2.1 Location of stress in Polish

In this section, I summarise previous analyses of Polish stress, which have characterised it as a bidirectional system with internal lapses. The specific theoretical claims made in these studies will lay the groundwork for the production study presented in later sections.

2.2 Acoustic manifestation of stress in Polish

Although Polish stress has been studied extensively from the theoretical perspective, not much work has been devoted to its acoustics. This section outlines previous work on the perception and production of Polish stress, and underscores the general lack of evidence for lower levels of stress. In this review, I identify the acoustic correlates of stress that will be considered in the production study.

Polish primary stress has been shown to be perceptually very robust. Peperkamp & Dupoux (Reference Peperkamp, Dupoux, Gussenhoven and Warner2002) found that Polish speakers were unique in their ability to accurately identify the placement of primary stress, despite its non-contrastive nature. In the experiment, three stress cues (pitch, duration and intensity) were manipulated simultaneously, which implies that at least one of them correlates with stress in Polish. In fact, all three parameters were identified as stress cues in independent perception studies. Pitch fluctuations were argued to be the central stress cue for Polish listeners in synthesised (Jassem et al. Reference Jassem, Morton, Steffen-Batog and Jassem1968) and natural speech (Awedyk Reference Awedyk1986). In addition, Jassem et al. (Reference Jassem, Morton, Steffen-Batog and Jassem1968) found increased intensity to be an effective cue, but only if the change was 6 db in magnitude or more. Lengthened vowels were also perceived as stressed, even if their duration was increased by as little as 20%.

Production studies have confirmed that these particular acoustic parameters are manipulated to signal stress. Both Jassem (Reference Jassem1962) and Dogil & Williams (Reference Dogil and Williams1999) pointed to fluctuations in pitch as a correlate of primary stress, while Jassem (Reference Jassem1962) also identified duration as having ‘an accessory role’. A more recent study, Łukaszewicz & Rozborski (Reference Łukaszewicz and Rozborski2008), argued that maximum intensity was the most reliable correlate. This finding is consistent with early impressionistic descriptions of stress, which often mentioned a slight rise in loudness on the penultimate syllable (Benni Reference Benni1914, Dłuska Reference Dłuska1950).

Unlike primary stress, lower levels of stress are not perceptually salient. Indeed, native Polish listeners have been found to be remarkably inconsistent in their identification and localisation of secondary stress (Steffen-Batogowa Reference Steffen-Batogowa2000). Steffen-Batogowa (Reference Steffen-Batogowa2000: 104) concludes that Polish secondary stress is perceptually very weak, and ascribes it to its non-contrastive nature. This hypothesis does not, however, account for why primary stress should be so perceptually salient, as it is also non-contrastive.

The difficulty in detecting secondary stress is matched by mixed results from the few production studies that exist. Dogil & Williams (Reference Dogil and Williams1999) failed to find conclusive evidence for the presence of secondary stress. They did, however, suggest that secondary stress on the initial syllable may be marked by a longer and more fully articulated vowel (1999: 286). This suggestion is in line with Dłuska (Reference Dłuska1974) 's impressionistic claim that Polish syllables form an alternating pattern of long and short vowels, mirroring the pattern of rhythmic stresses. In an older study, Dłuska (Reference Dłuska1932) argued that secondary stress correlated with a longer onset consonant. No statistical analysis was provided, however.

To sum up, while Polish primary stress is perceptually and acoustically robust, quite the opposite is true of lower levels of stress. The elusive nature of the latter is troubling, given the amount of attention the Polish stress system has received in the phonological literature. Below, I argue that the acoustic elusiveness of Polish secondary stress is in no sense an isolated phenomenon. Indeed, in a number of unrelated languages, rhythmic stresses posited on the basis of impressionistic evidence have not been confirmed acoustically.

2.3 Acoustic work on secondary stress in other languages

This section summarises previous work on the acoustics of secondary stress. A synthesis of findings from languages other than Polish will justify the design of the present study. In addition, a discussion of a sample of studies which failed to identify correlates of secondary stress will foreground the elusive character of lower degrees of stress.

Acoustic studies of stress have rarely investigated the phonetic manifestation of lower degree stresses, focusing instead on the opposition between primary stress and unstressed syllables. Moreover, among the studies that did analyse secondary stress, only a handful were successful at identifying its acoustic correlates. Although these studies have shown that secondary stress can be cued by a range of parameters, languages commonly seem to employ a fixed set of acoustic cues to signal all levels of stress. The differentiation of stress levels then comes from the strengthening of a particular cue. For example, Gordon (Reference Gordon2004) demonstrated that Chickasaw vowels with secondary stress displayed F0, duration and intensity values that were intermediate between those of primary stress and unstressed vowels. Similarly, in Estonian, the same acoustic cues were independently found to correlate with primary and secondary stress syllables: F0 contours and onset length (Lehiste Reference Lehiste1966, Eek Reference Eek1982, Gordon Reference Gordon1995, Reference Gordon, Lehiste and Ross1997). Estonian stress levels are differentiated by the former characteristic, i.e. syllables with primary stress exhibit F0 rises, while syllables with secondary stress manifest an interruption in the F0 decline (Eek Reference Eek1982, Gordon Reference Gordon1995). Cue trading has also been observed in English compounds appearing in various prosodic contexts (Morrill Adams Reference Morrill Adams, Trouvain and Barry2007).

Another strategy that languages frequently employ in marking lower levels of stress is to use a subset of the correlates associated with primary stress. The presence of a pitch accent on the strongest stress often accounts for this distinction. For example, it has been argued that English primary and secondary stress syllables are identical to each other on such measures as F0, intensity and spectral balance, unless the word is pitch-accented (Plag et al. Reference Plag, Kunter and Schramm2011), in which case the primary stress syllable exhibits a larger pitch slope. Other parameters have also been suggested as differentiating between stress levels in English. Fear et al. (Reference Fear, Cutler and Butterfield1995) found primary and secondary stress syllables to pattern uniformly in terms of intensity, pitch and spectral characteristics, but to differ in terms of vowel duration. Mattys (Reference Mattys2000) characterised syllables with primary stress as manifesting an increase in pitch and duration, but not in intensity, as compared to syllables with secondary stress. Although the results of these studies do not completely converge, they all suggest that secondary stress in English is realised by a subset of correlates systematically associated with primary stress. Similar claims have been put forth for other languages. In a study of two speakers of Indonesian, Adisasmito-Smith & Cohn (Reference Adisasmito-Smith and Cohn1996) argued that they manifested secondary stress differently, one with a rise in amplitude, and the other by lengthening of the vowel. Both of these correlates, together with F0, were also found to be present on syllables carrying primary stress.

The acoustic nature of secondary stress has also been investigated in compounds. For Dutch, van Heuven (Reference Heuven1987) showed that while intensity, duration and pitch values distinguish primary from secondary stress in disyllabic words, the two levels of stress seem to exhibit equal prominences in compounds, but only when unaccented. Similarly, Morrill Adams (Reference Morrill Adams, Trouvain and Barry2007) demonstrated that systematic differences occurred between phrasal and compound stress in English, across intonational environments. In particular, intensity stood out as a reliable cue to compound stress in all environments, except for questions. In the latter context, as well as in the continuation-rise pattern, duration emerged as the most reliable cue. Thus, although compounds can also exhibit levels of stress, the stress correlates they employ do not necessarily need to align with those of simple words.

The existence of studies which were successful at identifying correlates of secondary stress should not overshadow the large body of work which has grappled with the issue of the acoustic elusiveness of secondary stress. Secondary stress has been particularly hard to describe phonetically in languages which do not exhibit concomitant phonological phenomena, such as vowel reduction or aspiration of consonants, present for example in English. In addition, issues in identifying correlates of secondary stress are also more common in languages for which a rhythmic pattern of stresses has been proposed. Examples likes these include Hungarian (Blaho & Szeredi Reference Blaho and Szeredi2011), Brazilian Portuguese (Arantes & Barbosa Reference Arantes and Barbosa2008) and Spanish (Prieto & van Santen Reference Prieto, van Santen, Parodi, Quicoli, Saltarelli and Zubizarreta1996, Díaz-Campos Reference Díaz-Campos, Campos, Herburger, Morales-Front and Walsh2000).Footnote ⁴ Although none of these studies investigated an exhaustive range of possible stress parameters, the production results parallel the reported scepticism on the part of native speakers towards the perception of secondary stress. In addition, studies which did reveal a rhythmic pattern of stresses in ‘binary plus clash’ systems (Gordon Reference Gordon2002), were able to demonstrate acoustic differences in stress levels based on such fundamental correlates of stress as vowel duration, intensity and pitch (Gordon & Rose Reference Gordon and Rose2006 for Émérillon; Hintz Reference Hintz2006 for South Conchucos Quechua).

The review of previous acoustic work on secondary stress justifies the selection of acoustic parameters in the present study. Since previous studies have shown that correlates of secondary stress are largely derivative of those signalling primary stress, the acoustic analysis presented here will focus on parameters known to be affected by stress in Polish (i.e. vowel duration, maximum pitch, pitch change and maximum intensity). The same set of parameters is analysed in compounds, as these could express secondary stress through different parameters than simple words. Simple words are additionally analysed with respect to onset duration and pitch slope.

3.1 Predictions for simple words

3.2 Predictions for compounds

As all compounds featured in the word list are at least quadrisyllabic, I expect each stem to mirror the stress pattern of single-root words of the same length (see §2.1.2). Specifically, I expect both stems to exhibit robust penultimate stresses. In addition, I predict that the two stresses will differ in their degree of acoustic robustness, with the primary stress on the head of the compound (rightmost stem) being stronger than the secondary stress on the non-head penultimate syllable. In addition, tertiary stress is expected on the initial syllable of all stems with at least four syllables. As in the case of simple words, I will test whether the presence of degrees of stress is reflected in the measures of the acoustic correlates available for primary stress.

3.3 Hypothesis: Polish manifests degrees of stress

Analysing different syllable positions separately will allow me to assess the accuracy of the descriptions of Polish stress presented in the literature. Specifically, I will attempt to verify the representation of Polish stress as a bidirectional system with internal lapses (Rubach & Booij Reference Rubach and Booij1985, Kraska-Szlenk Reference Kraska-Szlenk2003). For this claim to be supported, the hypothesis that Polish manifests degrees of stress needs to be confirmed.

For Polish simple words to have four degrees of stress, the prominence hierarchy in (8a) should hold. In terms of the syllable distinctions I have assumed, this hierarchy of stress levels translates into the syllable positions ranked in (8b). The crucial distinction, however, is between unstressed syllables and all others, as it is possible for primary, secondary and tertiary stresses to be realised in a similar manner. A similar hierarchy for compounds is provided in (8c).

1. (8)

   

To be able to test this hypothesis, I conducted a production study with native speakers of Polish, aimed at eliciting lower levels of stress. The following section lays out the methodology of the experiment.

5.1 Coding

Six syllable positions were distinguished in the analysis of simple words, corresponding to the following degrees of stress: initial (secondary stress), odd-numbered (tertiary stress), even-numbered (unstressed), pre-tonic (unstressed), penultimate (primary stress) and final (unstressed). The final syllable was treated separately from other unstressed syllables because syllables in domain-final position have been shown to exhibit boundary-related lengthening (for English, Lehiste Reference Lehiste1973, Oller Reference Oller1973, Klatt Reference Klatt1975, Cooper & Paccia-Cooper Reference Cooper and Paccia-Cooper1980, Wightman et al. Reference Wightman, Shattuck-Hufnagel, Ostendorf and Price1992; for Tamil, Byrd et al. Reference Byrd, Kaun, Narayanan, Saltzman, Broe and Pierrehumbert2000). Since the initial, pre-tonic, penultimate and final syllables were given separate labels, they were never included in the odd- or even-numbered categories. This allowed for no overlap between the stress positions. (9) illustrates the way six-syllable words were coded.

1. (9)

   

5.2 Duration

5.2.1 Vowel duration

The mixed-effects model for vowel duration revealed significant effects of Syllable Position, Vowel Height and the interaction of Syllable Position and Vowel Height, with no effect of Gender (Fig. 1 and Table I). As in the case of all models presented in this section, the pattern is first illustrated in the form of a graph, followed by a table summarising the regression results and a discussion of the effects observed. In all regression tables in this section, the p-value is given in bold if significant below the 0·05 level. The intercept represents the baseline to which the remaining levels of a factor are compared. For Syllable Position, the penult was selected as the baseline, as it was expected to be markedly different from other syllables, being the locus of primary stress. In cases where a significant effect of Vowel Height is obtained, the baseline represents the average value of high vowels in the penultimate position. If Gender is also significant, the baseline stands for the mean for high vowels in the penultimate position and in female speech. The estimates for the remaining levels of each variable indicate the amount by which the mean is predicted to increase (if the number is positive) or decrease (if it is negative) for that particular level, as compared to the estimate for the intercept. Thus, in Table I, the estimate for the intercept provides the mean vowel duration for high vowels in penultimate syllables. Negative estimates for most of the remaining levels of Syllable Position indicate that the penult is longer than most syllables. The regression table also provides the standard error and t-values for each factor level, and two-tailed MCMC probabilities p.

Figure 1

Mean vowel duration (in ms) according to syllable position.

Table I

Fixed-effect coefficients in a mixed-effects model fitted to vowel duration in simple words.

The fact that an effect of Syllable Position was obtained implies that the length of vowels tends to vary depending on which syllable position they occupy. Regression results in Table I show that high vowels in the penultimate position are much longer than the same vowels in most positions in the word. This observation can be inferred from the negative values of the estimates provided for the remaining levels of Syllable Position, as compared to the baseline. The exception is the pre-tonic syllable, whose estimated mean value is positive (1·44 ms), and thus slightly higher than the penultimate syllable. Tukey post hoc tests confirmed that the effect of Syllable Position was due to a significantly longer vowel on the penultimate syllable as compared to other positions, except for the pre-tonic syllable (for all pairwise comparisons, p<0·05). The initial syllable, traditionally associated with secondary stress, is significantly shorter than all other syllables (all p<0·05). Other significant differences were also observed, i.e. even-numbered syllables were identified as longer than finals (p<0·001) and pre-tonic syllables as longer than all syllables, except for the penult (all p<0·01). These last two effects are likely spurious, and possibly result from the large sample size or unequal n's. Recall that the present data set combines six-syllable words, which include odd-numbered syllables, with four- and five-syllable words, which do not. Hence, odd-numbered syllables are underrepresented. In fact, increased duration on the even-numbered and pre-tonic syllables disappears if only six-syllable words are taken into account. Moreover, although these differences in length are statistically significant, they are not likely to be perceptually distinguishable. Klatt (Reference Klatt1976) reports 25 ms as the just-noticeable difference (JND) in vowel length for English listeners. While this threshold is certainly surpassed by the penultimate syllable, length differences between the other syllables are confined to the much lower range of 0·6–9 ms.

Vowel duration also manifests an effect of Vowel Height. In addition, the interaction of Syllable Position with Vowel Height emerges as significant. The estimates provided in Table I suggest that both mid and low vowels are much longer than penultimate high vowels, and incur an increase of 13 ms and 43 ms respectively. This effect can be ascribed to vowel-inherent differences. In the present data set, a negative correlation between vowel height and length is observed: the higher the vowel, the shorter it is. Hence, the low vowel /a/ is the longest (84·15 ms), followed by mid vowels /ε ɔ/ (68·4 ms) and high vowels /i ɨ u/ (59·14 ms). Tukey pairwise comparisons between these sets of vowels found all these differences to be highly significant (all p<0·0001). These vowel-specific durations also seem to account for the presence of the interaction effect between Vowel Height and Syllable Position. Figure 2 illustrates that durational differences are more pronounced with low and mid vowels as compared to high vowels. This effect may have to do with the fact that high vowels are the shortest, and thus do not have such a vast durational range as the much longer low vowel /a/. Crucially, all vowels replicate the pattern sketched in Fig. 1. Note that the initial syllable is consistently the shortest, which may point to vowel reduction in this position.

Figure 2

Mean vowel duration (in ms) for high, mid and low vowels according to syllable position.

5.2.2 Onset duration

A separate mixed-effects model was built for onset duration, with the fixed effect of Syllable Position and random intercepts of Speaker and Item (Fig. 3 and Table II). Since only the stop onset in [ba] syllables was investigated, the effect of Vowel Height was not examined.

Figure 3

Mean onset duration (in ms) according to syllable position.

The model for onset duration did not find Syllable Position to be a significant predictor. Regression results are summarised in Table II. Recall that the final syllable is not featured, as no instances of [ba] syllables occurred in this position.

Table II

Fixed-effect coefficients in a mixed-effects model fitted to onset duration in simple words.

Table III and Fig. 3 illustrate that onset duration generally oscillates around 76·6 ms, which is the average value for the penultimate syllable. Two syllable positions slightly depart from this mean: onsets of initial syllables are on average 8·5 ms longer than those of the baseline, while those of odd-numbered syllables are 8 ms shorter. These differences are not statistically significant, nor are they likely to be perceptually salient. Although sources vary on the JND for the duration of consonants, even the smallest increment proposed, that of 10 ms (Klatt Reference Klatt1976), is greater than the differences observed here.

Table III

Mean duration (in ms) of [ba] onsets according to syllable position.

5.3 Maximum intensity

The mixed-effects model for maximum intensity revealed significant effects of Syllable Position, Vowel Height and the interaction of Syllable Position with Vowel Height (Fig. 4 and Table IV). There was no effect of Gender.

Table IV

Fixed-effect coefficients in a mixed-effects model fitted to maximum intensity in simple words.

Figure 4

Mean maximum intensity (in dB) according to syllable position.

A significant effect of Syllable Position suggests that maximum intensity values vary depending on the position of the syllable within the word. Table IV illustrates that baseline high vowels have a higher maximum intensity (76·19 dB) on the penult than in any other position, except word-initially. The means provided in Table IX below illustrate that the initial and penultimate syllables have the highest maximum intensity values (initial: M=77·7 dB; penultimate: M=77·5 dB), while the final has the lowest (M=74·7 dB). These numbers suggest that intensity generally decreases towards the right edge of the word, only peaking again on the penultimate syllable. Tukey post hoc tests confirmed that the initial and penultimate syllables have significantly higher maximum intensity values than most syllables (all p<0·05), without being significantly different from each other (p=0·98). By contrast, final syllables are significantly lower in intensity than all other positions (all p<0·001), except for odd-numbered syllables (p=0·48). The downward trend in intensity is also visible in the fact that even-numbered syllables have a significantly greater maximum intensity than odd-numbered and final syllables (both p<0·05). Before culminating on the penult, maximum intensity starts rising on the pre-tonic syllable, which manifests higher intensity than the odd-numbered and final syllables (both p<0·05). Figure 4 presents the distribution of maximum intensity across syllable positions.

The effect of Vowel Height as well as the significant interaction of Vowel Height and Syllable Position are again attributable to inherent differences in intensity between vowels. Table IV shows that mid and low vowels have higher maximum intensities than the baseline high vowels, by 0·72 dB and 2 dB respectively. A negative correlation thus obtains: the higher the vowel, the lower its inherent intensity. Hence, the low vowel /a/ has the largest maximum intensity (78 dB), mid vowels /ε ɔ/ are in the middle of the range (77·1 dB), while high vowels /i ɨ u/ exhibit the lowest values of all (75·3 dB). Tukey tests established differences between all three vowel categories as statistically significant (all p<0·0001). Since vowels differ in terms of their maximum intensities, their intensity profiles are also expected to differ depending on their position in a word; hence the presence of the interaction effect. In fact, vowels of all heights exhibit the declination effect illustrated in Fig. 4, with a single peak coinciding with the penultimate syllable, but intensity fluctuations are more prominent for high vowels (see Fig. 5). Table IV shows that the interaction effect mostly concerns mid vowels and the first three syllable positions, which happen to exhibit close intensity values.

Figure 5

Mean maximum intensity (in dB) for high, mid and low vowels according to syllable position.

5.4 Maximum pitch (in semitones)

The model for maximum pitch showed significant effects of Syllable Position, Vowel Height and Gender, as well as a significant interaction of Syllable Position with Vowel Height (Fig. 6 and Table V).

Table V

Fixed-effect coefficients in a mixed-effects model fitted to maximum pitch in simple words.

Figure 6

Mean maximum pitch (in semitones) according to syllable position.

Figure 6 illustrates that values of maximum pitch vary greatly depending on Syllable Position. However, these values can be predicted based on whether a syllable is placed closer to the left or right edge of a word, as a result of a downward trend exhibited by this parameter. In particular, the regression results in Table V show that the estimated coefficients for the first two syllables in a word are positive, implying a higher pitch compared to the baseline penult. Syllables adjacent to the penult, however, have negative coefficients, which suggests that, at least in the case of the baseline high vowels, maximum pitch rises on the penultimate syllable. This general downward trend in maximum pitch is confirmed by Tukey post hoc tests. These pairwise comparisons established that syllables at the left edge of words (initial and even-numbered syllables), when compared to all others, exhibit significantly larger maximum pitch values (all p<0·05). On the other hand, the syllable at the right edge (final) manifests significantly lower maximum pitch (all p<0·001). Although, on average, the initial syllable has slightly higher pitch than even-numbered syllables, this difference is marginal (0·1 semitones), and indeed does not reach significance (p<0·08). Note that, except for the penultimate and final syllables, adjacent syllables are typically not significantly different from each other. This result is expected given that adjacent syllables' mean pitch values are very close to one another (see Table IX below) and do not exceed even the lower bound of the perception threshold reported for pitch, i.e. 1–3 semitones ('t Hart Reference Hart1981). Thus significant comparisons only arise between distant syllables, which seem to differ in pitch as a result of the observable pitch declination.

The significant effect of Vowel Height, as well as the significant interaction of Vowel Height with Syllable Position, suggests not only that vowels vary in their maximum pitch, but also that they display different pitch patterns across words. Table V shows that mid and low vowels on average have lower maximum pitch than the baseline high vowels, which is reflected in their negative coefficients. Overall means confirm this observation: high vowels /i ɨ u/ exhibit higher average maximum pitch (88·1 semitones) than mid or low vowels (for both, 88·06 semitones). Tukey pairwise comparisons conducted on the mixed-effects model confirmed that the pitch differences between high and low vowels were significant (p<0·01). The interaction effect can also be attributed to inherent pitch differences between vowels of different height. Figure 7 illustrates that for mid and low vowels, the decline in pitch across a word is gradual, while for high vowels pitch dips slightly on the odd-numbered syllable.

Figure 7

Mean maximum pitch (in semitones) for high, mid and low vowels according to syllable position.

The significant effect of Gender is fully expected, due to differences in male and female pitch ranges. Table V illustrates that maximum pitch is consistently higher in women's speech (baseline estimate), and decreases by 10·2 semitones in men's speech. Both genders, however, manifest a gradual declination of pitch throughout the length of the word, with men showing a slight peak on the penultimate syllable. Means for maximum pitch for each gender and across all syllable positions are listed in Table IX below.

5.5 Pitch change

The model for pitch change (Fig. 8 and Table VI) showed a significant effect of Syllable Position, Vowel Height and their interaction. Gender was not significant. This result is expected, given that pitch change represents the size of pitch excursions within a syllable, which is a relative measure, and thus should not be affected by differences in pitch ranges across genders.

Figure 8

Mean pitch change (in semitones) according to syllable position.

Table VI

Fixed-effect coefficients in a mixed-effects model fitted to pitch change in simple words.

The significant effect of Syllable Position implies that the syllable positions distinguished do not manifest uniform values of pitch change. Regression results in Table VI show that all syllable positions, with the exception of the final, have much lower values of pitch change than the baseline penultimate syllable. Indeed, Fig. 8 and Table VI illustrate that two of the syllables, the penultimate and the final, are characterised by large pitch changes, while the remaining syllables do not exhibit much fluctuation at all. Tukey tests confirmed that the penultimate and final syllables are statistically different from each of the remaining syllables (all p<0·001), as well as from each other (p<0·05). No further comparisons were significant.

The observed effect of Vowel Height and the interaction of Vowel Height with Syllable Position again appear to stem from inherent differences in the range of pitch fluctuations for vowels of different height. Table VI shows that low vowels in penultimate position exhibit a larger pitch change than the baseline high vowels (an increase of 0·1 semitones), while the mean value for mid vowels is almost identical to that of high vowels (a decrease of 0·01 semitones). Overall, high vowels exhibit significantly larger pitch changes (M: 0·44 semitones) than mid (M: 0·36 semitones) and low vowels (M: 0·38 semitones), possibly due to inherently higher maximum pitch. The coefficients provided in Table VI also indicate that the interaction effect only arises when high and low vowels are compared. Indeed, pitch change across syllable positions is consistently lower for low vowels than for high vowels, except for the penultimate position, where a greater pitch change is manifested by low vowels. Crucially, all vowels replicate the dichotomy between the penultimate and final on the one hand and the remaining syllables on the other, but in the case of the low vowel /a/ the penultimate is marked by a larger pitch change than the final (see Fig. 9).

Figure 9

Mean pitch change (in semitones) for high, mid and low vowels according to syllable position.

5.5.1 Direction of pitch excursions

As pitch change was not able to differentiate between the penultimate and final syllables, the direction of pitch excursions was investigated. A visual inspection of pitch contours for a subset of the data revealed that syllables were overwhelmingly unitonal, i.e. with pitch that was either falling or rising. Hence, pitch contours for all 12,256 vowels were determined, by subtracting the time at which minimum pitch occurred from the time at which maximum pitch was measured. A positive number indicated rising pitch, while a negative number signified falling pitch.

Falling pitch was found to be the majority pattern for all syllable positions, with rising pitch reported for 35% of all syllables analysed. For the penultimate and final syllables, the two pitch directions were associated with quite different pitch change values. In the case of the final, pitch rises (observed in 71% of finals) show a pitch change that is twice as large (1·08 semitones) as that of pitch falls (0·5 semitones). The reverse holds for penults, where the most common, falling pattern (observed in 78% of penults) is associated with a larger pitch change (0·71 semitones) than that of syllables with rising pitch (0·3 semitones). In the recordings of the ten speakers, pitch rises on the final syllable, reminiscent of a question-like intonation, were overwhelmingly observed in female speech. Thus, if the rising pitch pattern is disregarded, a marked difference arises between the penultimate and final syllables (see Table VII), suggesting that the penultimate syllable is typically more pronounced in terms of pitch change.

Table VII

Mean pitch change (in semitones) according to direction of pitch excursions.

5.6 Pitch slope

Tabel VIII

Fixed-effect coefficients in a mixed-effects model fitted to pitch slope in simple words.

The measure of pitch slope introduces a temporal dimension with respect to changes in pitch. That is, it divides pitch change by the temporal distance between pitch peak and trough. The mixed-effects regression model fitted for pitch slope identified as significant the same effects that occurred with pitch change: Syllable Position, Vowel Height and the interaction of Syllable Position with Vowel Height. Indeed, as can be seen from the means presented in Table IX, pitch slope yielded results that were qualitatively similar to those of pitch change. The penultimate and final syllables stand out as having larger pitch slopes than the remaining syllables, with the penult being more pronounced for high vowels. Regression results in Table VIII thus bear a resemblance to Table VI, and will not be discussed further.

Table IX

Means for all parameters according to syllable position in simple words.

5.7 Discussion: simple words

The results obtained from simple words indicate that there are substantial differences in the acoustic realisation of different syllable positions. The syllable that stands out the most is the penultimate, as a number of acoustic correlates are associated with it. In particular, the penult is the longest syllable, and one of the two loudest ones. In contrast to most other syllables, it manifests a large pitch change and a steep pitch slope. It is also marked by higher maximum pitch in subsets of the data, e.g. in women's speech, and for high vowels. This association of the penult with several acoustic correlates supports the claim that this syllable carries primary stress.

This finding is consistent with previous research, which showed that primary stress in Polish correlates with pitch fluctuations (Jassem Reference Jassem1962), maximum pitch (Dogil & Williams Reference Dogil and Williams1999) and maximum intensity (Łukaszewicz & Rozborski Reference Łukaszewicz and Rozborski2008). A novel result is that duration is also established as a stress correlate. To date, most studies have mentioned duration as having only an accessory role in marking stress. Such a weak effect is unexpected in light of findings from perception studies which showed that lengthening a vowel by as little as 20% of its duration is an effective cue to stress for Polish listeners (Jassem et al. Reference Jassem, Morton, Steffen-Batog and Jassem1968). The fact that the present study was able to find a significant effect of duration could be due to a different experimental design: I used word-list data, as opposed to data from spontaneous speech. This allowed for a much stricter control over sentence prosody and other sources of variation. Another methodological difference was that previous research customarily lumped all syllables into two groups: stressed (penultimate syllables) and unstressed (all others). The results from the previous section show that initials and finals differ significantly from other syllables, and thus grouping all non-penultimate syllables together is likely to have skewed the results. For example, final syllables tend to be lengthened phrase-finally (or PWd-finally, as will be seen in compounds). If these lengthened tokens, presumably more frequent in spontaneous speech, were included in previous analyses, a longer duration on the penult could have gone unnoticed.Footnote ⁶

Initial and odd-numbered syllables, traditionally believed to bear secondary and tertiary stress respectively, do not show much evidence of acoustic robustness. In particular, odd-numbered syllables are no more salient than the unstressed even-numbered syllables on any of the measures. On the contrary, odd-numbered syllables have shorter onsets, and exhibit lower maximum intensity and maximum pitch than even-numbered syllables. Some acoustic prominence is, however, observed on the initial syllable, traditionally associated with secondary stress. This position displays the largest maximum pitch and maximum intensity, and has the longest onset.

In terms of maximum pitch, the initial syllable is more prominent than all other positions, with the exception of the immediately following even-numbered syllable. The pitch differences between these two syllables are so small (0·1 semitones) that they are far below the just noticeable difference (JND) reported for speech, i.e. 1–3 semitones ('t Hart Reference Hart1981). If greater maximum pitch on the initial syllable were to signal secondary stress, one would expect the presence of a robust contrast with the adjacent syllable. I propose that, instead of signalling stress, the high maximum pitch value on the initial syllable results from ‘F0 declination’ – the cross-linguistic tendency for F0 to steadily decline across consecutive syllables in an utterance (Bolinger Reference Bolinger and Lunt1964, Cohen & 't Hart 1967, Maeda Reference Maeda1974, Ladd Reference Ladd1984).

A similar trend can be observed with respect to maximum intensity. On this measure, the initial syllable scores significantly higher than all other syllables, including the even-numbered (unstressed) position. Note, however, that even-numbered syllables have larger intensities than odd-numbered syllables, which suggests that intensity cannot be a marker of tertiary stress. Indeed, intensity values appear to progressively decrease towards the ends of words in Polish, and only rise on the penult to mark primary stress. This observation is in line with the trend reported for other languages, where intensity declines over utterances (Trouvain et al. Reference Trouvain, Barry, Nielsen and Andersen1998 for English and Dutch; Garnier et al. Reference Garnier, Dohen, Lœvenbruck, Welby, Bailly, Yehia, Demolin and Laboissiere2006 for French). As a reviewer points out, another possibility is that the increased pitch and intensity on the initial syllable result from the focus position of target words. Although the focus position is bound to result in more acoustic prominence on the initial syllable, the pitch and intensity profiles in whole utterances do point to declination effects in Polish. The spectrograms in the online supplementary materials give pitch and intensity patterns for target words in carrier phrases. These illustrate that both intensity and pitch values drop steadily over the length of entire phrases. The declination effect does not, however, preclude the possibility of the initial syllable carrying stress. After all, the difference in intensity between the initial and even-numbered syllables is large enough to be statistically significant, as opposed to the pattern observed for maximum pitch. Further experimentation, preferably investigating several prosodic environments, would be required to determine if increased intensity on the initial could signal stress. Data from compounds, presented in the next section, do not lend credence to this hypothesis, however. If intensity on the initial syllable were to mark stress, one would expect an intensity peak on the right-stem initial. Instead, only the left-stem initial exhibits prominence, and the overall intensity pattern unfolds as it would in simple words.

Finally, the initial syllable stands out in terms of onset duration. This could suggest that Polish is like Estonian in marking secondary stress by the lengthening of the onset (Gordon Reference Gordon, Lehiste and Ross1997). However, durational differences between the initial and the adjacent even-numbered syllables are not large enough to have perceptual significance. Sources vary on the JND for the duration of consonants, but even the smallest increment proposed, that of 10 ms (Klatt Reference Klatt1976), is almost twice the difference observed. Yet longer onsets are consistent with domain-initial strengthening that is independent of stress. It is thus more likely that longer word-initial onsets represent yet another word-boundary effect, well attested in the literature (Oller Reference Oller1973, Klatt Reference Klatt1974, Turk & Shattuck-Hufnagel Reference Turk and Shattuck-Hufnagel2000). Crucially, onsets in odd-numbered syllables, traditionally reported to bear tertiary stress, do not undergo lengthening. Their means are, in fact, shorter than those of other syllable positions. While this effect could point to syllable reduction, it is expected due to polysyllabic shortening – the length of all syllables undergoes compression in longer words (Turk & Shattuck-Hufnagel Reference Turk and Shattuck-Hufnagel2000). Recall that odd-numbered syllables were extracted from six-syllable words, while the remaining syllables belonged to four- and five-syllable words. Overall, the increase in onset duration visible on the initial syllable, as well as in F0 and maximum intensity, appear to be artefacts of this syllable's location at the left edge of words.

In what follows, I analyse data from compounds, which provide further support for the seemingly non-iterative nature of stress in Polish simple words. Specifically, a robust secondary stress is identified on the penultimate syllable of the non-head, leftmost stem in a subset of compounds. Moreover, the same domain-strengthening processes that operate in simple words are found to apply to the leftmost stem, which argues against the initial syllable being marked by secondary stress in simple words. Through a formal analysis, couched in Optimality Theory, I argue that the avoidance of adjacent feet largely accounts for the stress pattern observed in compounds.

6.1 All compounds

The following analysis combines data from all compounds in the word list. As outlined in §2.1.2, each compound stem in Polish is generally believed to mirror the stress pattern of simple words of the same length. Thus predictions as to the placement of levels of stress in compound stems are the same as the ones that applied in simple words. One additional distinction needs to be stipulated, however: the rightmost stem, being the head of the compound, is expected to be more prominent than the leftmost stem. Hence, evidence of prominence on the right-stem penult will be treated as primary stress, while prominence on the left-stem penult will be associated with secondary stress. Since levels of stress are shifted, tertiary stress is now expected to fall on the initial syllable of either stem.

6.1.2 Vowel duration

The mixed-effects regression model fitted for vowel duration (Fig. 10 and Table X) found significant effects of Syllable Position and Vowel Height, but not of Gender. The model included by-speaker slopes for Syllable Position, and by-word slopes for Vowel Height. In all figures in this section, stem boundaries are indicated by a dashed line. For all regression tables, the right-stem penult was selected as the baseline, as it was expected to carry primary stress. In any model which includes Vowel Height and/or Gender as a significant predictor, the baseline represents the mean value of the penult for high vowels and/or female speech.

Table X

Fixed-effect coefficients in a mixed-effects model fitted to vowel duration in compounds.

Figure 10

Mean vowel duration (in ms) according to syllable position in compounds.

The significant effect of Syllable Position implies that vowel duration is affected by the position a syllable occupies within a word. Table X shows that all syllable positions are significantly shorter than the baseline, i.e. the right-stem penultimate syllable. This is evidenced in the large negative coefficients calculated for each position. Note that the left-stem penult is the longest of the remaining syllables, as its coefficient is the smallest. Tukey post hoc tests confirmed that the effect of Syllable Position is due to a significantly greater duration of the two penults (left stem: M=69·8 ms; right stem: M=81·1 ms). Both penults are significantly longer than most other syllables (all p<0·01), with the right-stem penult still significantly longer than the left-stem penult (p<0·01). The only comparisons that did not reach significance were between the left-stem penult and both finals (for both, p=1). Another syllable which stands out in length is the left-stem final, found to be significantly longer than the initial and pre-tonic syllables sharing its stem (both p<0·05). This longer duration may be due to the syllable's position between PWds. Finally, the left-stem initial is significantly shorter than the right-stem initial position (p<0·05). Means for particular syllable positions are presented in Table XV below.

Vowel Height was identified as another significant predictor of vowel duration. As in simple words, vowel duration increases as vowel height decreases: mid and low vowels are longer than the baseline high vowels in the penultimate position, by 4·72 ms and 12·98 ms respectively. Tukey pairwise comparisons between these sets of vowels were again significant (all p<0·05).

6.1.3 Maximum intensity

The mixed-effects regression model fitted for maximum intensity (Fig. 11 and Table XI) found a significant effect of Syllable Position and Vowel Height, but not of Gender. By-speaker and by-word slopes for Vowel Height were included in the model.

Table XI

Fixed-effect coefficients in a mixed-effects model fitted to maximum intensity in compounds.

Figure 11

Mean maximum intensity (in dB) according to syllable position in compounds.

Syllable Position was found to be a significant predictor of maximum intensity values. In general, the largest maximum intensity is found on leftmost stems: the two highest average maximum intensity values belong to the left-stem initial (M=78·4 dB) and penultimate (M=78·1 dB) syllables. This pattern is reminiscent of simple words, where intensity declination was observed. Despite the fact that the left stem manifests greater intensity, penults in both stems have higher intensity values than adjacent syllable positions. In fact, the syllable with the highest maximum intensity in rightmost stems is again the penult (M=77 dB). Tukey pairwise comparisons support these observations. Specifically, the left-stem penult has significantly greater maximum intensity than all other syllable positions (all p<0·001), except for the left-stem initial (p=1). Maximum intensity on the latter also exceeds that of all remaining syllables (all p<0·05). A very similar pattern is observed on the rightmost stem. Here, the penult has a significantly greater maximum intensity than all other syllables that share the same stem (all p<0·001), except for the initial syllable (p=0·21). Although the right-stem initial position has a greater maximum intensity than other non-penultimate right-stem syllables (for all comparisons, p<0·05), it is almost equal in intensity to the adjacent left-stem final syllable (p=1).

Vowel Height is another significant predictor of maximum intensity. As in simple words, this effect can be attributed to different inherent intensities for vowels of different height. Table XI illustrates the trend already observed in simple words: maximum intensity increases as height decreases. Tukey pairwise comparisons established that mid and low vowels were significantly different from high vowels on this parameter (both p<0·0001), but were not different from each other (p=0·19).

6.1.4 Maximum pitch

The mixed-effects regression model fitted for maximum pitch (Fig. 12 and Table XII) found significant effects of Syllable Position and Gender, but not of Vowel Height. By-speaker and by-word slopes for Vowel Height were included in the model.

Table XII

Fixed-effect coefficients in a mixed-effects model fitted to maximum pitch measurements in compounds.

Figure 12

Mean maximum pitch (in semitones) in different syllable positions in compounds.

The significant effect of Syllable Position implies that maximum pitch values can be predicted based on the position of a syllable within a word. Table XII shows that in female speech all syllable positions are significantly different from the baseline right-stem penult. Positive coefficients for all positions except for the right-stem final suggest that maximum pitch values are typically higher on syllables leading up to the right-stem penult. This result likely reflects the same declination effect observed for maximum pitch in simple words. Note that the syllable position associated with the largest positive coefficient (highest maximum pitch) is the left-stem penult, traditionally described as carrying secondary stress. In addition, Table XIII shows that in men's speech, the highest maximum pitch on the right stem is also associated with the penultimate syllable. Post-hoc Tukey tests confirmed these differences to be statistically significant. In particular, the left-stem penult has a significantly larger maximum pitch than any other syllable position (all p<0·05), while the right-stem penult has significantly greater pitch than the following final syllable (p<0·001), but has significantly lower pitch than the preceding pre-tonic syllable (p<0·001). Lastly, a number of significant comparisons arise which seem to result from pitch declination. Specifically, the right-stem final's maximum pitch is significantly lower than that of any other syllable (all p<0·001). The left-stem final, in turn, has higher pitch than all syllables belonging to the rightmost stem (all p<0·01), as well as the left-stem pre-tonic syllable (p<0·01). The latter also manifests higher pitch than some right-stem syllables, i.e. the pre-tonic, the penultimate and final syllables (all p<0·01).

Table XIII

Means (in semitones) and sample sizes for maximum pitch in compounds.

Values of maximum pitch are also affected by Gender. Table XIII summarises the pitch pattern observed in male and female speech. Unsurprisingly, maximum pitch is consistently lower for men than for women. It is worth noting, however, that in men's speech, the right-stem penult displays a pitch peak absent from women's speech.

6.1.5 Pitch change

The model fitted for pitch change (Fig. 13 and Table XIV) reveals a significant effect of Syllable Position, with no effect of Vowel Height or Gender. The model presented here includes by-speaker slopes for Syllable Position and Vowel Height, and by-word slopes for Vowel Height, as they improved the overall fit.

Table XIV

Fixed-effect coefficients in a mixed-effects model fitted to pitch change in compounds.

Figure 13

Mean pitch change (in semitones) according to syllable position in compounds.

The significant effect of Syllable Position suggests that pitch change is not uniform across syllable positions. The negative coefficients observed in Table XIV for the various levels of this predictor indicate that almost all syllables exhibit a significantly lower pitch change than the baseline right-stem penultimate position. The only exception is the right-stem final, whose mean pitch change is very close to that of the right-stem penult. Post hoc Tukey tests confirm that the effect of Syllable Position is due to a significantly greater pitch change on the right-stem penult when compared to the left-stem initial and pre-tonic syllables (both p<0·05). No other comparisons were statistically significant. Note that the left-stem penult does not manifest a statistically greater pitch change than any other syllable.

Means for all parameters in compounds are given in Table XV.

Table XV

Means for all parameters according to syllable position in compounds.

6.2 Four-syllable compounds

I now turn to the analysis of four-syllable compounds with two disyllabic stems, such as /bɔgɔ-bɔjna/ ‘god-fearing’. These compounds have been described as constituting single PWds, based on the assumption that they exhibit the same stress pattern as simple words (primary stress on the penultimate, secondary stress on the initial syllable). I show that they do pattern with simple words, rather than compounds. This affinity, however, is due to the absence, rather than the presence, of secondary stress.

6.2.1 Vowel duration

The model fitted for vowel duration (Fig. 14 and Table XVI) found significant effects of Syllable Position and Vowel Height. In addition, by-speaker slopes for Syllable Position and Vowel Height, as well as by-word slopes for Syllable Position, improved the model, and were thus included in the best fit reported here.

Table XVI

Fixed-effect coefficients in a mixed-effects model fitted to vowel duration in four-syllable compounds.

Figure 14

Mean duration (in ms) according to syllable position in four-syllable compounds.

The significant effect of Syllable Position implies that vowel duration varies with the position of a syllable within a word. Table XVI shows that the baseline right-stem penult is significantly longer than all other syllable positions. In fact, all syllables are between 20 and 25 ms shorter than the right-stem penult, which is consistent with Klatt's (Reference Klatt1976) reported value for the perceptual threshold for vowel length, i.e. 25 ms. Tukey tests confirm the special status of the right-stem penult, as it is identified as significantly longer than all other positions (p<0·05). Although the left-stem penult is on average longer than the left-stem final, this difference is not statistically significant (p=0·37).

Vowel Height also emerges as a predictor of vowel duration. Table XVI illustrates that both mid and low vowels are longer than the baseline high vowels in the penultimate position, with the low vowel /a/ being the longest of all. Tukey tests identified the comparison of high and low vowels as statistically significant (all p<0·001).

6.2.2 Maximum intensity

The model fitted to maximum intensity measurements also established a significant effect of Syllable Position and Vowel Height. By-speaker slopes for Syllable Position and Vowel Height as well as by-word slopes for Syllable Position were included in the model. Regression results are provided in Table XVII.

Table XVII

Fixed-effect coefficients in a mixed-effects model fitted to maximum intensity in four-syllable compounds.

The fact that Syllable Position was identified as a predictor of maximum intensity suggests that intensity does not display a uniform pattern throughout the length of compounds. This is compatible with results reported in previous sections, where a declination effect was consistently observed. Figure 15 and Table XVII again show that all syllable positions are lower in intensity than the baseline right-stem penult. The left-stem penult (corresponding to the left-stem initial elsewhere in this paper) again exhibits higher intensity than all remaining positions, while the right-stem final's intensity is lowest of all. These observations are confirmed by Tukey tests, which identify the right-stem penult as having significantly greater intensity than the following final (p<0·001). Although mean maximum intensity is slightly greater on the right-stem penult (M=77·7 dB) than on the left-stem penult (M=77 dB), this difference is not statistically significant (p=0·9). Possibly due to the declination effect, the right-stem final was found to manifest significantly lower intensity than the left-stem penult (p<0·05).

Figure 15

Mean maximum intensity (in dB) according to syllable position in four-syllable compounds.

Vowel Height is another predictor of maximum intensity values. Table XVII shows that baseline high vowels overall have lower intensity (by about 2 dB) than mid and low vowels in the penultimate position. Tukey pairwise tests found the comparison between high and mid vowels, and high and low vowels, to be significant (in each case, p<0·01), as opposed to that between mid and low vowels (p=0·6).

Table XVIII

Fixed-effect coefficients in a mixed-effects model fitted to maximum pitch measurements in four-syllable compounds.

6.2.3 Maximum pitch

As in previous sections, the model for maximum pitch identified Syllable Position, Vowel Height and Gender as predictors of the values of this parameter. The model reported here includes by-speaker slopes for Syllable Position and by-word intercepts. Regression results are provided in Table XVIII.

Table XIX

Means (in semitones) and sample sizes for maximum pitch in four-syllable compounds.

As a significant predictor, Syllable Position reveals fluctuations in the pattern of maximum pitch displayed by compounds. Contrary to the analysis of all compounds, this effect was no longer due to greater pitch on the penults. Instead, the results reflect the declination pattern found for simple words. Specifically, Fig. 16 and Table XVIII illustrate that maximum pitch is highest on the left-stem penult (word-initial syllable) and decreases consecutively with every syllable. Tukey tests confirmed the left-stem penult to be significantly different from the right-stem syllables (p<0·01), but not the left-stem final (p=0·5). Although the right-stem penult has greater maximum pitch than the following final (p<0·05), its values are lower than those of left-stem syllables.

Figure 16

Mean maximum pitch (in semitones) in different syllable positions in four-syllable compounds.

Another significant predictor of maximum pitch values is Vowel Height. Table XVIII shows that high vowels on average have higher pitch than mid and low vowels in the baseline penultimate position. As the coefficients for mid and low vowels are close, and imply a decrease in pitch of 0·6 and 0·9 semitones respectively, one would not expect them to be significantly different from each other. In fact, Tukey tests confirmed that high vowels had greater pitch than mid and low vowels (for both, p<0·001), but a comparison of mid and low vowels did not yield significance (p=0·16).

Finally, the effect of Gender implies that men and women have different pitch ranges. Indeed, Table XVIII shows that for high vowels in the right-stem penultimate position maximum pitch is generally 10 semitones lower for men than for women. This finding is consistent with results reported in previous sections. Crucially, Table XIX illustrates that both genders exhibit the declination pattern evident in Fig. 16.

6.2.4 Pitch change

In the model for pitch change, syllable position emerged as the only significant predictor. In terms of random effects, only by-speaker slopes for Syllable Position were included in the model, as by-word slopes and intercepts did not improve the overall fit. Regression results are provided in Table XX.

Table XX

Fixed-effect coefficients in a mixed-effects model fitted to pitch change in four-syllable compounds.

Figure 17 illustrates that the significant effect of Syllable Position is due to strikingly different pitch change values for the two stems. Table XX identifies the right-stem penult as the syllable characterised by the largest pitch change, with the remaining positions associated with negative coefficients. Tukey tests confirmed these observations, establishing a significant contrast between the right-stem and the left-stem penults (p<0·05). However, no significant difference was observed between the right-stem penult and adjacent syllables: the left-stem final (p=0·19) and right-stem final (p=0·99).

Figure 17

Mean pitch change (in semitones) according to syllable position in four-syllable compounds.

Table XXI

Means for all parameters according to syllable position in four-syllable compounds.

6.3 Discussion: compounds

The analysis of compounds confirms the location and acoustic make-up of main stress in Polish, as well as identifying the presence of secondary stress. In line with theoretical descriptions of Polish compounds, the head stem shows a penultimate (primary) stress, while the non-head stem optionally shows another penultimate (secondary) stress. The two stresses are robust, in that they correlate with multiple acoustic parameters, i.e. maximum pitch, maximum intensity and duration. What differentiates them, however, is that the head penultimate syllable additionally correlates with a large pitch change. It is also substantially longer than the non-head penultimate syllable (81·1 ms as compared to 69·8 ms).

The presence of two stresses is not, however, confirmed for four-syllable compounds. Here, it is only the right-stem penult that is acoustically more prominent than the neighbouring vowels, and manifests greater pitch change, maximum intensity and duration. It is worth noting that this syllable position does not correlate with maximum pitch, as it did in the analysis of all compounds. Instead, pitch values are highest on the left-stem penult (compound-initial), and decrease progressively towards the ends of words. This pattern is reminiscent of what was observed for simple words. Thus, while longer compounds have penultimate stresses on each stem, four-syllable compounds are indistinguishable from simple words on the basis of stress correlates.

The number of stresses informs the prosodic analysis of compounds. It is typically assumed that the number of (primary) stresses in compounds coincides with the number of PWds present (Nespor Reference Nespor1999). A question arises, therefore, as to the nature of the second stress we have observed in Polish compounds. If it is in fact secondary, then Polish compounds should constitute single PWds. If it is a separate instantiation of primary stress, a prosodic boundary should be present between stems. I argue that the latter interpretation is most valid. First, the location of the second stress corresponds to the canonical position of primary stress in Polish, i.e. the penultimate syllable. Second, although its weaker prominence could point to it being a lower-degree stress, the head stem is expected to be stronger in compounds that form a phonological phrase (see Nespor Reference Nespor1999 for an analysis of prominence in [word]+[word] compounds in left- and right-branching languages). Lastly, the present data set points to phonetic evidence for the presence of a prosodic boundary between compound stems. Glottal stop insertion, which arguably applies across PWd boundaries in Polish in cases of vowel hiatus (Kraska-Szlenk Reference Kraska-Szlenk2003), was sometimes observed between stems in [samɔ-ɔbrɔna] ‘self-defence’ and [vɔdɔ-ɔtpɔrnɔɕtɕ] ‘water-proof (n)’. Hence, although the left-stem stress in compounds is phonetically weaker than the right-stem stress, it is more accurate to describe it as a separate instantiation of primary stress on phonological grounds.

By that token, Polish compounds differ in terms of their prosodic structure: some are two-PWd structures ([ʑεˈlɔnɔ][bεˈʒɔvɨ] ‘green-beige’), while others constitute single PWds ([bɔgɔ-ˈbɔjna] ‘god-fearing’). As such, the latter share the structure of compounds whose second constituent is monosyllabic, such as [duɕ-ˈi-grɔʃ] ‘penny pincher’ and [ʒuwt-ˈɔ-dʑup] ‘greenhorn’. For these compounds, Rubach & Booij (Reference Rubach and Booij1985) propose a rule of ‘demoting’, which erases the internal PWd (or mot) boundary to avoid stress clash. In the following section I reanalyse this pattern as an effect of the *FtFt constraint. Specifically, I argue that compound stems are parsed into separate PWds and assigned separate stresses only if the emergent trochees are non-adjacent.

The presence of a greater pitch change on the head penult also has implications for the structure of compounds. Jassem (Reference Jassem1962) identified pitch change as the most robust correlate of main stress in Polish, and claimed that Polish stress is essentially a pitch accent. If pitch change were the only, or main, correlate of stress, as Jassem (Reference Jassem1962) argues, then all compounds should be treated as single PWds. Yet the association of the non-head penult in longer compounds with such parameters as maximum pitch, maximum intensity and duration is undeniable, and cannot indicate anything other than stress. I would therefore like to suggest that while pitch change is associated with pitch accent in Polish, it does not in itself represent word-level stress. Rather, word-level stress is represented by greater maximum pitch, maximum intensity and duration. At the same time, the fact that a pitch accent falls on the penultimate syllable in Polish is not accidental, and results from the general tendency for pitch accents to be attracted to the strongest stress in an utterance (see Hayes Reference Hayes1995 for English). I believe that the same interaction of stress and pitch takes place in Polish – pitch accents are attracted to (stressed) penultimate syllables. Instead of indicating the presence of a single PWd, the greater pitch change on the head penultimate syllable may thus be a reflection of a recursive structure. To propose that Polish compounds represent recursive PWds is not at all controversial. For example, Rubach & Booij (Reference Rubach and Booij1985) argue that compound stems are mots embedded in another mot, while Kraska-Szlenk (Reference Kraska-Szlenk2003) analyses them as PWds embedded in a MWd. The recursive analysis is supported by the fact that the right-stem penult is acoustically more prominent. According to Nespor (Reference Nespor1999), this added prominence on the head stem is evidence that the compound stems together form a phonological phrase. Lastly, a pitch accent on the right-stem penult could potentially account for the perceived levels of stress in compounds. Perceptual evidence would be required to support this conjecture, however.

The analysis of compounds sheds new light on the issue of the absence of lower levels of stress in simple words. In simple words, tertiary stress was found to be non-existent, while possible evidence of secondary stress was recognised on the initial syllable. This position coincided with a larger maximum pitch, maximum intensity and onset duration, all of which could also be explained as boundary effects. In the analysis of compounds, tertiary stress was expected to arise on initials in all stems of at least four syllables in length (positions coded as Initial L and Initial R), but again, phonetic evidence is scarce. The only measures correlating with these positions were maximum pitch and maximum intensity. Both parameters have the highest values on the left-stem initial syllable, which could be due to its compound-initial position. Although the right-stem initial also exhibits a larger maximum intensity than the two following syllables, it is not significantly different from the syllable that precedes it (Final L). In fact, its mean maximum intensity is lower than that of the left-stem final. Thus, the correlation of maximum pitch and maximum intensity with left-stem initials again appears to result from the general downward trend in F0 and intensity.