Main prominence is conventionally described as being assigned to the final syllable of phrases in French, but previous quantitative and qualitative work has shown that this is not always the case. Using corpus data from Laurentian French (Saguenay, Quebec), we test the hypothesis that prominence is preferentially assigned to heavy syllables. Our results demonstrate that this is indeed the case, with both codas and heavy vowels attracting prominence away from final syllables, particularly when the final syllable is open. We infer two distinct types of prominence: lexical and phrasal. Lexical prominence, which is marked using duration and amplitude, variably attracts phrasal prominence, which is marked using pitch. We interpret these findings as indicating that the location of phrasal prominence is sensitive to syllable weight and that this prominence is best formally expressed as a pitch accent due to its attraction to lexically prominent syllables.
Final syllables are conventionally characterised as obligatorily bearing main prominence in French (e.g.
The observation that the domains of prominence assignment in French and English are different suggests that prominence serves different functions in the two languages. In French, prominence allows interlocutors to reliably recover the right edge of phrases and therefore reduce the risk of ambiguity (e.g.
The literature on French predominantly focuses on extragrammatical motivations for prominence shift. Dialectal substrates (
A plausible motivation for prominence shift is that speakers are enhancing word-level properties in addition to phrase edges. Heavy syllables commonly attract prominence across languages (
In this paper, we test the hypothesis that prominence assignment is sensitive to relative weight, which is consistent with how weight interacts with prominence cross-linguistically. We use mixed effects linear regression to test the effects of prosodic domains, vowel weight and coda weight on the realisation of prominence in read speech collected from the Saguenay (Quebec) survey (
In sum, this dialect is an ideal test case to systematically examine the relationship between weight and prominence shift in French. We will show that heavy syllables attract prominence, thereby affecting the same acoustic cues as those conventionally interpreted to mark phrase edges. Based on these results, we argue for a revised interpretation of prominence in (Laurentian) French: word-level prominence, which is marked using duration and amplitude, variably attracts phrase-level prominence, which is marked using pitch.
Typological work shows that weight and prominence often interact; in languages with lexical stress, for example, heavy syllables typically attract stress (e.g.
(1)
Hypothesis:
Final prominence assignment in French is sensitive to weight.
Languages vary in which types of rhymes count as heavy: codas may or may not be weight-bearing, while long vowels are consistently heavy (
Some authors have analysed word-final codas in French as onsets of empty-headed syllables (e.g.
Turning to vowels, French has a relatively large inventory, which is generally described as including both light (short) and heavy (long) vowels even though these contrasts are predominantly realised through quality differences in contemporary French (e.g.
We exemplify the weight contrast in
Phonological length of phonemes according to syllable profile in Laurentian French.
/a/ | light | short | short | short |
/ɛ/ | light | short | short | short |
/œ/ | light | absent | short | short |
/ɔ/ | light | absent | short | short |
/i/ | light | short | short | short |
/y/ | light | short | short | short |
/u/ | light | short | short | short |
/ɑ/ | heavy | short | long | long |
/e/ | heavy | short | absent | long |
/ø/ | heavy | short | long | long |
/o/ | heavy | short | long | long |
/ɛ:/ | heavy | absent | long | long |
/ɑ̃/ | heavy | short | long | long |
/ɛ̃/ | heavy | long | long | long |
/œ̃/ | heavy | long | long | long |
/ɔ̃/ | heavy | long | long | long |
Because we adopt the position that vowel quality differences reflect weight differences in French, we expect that prominence will shift inwards more often in /kote/
Expected prominence location based on final-syllable weight.
Coda weight | Closed | Final prominence favoured |
Final prominence favoured |
Open | Shift to penult favoured |
Shift to penult favoured |
Expected prominence location based on penult weight.
Coda weight | Closed | Shift to penult favoured |
Shift to penult favoured |
Open | Shift to penult favoured |
Final prominence favoured |
The acoustic profiles of syllables at the right edge of prosodic domains in French have been the subject of some debate, with pitch, duration and amplitude all having been discussed as possible cues. Authors typically agree upon the role played by pitch, with high and low tones marking phrase boundaries (e.g.
Prominent syllables are typically longer than non-prominent ones, which has led to debate about whether length is a primary (
We point out that duration is confounded with weight: phonologically long segments (heavy vowels) and additional segments (codas) are, of course, expected to affect rhyme durations because there is more content to pronounce independent of weight. We expect a trade-off between syllables, however, with the prominent syllable being lengthened and the non-prominent syllable compressed. The effect of weight on duration should therefore be particularly robust; we would not conclude that there is sufficient evidence to support our hypothesis that prominence assignment is sensitive to weight if a small change in duration is the only effect of increased weight.
Unlike pitch and duration, amplitude is not typically reported in acoustic studies on French prominence. Indeed, some authors propose that amplitude is not a possible cue to prominence in French because it is associated only with word-level prominence (
We finally consider the possibility that prominence cues may not pattern together. For example, the pitch peak may remain on a light final syllable while the heavy penult increases in duration, in which case we would have evidence that both word-level and phrase-level prominences are signalled simultaneously but on different syllables.
In order to illuminate the phonological contribution of word-level properties to prominence in Laurentian French, we must assess the effect of the word’s phrasal context. This is because, as previously noted, the smallest domain of prominence in French is not the word; only phrasal domains assign pitch targets. The model we adopt is that of Jun & Fougeron (
Many terms have been used to describe the domains under examination, but the smallest domain consistently groups together lexical words and their preceding syntactic dependents (e.g.
The right edges of assertive IPs are typically marked with a low boundary tone (L%), which replaces the high (H*) of the AP’s rise and makes the contour level or slightly falling (LL%).
Although pitch differences between APs and IPs have been studied extensively and it is known that AP-final syllables exhibit lengthening relative to AP-internal syllables, it remains unclear whether duration and amplitude are also manipulated to distinguish between these two prosodic domains.
Finally, the use of amplitude to mark different types of prosodic domains in Laurentian French, if manipulated at all, is not yet known. Based on the cross-linguistic observation that amplitude is available as a cue to mark lexical prominence but not phrasal prominence (e.g.
Based on previous literature, we expect that French, and particularly Laurentian French, will predominantly use pitch and duration to mark prosodic domains. We additionally expect the language to exhibit weight sensitivity, with both long vowels and closed syllables being heavy and therefore attracting prominence. These heavy syllables are expected to be marked with increased rhyme durations and greater amplitudes. Furthermore, we expect the tone target (H* in the AP’s LH*) to shift inwards towards heavy penults and away from open final syllables, leading to higher pitch maxima for penults compared to final syllables.
Expected pitch contours at end of APs depending on whether AP-final word is also IP-final (right) or not (left), and whether word undergoes prominence shift (bottom) or not (top).
The goal of this study is to quantitatively test whether phonologically heavy syllables attract final prominence in Laurentian French.
As mentioned, we draw our data from the Laurentian sub-corpus (Côté 2014) of the
As previously noted, we examine Laurentian French because it conserves a large number of vowel length contrasts, providing more opportunities for weight effects to be observed. Furthermore, diphthongisation enhances the salience of weight contrasts. This variety is consequently an optimal starting point for probing prosodic effects of syllable weight in French.
We specifically examine speakers from Chicoutimi, Quebec, located 200km north of Quebec City. This area was selected as it has limited contact with other languages and dialects. In the Saguenay area, which includes Chicoutimi, 98.3% of inhabitants report speaking French as a native language, and 98.9% speak only French at home; further, the rate of bilingualism, including French-English bilingualism, is relatively low: inhabitants aged 20–44 are the most likely to be bilingual (31.8%; nearly double the next highest group’s rate) (
The data in this study come from 11 native French speakers who were born and raised in Chicoutimi, with speakers spanning three generations (age 22–74) and being relatively well-balanced for sex (5 men, 6 women). All speakers were fluent readers.
Speakers in the Chicoutimi survey and their demographic factors (year of birth, age during data collection).
Young adult | fv1 (1984, 22) |
cl1 (1982, 24) |
Middle-aged | db1 (1954, 52) |
gm1 (1958, 48) |
Older | rt1 (1934, 72) | gt1 (1932, 74) |
The recordings were segmented using the SPLalign forced aligner created by Milne (
As we examine prominence shift to the penult, only words with at least two realised syllables were extracted. We only examine the last two syllables of each target word, regardless of how many syllables are in the word. We additionally restricted the words analysed to those that are at the end of an AP, given that this is the smallest prosodic domain described as assigning pitch targets in French (e.g.
The contexts included are shown in (2)–(4), where we see examples of tokens that are taken directly from the text analysed.
(2)
Contexts coded as AP-final without punctuation:
a.
…
… the
|
|
makes contact …
b.
…
… the government …
and
|
|
that …
c.
The
| …
| …
d.
…
… that
| …
| …
When punctuation was present, as in (3) and (4), we treated the prosodic context as being distinct from when there was no punctuation present (2) because the token is final within a domain larger than the AP, leaving us with three groups that were included in the statistical models: AP-final tokens not followed by punctuation, AP-final tokens followed by a comma, and assertive IP-final tokens (AP-final tokens followed by a period). Only two of these groups are under focus in the current study: AP-final tokens without punctuation (2), which, as mentioned earlier, we refer to as AP-final; and AP-final tokens followed by a period (4), which we refer to as IP-final.
(3)
Contexts coded as AP-final with punctuation:
|
| quite
|
| since the Prime
| …
| …
(4)
Contexts coded as IP-final:
… since the
| …
| …
For each syllable included in the analysis, a Praat script extracted the rhyme duration,
Maximum amplitude was preferred over mean amplitude because the mean is more affected by the segments present in a given syllable and because a shorter vowel would be expected to be at its maximum for a shorter period of time, thereby reducing the mean value without necessarily reflecting a lower amplitude target. Using maximum amplitude also meant that we could reliably measure through the rhyme instead of limiting ourselves to the vowel, which could have resulted in not including the point with the greatest amplitude if, for example, there was a sonorant consonant in the coda that was higher in amplitude than the immediately preceding vowel.
Acoustic cue realisations of the 2736 targeted syllables were analysed using the lme4 package (
The models take as their dependent variables not the raw acoustic measurements, but instead, the difference between the last two syllables’ values, which yields a relative value (RV) to provide some adjustment for context (e.g. speech rate), with the use of random intercepts simulating normalisation procedures to remove inter-speaker differences (
We use subtraction when calculating the RV for two reasons. First, cues are already log-scaled (manually for duration; by virtue of being in decibels for amplitude and in semitones for pitch). Second, this allows for a more intuitive interpretation of RVs: not only does the RV’s sign indicate which syllable has a greater cue measurement, but the RV can be interpreted as the change in measurement value. Because our models predict RVs, the model considered 1368 data points (one per word). Four tokens were excluded from the maximum pitch and maximum amplitude models due to excessive devoicing of high vowels.
(5)
Formula for calculating RVs:
RVcue = measurementpenult,cue – measurementfinal,cue
Syllable weight values and their associated RVs using hypothetical data.
Returning to the statistical analysis, all models include identical fixed and random effect structures to ensure comparability. With the exception of prosodic domain, all factors are binary and were both rescaled by two standard deviations and centred for better comparability with potential non-binary factor effects. The prosodic domain is a ternary factor (AP with no punctuation vs. AP with comma vs. IP) and was Helmert-coded so that the first prosodic domain factor in the model provides the difference between APs with no punctuation and IPs, the domain comparison we focus on in the analysis. In all cases, the following directions of effect are interpreted as an increase in prominence: increased pitch, amplitude and duration. The coefficients in the model can be interpreted as the size of the change in acoustic cue measurements.
We included one additional factor in the model that does not directly relate to the predictions under focus: morphological structure. A larger and more diverse dataset would be required to test the effects of morphological structure in detail since many properties of individual morphemes could play a role (e.g. consonant-initial vs. vowel-initial, derivation vs. inflection). However, preliminary data exploration revealed that we would need to control for morphological structure in our models, so we included a manually coded factor that identified whether the penult was base-final. We will point out where this factor was crucial in our description of results, but not treat it as a focus given that our dataset provides only a limited sample of distinct suffix types (inflectional, derivation) and suffix shapes (closed or open syllable; with or without onset).
In this section, we discuss predictions for the quantitative analysis that we undertake. These predictions stem from our hypothesis that final prominence in French is sensitive to weight. We begin with prosodic domains.
As evident from
Given the literature on final lengthening, we expect that duration RVs will be higher in APs than in IPs because the final syllable of IPs will be subject to greater phrase-final lengthening. However, this difference is likely to be small based on mixed results in the literature, and this study lacks the statistical power required to confidently conclude that no small effect exists. These expectations lead to the prediction that we will not find any significant effect of prosodic domain in duration.
Finally, we do not expect there to be a significant difference between APs and IPs with respect to amplitude RV because amplitude is not expected to be used as a cue to phrasal prominence, though again we will cautiously not conclude that no small effect exists if no statistically significant effect is found. We therefore do not predict any difference in amplitude resulting from the type of prosodic domain. We summarise our expectations in Prediction 1:
(6)
One source of weight we consider is codas, where closed syllables are heavy and open syllables (with short vowels) are light. Words exemplifying prominence shift from multiple varieties of French as well as previous findings that Laurentian French speakers associate closed syllables with increased prominence in perception (see Section 2.1) lead us to expect that syllables with codas will be more prominent than those without. As a result, our expectations for coda weight are straightforward: closed penults will attract prominence; closed final syllables will preserve prominence. These expectations lead to Prediction 2:
(7)
Recall that we model vowel weight as binary as determined by vowel behaviour in closed final syllables: heavy vowels surface as long in closed final syllables, while light vowels are short in this context. We expect that heavy vowels will attract prominence and therefore that final high tones will be attracted to heavy vowels; that heavy vowels will have significantly longer duration; and that heavy vowels will be marked with greater amplitude. Combining our expectations that final syllables must be closed to count as heavy and that vowel weight contributes to weight in closed final syllables, we expect that open final syllables will pattern as light even if they contain an underlyingly heavy vowel. These expectations lead to Prediction 3:
(8)
Thus far, we have treated the two syllables under focus as equally capable of hosting prominence. This, though, is not consistent with the literature where final syllables are standardly considered to be the default position for prominence in French. In view of this, we must modulate our predictions to ensure that the prominence-retaining properties of final syllables will have a greater effect than the prominence-attracting properties of penult syllables. This may manifest not only in the relative sizes of the predicted coefficients (larger for factors relating to final syllables than for those relating to penults), but also in the distributions. RVs may be visually compressed when plotting the data: values tend to be negative or near zero even where penult prominence is predicted because the cue values across the whole dataset remain biased in favour of the final syllable having higher values.
In this section, we discuss the results of our statistical models (included as Supplementary Materials 2–4). We present findings thematically, to directly compare each factor’s effect on acoustic cues. All figures will follow the same layout: panel A shows maximum pitch, panel B shows rhyme duration, and panel C shows maximum amplitude. Contra Lamontagne (
Prediction 1 stated that IPs would have higher pitch maximum RVs than APs because IP-final syllables receive a low boundary tone. We find that IPs have considerably higher values (β = 1.1817, p < 0.0001), which is consistent with IPs having a low tone rather than a high tone in the final syllable. Panel A of
Results for domain type.
Regarding duration RVs, we predicted lower values in IPs than in APs due to greater IP-final lengthening. Consistent with this and as shown in panel B of
As for amplitude, we predicted that we would find no significant difference between APs and IPs because amplitude is not expected to be manipulated to signal boundaries. Contrary to this, we find that IPs have significantly higher RVs for maximum amplitude (β = 1.9705, p = 0.0011), which indicates that IP-final syllables have much lower amplitude than AP-final ones. Panel C of
Overall, these results are consistent with APs being marked with a rising pitch contour and IPs being marked with low final pitch. We additionally find evidence that IP-final syllables are longer than AP-final ones and that IPs have lower final amplitude. These results suggest that Laurentian French patterns like other varieties of French in the tone targets used to mark phrasal domains. With the tone for APs established, we turn to the results for weight.
Prediction 2 stated that the RV for all acoustic cues would be higher in closed syllables, following from our hypothesis that closed syllables attract prominence.
We predicted that closed penults would have higher pitch maxima (i.e. a significantly higher pitch RV) because heavy penults attract H* from the final syllable. Our models support this prediction (β = 1.1533, p < 0.0001), but an examination of panel A in
Results for penult coda weight.
We predicted that closed penults would have significantly longer rhymes (i.e. higher RVs) because penults of this shape optimally attract prominence. Closed penults have significantly longer rhymes (β = 0.6992, p < 0.0001), and panel B of
In short, these results show that words with closed penults (e.g. /mɛʁsi/
Turning to coda weight in final syllables, we predicted that the final syllable being closed would be associated with that syllable likely preserving prominence and thereby being realised with a high tone (i.e. lower RVs). We see from panel A of
Results for final coda weight.
Also in line with our prediction that closed syllables are heavy, closed final syllables have significantly longer relative rhyme durations (i.e. lower RVs). While the increase in relative duration for the penult is large, we find an even larger effect for final syllables (β = –0.9202, p < 0.0001), as shown in panel B of
Finally, we find much lower RVs for amplitude when the final syllable is heavy (β = –3.1101, p < 0.0001), as illustrated in panel C of
In summation, we find that closed final syllables (e.g. /navɛt/
Prediction 3 stated that heavy vowels should attract prominence in penults, but that open final syllables should pattern as light and therefore vowel weight should not have a significant effect in this position because the underlying weight contrast that was used to code vowels is typically neutralised. We predicted that RVs would be higher for heavy penults, but unaffected by underlyingly heavy final-syllable vowels unless the final syllable is also closed, in which case the syllable would be more likely to attract prominence. We again begin by presenting the results for penults.
We predicted that heavy penults would be associated with higher RVs for pitch maxima consistent with increased prominence shift to the penult, but the results are marginal (β = 0.2493, p = 0.0720), as illustrated in panel A of
Results for penult vowel weight.
We find the predicted increase in duration RVs when the penult vowel is heavy (β = 0.1881, p = 0.0024), shown in panel B of
For final syllable vowel weight, we predicted no main effects with the possible exception of a small increase in final rhyme duration (i.e. lower duration RVs). The plot in this case is somewhat misleading: we appear to get higher RVs associated with heavy final vowels, which should indicate that the penult – rather than the final syllable – increases in prominence when the final syllable has greater (vowel) weight. This is a visual artefact of distributional trends in the data that does not confound the models due to the number of tokens for different word shapes; light vowels in final syllables are more likely to be followed by a coda, which, as shown previously, has a considerable effect on prominence. We return to the question of the interaction between coda and vowel weight in final syllables in Section 4.4.
Pitch maxima (panel A of
Results for final vowel weight.
Prediction 3 stated that the effects of a final syllable’s underlying vowel weight would primarily be observed in syllables that contain both a coda and a heavy vowel because vowel length is retained only in closed final syllables in French. We find no significant interaction for maximum pitch (see panel A of
Results for the interaction between coda weight (x-axis) and vowel weight (colours).
Turning to duration (panel B), we find a significant interaction between the presence of a coda and vowel weight in the final syllable (β = 0.4866, p = 0.0006). However, while it seems that closed syllables are longer when they contain a heavy vowel, data inspection reveals that the interaction predominantly reflects that the final syllable’s weight is instead affecting duration in the penult.
As for amplitude RVs, we find a large and significant interaction (β = 4.8909, p = 0.0007), revealing that light syllables pattern differently from heavy syllables. Panel C of
All acoustic cues are affected by weight and prosodic context. Our results suggest that the basic patterns for marking prosodic domains in Laurentian French match the patterns for other dialects. Crucially, our predictions hold for weight: heavy syllables are associated with greater prominence than light syllables. We discuss the implications of these results next.
The results of this study confirm that prominence shift does occur in Laurentian French, where the acoustic cues associated with prominence are realised on the penult. In the analysis that follows, we show that, although this phenomenon is probabilistic such that the location of prominence cannot consistently be predicted for any given token, it is not arbitrary once we examine the broader patterns.
As discussed in Section 2, it was necessary to consider how speakers mark prosodic domains in order to test our hypothesis that weight conditions the location of prominence assignment and that the same acoustic cues are modulated for lexical and phrasal properties. In Section 5.1.1, we detail how pitch is used to mark APs and assertive IPs. In Section 5.1.2, we discuss our results in relation to the mixed results found for duration in the literature. Finally, in Section 5.1.3, we contribute to the relatively limited information in the literature on the manipulation of amplitude to mark phrases in French.
We found that IP-final syllables have significantly lower maximum pitch than the AP’s final syllable does. This is consistent with Laurentian French speakers marking APs with a rising (LH*) bitonal unit and with IPs being assigned an additional low boundary tone (L%) that replaces any tone assigned to the AP-final syllable. This result suggests that, at least with respect to general pitch contours, Laurentian French follows the same system as other dialects. These results are also consistent with Prediction 1 that prosodic domains are distinguished at least in part by the pitch contours at their right edge.
We found only a very small rhyme duration difference between APs and IPs, with IPs tending to have longer final syllables compared to APs. That we did not find a robust result is consistent with the mixed results found in the literature for other dialects (see Section 2.3); the durational difference between APs and IPs may be very small, highly variable or non-existent, which leads to certain studies finding that final syllables in IPs get compressed, others finding that final syllables are further lengthened, and some studies not being able to conclude either way. Given that the results across studies are so mixed, and based on our relatively marginal result, our expectation that higher domains would show greater degrees of lengthening cannot be confidently confirmed (consistent with Prediction 1).
We conjecture, however, that the presence of mixed results across studies may be a consequence of the type of data analysed. In the present study (on read speech), speakers were not required to plan the content of upcoming phrases; they only had to retrieve lexical entries, potentially reducing the need to slow down at the end of an IP to facilitate planning the next prosodic domain. Future work should test the possibility that speech planning and discourse constraints are responsible for differences in the degree of phrase-final lengthening. If greater lengthening in IP-final tokens reflects planning limitations (with lengthening providing more time to plan upcoming words) or conversational cues (for example, signalling that the speaker is not ceding the floor), then perhaps IP-final lengthening is sensitive to speech context.
Lastly, we found that IPs have lower relative amplitude than APs do, seemingly contrary to Prediction 1 that amplitude would not be used as a cue to phrasal prominence. While this may suggest that amplitude could be directly manipulated by speakers as a cue because a gradual decrease in amplitude could signal that the right edge of the current IP has not yet been reached, cross-linguistic evidence leads us to believe that amplitude is not intentionally used by speakers to mark the right edge of prosodic domains.
Based on findings from German (
However, even if amplitude is not consciously manipulated by speakers, it could still be used as a perceptual cue by listeners. This proposal is not only consistent with the cross-linguistic acoustic work just mentioned, but we believe it is also supported by the results of a previous perceptual study on French speakers. Schwab & Llisterri (
Our results provide support for the hypothesis in (1) that prominence assignment in French is sensitive to weight. Only three studies to our knowledge have quantitatively examined the relationship between weight and prominence in French. The first (
Beginning with coda weight, which was expected to significantly attract prominence based on Prediction 2, we observe that a final coda increases the relative prominence of the final syllable, affecting pitch, amplitude, and duration. Similarly, closed penults show an increase in RV for these same cues. Our results suggest not only that these cues signal weight, but additionally, that only one syllable is targeted by these effects, and the other may even show
Our results suggest that heavy penult vowels attract prominence, consistent with heavy vowels contributing to syllable weight in Laurentian French, as we observe for codas. However, final syllables pattern differently from penults in this respect; a heavy vowel in the final syllable is not sufficient to retain prominence on that syllable. Instead, vowel weight only has a slight effect in enhancing final syllables that are heavy by virtue of being closed, either making that syllable more prominent or further decreasing the likelihood that prominence shifts to the penult. This suggests that closed syllables containing a heavy vowel may be phonologically heavier than closed syllables containing a light vowel and, thus, that the label superheavy may be an appropriate characterisation of the phonological behaviour of these syllables. This result is particularly noteworthy because it confirms that underlyingly heavy (oral) vowels in open final syllables pattern as short for prominence assignment, which accords with the unavailability of diphthongisation in that position (see Section 2.1).
In summation, we have found evidence of weight effects for vowels as well as codas for all three acoustic cues, consistent with our predictions, but we observe that these effects are not identical. Heavy vowels only pattern as heavy (i.e. attract prominence) when they are not in open final syllables, while codas show the same prominence-attracting property in both penultimate and final syllables. As such, while prominence assignment is probabilistic, the conditions under which prominence shift is most likely to occur are not arbitrary. Based on examples from other varieties of French (see Section 2.1), we expect that these conditions are not confined to Laurentian French; instead, weight sensitivity should contribute to prominence shift across varieties. A dialect’s propensity for prominence shift should, however, depend on its phoneme inventory. As proposed by Lamontagne (
Although Laurentian French may exhibit more frequent prominence shift because of its conservative phonemic inventory, our results demonstrate that the marking of prosodic domains in this dialect is consistent with what has been found for other dialects. The cues used to mark prosodic domains also signal weight, which means that these factors interact to produce the prominence patterns we observe. Heavy syllables attracting prominence therefore has important repercussions for our understanding of the prosodic system itself.
At the least, prominence, including the assignment of the AP’s H* tone, appears to play a different role in the grammar of French than conventionally proposed;
Signalling word-level factors (weight) using the same cues as those used to mark prosodic domains has implications beyond explaining otherwise surprising results in perceptual studies. In particular, it helps shed light on the type of prominence system that French employs. In Section 2.3, we mentioned that APs are marked with an LH* tone, described as a pitch accent, but we did not elaborate on how French prominence is best categorised. Authors differ in whether they label final prominence as stress (e.g.
As we have just noted, the formal description of obligatory final prominence in French is debated. On one hand, since many studies refer to the prominence in French as stress, it should be assigned at the level of the word and therefore its location should be sensitive to word-level properties, notably weight. On the other hand, since French prominence is often described as phrasal or post-lexical, it should not be sensitive to word-level properties (e.g.
Text of the read passage used in the
Model outputs for maximum pitch relative values (maximum pitch in the final rhyme subtracted from maximum pitch in the penult rhyme). P-values calculated using Satterthwaite approximation.
(Intercept) | –1.1343 | 0.2724 | 111.4 | –4.164 | 0.0001 | *** |
AP vs. IP | 1.1817 | 0.2105 | 99.7 | 5.612 | <0.0001 | *** |
AP&IP vs. Comma | –0.8607 | 0.1746 | 93.0 | –4.929 | <0.0001 | *** |
Base-final penult | 0.3047 | 0.1677 | 109.4 | 2.04 | 0.0464 | * |
Closed final syllable | –0.1440 | 0.3668 | 196.6 | –2.631 | 0.0139 | * |
Closed penult | 1.1533 | 0.1812 | 137.3 | 7.215 | <0.0001 | *** |
Heavy final vowel | –0.1859 | 0.3606 | 200.8 | –0.515 | 0.6070 | |
Heavy penult vowel | 0.2493 | 0.2749 | 91.5 | 1.817 | 0.0720 | . |
Superheavy final syllable | –0.1119 | 0.6721 | 209.6 | –0.366 | 0.7146 |
Model outputs for maximum amplitude relative values (maximum amplitude in the final rhyme subtracted from maximum amplitude in the penult rhyme). P-values calculated using Satterthwaite approximation.
(Intercept) | 1.3877 | 0.6574 | 111.4 | 2.111 | 0.0370 | * |
AP vs. IP | 1.9705 | 0.5863 | 99.7 | 3.361 | 0.0011 | ** |
AP&IP vs. Comma | –0.3995 | 0.4927 | 93.0 | –0.811 | 0.4196 | |
Base-final penult | 0.4822 | 0.4533 | 109.4 | 1.064 | 0.2900 | |
Closed final syllable | –3.1101 | 0.4576 | 196.6 | –7.31 | <0.0001 | *** |
Closed penult | 2.0582 | 0.7670 | 137.3 | 2.683 | 0.0084 | ** |
Heavy final vowel | –0.3573 | 0.8689 | 200.8 | –0.411 | 0.6813 | |
Heavy penult vowel | 0.9189 | 0.6288 | 91.5 | 1.461 | 0.1483 | |
Superheavy final syllable | 4.8909 | 1.4404 | 209.6 | 3.396 | 0.0007 | *** |
Model outputs for rhyme duration relative values (final rhyme duration subtracted from penult rhyme duration). P-values calculated using Satterthwaite approximation.
(Intercept) | –0.6097 | 0.0696 | 111.8 | –8.763 | <0.0001 | *** |
AP vs. IP | –0.1068 | 0.0570 | 100.0 | –1.873 | 0.0480 | * |
AP&IP vs. Comma | –0.0387 | 0.0479 | 93.3 | –0.808 | 0.4215 | |
Base-final penult | –0.2454 | 0.0438 | 109.7 | –3.317 | 0.0012 | ** |
Closed final syllable | –0.9202 | 0.0911 | 197.2 | –10.101 | <0.0001 | *** |
Closed penult | 0.6992 | 0.0710 | 137.7 | 9.845 | <0.0001 | *** |
Heavy final vowel | –0.2662 | 0.0864 | 201.4 | –2.376 | 0.0184 | * |
Heavy penult vowel | 0.1881 | 0.0602 | 91.8 | 3.126 | 0.0024 | ** |
Superheavy final syllable | 0.4866 | 0.1412 | 210.2 | 3.447 | 0.0006 | *** |
The corpus is available through the
Notably, focus is realised with a pitch contour distinct from the one associated with final prominence in the variety under examination (
Note that we contrast
We use
The patterns provided in
As we discuss both pitch as an acoustic cue (a phonetic measurement, here measured in semitones) and pitch targets (a phonological category), we use the term
Delais-Roussarie et al. (
We abstract away from IP variability here, but note that prior studies find some variation in the realisation or selection of pitch contours in these contexts (e.g.
Previous work does, however, demonstrate that duration differs between AP-final tokens at the right edge of intermediate phrases and AP-final tokens that are not at the right edge of intermediate phrases (
The translations correspond to the senses of the words as they appear in the text, taking into account portions of the text not included in the abridged examples. Where example passages also include prosodic boundaries of types other than the one being illustrated, only the boundaries targeted for the example have been marked.
AP-final tokens followed by a comma were excluded from the analysis for reasons of space and were coded as a separate prosodic level using Helmert contrast coding for statistical analysis (see further Section 3.3). In previous work (
Preliminary data analysis considered vowel duration; we do not report these results because the patterns under focus for the rhyme and vowel are the same.
Earlier analyses, such as those in Lamontagne et al. (
Our study does not have a sufficient number of speakers to confidently test inter-speaker variability, but the consistency within our sample suggests that weight effects are comparable across speakers. We leave examination of individual differences for future work.
While the factors are not too confounded for testing, there are distributional asymmetries in French which mean that the data are skewed towards having certain phonemic content in some contexts only (see
500 tokens were generated for each combination of the penult and final syllables being light or heavy. Prominent syllables were given a mean of 25, and non-prominent syllables, a mean of 15. If a final syllable was heavy (expected to preserve prominence) or the penult was light (not expected to attract prominence), the final syllable was prominent. To reflect the hypothesis that heavy penults will optionally attract prominence from light final vowels, the penult was treated as prominent (mean value of 25) in half of the cases where the penult syllable was heavy, with the final syllable thus not being prominent (mean value of 15). If the two syllables had equal weight and therefore more variation in prominence was expected to occur, the standard deviation for both syllables was set to 7, while it was set to 3 if the syllables differed in weight.
Note that the blue dots for final-syllable datapoints were added after the red dots for penult datapoints, meaning that the distribution of points visually underrepresents the number of penult datapoints in higher cue value ranges. Additionally, recall that cue RVs are not based on individual datapoints independently, but instead on the difference between a penult value and its associated final-syllable value. The consequence of this is that RVs better illustrate the shift in prominence because plotting syllables separately masks the fundamental relationship between the values in each of these simulated words.
The examples we provide to illustrate results are (near-)minimal pairs rather than words drawn from the text (see Supplementary Material 1) because they most clearly demonstrate the contrasts under discussion.
We thank Marie-Hélène Côté, Élisabeth Delais-Roussarie, Morgan Sonderegger, Francisco Torreira, Michael Wagner, and anonymous reviewers, who provided comments on earlier versions of this paper. We would also like to thank Marie-Hélène Côté for providing data access and Luc Baronian for the original on-site data collection.
This work was supported in part by the Social Sciences and Humanities Research Council of Canada [grant numbers 766-2013-0862 and 435-2015-0490].
The authors have no competing interests to declare.