

There is no literature on just-noticeable differences in duration (JNDs) specific to Ukrainian. In the analyses below we tentatively assume 10 ms as the minimal threshold because it is generally agreed that, in the range of durations between 30–300 ms, JNDs are between 10–40 ms, although the limit of perceptibility under optimal conditions may be much smaller (Lehiste (Reference Lehiste1970: 13); see also the more recent work of Fletcher (Reference Fletcher, Hardcastle, Laver and Gibbon2010: 526) and the literature cited therein).Footnote ^[11]

4.1 An overall analysis

The overall analysis was based on 1410 syllable tokens. All of the tokens were of the consonant–vowel type. All other syllable types were discarded.

The effect of stress on syllable duration was tested in terms of linear mixed-effects models, fitted in SPSS (version 23). Syllable duration may be affected considerably not only by prosodic position but also by other factors, such as inter-speaker differences in speech rate or word length. In order to control for such factors, the random effects of Speaker and Item were included. The competing models were compared in terms of likelihood ratio tests (e.g. Pinheiro & Bates Reference Pinheiro and Bates2000, Baayen Reference Baayen2008). The best fit was achieved by the linear mixed-effects model with random intercepts for Speaker and Item, and with Position as the fixed-effect term. The comparison of the nested models, using the standard chi-square reference, shows that the effect of stress is extremely significant ( $\unicode[STIX]{x1D712}^{\text{2}}(5)=1282.108$ , $p<.0001$ ). All positions exhibit significantly smaller duration compared to the baseline, i.e. the tonic position carrying lexical stress (see Table 2). The biggest distance is between the second and the tonic positions ( $\unicode[STIX]{x1D6FD}_{\text{2}}=-108.5$ , SE  $=$  2.45, $t=-44.22$ , $p<.0001$ .) The initial position (i.e. the secondary-stress carrying syllable) is the second longest syllable in the word ( $\unicode[STIX]{x1D6FD}_{\text{1}}=-78.9$ , SE  $=$  3.24, $t=-24.33$ , $p<.0001$ ). Table 3 depicts the predicted mean durations for the six positions (see also Figure 3). Post-hoc pairwise comparisons (see Appendix B) confirm that the tonic syllable is significantly longer and the second syllable significantly shorter than all the other positions. On average, the initial syllable is 29.6 ms longer than the second syllable. A significant increase in duration is also reported for the iterative position (the tertiary-stress carrying syllable), which turns out 21 ms longer than the preceding (i.e. second) syllable ( $p<.0001$ ).

Figure 3

Mean syllable duration depending on position. (Error bars represent 95% confidence interval for the mean.)

As illustrated in Figure 3, changes in syllable durations form a wave-like pattern, in which we can discern three peaks – in the initial, iterative and tonic positions. There is no clear pattern of peaks and valleys corresponding to alternating stresses in the two syllables preceding the tonic syllable (fourth and pretonic): both the downward slope in the fourth syllable and the upward slope in the pretonic position turn out statistically insignificant. As the latter is followed by a much bigger and statistically significant rise in the tonic positon, we interpret it as a lengthening effect in the vicinity of lexical stress. Notably, pretonic lengthening recurs regardless of the type of the word, i.e. it is also seen in the noniterative subset. That is why we take it to be unconnected with rhythm and connected with the tonic syllable. In other words, in the close vicinity of the syllable carrying lexical stress, some kind of anticipation of the lexical stress takes place and there is no space for an alternating pattern of smaller peaks and valleys characteristic of rhythm.

4.2 An analysis of segmentally identical syllables

In order to dismiss a possibility that the rise on the iterative syllable could be caused by intrinsic differences in segmental length, another linear mixed-effects model was fitted to compare segmentally identical syllables in different positions. For the purpose of this analysis, we identified three CV syllables (ka, ni, li), each of which recurred in four word-medial positions in the data: second, iterative, fourth, and pretonic;Footnote ^[12] see Appendix C for the list of words. The analysis was based on 180 tokens (12 words). (Each syllable in each position was represented by 15 tokens.) To make the three segmentally different syllables comparable, raw duration values were standardised (z-scores were calculated). The random intercept for Speaker was included to control for individual differences in speech rate.

As in the previous analysis, the effect of Position turned out statistically significant ( $\unicode[STIX]{x1D712}^{\text{2}}(3)=39.853$ , $p<.0001$ ). The pretonic position was longer than all other positions at $\unicode[STIX]{x1D6FC}=.05$ (see Table 4). Crucially, the biggest significant increase in syllable duration (by 0.665 standard deviation; see Appendix D) was found in the iterative position (compare the estimates for the mean standardised duration of the second and iterative syllables in Table 5). In raw duration terms, the iterative syllable was about 16 ms longer than the second syllable ( $p<.0001$ ). The fourth syllable was not statistically different from the iterative position (for post-hoc comparisons see Appendix D). Overall, these findings confirm the pattern obtained in the previous analysis (see also Figure 4).

Table 4

Estimates of fixed effects. Dependent variable: duration ( $z$ -scores).

Table 5

Estimates of means. Dependent variable: duration ( $z$ -scores).

Figure 4

Mean standardised syllable duration depending on position. (Error bars represent 95% confidence interval for the mean.)